drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include <linux/rculist.h>
  81 #include <net/flow_keys.h>
  82 #include <linux/reciprocal_div.h>
  83 #include "bonding.h"
  84 #include "bond_3ad.h"
  85 #include "bond_alb.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90 #define BOND_LINK_MON_INTERV    0
  91 #define BOND_LINK_ARP_INTERV    0
  92
  93 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  94 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  95 static int num_peer_notif = 1;
  96 static int miimon       = BOND_LINK_MON_INTERV;
  97 static int updelay;
  98 static int downdelay;
  99 static int use_carrier  = 1;
 100 static char *mode;
 101 static char *primary;
 102 static char *primary_reselect;
 103 static char *lacp_rate;
 104 static int min_links;
 105 static char *ad_select;
 106 static char *xmit_hash_policy;
 107 static int arp_interval = BOND_LINK_ARP_INTERV;
 108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 109 static char *arp_validate;
 110 static char *arp_all_targets;
 111 static char *fail_over_mac;
 112 static int all_slaves_active;
 113 static struct bond_params bonding_defaults;
 114 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 115 static int packets_per_slave = 1;
 116
 117 module_param(max_bonds, int, 0);
 118 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 119 module_param(tx_queues, int, 0);
 120 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 121 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 122 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 123                                "failover event (alias of num_unsol_na)");
 124 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 125 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 126                                "failover event (alias of num_grat_arp)");
 127 module_param(miimon, int, 0);
 128 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 129 module_param(updelay, int, 0);
 130 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 131 module_param(downdelay, int, 0);
 132 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 133                             "in milliseconds");
 134 module_param(use_carrier, int, 0);
 135 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 136                               "0 for off, 1 for on (default)");
 137 module_param(mode, charp, 0);
 138 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 139                        "1 for active-backup, 2 for balance-xor, "
 140                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 141                        "6 for balance-alb");
 142 module_param(primary, charp, 0);
 143 MODULE_PARM_DESC(primary, "Primary network device to use");
 144 module_param(primary_reselect, charp, 0);
 145 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 146                                    "once it comes up; "
 147                                    "0 for always (default), "
 148                                    "1 for only if speed of primary is "
 149                                    "better, "
 150                                    "2 for only on active slave "
 151                                    "failure");
 152 module_param(lacp_rate, charp, 0);
 153 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 154                             "0 for slow, 1 for fast");
 155 module_param(ad_select, charp, 0);
 156 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 157                             "0 for stable (default), 1 for bandwidth, "
 158                             "2 for count");
 159 module_param(min_links, int, 0);
 160 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 161
 162 module_param(xmit_hash_policy, charp, 0);
 163 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 164                                    "0 for layer 2 (default), 1 for layer 3+4, "
 165                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 166                                    "4 for encap layer 3+4");
 167 module_param(arp_interval, int, 0);
 168 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 169 module_param_array(arp_ip_target, charp, NULL, 0);
 170 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 171 module_param(arp_validate, charp, 0);
 172 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 173                                "0 for none (default), 1 for active, "
 174                                "2 for backup, 3 for all");
 175 module_param(arp_all_targets, charp, 0);
 176 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 177 module_param(fail_over_mac, charp, 0);
 178 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 179                                 "the same MAC; 0 for none (default), "
 180                                 "1 for active, 2 for follow");
 181 module_param(all_slaves_active, int, 0);
 182 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 183                                      "by setting active flag for all slaves; "
 184                                      "0 for never (default), 1 for always.");
 185 module_param(resend_igmp, int, 0);
 186 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 187                               "link failure");
 188 module_param(packets_per_slave, int, 0);
 189 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 190                                     "mode; 0 for a random slave, 1 packet per "
 191                                     "slave (default), >1 packets per slave.");
 192
 193 /*----------------------------- Global variables ----------------------------*/
 194
 195 #ifdef CONFIG_NET_POLL_CONTROLLER
 196 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 197 #endif
 198
 199 int bond_net_id __read_mostly;
 200
 201 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 202 static int arp_ip_count;
 203 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 204 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 205 static int lacp_fast;
 206
 207 const struct bond_parm_tbl bond_lacp_tbl[] = {
 208 {       "slow",         AD_LACP_SLOW},
 209 {       "fast",         AD_LACP_FAST},
 210 {       NULL,           -1},
 211 };
 212
 213 const struct bond_parm_tbl bond_mode_tbl[] = {
 214 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 215 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 216 {       "balance-xor",          BOND_MODE_XOR},
 217 {       "broadcast",            BOND_MODE_BROADCAST},
 218 {       "802.3ad",              BOND_MODE_8023AD},
 219 {       "balance-tlb",          BOND_MODE_TLB},
 220 {       "balance-alb",          BOND_MODE_ALB},
 221 {       NULL,                   -1},
 222 };
 223
 224 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 225 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 226 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 227 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 228 {       "encap2+3",             BOND_XMIT_POLICY_ENCAP23},
 229 {       "encap3+4",             BOND_XMIT_POLICY_ENCAP34},
 230 {       NULL,                   -1},
 231 };
 232
 233 const struct bond_parm_tbl arp_all_targets_tbl[] = {
 234 {       "any",                  BOND_ARP_TARGETS_ANY},
 235 {       "all",                  BOND_ARP_TARGETS_ALL},
 236 {       NULL,                   -1},
 237 };
 238
 239 const struct bond_parm_tbl arp_validate_tbl[] = {
 240 {       "none",                 BOND_ARP_VALIDATE_NONE},
 241 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 242 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 243 {       "all",                  BOND_ARP_VALIDATE_ALL},
 244 {       NULL,                   -1},
 245 };
 246
 247 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 248 {       "none",                 BOND_FOM_NONE},
 249 {       "active",               BOND_FOM_ACTIVE},
 250 {       "follow",               BOND_FOM_FOLLOW},
 251 {       NULL,                   -1},
 252 };
 253
 254 const struct bond_parm_tbl pri_reselect_tbl[] = {
 255 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 256 {       "better",               BOND_PRI_RESELECT_BETTER},
 257 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 258 {       NULL,                   -1},
 259 };
 260
 261 struct bond_parm_tbl ad_select_tbl[] = {
 262 {       "stable",       BOND_AD_STABLE},
 263 {       "bandwidth",    BOND_AD_BANDWIDTH},
 264 {       "count",        BOND_AD_COUNT},
 265 {       NULL,           -1},
 266 };
 267
 268 /*-------------------------- Forward declarations ---------------------------*/
 269
 270 static int bond_init(struct net_device *bond_dev);
 271 static void bond_uninit(struct net_device *bond_dev);
 272
 273 /*---------------------------- General routines -----------------------------*/
 274
 275 const char *bond_mode_name(int mode)
 276 {
 277         static const char *names[] = {
 278                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 279                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 280                 [BOND_MODE_XOR] = "load balancing (xor)",
 281                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 282                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 283                 [BOND_MODE_TLB] = "transmit load balancing",
 284                 [BOND_MODE_ALB] = "adaptive load balancing",
 285         };
 286
 287         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 288                 return "unknown";
 289
 290         return names[mode];
 291 }
 292
 293 /*---------------------------------- VLAN -----------------------------------*/
 294
 295 /**
 296  * bond_dev_queue_xmit - Prepare skb for xmit.
 297  *
 298  * @bond: bond device that got this skb for tx.
 299  * @skb: hw accel VLAN tagged skb to transmit
 300  * @slave_dev: slave that is supposed to xmit this skbuff
 301  */
 302 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 303                         struct net_device *slave_dev)
 304 {
 305         skb->dev = slave_dev;
 306
 307         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 308                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 309         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 310
 311         if (unlikely(netpoll_tx_running(bond->dev)))
 312                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 313         else
 314                 dev_queue_xmit(skb);
 315
 316         return 0;
 317 }
 318
 319 /*
 320  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 321  * We don't protect the slave list iteration with a lock because:
 322  * a. This operation is performed in IOCTL context,
 323  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 324  * c. Holding a lock with BH disabled while directly calling a base driver
 325  *    entry point is generally a BAD idea.
 326  *
 327  * The design of synchronization/protection for this operation in the 8021q
 328  * module is good for one or more VLAN devices over a single physical device
 329  * and cannot be extended for a teaming solution like bonding, so there is a
 330  * potential race condition here where a net device from the vlan group might
 331  * be referenced (either by a base driver or the 8021q code) while it is being
 332  * removed from the system. However, it turns out we're not making matters
 333  * worse, and if it works for regular VLAN usage it will work here too.
 334 */
 335
 336 /**
 337  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 338  * @bond_dev: bonding net device that got called
 339  * @vid: vlan id being added
 340  */
 341 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 342                                 __be16 proto, u16 vid)
 343 {
 344         struct bonding *bond = netdev_priv(bond_dev);
 345         struct slave *slave, *rollback_slave;
 346         struct list_head *iter;
 347         int res;
 348
 349         bond_for_each_slave(bond, slave, iter) {
 350                 res = vlan_vid_add(slave->dev, proto, vid);
 351                 if (res)
 352                         goto unwind;
 353         }
 354
 355         return 0;
 356
 357 unwind:
 358         /* unwind to the slave that failed */
 359         bond_for_each_slave(bond, rollback_slave, iter) {
 360                 if (rollback_slave == slave)
 361                         break;
 362
 363                 vlan_vid_del(rollback_slave->dev, proto, vid);
 364         }
 365
 366         return res;
 367 }
 368
 369 /**
 370  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 371  * @bond_dev: bonding net device that got called
 372  * @vid: vlan id being removed
 373  */
 374 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 375                                  __be16 proto, u16 vid)
 376 {
 377         struct bonding *bond = netdev_priv(bond_dev);
 378         struct list_head *iter;
 379         struct slave *slave;
 380
 381         bond_for_each_slave(bond, slave, iter)
 382                 vlan_vid_del(slave->dev, proto, vid);
 383
 384         if (bond_is_lb(bond))
 385                 bond_alb_clear_vlan(bond, vid);
 386
 387         return 0;
 388 }
 389
 390 /*------------------------------- Link status -------------------------------*/
 391
 392 /*
 393  * Set the carrier state for the master according to the state of its
 394  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 395  * do special 802.3ad magic.
 396  *
 397  * Returns zero if carrier state does not change, nonzero if it does.
 398  */
 399 static int bond_set_carrier(struct bonding *bond)
 400 {
 401         struct list_head *iter;
 402         struct slave *slave;
 403
 404         if (!bond_has_slaves(bond))
 405                 goto down;
 406
 407         if (bond->params.mode == BOND_MODE_8023AD)
 408                 return bond_3ad_set_carrier(bond);
 409
 410         bond_for_each_slave(bond, slave, iter) {
 411                 if (slave->link == BOND_LINK_UP) {
 412                         if (!netif_carrier_ok(bond->dev)) {
 413                                 netif_carrier_on(bond->dev);
 414                                 return 1;
 415                         }
 416                         return 0;
 417                 }
 418         }
 419
 420 down:
 421         if (netif_carrier_ok(bond->dev)) {
 422                 netif_carrier_off(bond->dev);
 423                 return 1;
 424         }
 425         return 0;
 426 }
 427
 428 /*
 429  * Get link speed and duplex from the slave's base driver
 430  * using ethtool. If for some reason the call fails or the
 431  * values are invalid, set speed and duplex to -1,
 432  * and return.
 433  */
 434 static void bond_update_speed_duplex(struct slave *slave)
 435 {
 436         struct net_device *slave_dev = slave->dev;
 437         struct ethtool_cmd ecmd;
 438         u32 slave_speed;
 439         int res;
 440
 441         slave->speed = SPEED_UNKNOWN;
 442         slave->duplex = DUPLEX_UNKNOWN;
 443
 444         res = __ethtool_get_settings(slave_dev, &ecmd);
 445         if (res < 0)
 446                 return;
 447
 448         slave_speed = ethtool_cmd_speed(&ecmd);
 449         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 450                 return;
 451
 452         switch (ecmd.duplex) {
 453         case DUPLEX_FULL:
 454         case DUPLEX_HALF:
 455                 break;
 456         default:
 457                 return;
 458         }
 459
 460         slave->speed = slave_speed;
 461         slave->duplex = ecmd.duplex;
 462
 463         return;
 464 }
 465
 466 /*
 467  * if <dev> supports MII link status reporting, check its link status.
 468  *
 469  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 470  * depending upon the setting of the use_carrier parameter.
 471  *
 472  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 473  * can't tell and just pretend it is), or 0, meaning that the link is
 474  * down.
 475  *
 476  * If reporting is non-zero, instead of faking link up, return -1 if
 477  * both ETHTOOL and MII ioctls fail (meaning the device does not
 478  * support them).  If use_carrier is set, return whatever it says.
 479  * It'd be nice if there was a good way to tell if a driver supports
 480  * netif_carrier, but there really isn't.
 481  */
 482 static int bond_check_dev_link(struct bonding *bond,
 483                                struct net_device *slave_dev, int reporting)
 484 {
 485         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 486         int (*ioctl)(struct net_device *, struct ifreq *, int);
 487         struct ifreq ifr;
 488         struct mii_ioctl_data *mii;
 489
 490         if (!reporting && !netif_running(slave_dev))
 491                 return 0;
 492
 493         if (bond->params.use_carrier)
 494                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 495
 496         /* Try to get link status using Ethtool first. */
 497         if (slave_dev->ethtool_ops->get_link)
 498                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 499                         BMSR_LSTATUS : 0;
 500
 501         /* Ethtool can't be used, fallback to MII ioctls. */
 502         ioctl = slave_ops->ndo_do_ioctl;
 503         if (ioctl) {
 504                 /* TODO: set pointer to correct ioctl on a per team member */
 505                 /*       bases to make this more efficient. that is, once  */
 506                 /*       we determine the correct ioctl, we will always    */
 507                 /*       call it and not the others for that team          */
 508                 /*       member.                                           */
 509
 510                 /*
 511                  * We cannot assume that SIOCGMIIPHY will also read a
 512                  * register; not all network drivers (e.g., e100)
 513                  * support that.
 514                  */
 515
 516                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 517                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 518                 mii = if_mii(&ifr);
 519                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 520                         mii->reg_num = MII_BMSR;
 521                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 522                                 return mii->val_out & BMSR_LSTATUS;
 523                 }
 524         }
 525
 526         /*
 527          * If reporting, report that either there's no dev->do_ioctl,
 528          * or both SIOCGMIIREG and get_link failed (meaning that we
 529          * cannot report link status).  If not reporting, pretend
 530          * we're ok.
 531          */
 532         return reporting ? -1 : BMSR_LSTATUS;
 533 }
 534
 535 /*----------------------------- Multicast list ------------------------------*/
 536
 537 /*
 538  * Push the promiscuity flag down to appropriate slaves
 539  */
 540 static int bond_set_promiscuity(struct bonding *bond, int inc)
 541 {
 542         struct list_head *iter;
 543         int err = 0;
 544
 545         if (USES_PRIMARY(bond->params.mode)) {
 546                 /* write lock already acquired */
 547                 if (bond->curr_active_slave) {
 548                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 549                                                   inc);
 550                 }
 551         } else {
 552                 struct slave *slave;
 553
 554                 bond_for_each_slave(bond, slave, iter) {
 555                         err = dev_set_promiscuity(slave->dev, inc);
 556                         if (err)
 557                                 return err;
 558                 }
 559         }
 560         return err;
 561 }
 562
 563 /*
 564  * Push the allmulti flag down to all slaves
 565  */
 566 static int bond_set_allmulti(struct bonding *bond, int inc)
 567 {
 568         struct list_head *iter;
 569         int err = 0;
 570
 571         if (USES_PRIMARY(bond->params.mode)) {
 572                 /* write lock already acquired */
 573                 if (bond->curr_active_slave) {
 574                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 575                                                inc);
 576                 }
 577         } else {
 578                 struct slave *slave;
 579
 580                 bond_for_each_slave(bond, slave, iter) {
 581                         err = dev_set_allmulti(slave->dev, inc);
 582                         if (err)
 583                                 return err;
 584                 }
 585         }
 586         return err;
 587 }
 588
 589 /*
 590  * Retrieve the list of registered multicast addresses for the bonding
 591  * device and retransmit an IGMP JOIN request to the current active
 592  * slave.
 593  */
 594 static void bond_resend_igmp_join_requests(struct bonding *bond)
 595 {
 596         if (!rtnl_trylock()) {
 597                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 598                 return;
 599         }
 600         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 601         rtnl_unlock();
 602
 603         /* We use curr_slave_lock to protect against concurrent access to
 604          * igmp_retrans from multiple running instances of this function and
 605          * bond_change_active_slave
 606          */
 607         write_lock_bh(&bond->curr_slave_lock);
 608         if (bond->igmp_retrans > 1) {
 609                 bond->igmp_retrans--;
 610                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 611         }
 612         write_unlock_bh(&bond->curr_slave_lock);
 613 }
 614
 615 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 616 {
 617         struct bonding *bond = container_of(work, struct bonding,
 618                                             mcast_work.work);
 619
 620         bond_resend_igmp_join_requests(bond);
 621 }
 622
 623 /* Flush bond's hardware addresses from slave
 624  */
 625 static void bond_hw_addr_flush(struct net_device *bond_dev,
 626                                struct net_device *slave_dev)
 627 {
 628         struct bonding *bond = netdev_priv(bond_dev);
 629
 630         dev_uc_unsync(slave_dev, bond_dev);
 631         dev_mc_unsync(slave_dev, bond_dev);
 632
 633         if (bond->params.mode == BOND_MODE_8023AD) {
 634                 /* del lacpdu mc addr from mc list */
 635                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 636
 637                 dev_mc_del(slave_dev, lacpdu_multicast);
 638         }
 639 }
 640
 641 /*--------------------------- Active slave change ---------------------------*/
 642
 643 /* Update the hardware address list and promisc/allmulti for the new and
 644  * old active slaves (if any).  Modes that are !USES_PRIMARY keep all
 645  * slaves up date at all times; only the USES_PRIMARY modes need to call
 646  * this function to swap these settings during a failover.
 647  */
 648 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 649                               struct slave *old_active)
 650 {
 651         ASSERT_RTNL();
 652
 653         if (old_active) {
 654                 if (bond->dev->flags & IFF_PROMISC)
 655                         dev_set_promiscuity(old_active->dev, -1);
 656
 657                 if (bond->dev->flags & IFF_ALLMULTI)
 658                         dev_set_allmulti(old_active->dev, -1);
 659
 660                 bond_hw_addr_flush(bond->dev, old_active->dev);
 661         }
 662
 663         if (new_active) {
 664                 /* FIXME: Signal errors upstream. */
 665                 if (bond->dev->flags & IFF_PROMISC)
 666                         dev_set_promiscuity(new_active->dev, 1);
 667
 668                 if (bond->dev->flags & IFF_ALLMULTI)
 669                         dev_set_allmulti(new_active->dev, 1);
 670
 671                 netif_addr_lock_bh(bond->dev);
 672                 dev_uc_sync(new_active->dev, bond->dev);
 673                 dev_mc_sync(new_active->dev, bond->dev);
 674                 netif_addr_unlock_bh(bond->dev);
 675         }
 676 }
 677
 678 /**
 679  * bond_set_dev_addr - clone slave's address to bond
 680  * @bond_dev: bond net device
 681  * @slave_dev: slave net device
 682  *
 683  * Should be called with RTNL held.
 684  */
 685 static void bond_set_dev_addr(struct net_device *bond_dev,
 686                               struct net_device *slave_dev)
 687 {
 688         pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 689                  bond_dev, slave_dev, slave_dev->addr_len);
 690         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 691         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 692         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 693 }
 694
 695 /*
 696  * bond_do_fail_over_mac
 697  *
 698  * Perform special MAC address swapping for fail_over_mac settings
 699  *
 700  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 701  */
 702 static void bond_do_fail_over_mac(struct bonding *bond,
 703                                   struct slave *new_active,
 704                                   struct slave *old_active)
 705         __releases(&bond->curr_slave_lock)
 706         __releases(&bond->lock)
 707         __acquires(&bond->lock)
 708         __acquires(&bond->curr_slave_lock)
 709 {
 710         u8 tmp_mac[ETH_ALEN];
 711         struct sockaddr saddr;
 712         int rv;
 713
 714         switch (bond->params.fail_over_mac) {
 715         case BOND_FOM_ACTIVE:
 716                 if (new_active) {
 717                         write_unlock_bh(&bond->curr_slave_lock);
 718                         read_unlock(&bond->lock);
 719                         bond_set_dev_addr(bond->dev, new_active->dev);
 720                         read_lock(&bond->lock);
 721                         write_lock_bh(&bond->curr_slave_lock);
 722                 }
 723                 break;
 724         case BOND_FOM_FOLLOW:
 725                 /*
 726                  * if new_active && old_active, swap them
 727                  * if just old_active, do nothing (going to no active slave)
 728                  * if just new_active, set new_active to bond's MAC
 729                  */
 730                 if (!new_active)
 731                         return;
 732
 733                 write_unlock_bh(&bond->curr_slave_lock);
 734                 read_unlock(&bond->lock);
 735
 736                 if (old_active) {
 737                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 738                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 739                                ETH_ALEN);
 740                         saddr.sa_family = new_active->dev->type;
 741                 } else {
 742                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 743                         saddr.sa_family = bond->dev->type;
 744                 }
 745
 746                 rv = dev_set_mac_address(new_active->dev, &saddr);
 747                 if (rv) {
 748                         pr_err("%s: Error %d setting MAC of slave %s\n",
 749                                bond->dev->name, -rv, new_active->dev->name);
 750                         goto out;
 751                 }
 752
 753                 if (!old_active)
 754                         goto out;
 755
 756                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 757                 saddr.sa_family = old_active->dev->type;
 758
 759                 rv = dev_set_mac_address(old_active->dev, &saddr);
 760                 if (rv)
 761                         pr_err("%s: Error %d setting MAC of slave %s\n",
 762                                bond->dev->name, -rv, new_active->dev->name);
 763 out:
 764                 read_lock(&bond->lock);
 765                 write_lock_bh(&bond->curr_slave_lock);
 766                 break;
 767         default:
 768                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 769                        bond->dev->name, bond->params.fail_over_mac);
 770                 break;
 771         }
 772
 773 }
 774
 775 static bool bond_should_change_active(struct bonding *bond)
 776 {
 777         struct slave *prim = bond->primary_slave;
 778         struct slave *curr = bond->curr_active_slave;
 779
 780         if (!prim || !curr || curr->link != BOND_LINK_UP)
 781                 return true;
 782         if (bond->force_primary) {
 783                 bond->force_primary = false;
 784                 return true;
 785         }
 786         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 787             (prim->speed < curr->speed ||
 788              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 789                 return false;
 790         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 791                 return false;
 792         return true;
 793 }
 794
 795 /**
 796  * find_best_interface - select the best available slave to be the active one
 797  * @bond: our bonding struct
 798  */
 799 static struct slave *bond_find_best_slave(struct bonding *bond)
 800 {
 801         struct slave *slave, *bestslave = NULL;
 802         struct list_head *iter;
 803         int mintime = bond->params.updelay;
 804
 805         if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
 806             bond_should_change_active(bond))
 807                 return bond->primary_slave;
 808
 809         bond_for_each_slave(bond, slave, iter) {
 810                 if (slave->link == BOND_LINK_UP)
 811                         return slave;
 812                 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
 813                     slave->delay < mintime) {
 814                         mintime = slave->delay;
 815                         bestslave = slave;
 816                 }
 817         }
 818
 819         return bestslave;
 820 }
 821
 822 static bool bond_should_notify_peers(struct bonding *bond)
 823 {
 824         struct slave *slave = bond->curr_active_slave;
 825
 826         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
 827                  bond->dev->name, slave ? slave->dev->name : "NULL");
 828
 829         if (!slave || !bond->send_peer_notif ||
 830             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 831                 return false;
 832
 833         return true;
 834 }
 835
 836 /**
 837  * change_active_interface - change the active slave into the specified one
 838  * @bond: our bonding struct
 839  * @new: the new slave to make the active one
 840  *
 841  * Set the new slave to the bond's settings and unset them on the old
 842  * curr_active_slave.
 843  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 844  *
 845  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 846  * because it is apparently the best available slave we have, even though its
 847  * updelay hasn't timed out yet.
 848  *
 849  * If new_active is not NULL, caller must hold bond->lock for read and
 850  * curr_slave_lock for write_bh.
 851  */
 852 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 853 {
 854         struct slave *old_active = bond->curr_active_slave;
 855
 856         if (old_active == new_active)
 857                 return;
 858
 859         if (new_active) {
 860                 new_active->jiffies = jiffies;
 861
 862                 if (new_active->link == BOND_LINK_BACK) {
 863                         if (USES_PRIMARY(bond->params.mode)) {
 864                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
 865                                         bond->dev->name, new_active->dev->name,
 866                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
 867                         }
 868
 869                         new_active->delay = 0;
 870                         new_active->link = BOND_LINK_UP;
 871
 872                         if (bond->params.mode == BOND_MODE_8023AD)
 873                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 874
 875                         if (bond_is_lb(bond))
 876                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 877                 } else {
 878                         if (USES_PRIMARY(bond->params.mode)) {
 879                                 pr_info("%s: making interface %s the new active one.\n",
 880                                         bond->dev->name, new_active->dev->name);
 881                         }
 882                 }
 883         }
 884
 885         if (USES_PRIMARY(bond->params.mode))
 886                 bond_hw_addr_swap(bond, new_active, old_active);
 887
 888         if (bond_is_lb(bond)) {
 889                 bond_alb_handle_active_change(bond, new_active);
 890                 if (old_active)
 891                         bond_set_slave_inactive_flags(old_active);
 892                 if (new_active)
 893                         bond_set_slave_active_flags(new_active);
 894         } else {
 895                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 896         }
 897
 898         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
 899                 if (old_active)
 900                         bond_set_slave_inactive_flags(old_active);
 901
 902                 if (new_active) {
 903                         bool should_notify_peers = false;
 904
 905                         bond_set_slave_active_flags(new_active);
 906
 907                         if (bond->params.fail_over_mac)
 908                                 bond_do_fail_over_mac(bond, new_active,
 909                                                       old_active);
 910
 911                         if (netif_running(bond->dev)) {
 912                                 bond->send_peer_notif =
 913                                         bond->params.num_peer_notif;
 914                                 should_notify_peers =
 915                                         bond_should_notify_peers(bond);
 916                         }
 917
 918                         write_unlock_bh(&bond->curr_slave_lock);
 919                         read_unlock(&bond->lock);
 920
 921                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 922                         if (should_notify_peers)
 923                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 924                                                          bond->dev);
 925
 926                         read_lock(&bond->lock);
 927                         write_lock_bh(&bond->curr_slave_lock);
 928                 }
 929         }
 930
 931         /* resend IGMP joins since active slave has changed or
 932          * all were sent on curr_active_slave.
 933          * resend only if bond is brought up with the affected
 934          * bonding modes and the retransmission is enabled */
 935         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 936             ((USES_PRIMARY(bond->params.mode) && new_active) ||
 937              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
 938                 bond->igmp_retrans = bond->params.resend_igmp;
 939                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 940         }
 941 }
 942
 943 /**
 944  * bond_select_active_slave - select a new active slave, if needed
 945  * @bond: our bonding struct
 946  *
 947  * This functions should be called when one of the following occurs:
 948  * - The old curr_active_slave has been released or lost its link.
 949  * - The primary_slave has got its link back.
 950  * - A slave has got its link back and there's no old curr_active_slave.
 951  *
 952  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
 953  */
 954 void bond_select_active_slave(struct bonding *bond)
 955 {
 956         struct slave *best_slave;
 957         int rv;
 958
 959         best_slave = bond_find_best_slave(bond);
 960         if (best_slave != bond->curr_active_slave) {
 961                 bond_change_active_slave(bond, best_slave);
 962                 rv = bond_set_carrier(bond);
 963                 if (!rv)
 964                         return;
 965
 966                 if (netif_carrier_ok(bond->dev)) {
 967                         pr_info("%s: first active interface up!\n",
 968                                 bond->dev->name);
 969                 } else {
 970                         pr_info("%s: now running without any active interface !\n",
 971                                 bond->dev->name);
 972                 }
 973         }
 974 }
 975
 976 #ifdef CONFIG_NET_POLL_CONTROLLER
 977 static inline int slave_enable_netpoll(struct slave *slave)
 978 {
 979         struct netpoll *np;
 980         int err = 0;
 981
 982         np = kzalloc(sizeof(*np), GFP_ATOMIC);
 983         err = -ENOMEM;
 984         if (!np)
 985                 goto out;
 986
 987         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
 988         if (err) {
 989                 kfree(np);
 990                 goto out;
 991         }
 992         slave->np = np;
 993 out:
 994         return err;
 995 }
 996 static inline void slave_disable_netpoll(struct slave *slave)
 997 {
 998         struct netpoll *np = slave->np;
 999
1000         if (!np)
1001                 return;
1002
1003         slave->np = NULL;
1004         __netpoll_free_async(np);
1005 }
1006 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1007 {
1008         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1009                 return false;
1010         if (!slave_dev->netdev_ops->ndo_poll_controller)
1011                 return false;
1012         return true;
1013 }
1014
1015 static void bond_poll_controller(struct net_device *bond_dev)
1016 {
1017 }
1018
1019 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1020 {
1021         struct bonding *bond = netdev_priv(bond_dev);
1022         struct list_head *iter;
1023         struct slave *slave;
1024
1025         bond_for_each_slave(bond, slave, iter)
1026                 if (IS_UP(slave->dev))
1027                         slave_disable_netpoll(slave);
1028 }
1029
1030 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1031 {
1032         struct bonding *bond = netdev_priv(dev);
1033         struct list_head *iter;
1034         struct slave *slave;
1035         int err = 0;
1036
1037         bond_for_each_slave(bond, slave, iter) {
1038                 err = slave_enable_netpoll(slave);
1039                 if (err) {
1040                         bond_netpoll_cleanup(dev);
1041                         break;
1042                 }
1043         }
1044         return err;
1045 }
1046 #else
1047 static inline int slave_enable_netpoll(struct slave *slave)
1048 {
1049         return 0;
1050 }
1051 static inline void slave_disable_netpoll(struct slave *slave)
1052 {
1053 }
1054 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1055 {
1056 }
1057 #endif
1058
1059 /*---------------------------------- IOCTL ----------------------------------*/
1060
1061 static netdev_features_t bond_fix_features(struct net_device *dev,
1062                                            netdev_features_t features)
1063 {
1064         struct bonding *bond = netdev_priv(dev);
1065         struct list_head *iter;
1066         netdev_features_t mask;
1067         struct slave *slave;
1068
1069         if (!bond_has_slaves(bond)) {
1070                 /* Disable adding VLANs to empty bond. But why? --mq */
1071                 features |= NETIF_F_VLAN_CHALLENGED;
1072                 return features;
1073         }
1074
1075         mask = features;
1076         features &= ~NETIF_F_ONE_FOR_ALL;
1077         features |= NETIF_F_ALL_FOR_ALL;
1078
1079         bond_for_each_slave(bond, slave, iter) {
1080                 features = netdev_increment_features(features,
1081                                                      slave->dev->features,
1082                                                      mask);
1083         }
1084         features = netdev_add_tso_features(features, mask);
1085
1086         return features;
1087 }
1088
1089 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1090                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1091                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1092
1093 static void bond_compute_features(struct bonding *bond)
1094 {
1095         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1096         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1097         struct net_device *bond_dev = bond->dev;
1098         struct list_head *iter;
1099         struct slave *slave;
1100         unsigned short max_hard_header_len = ETH_HLEN;
1101         unsigned int gso_max_size = GSO_MAX_SIZE;
1102         u16 gso_max_segs = GSO_MAX_SEGS;
1103
1104         if (!bond_has_slaves(bond))
1105                 goto done;
1106
1107         bond_for_each_slave(bond, slave, iter) {
1108                 vlan_features = netdev_increment_features(vlan_features,
1109                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1110
1111                 dst_release_flag &= slave->dev->priv_flags;
1112                 if (slave->dev->hard_header_len > max_hard_header_len)
1113                         max_hard_header_len = slave->dev->hard_header_len;
1114
1115                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1116                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1117         }
1118
1119 done:
1120         bond_dev->vlan_features = vlan_features;
1121         bond_dev->hard_header_len = max_hard_header_len;
1122         bond_dev->gso_max_segs = gso_max_segs;
1123         netif_set_gso_max_size(bond_dev, gso_max_size);
1124
1125         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1126         bond_dev->priv_flags = flags | dst_release_flag;
1127
1128         netdev_change_features(bond_dev);
1129 }
1130
1131 static void bond_setup_by_slave(struct net_device *bond_dev,
1132                                 struct net_device *slave_dev)
1133 {
1134         bond_dev->header_ops        = slave_dev->header_ops;
1135
1136         bond_dev->type              = slave_dev->type;
1137         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1138         bond_dev->addr_len          = slave_dev->addr_len;
1139
1140         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1141                 slave_dev->addr_len);
1142 }
1143
1144 /* On bonding slaves other than the currently active slave, suppress
1145  * duplicates except for alb non-mcast/bcast.
1146  */
1147 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1148                                             struct slave *slave,
1149                                             struct bonding *bond)
1150 {
1151         if (bond_is_slave_inactive(slave)) {
1152                 if (bond->params.mode == BOND_MODE_ALB &&
1153                     skb->pkt_type != PACKET_BROADCAST &&
1154                     skb->pkt_type != PACKET_MULTICAST)
1155                         return false;
1156                 return true;
1157         }
1158         return false;
1159 }
1160
1161 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1162 {
1163         struct sk_buff *skb = *pskb;
1164         struct slave *slave;
1165         struct bonding *bond;
1166         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1167                           struct slave *);
1168         int ret = RX_HANDLER_ANOTHER;
1169
1170         skb = skb_share_check(skb, GFP_ATOMIC);
1171         if (unlikely(!skb))
1172                 return RX_HANDLER_CONSUMED;
1173
1174         *pskb = skb;
1175
1176         slave = bond_slave_get_rcu(skb->dev);
1177         bond = slave->bond;
1178
1179         if (bond->params.arp_interval)
1180                 slave->dev->last_rx = jiffies;
1181
1182         recv_probe = ACCESS_ONCE(bond->recv_probe);
1183         if (recv_probe) {
1184                 ret = recv_probe(skb, bond, slave);
1185                 if (ret == RX_HANDLER_CONSUMED) {
1186                         consume_skb(skb);
1187                         return ret;
1188                 }
1189         }
1190
1191         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1192                 return RX_HANDLER_EXACT;
1193         }
1194
1195         skb->dev = bond->dev;
1196
1197         if (bond->params.mode == BOND_MODE_ALB &&
1198             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1199             skb->pkt_type == PACKET_HOST) {
1200
1201                 if (unlikely(skb_cow_head(skb,
1202                                           skb->data - skb_mac_header(skb)))) {
1203                         kfree_skb(skb);
1204                         return RX_HANDLER_CONSUMED;
1205                 }
1206                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1207         }
1208
1209         return ret;
1210 }
1211
1212 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1213                                       struct net_device *slave_dev,
1214                                       struct slave *slave)
1215 {
1216         int err;
1217
1218         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1219         if (err)
1220                 return err;
1221         slave_dev->flags |= IFF_SLAVE;
1222         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1223         return 0;
1224 }
1225
1226 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1227                                   struct net_device *slave_dev)
1228 {
1229         netdev_upper_dev_unlink(slave_dev, bond_dev);
1230         slave_dev->flags &= ~IFF_SLAVE;
1231         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1232 }
1233
1234 /* enslave device <slave> to bond device <master> */
1235 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1236 {
1237         struct bonding *bond = netdev_priv(bond_dev);
1238         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1239         struct slave *new_slave = NULL, *prev_slave;
1240         struct sockaddr addr;
1241         int link_reporting;
1242         int res = 0, i;
1243
1244         if (!bond->params.use_carrier &&
1245             slave_dev->ethtool_ops->get_link == NULL &&
1246             slave_ops->ndo_do_ioctl == NULL) {
1247                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1248                            bond_dev->name, slave_dev->name);
1249         }
1250
1251         /* already enslaved */
1252         if (slave_dev->flags & IFF_SLAVE) {
1253                 pr_debug("Error, Device was already enslaved\n");
1254                 return -EBUSY;
1255         }
1256
1257         /* vlan challenged mutual exclusion */
1258         /* no need to lock since we're protected by rtnl_lock */
1259         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1260                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1261                 if (vlan_uses_dev(bond_dev)) {
1262                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1263                                bond_dev->name, slave_dev->name, bond_dev->name);
1264                         return -EPERM;
1265                 } else {
1266                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1267                                    bond_dev->name, slave_dev->name,
1268                                    slave_dev->name, bond_dev->name);
1269                 }
1270         } else {
1271                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1272         }
1273
1274         /*
1275          * Old ifenslave binaries are no longer supported.  These can
1276          * be identified with moderate accuracy by the state of the slave:
1277          * the current ifenslave will set the interface down prior to
1278          * enslaving it; the old ifenslave will not.
1279          */
1280         if ((slave_dev->flags & IFF_UP)) {
1281                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1282                        slave_dev->name);
1283                 res = -EPERM;
1284                 goto err_undo_flags;
1285         }
1286
1287         /* set bonding device ether type by slave - bonding netdevices are
1288          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1289          * there is a need to override some of the type dependent attribs/funcs.
1290          *
1291          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1292          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1293          */
1294         if (!bond_has_slaves(bond)) {
1295                 if (bond_dev->type != slave_dev->type) {
1296                         pr_debug("%s: change device type from %d to %d\n",
1297                                  bond_dev->name,
1298                                  bond_dev->type, slave_dev->type);
1299
1300                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1301                                                        bond_dev);
1302                         res = notifier_to_errno(res);
1303                         if (res) {
1304                                 pr_err("%s: refused to change device type\n",
1305                                        bond_dev->name);
1306                                 res = -EBUSY;
1307                                 goto err_undo_flags;
1308                         }
1309
1310                         /* Flush unicast and multicast addresses */
1311                         dev_uc_flush(bond_dev);
1312                         dev_mc_flush(bond_dev);
1313
1314                         if (slave_dev->type != ARPHRD_ETHER)
1315                                 bond_setup_by_slave(bond_dev, slave_dev);
1316                         else {
1317                                 ether_setup(bond_dev);
1318                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1319                         }
1320
1321                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1322                                                  bond_dev);
1323                 }
1324         } else if (bond_dev->type != slave_dev->type) {
1325                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1326                        slave_dev->name,
1327                        slave_dev->type, bond_dev->type);
1328                 res = -EINVAL;
1329                 goto err_undo_flags;
1330         }
1331
1332         if (slave_ops->ndo_set_mac_address == NULL) {
1333                 if (!bond_has_slaves(bond)) {
1334                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1335                                    bond_dev->name);
1336                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1337                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1338                         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",
1339                                bond_dev->name);
1340                         res = -EOPNOTSUPP;
1341                         goto err_undo_flags;
1342                 }
1343         }
1344
1345         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1346
1347         /* If this is the first slave, then we need to set the master's hardware
1348          * address to be the same as the slave's. */
1349         if (!bond_has_slaves(bond) &&
1350             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1351                 bond_set_dev_addr(bond->dev, slave_dev);
1352
1353         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1354         if (!new_slave) {
1355                 res = -ENOMEM;
1356                 goto err_undo_flags;
1357         }
1358         /*
1359          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1360          * is set via sysfs or module option if desired.
1361          */
1362         new_slave->queue_id = 0;
1363
1364         /* Save slave's original mtu and then set it to match the bond */
1365         new_slave->original_mtu = slave_dev->mtu;
1366         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1367         if (res) {
1368                 pr_debug("Error %d calling dev_set_mtu\n", res);
1369                 goto err_free;
1370         }
1371
1372         /*
1373          * Save slave's original ("permanent") mac address for modes
1374          * that need it, and for restoring it upon release, and then
1375          * set it to the master's address
1376          */
1377         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1378
1379         if (!bond->params.fail_over_mac) {
1380                 /*
1381                  * Set slave to master's mac address.  The application already
1382                  * set the master's mac address to that of the first slave
1383                  */
1384                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1385                 addr.sa_family = slave_dev->type;
1386                 res = dev_set_mac_address(slave_dev, &addr);
1387                 if (res) {
1388                         pr_debug("Error %d calling set_mac_address\n", res);
1389                         goto err_restore_mtu;
1390                 }
1391         }
1392
1393         /* open the slave since the application closed it */
1394         res = dev_open(slave_dev);
1395         if (res) {
1396                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1397                 goto err_restore_mac;
1398         }
1399
1400         new_slave->bond = bond;
1401         new_slave->dev = slave_dev;
1402         slave_dev->priv_flags |= IFF_BONDING;
1403
1404         if (bond_is_lb(bond)) {
1405                 /* bond_alb_init_slave() must be called before all other stages since
1406                  * it might fail and we do not want to have to undo everything
1407                  */
1408                 res = bond_alb_init_slave(bond, new_slave);
1409                 if (res)
1410                         goto err_close;
1411         }
1412
1413         /* If the mode USES_PRIMARY, then the following is handled by
1414          * bond_change_active_slave().
1415          */
1416         if (!USES_PRIMARY(bond->params.mode)) {
1417                 /* set promiscuity level to new slave */
1418                 if (bond_dev->flags & IFF_PROMISC) {
1419                         res = dev_set_promiscuity(slave_dev, 1);
1420                         if (res)
1421                                 goto err_close;
1422                 }
1423
1424                 /* set allmulti level to new slave */
1425                 if (bond_dev->flags & IFF_ALLMULTI) {
1426                         res = dev_set_allmulti(slave_dev, 1);
1427                         if (res)
1428                                 goto err_close;
1429                 }
1430
1431                 netif_addr_lock_bh(bond_dev);
1432
1433                 dev_mc_sync_multiple(slave_dev, bond_dev);
1434                 dev_uc_sync_multiple(slave_dev, bond_dev);
1435
1436                 netif_addr_unlock_bh(bond_dev);
1437         }
1438
1439         if (bond->params.mode == BOND_MODE_8023AD) {
1440                 /* add lacpdu mc addr to mc list */
1441                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1442
1443                 dev_mc_add(slave_dev, lacpdu_multicast);
1444         }
1445
1446         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1447         if (res) {
1448                 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1449                        bond_dev->name, slave_dev->name);
1450                 goto err_close;
1451         }
1452
1453         prev_slave = bond_last_slave(bond);
1454
1455         new_slave->delay = 0;
1456         new_slave->link_failure_count = 0;
1457
1458         bond_update_speed_duplex(new_slave);
1459
1460         new_slave->last_arp_rx = jiffies -
1461                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1462         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1463                 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1464
1465         if (bond->params.miimon && !bond->params.use_carrier) {
1466                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1467
1468                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1469                         /*
1470                          * miimon is set but a bonded network driver
1471                          * does not support ETHTOOL/MII and
1472                          * arp_interval is not set.  Note: if
1473                          * use_carrier is enabled, we will never go
1474                          * here (because netif_carrier is always
1475                          * supported); thus, we don't need to change
1476                          * the messages for netif_carrier.
1477                          */
1478                         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",
1479                                bond_dev->name, slave_dev->name);
1480                 } else if (link_reporting == -1) {
1481                         /* unable get link status using mii/ethtool */
1482                         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",
1483                                    bond_dev->name, slave_dev->name);
1484                 }
1485         }
1486
1487         /* check for initial state */
1488         if (bond->params.miimon) {
1489                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1490                         if (bond->params.updelay) {
1491                                 new_slave->link = BOND_LINK_BACK;
1492                                 new_slave->delay = bond->params.updelay;
1493                         } else {
1494                                 new_slave->link = BOND_LINK_UP;
1495                         }
1496                 } else {
1497                         new_slave->link = BOND_LINK_DOWN;
1498                 }
1499         } else if (bond->params.arp_interval) {
1500                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1501                         BOND_LINK_UP : BOND_LINK_DOWN);
1502         } else {
1503                 new_slave->link = BOND_LINK_UP;
1504         }
1505
1506         if (new_slave->link != BOND_LINK_DOWN)
1507                 new_slave->jiffies = jiffies;
1508         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1509                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1510                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1511
1512         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1513                 /* if there is a primary slave, remember it */
1514                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1515                         bond->primary_slave = new_slave;
1516                         bond->force_primary = true;
1517                 }
1518         }
1519
1520         switch (bond->params.mode) {
1521         case BOND_MODE_ACTIVEBACKUP:
1522                 bond_set_slave_inactive_flags(new_slave);
1523                 break;
1524         case BOND_MODE_8023AD:
1525                 /* in 802.3ad mode, the internal mechanism
1526                  * will activate the slaves in the selected
1527                  * aggregator
1528                  */
1529                 bond_set_slave_inactive_flags(new_slave);
1530                 /* if this is the first slave */
1531                 if (!prev_slave) {
1532                         SLAVE_AD_INFO(new_slave).id = 1;
1533                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1534                          * can be called only after the mac address of the bond is set
1535                          */
1536                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1537                 } else {
1538                         SLAVE_AD_INFO(new_slave).id =
1539                                 SLAVE_AD_INFO(prev_slave).id + 1;
1540                 }
1541
1542                 bond_3ad_bind_slave(new_slave);
1543                 break;
1544         case BOND_MODE_TLB:
1545         case BOND_MODE_ALB:
1546                 bond_set_active_slave(new_slave);
1547                 bond_set_slave_inactive_flags(new_slave);
1548                 break;
1549         default:
1550                 pr_debug("This slave is always active in trunk mode\n");
1551
1552                 /* always active in trunk mode */
1553                 bond_set_active_slave(new_slave);
1554
1555                 /* In trunking mode there is little meaning to curr_active_slave
1556                  * anyway (it holds no special properties of the bond device),
1557                  * so we can change it without calling change_active_interface()
1558                  */
1559                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1560                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1561
1562                 break;
1563         } /* switch(bond_mode) */
1564
1565 #ifdef CONFIG_NET_POLL_CONTROLLER
1566         slave_dev->npinfo = bond->dev->npinfo;
1567         if (slave_dev->npinfo) {
1568                 if (slave_enable_netpoll(new_slave)) {
1569                         read_unlock(&bond->lock);
1570                         pr_info("Error, %s: master_dev is using netpoll, "
1571                                  "but new slave device does not support netpoll.\n",
1572                                  bond_dev->name);
1573                         res = -EBUSY;
1574                         goto err_detach;
1575                 }
1576         }
1577 #endif
1578
1579         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1580                                          new_slave);
1581         if (res) {
1582                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1583                 goto err_detach;
1584         }
1585
1586         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1587         if (res) {
1588                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1589                 goto err_unregister;
1590         }
1591
1592         bond->slave_cnt++;
1593         bond_compute_features(bond);
1594         bond_set_carrier(bond);
1595
1596         if (USES_PRIMARY(bond->params.mode)) {
1597                 read_lock(&bond->lock);
1598                 write_lock_bh(&bond->curr_slave_lock);
1599                 bond_select_active_slave(bond);
1600                 write_unlock_bh(&bond->curr_slave_lock);
1601                 read_unlock(&bond->lock);
1602         }
1603
1604         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1605                 bond_dev->name, slave_dev->name,
1606                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1607                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1608
1609         /* enslave is successful */
1610         return 0;
1611
1612 /* Undo stages on error */
1613 err_unregister:
1614         netdev_rx_handler_unregister(slave_dev);
1615
1616 err_detach:
1617         if (!USES_PRIMARY(bond->params.mode))
1618                 bond_hw_addr_flush(bond_dev, slave_dev);
1619
1620         vlan_vids_del_by_dev(slave_dev, bond_dev);
1621         write_lock_bh(&bond->lock);
1622         if (bond->primary_slave == new_slave)
1623                 bond->primary_slave = NULL;
1624         if (bond->curr_active_slave == new_slave) {
1625                 bond_change_active_slave(bond, NULL);
1626                 write_unlock_bh(&bond->lock);
1627                 read_lock(&bond->lock);
1628                 write_lock_bh(&bond->curr_slave_lock);
1629                 bond_select_active_slave(bond);
1630                 write_unlock_bh(&bond->curr_slave_lock);
1631                 read_unlock(&bond->lock);
1632         } else {
1633                 write_unlock_bh(&bond->lock);
1634         }
1635         slave_disable_netpoll(new_slave);
1636
1637 err_close:
1638         slave_dev->priv_flags &= ~IFF_BONDING;
1639         dev_close(slave_dev);
1640
1641 err_restore_mac:
1642         if (!bond->params.fail_over_mac) {
1643                 /* XXX TODO - fom follow mode needs to change master's
1644                  * MAC if this slave's MAC is in use by the bond, or at
1645                  * least print a warning.
1646                  */
1647                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1648                 addr.sa_family = slave_dev->type;
1649                 dev_set_mac_address(slave_dev, &addr);
1650         }
1651
1652 err_restore_mtu:
1653         dev_set_mtu(slave_dev, new_slave->original_mtu);
1654
1655 err_free:
1656         kfree(new_slave);
1657
1658 err_undo_flags:
1659         /* Enslave of first slave has failed and we need to fix master's mac */
1660         if (!bond_has_slaves(bond) &&
1661             ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1662                 eth_hw_addr_random(bond_dev);
1663
1664         return res;
1665 }
1666
1667 /*
1668  * Try to release the slave device <slave> from the bond device <master>
1669  * It is legal to access curr_active_slave without a lock because all the function
1670  * is write-locked. If "all" is true it means that the function is being called
1671  * while destroying a bond interface and all slaves are being released.
1672  *
1673  * The rules for slave state should be:
1674  *   for Active/Backup:
1675  *     Active stays on all backups go down
1676  *   for Bonded connections:
1677  *     The first up interface should be left on and all others downed.
1678  */
1679 static int __bond_release_one(struct net_device *bond_dev,
1680                               struct net_device *slave_dev,
1681                               bool all)
1682 {
1683         struct bonding *bond = netdev_priv(bond_dev);
1684         struct slave *slave, *oldcurrent;
1685         struct sockaddr addr;
1686         int old_flags = bond_dev->flags;
1687         netdev_features_t old_features = bond_dev->features;
1688
1689         /* slave is not a slave or master is not master of this slave */
1690         if (!(slave_dev->flags & IFF_SLAVE) ||
1691             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1692                 pr_err("%s: Error: cannot release %s.\n",
1693                        bond_dev->name, slave_dev->name);
1694                 return -EINVAL;
1695         }
1696
1697         block_netpoll_tx();
1698         write_lock_bh(&bond->lock);
1699
1700         slave = bond_get_slave_by_dev(bond, slave_dev);
1701         if (!slave) {
1702                 /* not a slave of this bond */
1703                 pr_info("%s: %s not enslaved\n",
1704                         bond_dev->name, slave_dev->name);
1705                 write_unlock_bh(&bond->lock);
1706                 unblock_netpoll_tx();
1707                 return -EINVAL;
1708         }
1709
1710         write_unlock_bh(&bond->lock);
1711
1712         /* release the slave from its bond */
1713         bond->slave_cnt--;
1714
1715         bond_upper_dev_unlink(bond_dev, slave_dev);
1716         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1717          * for this slave anymore.
1718          */
1719         netdev_rx_handler_unregister(slave_dev);
1720         write_lock_bh(&bond->lock);
1721
1722         /* Inform AD package of unbinding of slave. */
1723         if (bond->params.mode == BOND_MODE_8023AD) {
1724                 /* must be called before the slave is
1725                  * detached from the list
1726                  */
1727                 bond_3ad_unbind_slave(slave);
1728         }
1729
1730         pr_info("%s: releasing %s interface %s\n",
1731                 bond_dev->name,
1732                 bond_is_active_slave(slave) ? "active" : "backup",
1733                 slave_dev->name);
1734
1735         oldcurrent = bond->curr_active_slave;
1736
1737         bond->current_arp_slave = NULL;
1738
1739         if (!all && !bond->params.fail_over_mac) {
1740                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1741                     bond_has_slaves(bond))
1742                         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",
1743                                    bond_dev->name, slave_dev->name,
1744                                    slave->perm_hwaddr,
1745                                    bond_dev->name, slave_dev->name);
1746         }
1747
1748         if (bond->primary_slave == slave)
1749                 bond->primary_slave = NULL;
1750
1751         if (oldcurrent == slave)
1752                 bond_change_active_slave(bond, NULL);
1753
1754         if (bond_is_lb(bond)) {
1755                 /* Must be called only after the slave has been
1756                  * detached from the list and the curr_active_slave
1757                  * has been cleared (if our_slave == old_current),
1758                  * but before a new active slave is selected.
1759                  */
1760                 write_unlock_bh(&bond->lock);
1761                 bond_alb_deinit_slave(bond, slave);
1762                 write_lock_bh(&bond->lock);
1763         }
1764
1765         if (all) {
1766                 rcu_assign_pointer(bond->curr_active_slave, NULL);
1767         } else if (oldcurrent == slave) {
1768                 /*
1769                  * Note that we hold RTNL over this sequence, so there
1770                  * is no concern that another slave add/remove event
1771                  * will interfere.
1772                  */
1773                 write_unlock_bh(&bond->lock);
1774                 read_lock(&bond->lock);
1775                 write_lock_bh(&bond->curr_slave_lock);
1776
1777                 bond_select_active_slave(bond);
1778
1779                 write_unlock_bh(&bond->curr_slave_lock);
1780                 read_unlock(&bond->lock);
1781                 write_lock_bh(&bond->lock);
1782         }
1783
1784         if (!bond_has_slaves(bond)) {
1785                 bond_set_carrier(bond);
1786                 eth_hw_addr_random(bond_dev);
1787
1788                 if (vlan_uses_dev(bond_dev)) {
1789                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1790                                    bond_dev->name, bond_dev->name);
1791                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1792                                    bond_dev->name);
1793                 }
1794         }
1795
1796         write_unlock_bh(&bond->lock);
1797         unblock_netpoll_tx();
1798         synchronize_rcu();
1799
1800         if (!bond_has_slaves(bond)) {
1801                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1802                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1803         }
1804
1805         bond_compute_features(bond);
1806         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1807             (old_features & NETIF_F_VLAN_CHALLENGED))
1808                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1809                         bond_dev->name, slave_dev->name, bond_dev->name);
1810
1811         /* must do this from outside any spinlocks */
1812         vlan_vids_del_by_dev(slave_dev, bond_dev);
1813
1814         /* If the mode USES_PRIMARY, then this cases was handled above by
1815          * bond_change_active_slave(..., NULL)
1816          */
1817         if (!USES_PRIMARY(bond->params.mode)) {
1818                 /* unset promiscuity level from slave
1819                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1820                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1821                  * value of that flag before that change, as that was the value
1822                  * when this slave was attached, so we cache at the start of the
1823                  * function and use it here. Same goes for ALLMULTI below
1824                  */
1825                 if (old_flags & IFF_PROMISC)
1826                         dev_set_promiscuity(slave_dev, -1);
1827
1828                 /* unset allmulti level from slave */
1829                 if (old_flags & IFF_ALLMULTI)
1830                         dev_set_allmulti(slave_dev, -1);
1831
1832                 bond_hw_addr_flush(bond_dev, slave_dev);
1833         }
1834
1835         slave_disable_netpoll(slave);
1836
1837         /* close slave before restoring its mac address */
1838         dev_close(slave_dev);
1839
1840         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1841                 /* restore original ("permanent") mac address */
1842                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1843                 addr.sa_family = slave_dev->type;
1844                 dev_set_mac_address(slave_dev, &addr);
1845         }
1846
1847         dev_set_mtu(slave_dev, slave->original_mtu);
1848
1849         slave_dev->priv_flags &= ~IFF_BONDING;
1850
1851         kfree(slave);
1852
1853         return 0;  /* deletion OK */
1854 }
1855
1856 /* A wrapper used because of ndo_del_link */
1857 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1858 {
1859         return __bond_release_one(bond_dev, slave_dev, false);
1860 }
1861
1862 /*
1863 * First release a slave and then destroy the bond if no more slaves are left.
1864 * Must be under rtnl_lock when this function is called.
1865 */
1866 static int  bond_release_and_destroy(struct net_device *bond_dev,
1867                                      struct net_device *slave_dev)
1868 {
1869         struct bonding *bond = netdev_priv(bond_dev);
1870         int ret;
1871
1872         ret = bond_release(bond_dev, slave_dev);
1873         if (ret == 0 && !bond_has_slaves(bond)) {
1874                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1875                 pr_info("%s: destroying bond %s.\n",
1876                         bond_dev->name, bond_dev->name);
1877                 unregister_netdevice(bond_dev);
1878         }
1879         return ret;
1880 }
1881
1882 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1883 {
1884         struct bonding *bond = netdev_priv(bond_dev);
1885
1886         info->bond_mode = bond->params.mode;
1887         info->miimon = bond->params.miimon;
1888
1889         read_lock(&bond->lock);
1890         info->num_slaves = bond->slave_cnt;
1891         read_unlock(&bond->lock);
1892
1893         return 0;
1894 }
1895
1896 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1897 {
1898         struct bonding *bond = netdev_priv(bond_dev);
1899         struct list_head *iter;
1900         int i = 0, res = -ENODEV;
1901         struct slave *slave;
1902
1903         read_lock(&bond->lock);
1904         bond_for_each_slave(bond, slave, iter) {
1905                 if (i++ == (int)info->slave_id) {
1906                         res = 0;
1907                         strcpy(info->slave_name, slave->dev->name);
1908                         info->link = slave->link;
1909                         info->state = bond_slave_state(slave);
1910                         info->link_failure_count = slave->link_failure_count;
1911                         break;
1912                 }
1913         }
1914         read_unlock(&bond->lock);
1915
1916         return res;
1917 }
1918
1919 /*-------------------------------- Monitoring -------------------------------*/
1920
1921
1922 static int bond_miimon_inspect(struct bonding *bond)
1923 {
1924         int link_state, commit = 0;
1925         struct list_head *iter;
1926         struct slave *slave;
1927         bool ignore_updelay;
1928
1929         ignore_updelay = !bond->curr_active_slave ? true : false;
1930
1931         bond_for_each_slave(bond, slave, iter) {
1932                 slave->new_link = BOND_LINK_NOCHANGE;
1933
1934                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1935
1936                 switch (slave->link) {
1937                 case BOND_LINK_UP:
1938                         if (link_state)
1939                                 continue;
1940
1941                         slave->link = BOND_LINK_FAIL;
1942                         slave->delay = bond->params.downdelay;
1943                         if (slave->delay) {
1944                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1945                                         bond->dev->name,
1946                                         (bond->params.mode ==
1947                                          BOND_MODE_ACTIVEBACKUP) ?
1948                                         (bond_is_active_slave(slave) ?
1949                                          "active " : "backup ") : "",
1950                                         slave->dev->name,
1951                                         bond->params.downdelay * bond->params.miimon);
1952                         }
1953                         /*FALLTHRU*/
1954                 case BOND_LINK_FAIL:
1955                         if (link_state) {
1956                                 /*
1957                                  * recovered before downdelay expired
1958                                  */
1959                                 slave->link = BOND_LINK_UP;
1960                                 slave->jiffies = jiffies;
1961                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
1962                                         bond->dev->name,
1963                                         (bond->params.downdelay - slave->delay) *
1964                                         bond->params.miimon,
1965                                         slave->dev->name);
1966                                 continue;
1967                         }
1968
1969                         if (slave->delay <= 0) {
1970                                 slave->new_link = BOND_LINK_DOWN;
1971                                 commit++;
1972                                 continue;
1973                         }
1974
1975                         slave->delay--;
1976                         break;
1977
1978                 case BOND_LINK_DOWN:
1979                         if (!link_state)
1980                                 continue;
1981
1982                         slave->link = BOND_LINK_BACK;
1983                         slave->delay = bond->params.updelay;
1984
1985                         if (slave->delay) {
1986                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1987                                         bond->dev->name, slave->dev->name,
1988                                         ignore_updelay ? 0 :
1989                                         bond->params.updelay *
1990                                         bond->params.miimon);
1991                         }
1992                         /*FALLTHRU*/
1993                 case BOND_LINK_BACK:
1994                         if (!link_state) {
1995                                 slave->link = BOND_LINK_DOWN;
1996                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
1997                                         bond->dev->name,
1998                                         (bond->params.updelay - slave->delay) *
1999                                         bond->params.miimon,
2000                                         slave->dev->name);
2001
2002                                 continue;
2003                         }
2004
2005                         if (ignore_updelay)
2006                                 slave->delay = 0;
2007
2008                         if (slave->delay <= 0) {
2009                                 slave->new_link = BOND_LINK_UP;
2010                                 commit++;
2011                                 ignore_updelay = false;
2012                                 continue;
2013                         }
2014
2015                         slave->delay--;
2016                         break;
2017                 }
2018         }
2019
2020         return commit;
2021 }
2022
2023 static void bond_miimon_commit(struct bonding *bond)
2024 {
2025         struct list_head *iter;
2026         struct slave *slave;
2027
2028         bond_for_each_slave(bond, slave, iter) {
2029                 switch (slave->new_link) {
2030                 case BOND_LINK_NOCHANGE:
2031                         continue;
2032
2033                 case BOND_LINK_UP:
2034                         slave->link = BOND_LINK_UP;
2035                         slave->jiffies = jiffies;
2036
2037                         if (bond->params.mode == BOND_MODE_8023AD) {
2038                                 /* prevent it from being the active one */
2039                                 bond_set_backup_slave(slave);
2040                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2041                                 /* make it immediately active */
2042                                 bond_set_active_slave(slave);
2043                         } else if (slave != bond->primary_slave) {
2044                                 /* prevent it from being the active one */
2045                                 bond_set_backup_slave(slave);
2046                         }
2047
2048                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2049                                 bond->dev->name, slave->dev->name,
2050                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2051                                 slave->duplex ? "full" : "half");
2052
2053                         /* notify ad that the link status has changed */
2054                         if (bond->params.mode == BOND_MODE_8023AD)
2055                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2056
2057                         if (bond_is_lb(bond))
2058                                 bond_alb_handle_link_change(bond, slave,
2059                                                             BOND_LINK_UP);
2060
2061                         if (!bond->curr_active_slave ||
2062                             (slave == bond->primary_slave))
2063                                 goto do_failover;
2064
2065                         continue;
2066
2067                 case BOND_LINK_DOWN:
2068                         if (slave->link_failure_count < UINT_MAX)
2069                                 slave->link_failure_count++;
2070
2071                         slave->link = BOND_LINK_DOWN;
2072
2073                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2074                             bond->params.mode == BOND_MODE_8023AD)
2075                                 bond_set_slave_inactive_flags(slave);
2076
2077                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2078                                 bond->dev->name, slave->dev->name);
2079
2080                         if (bond->params.mode == BOND_MODE_8023AD)
2081                                 bond_3ad_handle_link_change(slave,
2082                                                             BOND_LINK_DOWN);
2083
2084                         if (bond_is_lb(bond))
2085                                 bond_alb_handle_link_change(bond, slave,
2086                                                             BOND_LINK_DOWN);
2087
2088                         if (slave == bond->curr_active_slave)
2089                                 goto do_failover;
2090
2091                         continue;
2092
2093                 default:
2094                         pr_err("%s: invalid new link %d on slave %s\n",
2095                                bond->dev->name, slave->new_link,
2096                                slave->dev->name);
2097                         slave->new_link = BOND_LINK_NOCHANGE;
2098
2099                         continue;
2100                 }
2101
2102 do_failover:
2103                 ASSERT_RTNL();
2104                 block_netpoll_tx();
2105                 write_lock_bh(&bond->curr_slave_lock);
2106                 bond_select_active_slave(bond);
2107                 write_unlock_bh(&bond->curr_slave_lock);
2108                 unblock_netpoll_tx();
2109         }
2110
2111         bond_set_carrier(bond);
2112 }
2113
2114 /*
2115  * bond_mii_monitor
2116  *
2117  * Really a wrapper that splits the mii monitor into two phases: an
2118  * inspection, then (if inspection indicates something needs to be done)
2119  * an acquisition of appropriate locks followed by a commit phase to
2120  * implement whatever link state changes are indicated.
2121  */
2122 void bond_mii_monitor(struct work_struct *work)
2123 {
2124         struct bonding *bond = container_of(work, struct bonding,
2125                                             mii_work.work);
2126         bool should_notify_peers = false;
2127         unsigned long delay;
2128
2129         read_lock(&bond->lock);
2130
2131         delay = msecs_to_jiffies(bond->params.miimon);
2132
2133         if (!bond_has_slaves(bond))
2134                 goto re_arm;
2135
2136         should_notify_peers = bond_should_notify_peers(bond);
2137
2138         if (bond_miimon_inspect(bond)) {
2139                 read_unlock(&bond->lock);
2140
2141                 /* Race avoidance with bond_close cancel of workqueue */
2142                 if (!rtnl_trylock()) {
2143                         read_lock(&bond->lock);
2144                         delay = 1;
2145                         should_notify_peers = false;
2146                         goto re_arm;
2147                 }
2148
2149                 read_lock(&bond->lock);
2150
2151                 bond_miimon_commit(bond);
2152
2153                 read_unlock(&bond->lock);
2154                 rtnl_unlock();  /* might sleep, hold no other locks */
2155                 read_lock(&bond->lock);
2156         }
2157
2158 re_arm:
2159         if (bond->params.miimon)
2160                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2161
2162         read_unlock(&bond->lock);
2163
2164         if (should_notify_peers) {
2165                 if (!rtnl_trylock())
2166                         return;
2167                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2168                 rtnl_unlock();
2169         }
2170 }
2171
2172 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2173 {
2174         struct net_device *upper;
2175         struct list_head *iter;
2176         bool ret = false;
2177
2178         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2179                 return true;
2180
2181         rcu_read_lock();
2182         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2183                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2184                         ret = true;
2185                         break;
2186                 }
2187         }
2188         rcu_read_unlock();
2189
2190         return ret;
2191 }
2192
2193 /*
2194  * We go to the (large) trouble of VLAN tagging ARP frames because
2195  * switches in VLAN mode (especially if ports are configured as
2196  * "native" to a VLAN) might not pass non-tagged frames.
2197  */
2198 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2199 {
2200         struct sk_buff *skb;
2201
2202         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2203                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2204
2205         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2206                          NULL, slave_dev->dev_addr, NULL);
2207
2208         if (!skb) {
2209                 pr_err("ARP packet allocation failed\n");
2210                 return;
2211         }
2212         if (vlan_id) {
2213                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2214                 if (!skb) {
2215                         pr_err("failed to insert VLAN tag\n");
2216                         return;
2217                 }
2218         }
2219         arp_xmit(skb);
2220 }
2221
2222
2223 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2224 {
2225         struct net_device *upper, *vlan_upper;
2226         struct list_head *iter, *vlan_iter;
2227         struct rtable *rt;
2228         __be32 *targets = bond->params.arp_targets, addr;
2229         int i, vlan_id;
2230
2231         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2232                 pr_debug("basa: target %pI4\n", &targets[i]);
2233
2234                 /* Find out through which dev should the packet go */
2235                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2236                                      RTO_ONLINK, 0);
2237                 if (IS_ERR(rt)) {
2238                         pr_debug("%s: no route to arp_ip_target %pI4\n",
2239                                  bond->dev->name, &targets[i]);
2240                         continue;
2241                 }
2242
2243                 vlan_id = 0;
2244
2245                 /* bond device itself */
2246                 if (rt->dst.dev == bond->dev)
2247                         goto found;
2248
2249                 rcu_read_lock();
2250                 /* first we search only for vlan devices. for every vlan
2251                  * found we verify its upper dev list, searching for the
2252                  * rt->dst.dev. If found we save the tag of the vlan and
2253                  * proceed to send the packet.
2254                  *
2255                  * TODO: QinQ?
2256                  */
2257                 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2258                                                   vlan_iter) {
2259                         if (!is_vlan_dev(vlan_upper))
2260                                 continue;
2261                         netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2262                                                           iter) {
2263                                 if (upper == rt->dst.dev) {
2264                                         vlan_id = vlan_dev_vlan_id(vlan_upper);
2265                                         rcu_read_unlock();
2266                                         goto found;
2267                                 }
2268                         }
2269                 }
2270
2271                 /* if the device we're looking for is not on top of any of
2272                  * our upper vlans, then just search for any dev that
2273                  * matches, and in case it's a vlan - save the id
2274                  */
2275                 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2276                         if (upper == rt->dst.dev) {
2277                                 /* if it's a vlan - get its VID */
2278                                 if (is_vlan_dev(upper))
2279                                         vlan_id = vlan_dev_vlan_id(upper);
2280
2281                                 rcu_read_unlock();
2282                                 goto found;
2283                         }
2284                 }
2285                 rcu_read_unlock();
2286
2287                 /* Not our device - skip */
2288                 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2289                          bond->dev->name, &targets[i],
2290                          rt->dst.dev ? rt->dst.dev->name : "NULL");
2291
2292                 ip_rt_put(rt);
2293                 continue;
2294
2295 found:
2296                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2297                 ip_rt_put(rt);
2298                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2299                               addr, vlan_id);
2300         }
2301 }
2302
2303 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2304 {
2305         int i;
2306
2307         if (!sip || !bond_has_this_ip(bond, tip)) {
2308                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2309                 return;
2310         }
2311
2312         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2313         if (i == -1) {
2314                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2315                 return;
2316         }
2317         slave->last_arp_rx = jiffies;
2318         slave->target_last_arp_rx[i] = jiffies;
2319 }
2320
2321 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2322                  struct slave *slave)
2323 {
2324         struct arphdr *arp = (struct arphdr *)skb->data;
2325         unsigned char *arp_ptr;
2326         __be32 sip, tip;
2327         int alen;
2328
2329         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2330                 return RX_HANDLER_ANOTHER;
2331
2332         read_lock(&bond->lock);
2333
2334         if (!slave_do_arp_validate(bond, slave))
2335                 goto out_unlock;
2336
2337         alen = arp_hdr_len(bond->dev);
2338
2339         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2340                  bond->dev->name, skb->dev->name);
2341
2342         if (alen > skb_headlen(skb)) {
2343                 arp = kmalloc(alen, GFP_ATOMIC);
2344                 if (!arp)
2345                         goto out_unlock;
2346                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2347                         goto out_unlock;
2348         }
2349
2350         if (arp->ar_hln != bond->dev->addr_len ||
2351             skb->pkt_type == PACKET_OTHERHOST ||
2352             skb->pkt_type == PACKET_LOOPBACK ||
2353             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2354             arp->ar_pro != htons(ETH_P_IP) ||
2355             arp->ar_pln != 4)
2356                 goto out_unlock;
2357
2358         arp_ptr = (unsigned char *)(arp + 1);
2359         arp_ptr += bond->dev->addr_len;
2360         memcpy(&sip, arp_ptr, 4);
2361         arp_ptr += 4 + bond->dev->addr_len;
2362         memcpy(&tip, arp_ptr, 4);
2363
2364         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2365                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2366                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2367                  &sip, &tip);
2368
2369         /*
2370          * Backup slaves won't see the ARP reply, but do come through
2371          * here for each ARP probe (so we swap the sip/tip to validate
2372          * the probe).  In a "redundant switch, common router" type of
2373          * configuration, the ARP probe will (hopefully) travel from
2374          * the active, through one switch, the router, then the other
2375          * switch before reaching the backup.
2376          *
2377          * We 'trust' the arp requests if there is an active slave and
2378          * it received valid arp reply(s) after it became active. This
2379          * is done to avoid endless looping when we can't reach the
2380          * arp_ip_target and fool ourselves with our own arp requests.
2381          */
2382         if (bond_is_active_slave(slave))
2383                 bond_validate_arp(bond, slave, sip, tip);
2384         else if (bond->curr_active_slave &&
2385                  time_after(slave_last_rx(bond, bond->curr_active_slave),
2386                             bond->curr_active_slave->jiffies))
2387                 bond_validate_arp(bond, slave, tip, sip);
2388
2389 out_unlock:
2390         read_unlock(&bond->lock);
2391         if (arp != (struct arphdr *)skb->data)
2392                 kfree(arp);
2393         return RX_HANDLER_ANOTHER;
2394 }
2395
2396 /* function to verify if we're in the arp_interval timeslice, returns true if
2397  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2398  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2399  */
2400 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2401                                   int mod)
2402 {
2403         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2404
2405         return time_in_range(jiffies,
2406                              last_act - delta_in_ticks,
2407                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2408 }
2409
2410 /*
2411  * this function is called regularly to monitor each slave's link
2412  * ensuring that traffic is being sent and received when arp monitoring
2413  * is used in load-balancing mode. if the adapter has been dormant, then an
2414  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2415  * arp monitoring in active backup mode.
2416  */
2417 void bond_loadbalance_arp_mon(struct work_struct *work)
2418 {
2419         struct bonding *bond = container_of(work, struct bonding,
2420                                             arp_work.work);
2421         struct slave *slave, *oldcurrent;
2422         struct list_head *iter;
2423         int do_failover = 0;
2424
2425         read_lock(&bond->lock);
2426
2427         if (!bond_has_slaves(bond))
2428                 goto re_arm;
2429
2430         oldcurrent = bond->curr_active_slave;
2431         /* see if any of the previous devices are up now (i.e. they have
2432          * xmt and rcv traffic). the curr_active_slave does not come into
2433          * the picture unless it is null. also, slave->jiffies is not needed
2434          * here because we send an arp on each slave and give a slave as
2435          * long as it needs to get the tx/rx within the delta.
2436          * TODO: what about up/down delay in arp mode? it wasn't here before
2437          *       so it can wait
2438          */
2439         bond_for_each_slave(bond, slave, iter) {
2440                 unsigned long trans_start = dev_trans_start(slave->dev);
2441
2442                 if (slave->link != BOND_LINK_UP) {
2443                         if (bond_time_in_interval(bond, trans_start, 1) &&
2444                             bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2445
2446                                 slave->link  = BOND_LINK_UP;
2447                                 bond_set_active_slave(slave);
2448
2449                                 /* primary_slave has no meaning in round-robin
2450                                  * mode. the window of a slave being up and
2451                                  * curr_active_slave being null after enslaving
2452                                  * is closed.
2453                                  */
2454                                 if (!oldcurrent) {
2455                                         pr_info("%s: link status definitely up for interface %s, ",
2456                                                 bond->dev->name,
2457                                                 slave->dev->name);
2458                                         do_failover = 1;
2459                                 } else {
2460                                         pr_info("%s: interface %s is now up\n",
2461                                                 bond->dev->name,
2462                                                 slave->dev->name);
2463                                 }
2464                         }
2465                 } else {
2466                         /* slave->link == BOND_LINK_UP */
2467
2468                         /* not all switches will respond to an arp request
2469                          * when the source ip is 0, so don't take the link down
2470                          * if we don't know our ip yet
2471                          */
2472                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2473                             !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2474
2475                                 slave->link  = BOND_LINK_DOWN;
2476                                 bond_set_backup_slave(slave);
2477
2478                                 if (slave->link_failure_count < UINT_MAX)
2479                                         slave->link_failure_count++;
2480
2481                                 pr_info("%s: interface %s is now down.\n",
2482                                         bond->dev->name,
2483                                         slave->dev->name);
2484
2485                                 if (slave == oldcurrent)
2486                                         do_failover = 1;
2487                         }
2488                 }
2489
2490                 /* note: if switch is in round-robin mode, all links
2491                  * must tx arp to ensure all links rx an arp - otherwise
2492                  * links may oscillate or not come up at all; if switch is
2493                  * in something like xor mode, there is nothing we can
2494                  * do - all replies will be rx'ed on same link causing slaves
2495                  * to be unstable during low/no traffic periods
2496                  */
2497                 if (IS_UP(slave->dev))
2498                         bond_arp_send_all(bond, slave);
2499         }
2500
2501         if (do_failover) {
2502                 block_netpoll_tx();
2503                 write_lock_bh(&bond->curr_slave_lock);
2504
2505                 bond_select_active_slave(bond);
2506
2507                 write_unlock_bh(&bond->curr_slave_lock);
2508                 unblock_netpoll_tx();
2509         }
2510
2511 re_arm:
2512         if (bond->params.arp_interval)
2513                 queue_delayed_work(bond->wq, &bond->arp_work,
2514                                    msecs_to_jiffies(bond->params.arp_interval));
2515
2516         read_unlock(&bond->lock);
2517 }
2518
2519 /*
2520  * Called to inspect slaves for active-backup mode ARP monitor link state
2521  * changes.  Sets new_link in slaves to specify what action should take
2522  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2523  * to link states must be committed.
2524  *
2525  * Called with bond->lock held for read.
2526  */
2527 static int bond_ab_arp_inspect(struct bonding *bond)
2528 {
2529         unsigned long trans_start, last_rx;
2530         struct list_head *iter;
2531         struct slave *slave;
2532         int commit = 0;
2533
2534         bond_for_each_slave(bond, slave, iter) {
2535                 slave->new_link = BOND_LINK_NOCHANGE;
2536                 last_rx = slave_last_rx(bond, slave);
2537
2538                 if (slave->link != BOND_LINK_UP) {
2539                         if (bond_time_in_interval(bond, last_rx, 1)) {
2540                                 slave->new_link = BOND_LINK_UP;
2541                                 commit++;
2542                         }
2543                         continue;
2544                 }
2545
2546                 /*
2547                  * Give slaves 2*delta after being enslaved or made
2548                  * active.  This avoids bouncing, as the last receive
2549                  * times need a full ARP monitor cycle to be updated.
2550                  */
2551                 if (bond_time_in_interval(bond, slave->jiffies, 2))
2552                         continue;
2553
2554                 /*
2555                  * Backup slave is down if:
2556                  * - No current_arp_slave AND
2557                  * - more than 3*delta since last receive AND
2558                  * - the bond has an IP address
2559                  *
2560                  * Note: a non-null current_arp_slave indicates
2561                  * the curr_active_slave went down and we are
2562                  * searching for a new one; under this condition
2563                  * we only take the curr_active_slave down - this
2564                  * gives each slave a chance to tx/rx traffic
2565                  * before being taken out
2566                  */
2567                 if (!bond_is_active_slave(slave) &&
2568                     !bond->current_arp_slave &&
2569                     !bond_time_in_interval(bond, last_rx, 3)) {
2570                         slave->new_link = BOND_LINK_DOWN;
2571                         commit++;
2572                 }
2573
2574                 /*
2575                  * Active slave is down if:
2576                  * - more than 2*delta since transmitting OR
2577                  * - (more than 2*delta since receive AND
2578                  *    the bond has an IP address)
2579                  */
2580                 trans_start = dev_trans_start(slave->dev);
2581                 if (bond_is_active_slave(slave) &&
2582                     (!bond_time_in_interval(bond, trans_start, 2) ||
2583                      !bond_time_in_interval(bond, last_rx, 2))) {
2584                         slave->new_link = BOND_LINK_DOWN;
2585                         commit++;
2586                 }
2587         }
2588
2589         return commit;
2590 }
2591
2592 /*
2593  * Called to commit link state changes noted by inspection step of
2594  * active-backup mode ARP monitor.
2595  *
2596  * Called with RTNL and bond->lock for read.
2597  */
2598 static void bond_ab_arp_commit(struct bonding *bond)
2599 {
2600         unsigned long trans_start;
2601         struct list_head *iter;
2602         struct slave *slave;
2603
2604         bond_for_each_slave(bond, slave, iter) {
2605                 switch (slave->new_link) {
2606                 case BOND_LINK_NOCHANGE:
2607                         continue;
2608
2609                 case BOND_LINK_UP:
2610                         trans_start = dev_trans_start(slave->dev);
2611                         if (bond->curr_active_slave != slave ||
2612                             (!bond->curr_active_slave &&
2613                              bond_time_in_interval(bond, trans_start, 1))) {
2614                                 slave->link = BOND_LINK_UP;
2615                                 if (bond->current_arp_slave) {
2616                                         bond_set_slave_inactive_flags(
2617                                                 bond->current_arp_slave);
2618                                         bond->current_arp_slave = NULL;
2619                                 }
2620
2621                                 pr_info("%s: link status definitely up for interface %s.\n",
2622                                         bond->dev->name, slave->dev->name);
2623
2624                                 if (!bond->curr_active_slave ||
2625                                     (slave == bond->primary_slave))
2626                                         goto do_failover;
2627
2628                         }
2629
2630                         continue;
2631
2632                 case BOND_LINK_DOWN:
2633                         if (slave->link_failure_count < UINT_MAX)
2634                                 slave->link_failure_count++;
2635
2636                         slave->link = BOND_LINK_DOWN;
2637                         bond_set_slave_inactive_flags(slave);
2638
2639                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2640                                 bond->dev->name, slave->dev->name);
2641
2642                         if (slave == bond->curr_active_slave) {
2643                                 bond->current_arp_slave = NULL;
2644                                 goto do_failover;
2645                         }
2646
2647                         continue;
2648
2649                 default:
2650                         pr_err("%s: impossible: new_link %d on slave %s\n",
2651                                bond->dev->name, slave->new_link,
2652                                slave->dev->name);
2653                         continue;
2654                 }
2655
2656 do_failover:
2657                 ASSERT_RTNL();
2658                 block_netpoll_tx();
2659                 write_lock_bh(&bond->curr_slave_lock);
2660                 bond_select_active_slave(bond);
2661                 write_unlock_bh(&bond->curr_slave_lock);
2662                 unblock_netpoll_tx();
2663         }
2664
2665         bond_set_carrier(bond);
2666 }
2667
2668 /*
2669  * Send ARP probes for active-backup mode ARP monitor.
2670  *
2671  * Called with bond->lock held for read.
2672  */
2673 static void bond_ab_arp_probe(struct bonding *bond)
2674 {
2675         struct slave *slave, *before = NULL, *new_slave = NULL;
2676         struct list_head *iter;
2677         bool found = false;
2678
2679         read_lock(&bond->curr_slave_lock);
2680
2681         if (bond->current_arp_slave && bond->curr_active_slave)
2682                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2683                         bond->current_arp_slave->dev->name,
2684                         bond->curr_active_slave->dev->name);
2685
2686         if (bond->curr_active_slave) {
2687                 bond_arp_send_all(bond, bond->curr_active_slave);
2688                 read_unlock(&bond->curr_slave_lock);
2689                 return;
2690         }
2691
2692         read_unlock(&bond->curr_slave_lock);
2693
2694         /* if we don't have a curr_active_slave, search for the next available
2695          * backup slave from the current_arp_slave and make it the candidate
2696          * for becoming the curr_active_slave
2697          */
2698
2699         if (!bond->current_arp_slave) {
2700                 bond->current_arp_slave = bond_first_slave(bond);
2701                 if (!bond->current_arp_slave)
2702                         return;
2703         }
2704
2705         bond_set_slave_inactive_flags(bond->current_arp_slave);
2706
2707         bond_for_each_slave(bond, slave, iter) {
2708                 if (!found && !before && IS_UP(slave->dev))
2709                         before = slave;
2710
2711                 if (found && !new_slave && IS_UP(slave->dev))
2712                         new_slave = slave;
2713                 /* if the link state is up at this point, we
2714                  * mark it down - this can happen if we have
2715                  * simultaneous link failures and
2716                  * reselect_active_interface doesn't make this
2717                  * one the current slave so it is still marked
2718                  * up when it is actually down
2719                  */
2720                 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2721                         slave->link = BOND_LINK_DOWN;
2722                         if (slave->link_failure_count < UINT_MAX)
2723                                 slave->link_failure_count++;
2724
2725                         bond_set_slave_inactive_flags(slave);
2726
2727                         pr_info("%s: backup interface %s is now down.\n",
2728                                 bond->dev->name, slave->dev->name);
2729                 }
2730                 if (slave == bond->current_arp_slave)
2731                         found = true;
2732         }
2733
2734         if (!new_slave && before)
2735                 new_slave = before;
2736
2737         if (!new_slave)
2738                 return;
2739
2740         new_slave->link = BOND_LINK_BACK;
2741         bond_set_slave_active_flags(new_slave);
2742         bond_arp_send_all(bond, new_slave);
2743         new_slave->jiffies = jiffies;
2744         bond->current_arp_slave = new_slave;
2745
2746 }
2747
2748 void bond_activebackup_arp_mon(struct work_struct *work)
2749 {
2750         struct bonding *bond = container_of(work, struct bonding,
2751                                             arp_work.work);
2752         bool should_notify_peers = false;
2753         int delta_in_ticks;
2754
2755         read_lock(&bond->lock);
2756
2757         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2758
2759         if (!bond_has_slaves(bond))
2760                 goto re_arm;
2761
2762         should_notify_peers = bond_should_notify_peers(bond);
2763
2764         if (bond_ab_arp_inspect(bond)) {
2765                 read_unlock(&bond->lock);
2766
2767                 /* Race avoidance with bond_close flush of workqueue */
2768                 if (!rtnl_trylock()) {
2769                         read_lock(&bond->lock);
2770                         delta_in_ticks = 1;
2771                         should_notify_peers = false;
2772                         goto re_arm;
2773                 }
2774
2775                 read_lock(&bond->lock);
2776
2777                 bond_ab_arp_commit(bond);
2778
2779                 read_unlock(&bond->lock);
2780                 rtnl_unlock();
2781                 read_lock(&bond->lock);
2782         }
2783
2784         bond_ab_arp_probe(bond);
2785
2786 re_arm:
2787         if (bond->params.arp_interval)
2788                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2789
2790         read_unlock(&bond->lock);
2791
2792         if (should_notify_peers) {
2793                 if (!rtnl_trylock())
2794                         return;
2795                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2796                 rtnl_unlock();
2797         }
2798 }
2799
2800 /*-------------------------- netdev event handling --------------------------*/
2801
2802 /*
2803  * Change device name
2804  */
2805 static int bond_event_changename(struct bonding *bond)
2806 {
2807         bond_remove_proc_entry(bond);
2808         bond_create_proc_entry(bond);
2809
2810         bond_debug_reregister(bond);
2811
2812         return NOTIFY_DONE;
2813 }
2814
2815 static int bond_master_netdev_event(unsigned long event,
2816                                     struct net_device *bond_dev)
2817 {
2818         struct bonding *event_bond = netdev_priv(bond_dev);
2819
2820         switch (event) {
2821         case NETDEV_CHANGENAME:
2822                 return bond_event_changename(event_bond);
2823         case NETDEV_UNREGISTER:
2824                 bond_remove_proc_entry(event_bond);
2825                 break;
2826         case NETDEV_REGISTER:
2827                 bond_create_proc_entry(event_bond);
2828                 break;
2829         case NETDEV_NOTIFY_PEERS:
2830                 if (event_bond->send_peer_notif)
2831                         event_bond->send_peer_notif--;
2832                 break;
2833         default:
2834                 break;
2835         }
2836
2837         return NOTIFY_DONE;
2838 }
2839
2840 static int bond_slave_netdev_event(unsigned long event,
2841                                    struct net_device *slave_dev)
2842 {
2843         struct slave *slave = bond_slave_get_rtnl(slave_dev);
2844         struct bonding *bond;
2845         struct net_device *bond_dev;
2846         u32 old_speed;
2847         u8 old_duplex;
2848
2849         /* A netdev event can be generated while enslaving a device
2850          * before netdev_rx_handler_register is called in which case
2851          * slave will be NULL
2852          */
2853         if (!slave)
2854                 return NOTIFY_DONE;
2855         bond_dev = slave->bond->dev;
2856         bond = slave->bond;
2857
2858         switch (event) {
2859         case NETDEV_UNREGISTER:
2860                 if (bond_dev->type != ARPHRD_ETHER)
2861                         bond_release_and_destroy(bond_dev, slave_dev);
2862                 else
2863                         bond_release(bond_dev, slave_dev);
2864                 break;
2865         case NETDEV_UP:
2866         case NETDEV_CHANGE:
2867                 old_speed = slave->speed;
2868                 old_duplex = slave->duplex;
2869
2870                 bond_update_speed_duplex(slave);
2871
2872                 if (bond->params.mode == BOND_MODE_8023AD) {
2873                         if (old_speed != slave->speed)
2874                                 bond_3ad_adapter_speed_changed(slave);
2875                         if (old_duplex != slave->duplex)
2876                                 bond_3ad_adapter_duplex_changed(slave);
2877                 }
2878                 break;
2879         case NETDEV_DOWN:
2880                 /*
2881                  * ... Or is it this?
2882                  */
2883                 break;
2884         case NETDEV_CHANGEMTU:
2885                 /*
2886                  * TODO: Should slaves be allowed to
2887                  * independently alter their MTU?  For
2888                  * an active-backup bond, slaves need
2889                  * not be the same type of device, so
2890                  * MTUs may vary.  For other modes,
2891                  * slaves arguably should have the
2892                  * same MTUs. To do this, we'd need to
2893                  * take over the slave's change_mtu
2894                  * function for the duration of their
2895                  * servitude.
2896                  */
2897                 break;
2898         case NETDEV_CHANGENAME:
2899                 /*
2900                  * TODO: handle changing the primary's name
2901                  */
2902                 break;
2903         case NETDEV_FEAT_CHANGE:
2904                 bond_compute_features(bond);
2905                 break;
2906         case NETDEV_RESEND_IGMP:
2907                 /* Propagate to master device */
2908                 call_netdevice_notifiers(event, slave->bond->dev);
2909                 break;
2910         default:
2911                 break;
2912         }
2913
2914         return NOTIFY_DONE;
2915 }
2916
2917 /*
2918  * bond_netdev_event: handle netdev notifier chain events.
2919  *
2920  * This function receives events for the netdev chain.  The caller (an
2921  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2922  * locks for us to safely manipulate the slave devices (RTNL lock,
2923  * dev_probe_lock).
2924  */
2925 static int bond_netdev_event(struct notifier_block *this,
2926                              unsigned long event, void *ptr)
2927 {
2928         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2929
2930         pr_debug("event_dev: %s, event: %lx\n",
2931                  event_dev ? event_dev->name : "None",
2932                  event);
2933
2934         if (!(event_dev->priv_flags & IFF_BONDING))
2935                 return NOTIFY_DONE;
2936
2937         if (event_dev->flags & IFF_MASTER) {
2938                 pr_debug("IFF_MASTER\n");
2939                 return bond_master_netdev_event(event, event_dev);
2940         }
2941
2942         if (event_dev->flags & IFF_SLAVE) {
2943                 pr_debug("IFF_SLAVE\n");
2944                 return bond_slave_netdev_event(event, event_dev);
2945         }
2946
2947         return NOTIFY_DONE;
2948 }
2949
2950 static struct notifier_block bond_netdev_notifier = {
2951         .notifier_call = bond_netdev_event,
2952 };
2953
2954 /*---------------------------- Hashing Policies -----------------------------*/
2955
2956 /* L2 hash helper */
2957 static inline u32 bond_eth_hash(struct sk_buff *skb)
2958 {
2959         struct ethhdr *data = (struct ethhdr *)skb->data;
2960
2961         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2962                 return data->h_dest[5] ^ data->h_source[5];
2963
2964         return 0;
2965 }
2966
2967 /* Extract the appropriate headers based on bond's xmit policy */
2968 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2969                               struct flow_keys *fk)
2970 {
2971         const struct ipv6hdr *iph6;
2972         const struct iphdr *iph;
2973         int noff, proto = -1;
2974
2975         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2976                 return skb_flow_dissect(skb, fk);
2977
2978         fk->ports = 0;
2979         noff = skb_network_offset(skb);
2980         if (skb->protocol == htons(ETH_P_IP)) {
2981                 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2982                         return false;
2983                 iph = ip_hdr(skb);
2984                 fk->src = iph->saddr;
2985                 fk->dst = iph->daddr;
2986                 noff += iph->ihl << 2;
2987                 if (!ip_is_fragment(iph))
2988                         proto = iph->protocol;
2989         } else if (skb->protocol == htons(ETH_P_IPV6)) {
2990                 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2991                         return false;
2992                 iph6 = ipv6_hdr(skb);
2993                 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2994                 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2995                 noff += sizeof(*iph6);
2996                 proto = iph6->nexthdr;
2997         } else {
2998                 return false;
2999         }
3000         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3001                 fk->ports = skb_flow_get_ports(skb, noff, proto);
3002
3003         return true;
3004 }
3005
3006 /**
3007  * bond_xmit_hash - generate a hash value based on the xmit policy
3008  * @bond: bonding device
3009  * @skb: buffer to use for headers
3010  * @count: modulo value
3011  *
3012  * This function will extract the necessary headers from the skb buffer and use
3013  * them to generate a hash based on the xmit_policy set in the bonding device
3014  * which will be reduced modulo count before returning.
3015  */
3016 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
3017 {
3018         struct flow_keys flow;
3019         u32 hash;
3020
3021         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3022             !bond_flow_dissect(bond, skb, &flow))
3023                 return bond_eth_hash(skb) % count;
3024
3025         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3026             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3027                 hash = bond_eth_hash(skb);
3028         else
3029                 hash = (__force u32)flow.ports;
3030         hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3031         hash ^= (hash >> 16);
3032         hash ^= (hash >> 8);
3033
3034         return hash % count;
3035 }
3036
3037 /*-------------------------- Device entry points ----------------------------*/
3038
3039 static void bond_work_init_all(struct bonding *bond)
3040 {
3041         INIT_DELAYED_WORK(&bond->mcast_work,
3042                           bond_resend_igmp_join_requests_delayed);
3043         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3044         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3045         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3046                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3047         else
3048                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3049         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3050 }
3051
3052 static void bond_work_cancel_all(struct bonding *bond)
3053 {
3054         cancel_delayed_work_sync(&bond->mii_work);
3055         cancel_delayed_work_sync(&bond->arp_work);
3056         cancel_delayed_work_sync(&bond->alb_work);
3057         cancel_delayed_work_sync(&bond->ad_work);
3058         cancel_delayed_work_sync(&bond->mcast_work);
3059 }
3060
3061 static int bond_open(struct net_device *bond_dev)
3062 {
3063         struct bonding *bond = netdev_priv(bond_dev);
3064         struct list_head *iter;
3065         struct slave *slave;
3066
3067         /* reset slave->backup and slave->inactive */
3068         read_lock(&bond->lock);
3069         if (bond_has_slaves(bond)) {
3070                 read_lock(&bond->curr_slave_lock);
3071                 bond_for_each_slave(bond, slave, iter) {
3072                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3073                                 && (slave != bond->curr_active_slave)) {
3074                                 bond_set_slave_inactive_flags(slave);
3075                         } else {
3076                                 bond_set_slave_active_flags(slave);
3077                         }
3078                 }
3079                 read_unlock(&bond->curr_slave_lock);
3080         }
3081         read_unlock(&bond->lock);
3082
3083         bond_work_init_all(bond);
3084
3085         if (bond_is_lb(bond)) {
3086                 /* bond_alb_initialize must be called before the timer
3087                  * is started.
3088                  */
3089                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3090                         return -ENOMEM;
3091                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3092         }
3093
3094         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3095                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3096
3097         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3098                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3099                 if (bond->params.arp_validate)
3100                         bond->recv_probe = bond_arp_rcv;
3101         }
3102
3103         if (bond->params.mode == BOND_MODE_8023AD) {
3104                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3105                 /* register to receive LACPDUs */
3106                 bond->recv_probe = bond_3ad_lacpdu_recv;
3107                 bond_3ad_initiate_agg_selection(bond, 1);
3108         }
3109
3110         return 0;
3111 }
3112
3113 static int bond_close(struct net_device *bond_dev)
3114 {
3115         struct bonding *bond = netdev_priv(bond_dev);
3116
3117         bond_work_cancel_all(bond);
3118         bond->send_peer_notif = 0;
3119         if (bond_is_lb(bond))
3120                 bond_alb_deinitialize(bond);
3121         bond->recv_probe = NULL;
3122
3123         return 0;
3124 }
3125
3126 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3127                                                 struct rtnl_link_stats64 *stats)
3128 {
3129         struct bonding *bond = netdev_priv(bond_dev);
3130         struct rtnl_link_stats64 temp;
3131         struct list_head *iter;
3132         struct slave *slave;
3133
3134         memset(stats, 0, sizeof(*stats));
3135
3136         read_lock_bh(&bond->lock);
3137         bond_for_each_slave(bond, slave, iter) {
3138                 const struct rtnl_link_stats64 *sstats =
3139                         dev_get_stats(slave->dev, &temp);
3140
3141                 stats->rx_packets += sstats->rx_packets;
3142                 stats->rx_bytes += sstats->rx_bytes;
3143                 stats->rx_errors += sstats->rx_errors;
3144                 stats->rx_dropped += sstats->rx_dropped;
3145
3146                 stats->tx_packets += sstats->tx_packets;
3147                 stats->tx_bytes += sstats->tx_bytes;
3148                 stats->tx_errors += sstats->tx_errors;
3149                 stats->tx_dropped += sstats->tx_dropped;
3150
3151                 stats->multicast += sstats->multicast;
3152                 stats->collisions += sstats->collisions;
3153
3154                 stats->rx_length_errors += sstats->rx_length_errors;
3155                 stats->rx_over_errors += sstats->rx_over_errors;
3156                 stats->rx_crc_errors += sstats->rx_crc_errors;
3157                 stats->rx_frame_errors += sstats->rx_frame_errors;
3158                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3159                 stats->rx_missed_errors += sstats->rx_missed_errors;
3160
3161                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3162                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3163                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3164                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3165                 stats->tx_window_errors += sstats->tx_window_errors;
3166         }
3167         read_unlock_bh(&bond->lock);
3168
3169         return stats;
3170 }
3171
3172 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3173 {
3174         struct bonding *bond = netdev_priv(bond_dev);
3175         struct net_device *slave_dev = NULL;
3176         struct ifbond k_binfo;
3177         struct ifbond __user *u_binfo = NULL;
3178         struct ifslave k_sinfo;
3179         struct ifslave __user *u_sinfo = NULL;
3180         struct mii_ioctl_data *mii = NULL;
3181         struct net *net;
3182         int res = 0;
3183
3184         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3185
3186         switch (cmd) {
3187         case SIOCGMIIPHY:
3188                 mii = if_mii(ifr);
3189                 if (!mii)
3190                         return -EINVAL;
3191
3192                 mii->phy_id = 0;
3193                 /* Fall Through */
3194         case SIOCGMIIREG:
3195                 /*
3196                  * We do this again just in case we were called by SIOCGMIIREG
3197                  * instead of SIOCGMIIPHY.
3198                  */
3199                 mii = if_mii(ifr);
3200                 if (!mii)
3201                         return -EINVAL;
3202
3203
3204                 if (mii->reg_num == 1) {
3205                         mii->val_out = 0;
3206                         read_lock(&bond->lock);
3207                         read_lock(&bond->curr_slave_lock);
3208                         if (netif_carrier_ok(bond->dev))
3209                                 mii->val_out = BMSR_LSTATUS;
3210
3211                         read_unlock(&bond->curr_slave_lock);
3212                         read_unlock(&bond->lock);
3213                 }
3214
3215                 return 0;
3216         case BOND_INFO_QUERY_OLD:
3217         case SIOCBONDINFOQUERY:
3218                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3219
3220                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3221                         return -EFAULT;
3222
3223                 res = bond_info_query(bond_dev, &k_binfo);
3224                 if (res == 0 &&
3225                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3226                         return -EFAULT;
3227
3228                 return res;
3229         case BOND_SLAVE_INFO_QUERY_OLD:
3230         case SIOCBONDSLAVEINFOQUERY:
3231                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3232
3233                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3234                         return -EFAULT;
3235
3236                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3237                 if (res == 0 &&
3238                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3239                         return -EFAULT;
3240
3241                 return res;
3242         default:
3243                 /* Go on */
3244                 break;
3245         }
3246
3247         net = dev_net(bond_dev);
3248
3249         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3250                 return -EPERM;
3251
3252         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3253
3254         pr_debug("slave_dev=%p:\n", slave_dev);
3255
3256         if (!slave_dev)
3257                 res = -ENODEV;
3258         else {
3259                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3260                 switch (cmd) {
3261                 case BOND_ENSLAVE_OLD:
3262                 case SIOCBONDENSLAVE:
3263                         res = bond_enslave(bond_dev, slave_dev);
3264                         break;
3265                 case BOND_RELEASE_OLD:
3266                 case SIOCBONDRELEASE:
3267                         res = bond_release(bond_dev, slave_dev);
3268                         break;
3269                 case BOND_SETHWADDR_OLD:
3270                 case SIOCBONDSETHWADDR:
3271                         bond_set_dev_addr(bond_dev, slave_dev);
3272                         res = 0;
3273                         break;
3274                 case BOND_CHANGE_ACTIVE_OLD:
3275                 case SIOCBONDCHANGEACTIVE:
3276                         res = bond_option_active_slave_set(bond, slave_dev);
3277                         break;
3278                 default:
3279                         res = -EOPNOTSUPP;
3280                 }
3281
3282                 dev_put(slave_dev);
3283         }
3284
3285         return res;
3286 }
3287
3288 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3289 {
3290         struct bonding *bond = netdev_priv(bond_dev);
3291
3292         if (change & IFF_PROMISC)
3293                 bond_set_promiscuity(bond,
3294                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3295
3296         if (change & IFF_ALLMULTI)
3297                 bond_set_allmulti(bond,
3298                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3299 }
3300
3301 static void bond_set_rx_mode(struct net_device *bond_dev)
3302 {
3303         struct bonding *bond = netdev_priv(bond_dev);
3304         struct list_head *iter;
3305         struct slave *slave;
3306
3307
3308         rcu_read_lock();
3309         if (USES_PRIMARY(bond->params.mode)) {
3310                 slave = rcu_dereference(bond->curr_active_slave);
3311                 if (slave) {
3312                         dev_uc_sync(slave->dev, bond_dev);
3313                         dev_mc_sync(slave->dev, bond_dev);
3314                 }
3315         } else {
3316                 bond_for_each_slave_rcu(bond, slave, iter) {
3317                         dev_uc_sync_multiple(slave->dev, bond_dev);
3318                         dev_mc_sync_multiple(slave->dev, bond_dev);
3319                 }
3320         }
3321         rcu_read_unlock();
3322 }
3323
3324 static int bond_neigh_init(struct neighbour *n)
3325 {
3326         struct bonding *bond = netdev_priv(n->dev);
3327         const struct net_device_ops *slave_ops;
3328         struct neigh_parms parms;
3329         struct slave *slave;
3330         int ret;
3331
3332         slave = bond_first_slave(bond);
3333         if (!slave)
3334                 return 0;
3335         slave_ops = slave->dev->netdev_ops;
3336         if (!slave_ops->ndo_neigh_setup)
3337                 return 0;
3338
3339         parms.neigh_setup = NULL;
3340         parms.neigh_cleanup = NULL;
3341         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3342         if (ret)
3343                 return ret;
3344
3345         /*
3346          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3347          * after the last slave has been detached.  Assumes that all slaves
3348          * utilize the same neigh_cleanup (true at this writing as only user
3349          * is ipoib).
3350          */
3351         n->parms->neigh_cleanup = parms.neigh_cleanup;
3352
3353         if (!parms.neigh_setup)
3354                 return 0;
3355
3356         return parms.neigh_setup(n);
3357 }
3358
3359 /*
3360  * The bonding ndo_neigh_setup is called at init time beofre any
3361  * slave exists. So we must declare proxy setup function which will
3362  * be used at run time to resolve the actual slave neigh param setup.
3363  *
3364  * It's also called by master devices (such as vlans) to setup their
3365  * underlying devices. In that case - do nothing, we're already set up from
3366  * our init.
3367  */
3368 static int bond_neigh_setup(struct net_device *dev,
3369                             struct neigh_parms *parms)
3370 {
3371         /* modify only our neigh_parms */
3372         if (parms->dev == dev)
3373                 parms->neigh_setup = bond_neigh_init;
3374
3375         return 0;
3376 }
3377
3378 /*
3379  * Change the MTU of all of a master's slaves to match the master
3380  */
3381 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3382 {
3383         struct bonding *bond = netdev_priv(bond_dev);
3384         struct slave *slave, *rollback_slave;
3385         struct list_head *iter;
3386         int res = 0;
3387
3388         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3389                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3390
3391         /* Can't hold bond->lock with bh disabled here since
3392          * some base drivers panic. On the other hand we can't
3393          * hold bond->lock without bh disabled because we'll
3394          * deadlock. The only solution is to rely on the fact
3395          * that we're under rtnl_lock here, and the slaves
3396          * list won't change. This doesn't solve the problem
3397          * of setting the slave's MTU while it is
3398          * transmitting, but the assumption is that the base
3399          * driver can handle that.
3400          *
3401          * TODO: figure out a way to safely iterate the slaves
3402          * list, but without holding a lock around the actual
3403          * call to the base driver.
3404          */
3405
3406         bond_for_each_slave(bond, slave, iter) {
3407                 pr_debug("s %p c_m %p\n",
3408                          slave,
3409                          slave->dev->netdev_ops->ndo_change_mtu);
3410
3411                 res = dev_set_mtu(slave->dev, new_mtu);
3412
3413                 if (res) {
3414                         /* If we failed to set the slave's mtu to the new value
3415                          * we must abort the operation even in ACTIVE_BACKUP
3416                          * mode, because if we allow the backup slaves to have
3417                          * different mtu values than the active slave we'll
3418                          * need to change their mtu when doing a failover. That
3419                          * means changing their mtu from timer context, which
3420                          * is probably not a good idea.
3421                          */
3422                         pr_debug("err %d %s\n", res, slave->dev->name);
3423                         goto unwind;
3424                 }
3425         }
3426
3427         bond_dev->mtu = new_mtu;
3428
3429         return 0;
3430
3431 unwind:
3432         /* unwind from head to the slave that failed */
3433         bond_for_each_slave(bond, rollback_slave, iter) {
3434                 int tmp_res;
3435
3436                 if (rollback_slave == slave)
3437                         break;
3438
3439                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3440                 if (tmp_res) {
3441                         pr_debug("unwind err %d dev %s\n",
3442                                  tmp_res, rollback_slave->dev->name);
3443                 }
3444         }
3445
3446         return res;
3447 }
3448
3449 /*
3450  * Change HW address
3451  *
3452  * Note that many devices must be down to change the HW address, and
3453  * downing the master releases all slaves.  We can make bonds full of
3454  * bonding devices to test this, however.
3455  */
3456 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3457 {
3458         struct bonding *bond = netdev_priv(bond_dev);
3459         struct slave *slave, *rollback_slave;
3460         struct sockaddr *sa = addr, tmp_sa;
3461         struct list_head *iter;
3462         int res = 0;
3463
3464         if (bond->params.mode == BOND_MODE_ALB)
3465                 return bond_alb_set_mac_address(bond_dev, addr);
3466
3467
3468         pr_debug("bond=%p, name=%s\n",
3469                  bond, bond_dev ? bond_dev->name : "None");
3470
3471         /* If fail_over_mac is enabled, do nothing and return success.
3472          * Returning an error causes ifenslave to fail.
3473          */
3474         if (bond->params.fail_over_mac)
3475                 return 0;
3476
3477         if (!is_valid_ether_addr(sa->sa_data))
3478                 return -EADDRNOTAVAIL;
3479
3480         /* Can't hold bond->lock with bh disabled here since
3481          * some base drivers panic. On the other hand we can't
3482          * hold bond->lock without bh disabled because we'll
3483          * deadlock. The only solution is to rely on the fact
3484          * that we're under rtnl_lock here, and the slaves
3485          * list won't change. This doesn't solve the problem
3486          * of setting the slave's hw address while it is
3487          * transmitting, but the assumption is that the base
3488          * driver can handle that.
3489          *
3490          * TODO: figure out a way to safely iterate the slaves
3491          * list, but without holding a lock around the actual
3492          * call to the base driver.
3493          */
3494
3495         bond_for_each_slave(bond, slave, iter) {
3496                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3497                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3498
3499                 if (slave_ops->ndo_set_mac_address == NULL) {
3500                         res = -EOPNOTSUPP;
3501                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3502                         goto unwind;
3503                 }
3504
3505                 res = dev_set_mac_address(slave->dev, addr);
3506                 if (res) {
3507                         /* TODO: consider downing the slave
3508                          * and retry ?
3509                          * User should expect communications
3510                          * breakage anyway until ARP finish
3511                          * updating, so...
3512                          */
3513                         pr_debug("err %d %s\n", res, slave->dev->name);
3514                         goto unwind;
3515                 }
3516         }
3517
3518         /* success */
3519         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3520         return 0;
3521
3522 unwind:
3523         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3524         tmp_sa.sa_family = bond_dev->type;
3525
3526         /* unwind from head to the slave that failed */
3527         bond_for_each_slave(bond, rollback_slave, iter) {
3528                 int tmp_res;
3529
3530                 if (rollback_slave == slave)
3531                         break;
3532
3533                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3534                 if (tmp_res) {
3535                         pr_debug("unwind err %d dev %s\n",
3536                                  tmp_res, rollback_slave->dev->name);
3537                 }
3538         }
3539
3540         return res;
3541 }
3542
3543 /**
3544  * bond_xmit_slave_id - transmit skb through slave with slave_id
3545  * @bond: bonding device that is transmitting
3546  * @skb: buffer to transmit
3547  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3548  *
3549  * This function tries to transmit through slave with slave_id but in case
3550  * it fails, it tries to find the first available slave for transmission.
3551  * The skb is consumed in all cases, thus the function is void.
3552  */
3553 void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3554 {
3555         struct list_head *iter;
3556         struct slave *slave;
3557         int i = slave_id;
3558
3559         /* Here we start from the slave with slave_id */
3560         bond_for_each_slave_rcu(bond, slave, iter) {
3561                 if (--i < 0) {
3562                         if (slave_can_tx(slave)) {
3563                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3564                                 return;
3565                         }
3566                 }
3567         }
3568
3569         /* Here we start from the first slave up to slave_id */
3570         i = slave_id;
3571         bond_for_each_slave_rcu(bond, slave, iter) {
3572                 if (--i < 0)
3573                         break;
3574                 if (slave_can_tx(slave)) {
3575                         bond_dev_queue_xmit(bond, skb, slave->dev);
3576                         return;
3577                 }
3578         }
3579         /* no slave that can tx has been found */
3580         kfree_skb(skb);
3581 }
3582
3583 /**
3584  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3585  * @bond: bonding device to use
3586  *
3587  * Based on the value of the bonding device's packets_per_slave parameter
3588  * this function generates a slave id, which is usually used as the next
3589  * slave to transmit through.
3590  */
3591 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3592 {
3593         int packets_per_slave = bond->params.packets_per_slave;
3594         u32 slave_id;
3595
3596         switch (packets_per_slave) {
3597         case 0:
3598                 slave_id = prandom_u32();
3599                 break;
3600         case 1:
3601                 slave_id = bond->rr_tx_counter;
3602                 break;
3603         default:
3604                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3605                                              packets_per_slave);
3606                 break;
3607         }
3608         bond->rr_tx_counter++;
3609
3610         return slave_id;
3611 }
3612
3613 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3614 {
3615         struct bonding *bond = netdev_priv(bond_dev);
3616         struct iphdr *iph = ip_hdr(skb);
3617         struct slave *slave;
3618         u32 slave_id;
3619
3620         /* Start with the curr_active_slave that joined the bond as the
3621          * default for sending IGMP traffic.  For failover purposes one
3622          * needs to maintain some consistency for the interface that will
3623          * send the join/membership reports.  The curr_active_slave found
3624          * will send all of this type of traffic.
3625          */
3626         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3627                 slave = rcu_dereference(bond->curr_active_slave);
3628                 if (slave && slave_can_tx(slave))
3629                         bond_dev_queue_xmit(bond, skb, slave->dev);
3630                 else
3631                         bond_xmit_slave_id(bond, skb, 0);
3632         } else {
3633                 slave_id = bond_rr_gen_slave_id(bond);
3634                 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3635         }
3636
3637         return NETDEV_TX_OK;
3638 }
3639
3640 /*
3641  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3642  * the bond has a usable interface.
3643  */
3644 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3645 {
3646         struct bonding *bond = netdev_priv(bond_dev);
3647         struct slave *slave;
3648
3649         slave = rcu_dereference(bond->curr_active_slave);
3650         if (slave)
3651                 bond_dev_queue_xmit(bond, skb, slave->dev);
3652         else
3653                 kfree_skb(skb);
3654
3655         return NETDEV_TX_OK;
3656 }
3657
3658 /* In bond_xmit_xor() , we determine the output device by using a pre-
3659  * determined xmit_hash_policy(), If the selected device is not enabled,
3660  * find the next active slave.
3661  */
3662 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3663 {
3664         struct bonding *bond = netdev_priv(bond_dev);
3665
3666         bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3667
3668         return NETDEV_TX_OK;
3669 }
3670
3671 /* in broadcast mode, we send everything to all usable interfaces. */
3672 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3673 {
3674         struct bonding *bond = netdev_priv(bond_dev);
3675         struct slave *slave = NULL;
3676         struct list_head *iter;
3677
3678         bond_for_each_slave_rcu(bond, slave, iter) {
3679                 if (bond_is_last_slave(bond, slave))
3680                         break;
3681                 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3682                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3683
3684                         if (!skb2) {
3685                                 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3686                                        bond_dev->name);
3687                                 continue;
3688                         }
3689                         /* bond_dev_queue_xmit always returns 0 */
3690                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3691                 }
3692         }
3693         if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3694                 bond_dev_queue_xmit(bond, skb, slave->dev);
3695         else
3696                 kfree_skb(skb);
3697
3698         return NETDEV_TX_OK;
3699 }
3700
3701 /*------------------------- Device initialization ---------------------------*/
3702
3703 /*
3704  * Lookup the slave that corresponds to a qid
3705  */
3706 static inline int bond_slave_override(struct bonding *bond,
3707                                       struct sk_buff *skb)
3708 {
3709         struct slave *slave = NULL;
3710         struct slave *check_slave;
3711         struct list_head *iter;
3712         int res = 1;
3713
3714         if (!skb->queue_mapping)
3715                 return 1;
3716
3717         /* Find out if any slaves have the same mapping as this skb. */
3718         bond_for_each_slave_rcu(bond, check_slave, iter) {
3719                 if (check_slave->queue_id == skb->queue_mapping) {
3720                         slave = check_slave;
3721                         break;
3722                 }
3723         }
3724
3725         /* If the slave isn't UP, use default transmit policy. */
3726         if (slave && slave->queue_id && IS_UP(slave->dev) &&
3727             (slave->link == BOND_LINK_UP)) {
3728                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3729         }
3730
3731         return res;
3732 }
3733
3734
3735 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3736 {
3737         /*
3738          * This helper function exists to help dev_pick_tx get the correct
3739          * destination queue.  Using a helper function skips a call to
3740          * skb_tx_hash and will put the skbs in the queue we expect on their
3741          * way down to the bonding driver.
3742          */
3743         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3744
3745         /*
3746          * Save the original txq to restore before passing to the driver
3747          */
3748         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3749
3750         if (unlikely(txq >= dev->real_num_tx_queues)) {
3751                 do {
3752                         txq -= dev->real_num_tx_queues;
3753                 } while (txq >= dev->real_num_tx_queues);
3754         }
3755         return txq;
3756 }
3757
3758 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3759 {
3760         struct bonding *bond = netdev_priv(dev);
3761
3762         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3763                 if (!bond_slave_override(bond, skb))
3764                         return NETDEV_TX_OK;
3765         }
3766
3767         switch (bond->params.mode) {
3768         case BOND_MODE_ROUNDROBIN:
3769                 return bond_xmit_roundrobin(skb, dev);
3770         case BOND_MODE_ACTIVEBACKUP:
3771                 return bond_xmit_activebackup(skb, dev);
3772         case BOND_MODE_XOR:
3773                 return bond_xmit_xor(skb, dev);
3774         case BOND_MODE_BROADCAST:
3775                 return bond_xmit_broadcast(skb, dev);
3776         case BOND_MODE_8023AD:
3777                 return bond_3ad_xmit_xor(skb, dev);
3778         case BOND_MODE_ALB:
3779         case BOND_MODE_TLB:
3780                 return bond_alb_xmit(skb, dev);
3781         default:
3782                 /* Should never happen, mode already checked */
3783                 pr_err("%s: Error: Unknown bonding mode %d\n",
3784                        dev->name, bond->params.mode);
3785                 WARN_ON_ONCE(1);
3786                 kfree_skb(skb);
3787                 return NETDEV_TX_OK;
3788         }
3789 }
3790
3791 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3792 {
3793         struct bonding *bond = netdev_priv(dev);
3794         netdev_tx_t ret = NETDEV_TX_OK;
3795
3796         /*
3797          * If we risk deadlock from transmitting this in the
3798          * netpoll path, tell netpoll to queue the frame for later tx
3799          */
3800         if (is_netpoll_tx_blocked(dev))
3801                 return NETDEV_TX_BUSY;
3802
3803         rcu_read_lock();
3804         if (bond_has_slaves(bond))
3805                 ret = __bond_start_xmit(skb, dev);
3806         else
3807                 kfree_skb(skb);
3808         rcu_read_unlock();
3809
3810         return ret;
3811 }
3812
3813 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3814                                      struct ethtool_cmd *ecmd)
3815 {
3816         struct bonding *bond = netdev_priv(bond_dev);
3817         unsigned long speed = 0;
3818         struct list_head *iter;
3819         struct slave *slave;
3820
3821         ecmd->duplex = DUPLEX_UNKNOWN;
3822         ecmd->port = PORT_OTHER;
3823
3824         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3825          * do not need to check mode.  Though link speed might not represent
3826          * the true receive or transmit bandwidth (not all modes are symmetric)
3827          * this is an accurate maximum.
3828          */
3829         read_lock(&bond->lock);
3830         bond_for_each_slave(bond, slave, iter) {
3831                 if (SLAVE_IS_OK(slave)) {
3832                         if (slave->speed != SPEED_UNKNOWN)
3833                                 speed += slave->speed;
3834                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
3835                             slave->duplex != DUPLEX_UNKNOWN)
3836                                 ecmd->duplex = slave->duplex;
3837                 }
3838         }
3839         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3840         read_unlock(&bond->lock);
3841
3842         return 0;
3843 }
3844
3845 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3846                                      struct ethtool_drvinfo *drvinfo)
3847 {
3848         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3849         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3850         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3851                  BOND_ABI_VERSION);
3852 }
3853
3854 static const struct ethtool_ops bond_ethtool_ops = {
3855         .get_drvinfo            = bond_ethtool_get_drvinfo,
3856         .get_settings           = bond_ethtool_get_settings,
3857         .get_link               = ethtool_op_get_link,
3858 };
3859
3860 static const struct net_device_ops bond_netdev_ops = {
3861         .ndo_init               = bond_init,
3862         .ndo_uninit             = bond_uninit,
3863         .ndo_open               = bond_open,
3864         .ndo_stop               = bond_close,
3865         .ndo_start_xmit         = bond_start_xmit,
3866         .ndo_select_queue       = bond_select_queue,
3867         .ndo_get_stats64        = bond_get_stats,
3868         .ndo_do_ioctl           = bond_do_ioctl,
3869         .ndo_change_rx_flags    = bond_change_rx_flags,
3870         .ndo_set_rx_mode        = bond_set_rx_mode,
3871         .ndo_change_mtu         = bond_change_mtu,
3872         .ndo_set_mac_address    = bond_set_mac_address,
3873         .ndo_neigh_setup        = bond_neigh_setup,
3874         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
3875         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
3876 #ifdef CONFIG_NET_POLL_CONTROLLER
3877         .ndo_netpoll_setup      = bond_netpoll_setup,
3878         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
3879         .ndo_poll_controller    = bond_poll_controller,
3880 #endif
3881         .ndo_add_slave          = bond_enslave,
3882         .ndo_del_slave          = bond_release,
3883         .ndo_fix_features       = bond_fix_features,
3884 };
3885
3886 static const struct device_type bond_type = {
3887         .name = "bond",
3888 };
3889
3890 static void bond_destructor(struct net_device *bond_dev)
3891 {
3892         struct bonding *bond = netdev_priv(bond_dev);
3893         if (bond->wq)
3894                 destroy_workqueue(bond->wq);
3895         free_netdev(bond_dev);
3896 }
3897
3898 void bond_setup(struct net_device *bond_dev)
3899 {
3900         struct bonding *bond = netdev_priv(bond_dev);
3901
3902         /* initialize rwlocks */
3903         rwlock_init(&bond->lock);
3904         rwlock_init(&bond->curr_slave_lock);
3905         bond->params = bonding_defaults;
3906
3907         /* Initialize pointers */
3908         bond->dev = bond_dev;
3909
3910         /* Initialize the device entry points */
3911         ether_setup(bond_dev);
3912         bond_dev->netdev_ops = &bond_netdev_ops;
3913         bond_dev->ethtool_ops = &bond_ethtool_ops;
3914
3915         bond_dev->destructor = bond_destructor;
3916
3917         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3918
3919         /* Initialize the device options */
3920         bond_dev->tx_queue_len = 0;
3921         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3922         bond_dev->priv_flags |= IFF_BONDING;
3923         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3924
3925         /* At first, we block adding VLANs. That's the only way to
3926          * prevent problems that occur when adding VLANs over an
3927          * empty bond. The block will be removed once non-challenged
3928          * slaves are enslaved.
3929          */
3930         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3931
3932         /* don't acquire bond device's netif_tx_lock when
3933          * transmitting */
3934         bond_dev->features |= NETIF_F_LLTX;
3935
3936         /* By default, we declare the bond to be fully
3937          * VLAN hardware accelerated capable. Special
3938          * care is taken in the various xmit functions
3939          * when there are slaves that are not hw accel
3940          * capable
3941          */
3942
3943         bond_dev->hw_features = BOND_VLAN_FEATURES |
3944                                 NETIF_F_HW_VLAN_CTAG_TX |
3945                                 NETIF_F_HW_VLAN_CTAG_RX |
3946                                 NETIF_F_HW_VLAN_CTAG_FILTER;
3947
3948         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3949         bond_dev->features |= bond_dev->hw_features;
3950 }
3951
3952 /*
3953 * Destroy a bonding device.
3954 * Must be under rtnl_lock when this function is called.
3955 */
3956 static void bond_uninit(struct net_device *bond_dev)
3957 {
3958         struct bonding *bond = netdev_priv(bond_dev);
3959         struct list_head *iter;
3960         struct slave *slave;
3961
3962         bond_netpoll_cleanup(bond_dev);
3963
3964         /* Release the bonded slaves */
3965         bond_for_each_slave(bond, slave, iter)
3966                 __bond_release_one(bond_dev, slave->dev, true);
3967         pr_info("%s: released all slaves\n", bond_dev->name);
3968
3969         list_del(&bond->bond_list);
3970
3971         bond_debug_unregister(bond);
3972 }
3973
3974 /*------------------------- Module initialization ---------------------------*/
3975
3976 /*
3977  * Convert string input module parms.  Accept either the
3978  * number of the mode or its string name.  A bit complicated because
3979  * some mode names are substrings of other names, and calls from sysfs
3980  * may have whitespace in the name (trailing newlines, for example).
3981  */
3982 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
3983 {
3984         int modeint = -1, i, rv;
3985         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
3986
3987         for (p = (char *)buf; *p; p++)
3988                 if (!(isdigit(*p) || isspace(*p)))
3989                         break;
3990
3991         if (*p)
3992                 rv = sscanf(buf, "%20s", modestr);
3993         else
3994                 rv = sscanf(buf, "%d", &modeint);
3995
3996         if (!rv)
3997                 return -1;
3998
3999         for (i = 0; tbl[i].modename; i++) {
4000                 if (modeint == tbl[i].mode)
4001                         return tbl[i].mode;
4002                 if (strcmp(modestr, tbl[i].modename) == 0)
4003                         return tbl[i].mode;
4004         }
4005
4006         return -1;
4007 }
4008
4009 static int bond_check_params(struct bond_params *params)
4010 {
4011         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4012         int arp_all_targets_value;
4013
4014         /*
4015          * Convert string parameters.
4016          */
4017         if (mode) {
4018                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4019                 if (bond_mode == -1) {
4020                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4021                                mode == NULL ? "NULL" : mode);
4022                         return -EINVAL;
4023                 }
4024         }
4025
4026         if (xmit_hash_policy) {
4027                 if ((bond_mode != BOND_MODE_XOR) &&
4028                     (bond_mode != BOND_MODE_8023AD)) {
4029                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4030                                bond_mode_name(bond_mode));
4031                 } else {
4032                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4033                                                         xmit_hashtype_tbl);
4034                         if (xmit_hashtype == -1) {
4035                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4036                                        xmit_hash_policy == NULL ? "NULL" :
4037                                        xmit_hash_policy);
4038                                 return -EINVAL;
4039                         }
4040                 }
4041         }
4042
4043         if (lacp_rate) {
4044                 if (bond_mode != BOND_MODE_8023AD) {
4045                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4046                                 bond_mode_name(bond_mode));
4047                 } else {
4048                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4049                         if (lacp_fast == -1) {
4050                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4051                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4052                                 return -EINVAL;
4053                         }
4054                 }
4055         }
4056
4057         if (ad_select) {
4058                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4059                 if (params->ad_select == -1) {
4060                         pr_err("Error: Invalid ad_select \"%s\"\n",
4061                                ad_select == NULL ? "NULL" : ad_select);
4062                         return -EINVAL;
4063                 }
4064
4065                 if (bond_mode != BOND_MODE_8023AD) {
4066                         pr_warning("ad_select param only affects 802.3ad mode\n");
4067                 }
4068         } else {
4069                 params->ad_select = BOND_AD_STABLE;
4070         }
4071
4072         if (max_bonds < 0) {
4073                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4074                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4075                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4076         }
4077
4078         if (miimon < 0) {
4079                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4080                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4081                 miimon = BOND_LINK_MON_INTERV;
4082         }
4083
4084         if (updelay < 0) {
4085                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4086                            updelay, INT_MAX);
4087                 updelay = 0;
4088         }
4089
4090         if (downdelay < 0) {
4091                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4092                            downdelay, INT_MAX);
4093                 downdelay = 0;
4094         }
4095
4096         if ((use_carrier != 0) && (use_carrier != 1)) {
4097                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4098                            use_carrier);
4099                 use_carrier = 1;
4100         }
4101
4102         if (num_peer_notif < 0 || num_peer_notif > 255) {
4103                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4104                            num_peer_notif);
4105                 num_peer_notif = 1;
4106         }
4107
4108         /* reset values for 802.3ad */
4109         if (bond_mode == BOND_MODE_8023AD) {
4110                 if (!miimon) {
4111                         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");
4112                         pr_warning("Forcing miimon to 100msec\n");
4113                         miimon = BOND_DEFAULT_MIIMON;
4114                 }
4115         }
4116
4117         if (tx_queues < 1 || tx_queues > 255) {
4118                 pr_warning("Warning: tx_queues (%d) should be between "
4119                            "1 and 255, resetting to %d\n",
4120                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4121                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4122         }
4123
4124         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4125                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4126                            "not of valid value (0/1), so it was set to "
4127                            "0\n", all_slaves_active);
4128                 all_slaves_active = 0;
4129         }
4130
4131         if (resend_igmp < 0 || resend_igmp > 255) {
4132                 pr_warning("Warning: resend_igmp (%d) should be between "
4133                            "0 and 255, resetting to %d\n",
4134                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4135                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4136         }
4137
4138         if (packets_per_slave < 0 || packets_per_slave > USHRT_MAX) {
4139                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4140                         packets_per_slave, USHRT_MAX);
4141                 packets_per_slave = 1;
4142         }
4143
4144         /* reset values for TLB/ALB */
4145         if ((bond_mode == BOND_MODE_TLB) ||
4146             (bond_mode == BOND_MODE_ALB)) {
4147                 if (!miimon) {
4148                         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");
4149                         pr_warning("Forcing miimon to 100msec\n");
4150                         miimon = BOND_DEFAULT_MIIMON;
4151                 }
4152         }
4153
4154         if (bond_mode == BOND_MODE_ALB) {
4155                 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",
4156                           updelay);
4157         }
4158
4159         if (!miimon) {
4160                 if (updelay || downdelay) {
4161                         /* just warn the user the up/down delay will have
4162                          * no effect since miimon is zero...
4163                          */
4164                         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",
4165                                    updelay, downdelay);
4166                 }
4167         } else {
4168                 /* don't allow arp monitoring */
4169                 if (arp_interval) {
4170                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4171                                    miimon, arp_interval);
4172                         arp_interval = 0;
4173                 }
4174
4175                 if ((updelay % miimon) != 0) {
4176                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4177                                    updelay, miimon,
4178                                    (updelay / miimon) * miimon);
4179                 }
4180
4181                 updelay /= miimon;
4182
4183                 if ((downdelay % miimon) != 0) {
4184                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4185                                    downdelay, miimon,
4186                                    (downdelay / miimon) * miimon);
4187                 }
4188
4189                 downdelay /= miimon;
4190         }
4191
4192         if (arp_interval < 0) {
4193                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4194                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4195                 arp_interval = BOND_LINK_ARP_INTERV;
4196         }
4197
4198         for (arp_ip_count = 0, i = 0;
4199              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4200                 /* not complete check, but should be good enough to
4201                    catch mistakes */
4202                 __be32 ip;
4203                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4204                     IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4205                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4206                                    arp_ip_target[i]);
4207                         arp_interval = 0;
4208                 } else {
4209                         if (bond_get_targets_ip(arp_target, ip) == -1)
4210                                 arp_target[arp_ip_count++] = ip;
4211                         else
4212                                 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4213                                            &ip);
4214                 }
4215         }
4216
4217         if (arp_interval && !arp_ip_count) {
4218                 /* don't allow arping if no arp_ip_target given... */
4219                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4220                            arp_interval);
4221                 arp_interval = 0;
4222         }
4223
4224         if (arp_validate) {
4225                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4226                         pr_err("arp_validate only supported in active-backup mode\n");
4227                         return -EINVAL;
4228                 }
4229                 if (!arp_interval) {
4230                         pr_err("arp_validate requires arp_interval\n");
4231                         return -EINVAL;
4232                 }
4233
4234                 arp_validate_value = bond_parse_parm(arp_validate,
4235                                                      arp_validate_tbl);
4236                 if (arp_validate_value == -1) {
4237                         pr_err("Error: invalid arp_validate \"%s\"\n",
4238                                arp_validate == NULL ? "NULL" : arp_validate);
4239                         return -EINVAL;
4240                 }
4241         } else
4242                 arp_validate_value = 0;
4243
4244         arp_all_targets_value = 0;
4245         if (arp_all_targets) {
4246                 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4247                                                         arp_all_targets_tbl);
4248
4249                 if (arp_all_targets_value == -1) {
4250                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4251                                arp_all_targets);
4252                         arp_all_targets_value = 0;
4253                 }
4254         }
4255
4256         if (miimon) {
4257                 pr_info("MII link monitoring set to %d ms\n", miimon);
4258         } else if (arp_interval) {
4259                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4260                         arp_interval,
4261                         arp_validate_tbl[arp_validate_value].modename,
4262                         arp_ip_count);
4263
4264                 for (i = 0; i < arp_ip_count; i++)
4265                         pr_info(" %s", arp_ip_target[i]);
4266
4267                 pr_info("\n");
4268
4269         } else if (max_bonds) {
4270                 /* miimon and arp_interval not set, we need one so things
4271                  * work as expected, see bonding.txt for details
4272                  */
4273                 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");
4274         }
4275
4276         if (primary && !USES_PRIMARY(bond_mode)) {
4277                 /* currently, using a primary only makes sense
4278                  * in active backup, TLB or ALB modes
4279                  */
4280                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4281                            primary, bond_mode_name(bond_mode));
4282                 primary = NULL;
4283         }
4284
4285         if (primary && primary_reselect) {
4286                 primary_reselect_value = bond_parse_parm(primary_reselect,
4287                                                          pri_reselect_tbl);
4288                 if (primary_reselect_value == -1) {
4289                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4290                                primary_reselect ==
4291                                         NULL ? "NULL" : primary_reselect);
4292                         return -EINVAL;
4293                 }
4294         } else {
4295                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4296         }
4297
4298         if (fail_over_mac) {
4299                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4300                                                       fail_over_mac_tbl);
4301                 if (fail_over_mac_value == -1) {
4302                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4303                                arp_validate == NULL ? "NULL" : arp_validate);
4304                         return -EINVAL;
4305                 }
4306
4307                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4308                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4309         } else {
4310                 fail_over_mac_value = BOND_FOM_NONE;
4311         }
4312
4313         /* fill params struct with the proper values */
4314         params->mode = bond_mode;
4315         params->xmit_policy = xmit_hashtype;
4316         params->miimon = miimon;
4317         params->num_peer_notif = num_peer_notif;
4318         params->arp_interval = arp_interval;
4319         params->arp_validate = arp_validate_value;
4320         params->arp_all_targets = arp_all_targets_value;
4321         params->updelay = updelay;
4322         params->downdelay = downdelay;
4323         params->use_carrier = use_carrier;
4324         params->lacp_fast = lacp_fast;
4325         params->primary[0] = 0;
4326         params->primary_reselect = primary_reselect_value;
4327         params->fail_over_mac = fail_over_mac_value;
4328         params->tx_queues = tx_queues;
4329         params->all_slaves_active = all_slaves_active;
4330         params->resend_igmp = resend_igmp;
4331         params->min_links = min_links;
4332         params->lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4333         if (packets_per_slave > 1)
4334                 params->packets_per_slave = reciprocal_value(packets_per_slave);
4335         else
4336                 params->packets_per_slave = packets_per_slave;
4337         if (primary) {
4338                 strncpy(params->primary, primary, IFNAMSIZ);
4339                 params->primary[IFNAMSIZ - 1] = 0;
4340         }
4341
4342         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4343
4344         return 0;
4345 }
4346
4347 static struct lock_class_key bonding_netdev_xmit_lock_key;
4348 static struct lock_class_key bonding_netdev_addr_lock_key;
4349 static struct lock_class_key bonding_tx_busylock_key;
4350
4351 static void bond_set_lockdep_class_one(struct net_device *dev,
4352                                        struct netdev_queue *txq,
4353                                        void *_unused)
4354 {
4355         lockdep_set_class(&txq->_xmit_lock,
4356                           &bonding_netdev_xmit_lock_key);
4357 }
4358
4359 static void bond_set_lockdep_class(struct net_device *dev)
4360 {
4361         lockdep_set_class(&dev->addr_list_lock,
4362                           &bonding_netdev_addr_lock_key);
4363         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4364         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4365 }
4366
4367 /*
4368  * Called from registration process
4369  */
4370 static int bond_init(struct net_device *bond_dev)
4371 {
4372         struct bonding *bond = netdev_priv(bond_dev);
4373         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4374         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4375
4376         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4377
4378         /*
4379          * Initialize locks that may be required during
4380          * en/deslave operations.  All of the bond_open work
4381          * (of which this is part) should really be moved to
4382          * a phase prior to dev_open
4383          */
4384         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4385         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4386
4387         bond->wq = create_singlethread_workqueue(bond_dev->name);
4388         if (!bond->wq)
4389                 return -ENOMEM;
4390
4391         bond_set_lockdep_class(bond_dev);
4392
4393         list_add_tail(&bond->bond_list, &bn->dev_list);
4394
4395         bond_prepare_sysfs_group(bond);
4396
4397         bond_debug_register(bond);
4398
4399         /* Ensure valid dev_addr */
4400         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4401             bond_dev->addr_assign_type == NET_ADDR_PERM)
4402                 eth_hw_addr_random(bond_dev);
4403
4404         return 0;
4405 }
4406
4407 unsigned int bond_get_num_tx_queues(void)
4408 {
4409         return tx_queues;
4410 }
4411
4412 /* Create a new bond based on the specified name and bonding parameters.
4413  * If name is NULL, obtain a suitable "bond%d" name for us.
4414  * Caller must NOT hold rtnl_lock; we need to release it here before we
4415  * set up our sysfs entries.
4416  */
4417 int bond_create(struct net *net, const char *name)
4418 {
4419         struct net_device *bond_dev;
4420         int res;
4421
4422         rtnl_lock();
4423
4424         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4425                                    name ? name : "bond%d",
4426                                    bond_setup, tx_queues);
4427         if (!bond_dev) {
4428                 pr_err("%s: eek! can't alloc netdev!\n", name);
4429                 rtnl_unlock();
4430                 return -ENOMEM;
4431         }
4432
4433         dev_net_set(bond_dev, net);
4434         bond_dev->rtnl_link_ops = &bond_link_ops;
4435
4436         res = register_netdevice(bond_dev);
4437
4438         netif_carrier_off(bond_dev);
4439
4440         rtnl_unlock();
4441         if (res < 0)
4442                 bond_destructor(bond_dev);
4443         return res;
4444 }
4445
4446 static int __net_init bond_net_init(struct net *net)
4447 {
4448         struct bond_net *bn = net_generic(net, bond_net_id);
4449
4450         bn->net = net;
4451         INIT_LIST_HEAD(&bn->dev_list);
4452
4453         bond_create_proc_dir(bn);
4454         bond_create_sysfs(bn);
4455
4456         return 0;
4457 }
4458
4459 static void __net_exit bond_net_exit(struct net *net)
4460 {
4461         struct bond_net *bn = net_generic(net, bond_net_id);
4462         struct bonding *bond, *tmp_bond;
4463         LIST_HEAD(list);
4464
4465         bond_destroy_sysfs(bn);
4466         bond_destroy_proc_dir(bn);
4467
4468         /* Kill off any bonds created after unregistering bond rtnl ops */
4469         rtnl_lock();
4470         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4471                 unregister_netdevice_queue(bond->dev, &list);
4472         unregister_netdevice_many(&list);
4473         rtnl_unlock();
4474 }
4475
4476 static struct pernet_operations bond_net_ops = {
4477         .init = bond_net_init,
4478         .exit = bond_net_exit,
4479         .id   = &bond_net_id,
4480         .size = sizeof(struct bond_net),
4481 };
4482
4483 static int __init bonding_init(void)
4484 {
4485         int i;
4486         int res;
4487
4488         pr_info("%s", bond_version);
4489
4490         res = bond_check_params(&bonding_defaults);
4491         if (res)
4492                 goto out;
4493
4494         res = register_pernet_subsys(&bond_net_ops);
4495         if (res)
4496                 goto out;
4497
4498         res = bond_netlink_init();
4499         if (res)
4500                 goto err_link;
4501
4502         bond_create_debugfs();
4503
4504         for (i = 0; i < max_bonds; i++) {
4505                 res = bond_create(&init_net, NULL);
4506                 if (res)
4507                         goto err;
4508         }
4509
4510         register_netdevice_notifier(&bond_netdev_notifier);
4511 out:
4512         return res;
4513 err:
4514         bond_netlink_fini();
4515 err_link:
4516         unregister_pernet_subsys(&bond_net_ops);
4517         goto out;
4518
4519 }
4520
4521 static void __exit bonding_exit(void)
4522 {
4523         unregister_netdevice_notifier(&bond_netdev_notifier);
4524
4525         bond_destroy_debugfs();
4526
4527         bond_netlink_fini();
4528         unregister_pernet_subsys(&bond_net_ops);
4529
4530 #ifdef CONFIG_NET_POLL_CONTROLLER
4531         /*
4532          * Make sure we don't have an imbalance on our netpoll blocking
4533          */
4534         WARN_ON(atomic_read(&netpoll_block_tx));
4535 #endif
4536 }
4537
4538 module_init(bonding_init);
4539 module_exit(bonding_exit);
4540 MODULE_LICENSE("GPL");
4541 MODULE_VERSION(DRV_VERSION);
4542 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4543 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");