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 "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon       = BOND_LINK_MON_INTERV;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static int all_slaves_active = 0;
 109 static struct bond_params bonding_defaults;
 110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 111
 112 module_param(max_bonds, int, 0);
 113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 114 module_param(tx_queues, int, 0);
 115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 118                                "failover event (alias of num_unsol_na)");
 119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 121                                "failover event (alias of num_grat_arp)");
 122 module_param(miimon, int, 0);
 123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 124 module_param(updelay, int, 0);
 125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 126 module_param(downdelay, int, 0);
 127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 128                             "in milliseconds");
 129 module_param(use_carrier, int, 0);
 130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 131                               "0 for off, 1 for on (default)");
 132 module_param(mode, charp, 0);
 133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 134                        "1 for active-backup, 2 for balance-xor, "
 135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 136                        "6 for balance-alb");
 137 module_param(primary, charp, 0);
 138 MODULE_PARM_DESC(primary, "Primary network device to use");
 139 module_param(primary_reselect, charp, 0);
 140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 141                                    "once it comes up; "
 142                                    "0 for always (default), "
 143                                    "1 for only if speed of primary is "
 144                                    "better, "
 145                                    "2 for only on active slave "
 146                                    "failure");
 147 module_param(lacp_rate, charp, 0);
 148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 149                             "0 for slow, 1 for fast");
 150 module_param(ad_select, charp, 0);
 151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 152                             "0 for stable (default), 1 for bandwidth, "
 153                             "2 for count");
 154 module_param(min_links, int, 0);
 155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 156
 157 module_param(xmit_hash_policy, charp, 0);
 158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 159                                    "0 for layer 2 (default), 1 for layer 3+4, "
 160                                    "2 for layer 2+3");
 161 module_param(arp_interval, int, 0);
 162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 163 module_param_array(arp_ip_target, charp, NULL, 0);
 164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 165 module_param(arp_validate, charp, 0);
 166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 167                                "0 for none (default), 1 for active, "
 168                                "2 for backup, 3 for all");
 169 module_param(fail_over_mac, charp, 0);
 170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 171                                 "the same MAC; 0 for none (default), "
 172                                 "1 for active, 2 for follow");
 173 module_param(all_slaves_active, int, 0);
 174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 175                                      "by setting active flag for all slaves; "
 176                                      "0 for never (default), 1 for always.");
 177 module_param(resend_igmp, int, 0);
 178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 179                               "link failure");
 180
 181 /*----------------------------- Global variables ----------------------------*/
 182
 183 #ifdef CONFIG_NET_POLL_CONTROLLER
 184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 185 #endif
 186
 187 int bond_net_id __read_mostly;
 188
 189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 190 static int arp_ip_count;
 191 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 193 static int lacp_fast;
 194
 195 const struct bond_parm_tbl bond_lacp_tbl[] = {
 196 {       "slow",         AD_LACP_SLOW},
 197 {       "fast",         AD_LACP_FAST},
 198 {       NULL,           -1},
 199 };
 200
 201 const struct bond_parm_tbl bond_mode_tbl[] = {
 202 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 203 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 204 {       "balance-xor",          BOND_MODE_XOR},
 205 {       "broadcast",            BOND_MODE_BROADCAST},
 206 {       "802.3ad",              BOND_MODE_8023AD},
 207 {       "balance-tlb",          BOND_MODE_TLB},
 208 {       "balance-alb",          BOND_MODE_ALB},
 209 {       NULL,                   -1},
 210 };
 211
 212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 213 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 214 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 215 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 216 {       NULL,                   -1},
 217 };
 218
 219 const struct bond_parm_tbl arp_validate_tbl[] = {
 220 {       "none",                 BOND_ARP_VALIDATE_NONE},
 221 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 222 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 223 {       "all",                  BOND_ARP_VALIDATE_ALL},
 224 {       NULL,                   -1},
 225 };
 226
 227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 228 {       "none",                 BOND_FOM_NONE},
 229 {       "active",               BOND_FOM_ACTIVE},
 230 {       "follow",               BOND_FOM_FOLLOW},
 231 {       NULL,                   -1},
 232 };
 233
 234 const struct bond_parm_tbl pri_reselect_tbl[] = {
 235 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 236 {       "better",               BOND_PRI_RESELECT_BETTER},
 237 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 238 {       NULL,                   -1},
 239 };
 240
 241 struct bond_parm_tbl ad_select_tbl[] = {
 242 {       "stable",       BOND_AD_STABLE},
 243 {       "bandwidth",    BOND_AD_BANDWIDTH},
 244 {       "count",        BOND_AD_COUNT},
 245 {       NULL,           -1},
 246 };
 247
 248 /*-------------------------- Forward declarations ---------------------------*/
 249
 250 static int bond_init(struct net_device *bond_dev);
 251 static void bond_uninit(struct net_device *bond_dev);
 252
 253 /*---------------------------- General routines -----------------------------*/
 254
 255 const char *bond_mode_name(int mode)
 256 {
 257         static const char *names[] = {
 258                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 259                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 260                 [BOND_MODE_XOR] = "load balancing (xor)",
 261                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 262                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 263                 [BOND_MODE_TLB] = "transmit load balancing",
 264                 [BOND_MODE_ALB] = "adaptive load balancing",
 265         };
 266
 267         if (mode < 0 || mode > BOND_MODE_ALB)
 268                 return "unknown";
 269
 270         return names[mode];
 271 }
 272
 273 /*---------------------------------- VLAN -----------------------------------*/
 274
 275 /**
 276  * bond_add_vlan - add a new vlan id on bond
 277  * @bond: bond that got the notification
 278  * @vlan_id: the vlan id to add
 279  *
 280  * Returns -ENOMEM if allocation failed.
 281  */
 282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 283 {
 284         struct vlan_entry *vlan;
 285
 286         pr_debug("bond: %s, vlan id %d\n",
 287                  (bond ? bond->dev->name : "None"), vlan_id);
 288
 289         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 290         if (!vlan)
 291                 return -ENOMEM;
 292
 293         INIT_LIST_HEAD(&vlan->vlan_list);
 294         vlan->vlan_id = vlan_id;
 295
 296         write_lock_bh(&bond->lock);
 297
 298         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 299
 300         write_unlock_bh(&bond->lock);
 301
 302         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 303
 304         return 0;
 305 }
 306
 307 /**
 308  * bond_del_vlan - delete a vlan id from bond
 309  * @bond: bond that got the notification
 310  * @vlan_id: the vlan id to delete
 311  *
 312  * returns -ENODEV if @vlan_id was not found in @bond.
 313  */
 314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 315 {
 316         struct vlan_entry *vlan;
 317         int res = -ENODEV;
 318
 319         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 320
 321         block_netpoll_tx();
 322         write_lock_bh(&bond->lock);
 323
 324         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 325                 if (vlan->vlan_id == vlan_id) {
 326                         list_del(&vlan->vlan_list);
 327
 328                         if (bond_is_lb(bond))
 329                                 bond_alb_clear_vlan(bond, vlan_id);
 330
 331                         pr_debug("removed VLAN ID %d from bond %s\n",
 332                                  vlan_id, bond->dev->name);
 333
 334                         kfree(vlan);
 335
 336                         res = 0;
 337                         goto out;
 338                 }
 339         }
 340
 341         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 342                  vlan_id, bond->dev->name);
 343
 344 out:
 345         write_unlock_bh(&bond->lock);
 346         unblock_netpoll_tx();
 347         return res;
 348 }
 349
 350 /**
 351  * bond_next_vlan - safely skip to the next item in the vlans list.
 352  * @bond: the bond we're working on
 353  * @curr: item we're advancing from
 354  *
 355  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 356  * or @curr->next otherwise (even if it is @curr itself again).
 357  *
 358  * Caller must hold bond->lock
 359  */
 360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 361 {
 362         struct vlan_entry *next, *last;
 363
 364         if (list_empty(&bond->vlan_list))
 365                 return NULL;
 366
 367         if (!curr) {
 368                 next = list_entry(bond->vlan_list.next,
 369                                   struct vlan_entry, vlan_list);
 370         } else {
 371                 last = list_entry(bond->vlan_list.prev,
 372                                   struct vlan_entry, vlan_list);
 373                 if (last == curr) {
 374                         next = list_entry(bond->vlan_list.next,
 375                                           struct vlan_entry, vlan_list);
 376                 } else {
 377                         next = list_entry(curr->vlan_list.next,
 378                                           struct vlan_entry, vlan_list);
 379                 }
 380         }
 381
 382         return next;
 383 }
 384
 385 /**
 386  * bond_dev_queue_xmit - Prepare skb for xmit.
 387  *
 388  * @bond: bond device that got this skb for tx.
 389  * @skb: hw accel VLAN tagged skb to transmit
 390  * @slave_dev: slave that is supposed to xmit this skbuff
 391  */
 392 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 393                         struct net_device *slave_dev)
 394 {
 395         skb->dev = slave_dev;
 396
 397         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 398                      sizeof(qdisc_skb_cb(skb)->bond_queue_mapping));
 399         skb->queue_mapping = qdisc_skb_cb(skb)->bond_queue_mapping;
 400
 401         if (unlikely(netpoll_tx_running(slave_dev)))
 402                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 403         else
 404                 dev_queue_xmit(skb);
 405
 406         return 0;
 407 }
 408
 409 /*
 410  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 411  * We don't protect the slave list iteration with a lock because:
 412  * a. This operation is performed in IOCTL context,
 413  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 414  * c. Holding a lock with BH disabled while directly calling a base driver
 415  *    entry point is generally a BAD idea.
 416  *
 417  * The design of synchronization/protection for this operation in the 8021q
 418  * module is good for one or more VLAN devices over a single physical device
 419  * and cannot be extended for a teaming solution like bonding, so there is a
 420  * potential race condition here where a net device from the vlan group might
 421  * be referenced (either by a base driver or the 8021q code) while it is being
 422  * removed from the system. However, it turns out we're not making matters
 423  * worse, and if it works for regular VLAN usage it will work here too.
 424 */
 425
 426 /**
 427  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 428  * @bond_dev: bonding net device that got called
 429  * @vid: vlan id being added
 430  */
 431 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 432 {
 433         struct bonding *bond = netdev_priv(bond_dev);
 434         struct slave *slave, *stop_at;
 435         int i, res;
 436
 437         bond_for_each_slave(bond, slave, i) {
 438                 res = vlan_vid_add(slave->dev, vid);
 439                 if (res)
 440                         goto unwind;
 441         }
 442
 443         res = bond_add_vlan(bond, vid);
 444         if (res) {
 445                 pr_err("%s: Error: Failed to add vlan id %d\n",
 446                        bond_dev->name, vid);
 447                 return res;
 448         }
 449
 450         return 0;
 451
 452 unwind:
 453         /* unwind from head to the slave that failed */
 454         stop_at = slave;
 455         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
 456                 vlan_vid_del(slave->dev, vid);
 457
 458         return res;
 459 }
 460
 461 /**
 462  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 463  * @bond_dev: bonding net device that got called
 464  * @vid: vlan id being removed
 465  */
 466 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 467 {
 468         struct bonding *bond = netdev_priv(bond_dev);
 469         struct slave *slave;
 470         int i, res;
 471
 472         bond_for_each_slave(bond, slave, i)
 473                 vlan_vid_del(slave->dev, vid);
 474
 475         res = bond_del_vlan(bond, vid);
 476         if (res) {
 477                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 478                        bond_dev->name, vid);
 479                 return res;
 480         }
 481
 482         return 0;
 483 }
 484
 485 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 486 {
 487         struct vlan_entry *vlan;
 488         int res;
 489
 490         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 491                 res = vlan_vid_add(slave_dev, vlan->vlan_id);
 492                 if (res)
 493                         pr_warning("%s: Failed to add vlan id %d to device %s\n",
 494                                    bond->dev->name, vlan->vlan_id,
 495                                    slave_dev->name);
 496         }
 497 }
 498
 499 static void bond_del_vlans_from_slave(struct bonding *bond,
 500                                       struct net_device *slave_dev)
 501 {
 502         struct vlan_entry *vlan;
 503
 504         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 505                 if (!vlan->vlan_id)
 506                         continue;
 507                 vlan_vid_del(slave_dev, vlan->vlan_id);
 508         }
 509 }
 510
 511 /*------------------------------- Link status -------------------------------*/
 512
 513 /*
 514  * Set the carrier state for the master according to the state of its
 515  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 516  * do special 802.3ad magic.
 517  *
 518  * Returns zero if carrier state does not change, nonzero if it does.
 519  */
 520 static int bond_set_carrier(struct bonding *bond)
 521 {
 522         struct slave *slave;
 523         int i;
 524
 525         if (bond->slave_cnt == 0)
 526                 goto down;
 527
 528         if (bond->params.mode == BOND_MODE_8023AD)
 529                 return bond_3ad_set_carrier(bond);
 530
 531         bond_for_each_slave(bond, slave, i) {
 532                 if (slave->link == BOND_LINK_UP) {
 533                         if (!netif_carrier_ok(bond->dev)) {
 534                                 netif_carrier_on(bond->dev);
 535                                 return 1;
 536                         }
 537                         return 0;
 538                 }
 539         }
 540
 541 down:
 542         if (netif_carrier_ok(bond->dev)) {
 543                 netif_carrier_off(bond->dev);
 544                 return 1;
 545         }
 546         return 0;
 547 }
 548
 549 /*
 550  * Get link speed and duplex from the slave's base driver
 551  * using ethtool. If for some reason the call fails or the
 552  * values are invalid, set speed and duplex to -1,
 553  * and return.
 554  */
 555 static void bond_update_speed_duplex(struct slave *slave)
 556 {
 557         struct net_device *slave_dev = slave->dev;
 558         struct ethtool_cmd ecmd;
 559         u32 slave_speed;
 560         int res;
 561
 562         slave->speed = SPEED_UNKNOWN;
 563         slave->duplex = DUPLEX_UNKNOWN;
 564
 565         res = __ethtool_get_settings(slave_dev, &ecmd);
 566         if (res < 0)
 567                 return;
 568
 569         slave_speed = ethtool_cmd_speed(&ecmd);
 570         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 571                 return;
 572
 573         switch (ecmd.duplex) {
 574         case DUPLEX_FULL:
 575         case DUPLEX_HALF:
 576                 break;
 577         default:
 578                 return;
 579         }
 580
 581         slave->speed = slave_speed;
 582         slave->duplex = ecmd.duplex;
 583
 584         return;
 585 }
 586
 587 /*
 588  * if <dev> supports MII link status reporting, check its link status.
 589  *
 590  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 591  * depending upon the setting of the use_carrier parameter.
 592  *
 593  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 594  * can't tell and just pretend it is), or 0, meaning that the link is
 595  * down.
 596  *
 597  * If reporting is non-zero, instead of faking link up, return -1 if
 598  * both ETHTOOL and MII ioctls fail (meaning the device does not
 599  * support them).  If use_carrier is set, return whatever it says.
 600  * It'd be nice if there was a good way to tell if a driver supports
 601  * netif_carrier, but there really isn't.
 602  */
 603 static int bond_check_dev_link(struct bonding *bond,
 604                                struct net_device *slave_dev, int reporting)
 605 {
 606         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 607         int (*ioctl)(struct net_device *, struct ifreq *, int);
 608         struct ifreq ifr;
 609         struct mii_ioctl_data *mii;
 610
 611         if (!reporting && !netif_running(slave_dev))
 612                 return 0;
 613
 614         if (bond->params.use_carrier)
 615                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 616
 617         /* Try to get link status using Ethtool first. */
 618         if (slave_dev->ethtool_ops) {
 619                 if (slave_dev->ethtool_ops->get_link) {
 620                         u32 link;
 621
 622                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 623
 624                         return link ? BMSR_LSTATUS : 0;
 625                 }
 626         }
 627
 628         /* Ethtool can't be used, fallback to MII ioctls. */
 629         ioctl = slave_ops->ndo_do_ioctl;
 630         if (ioctl) {
 631                 /* TODO: set pointer to correct ioctl on a per team member */
 632                 /*       bases to make this more efficient. that is, once  */
 633                 /*       we determine the correct ioctl, we will always    */
 634                 /*       call it and not the others for that team          */
 635                 /*       member.                                           */
 636
 637                 /*
 638                  * We cannot assume that SIOCGMIIPHY will also read a
 639                  * register; not all network drivers (e.g., e100)
 640                  * support that.
 641                  */
 642
 643                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 644                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 645                 mii = if_mii(&ifr);
 646                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 647                         mii->reg_num = MII_BMSR;
 648                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 649                                 return mii->val_out & BMSR_LSTATUS;
 650                 }
 651         }
 652
 653         /*
 654          * If reporting, report that either there's no dev->do_ioctl,
 655          * or both SIOCGMIIREG and get_link failed (meaning that we
 656          * cannot report link status).  If not reporting, pretend
 657          * we're ok.
 658          */
 659         return reporting ? -1 : BMSR_LSTATUS;
 660 }
 661
 662 /*----------------------------- Multicast list ------------------------------*/
 663
 664 /*
 665  * Push the promiscuity flag down to appropriate slaves
 666  */
 667 static int bond_set_promiscuity(struct bonding *bond, int inc)
 668 {
 669         int err = 0;
 670         if (USES_PRIMARY(bond->params.mode)) {
 671                 /* write lock already acquired */
 672                 if (bond->curr_active_slave) {
 673                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 674                                                   inc);
 675                 }
 676         } else {
 677                 struct slave *slave;
 678                 int i;
 679                 bond_for_each_slave(bond, slave, i) {
 680                         err = dev_set_promiscuity(slave->dev, inc);
 681                         if (err)
 682                                 return err;
 683                 }
 684         }
 685         return err;
 686 }
 687
 688 /*
 689  * Push the allmulti flag down to all slaves
 690  */
 691 static int bond_set_allmulti(struct bonding *bond, int inc)
 692 {
 693         int err = 0;
 694         if (USES_PRIMARY(bond->params.mode)) {
 695                 /* write lock already acquired */
 696                 if (bond->curr_active_slave) {
 697                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 698                                                inc);
 699                 }
 700         } else {
 701                 struct slave *slave;
 702                 int i;
 703                 bond_for_each_slave(bond, slave, i) {
 704                         err = dev_set_allmulti(slave->dev, inc);
 705                         if (err)
 706                                 return err;
 707                 }
 708         }
 709         return err;
 710 }
 711
 712 /*
 713  * Add a Multicast address to slaves
 714  * according to mode
 715  */
 716 static void bond_mc_add(struct bonding *bond, void *addr)
 717 {
 718         if (USES_PRIMARY(bond->params.mode)) {
 719                 /* write lock already acquired */
 720                 if (bond->curr_active_slave)
 721                         dev_mc_add(bond->curr_active_slave->dev, addr);
 722         } else {
 723                 struct slave *slave;
 724                 int i;
 725
 726                 bond_for_each_slave(bond, slave, i)
 727                         dev_mc_add(slave->dev, addr);
 728         }
 729 }
 730
 731 /*
 732  * Remove a multicast address from slave
 733  * according to mode
 734  */
 735 static void bond_mc_del(struct bonding *bond, void *addr)
 736 {
 737         if (USES_PRIMARY(bond->params.mode)) {
 738                 /* write lock already acquired */
 739                 if (bond->curr_active_slave)
 740                         dev_mc_del(bond->curr_active_slave->dev, addr);
 741         } else {
 742                 struct slave *slave;
 743                 int i;
 744                 bond_for_each_slave(bond, slave, i) {
 745                         dev_mc_del(slave->dev, addr);
 746                 }
 747         }
 748 }
 749
 750
 751 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 752 {
 753         struct in_device *in_dev;
 754
 755         rcu_read_lock();
 756         in_dev = __in_dev_get_rcu(dev);
 757         if (in_dev)
 758                 ip_mc_rejoin_groups(in_dev);
 759         rcu_read_unlock();
 760 }
 761
 762 /*
 763  * Retrieve the list of registered multicast addresses for the bonding
 764  * device and retransmit an IGMP JOIN request to the current active
 765  * slave.
 766  */
 767 static void bond_resend_igmp_join_requests(struct bonding *bond)
 768 {
 769         struct net_device *bond_dev, *vlan_dev, *master_dev;
 770         struct vlan_entry *vlan;
 771
 772         read_lock(&bond->lock);
 773
 774         bond_dev = bond->dev;
 775
 776         /* rejoin all groups on bond device */
 777         __bond_resend_igmp_join_requests(bond_dev);
 778
 779         /*
 780          * if bond is enslaved to a bridge,
 781          * then rejoin all groups on its master
 782          */
 783         master_dev = bond_dev->master;
 784         if (master_dev)
 785                 if ((master_dev->priv_flags & IFF_EBRIDGE)
 786                         && (bond_dev->priv_flags & IFF_BRIDGE_PORT))
 787                         __bond_resend_igmp_join_requests(master_dev);
 788
 789         /* rejoin all groups on vlan devices */
 790         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 791                 rcu_read_lock();
 792                 vlan_dev = __vlan_find_dev_deep(bond_dev,
 793                                                 vlan->vlan_id);
 794                 rcu_read_unlock();
 795                 if (vlan_dev)
 796                         __bond_resend_igmp_join_requests(vlan_dev);
 797         }
 798
 799         if (--bond->igmp_retrans > 0)
 800                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 801
 802         read_unlock(&bond->lock);
 803 }
 804
 805 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 806 {
 807         struct bonding *bond = container_of(work, struct bonding,
 808                                             mcast_work.work);
 809         bond_resend_igmp_join_requests(bond);
 810 }
 811
 812 /*
 813  * flush all members of flush->mc_list from device dev->mc_list
 814  */
 815 static void bond_mc_list_flush(struct net_device *bond_dev,
 816                                struct net_device *slave_dev)
 817 {
 818         struct bonding *bond = netdev_priv(bond_dev);
 819         struct netdev_hw_addr *ha;
 820
 821         netdev_for_each_mc_addr(ha, bond_dev)
 822                 dev_mc_del(slave_dev, ha->addr);
 823
 824         if (bond->params.mode == BOND_MODE_8023AD) {
 825                 /* del lacpdu mc addr from mc list */
 826                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 827
 828                 dev_mc_del(slave_dev, lacpdu_multicast);
 829         }
 830 }
 831
 832 /*--------------------------- Active slave change ---------------------------*/
 833
 834 /*
 835  * Update the mc list and multicast-related flags for the new and
 836  * old active slaves (if any) according to the multicast mode, and
 837  * promiscuous flags unconditionally.
 838  */
 839 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 840                          struct slave *old_active)
 841 {
 842         struct netdev_hw_addr *ha;
 843
 844         if (!USES_PRIMARY(bond->params.mode))
 845                 /* nothing to do -  mc list is already up-to-date on
 846                  * all slaves
 847                  */
 848                 return;
 849
 850         if (old_active) {
 851                 if (bond->dev->flags & IFF_PROMISC)
 852                         dev_set_promiscuity(old_active->dev, -1);
 853
 854                 if (bond->dev->flags & IFF_ALLMULTI)
 855                         dev_set_allmulti(old_active->dev, -1);
 856
 857                 netdev_for_each_mc_addr(ha, bond->dev)
 858                         dev_mc_del(old_active->dev, ha->addr);
 859         }
 860
 861         if (new_active) {
 862                 /* FIXME: Signal errors upstream. */
 863                 if (bond->dev->flags & IFF_PROMISC)
 864                         dev_set_promiscuity(new_active->dev, 1);
 865
 866                 if (bond->dev->flags & IFF_ALLMULTI)
 867                         dev_set_allmulti(new_active->dev, 1);
 868
 869                 netdev_for_each_mc_addr(ha, bond->dev)
 870                         dev_mc_add(new_active->dev, ha->addr);
 871         }
 872 }
 873
 874 /*
 875  * bond_do_fail_over_mac
 876  *
 877  * Perform special MAC address swapping for fail_over_mac settings
 878  *
 879  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 880  */
 881 static void bond_do_fail_over_mac(struct bonding *bond,
 882                                   struct slave *new_active,
 883                                   struct slave *old_active)
 884         __releases(&bond->curr_slave_lock)
 885         __releases(&bond->lock)
 886         __acquires(&bond->lock)
 887         __acquires(&bond->curr_slave_lock)
 888 {
 889         u8 tmp_mac[ETH_ALEN];
 890         struct sockaddr saddr;
 891         int rv;
 892
 893         switch (bond->params.fail_over_mac) {
 894         case BOND_FOM_ACTIVE:
 895                 if (new_active) {
 896                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 897                                new_active->dev->addr_len);
 898                         write_unlock_bh(&bond->curr_slave_lock);
 899                         read_unlock(&bond->lock);
 900                         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
 901                         read_lock(&bond->lock);
 902                         write_lock_bh(&bond->curr_slave_lock);
 903                 }
 904                 break;
 905         case BOND_FOM_FOLLOW:
 906                 /*
 907                  * if new_active && old_active, swap them
 908                  * if just old_active, do nothing (going to no active slave)
 909                  * if just new_active, set new_active to bond's MAC
 910                  */
 911                 if (!new_active)
 912                         return;
 913
 914                 write_unlock_bh(&bond->curr_slave_lock);
 915                 read_unlock(&bond->lock);
 916
 917                 if (old_active) {
 918                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 919                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 920                                ETH_ALEN);
 921                         saddr.sa_family = new_active->dev->type;
 922                 } else {
 923                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 924                         saddr.sa_family = bond->dev->type;
 925                 }
 926
 927                 rv = dev_set_mac_address(new_active->dev, &saddr);
 928                 if (rv) {
 929                         pr_err("%s: Error %d setting MAC of slave %s\n",
 930                                bond->dev->name, -rv, new_active->dev->name);
 931                         goto out;
 932                 }
 933
 934                 if (!old_active)
 935                         goto out;
 936
 937                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 938                 saddr.sa_family = old_active->dev->type;
 939
 940                 rv = dev_set_mac_address(old_active->dev, &saddr);
 941                 if (rv)
 942                         pr_err("%s: Error %d setting MAC of slave %s\n",
 943                                bond->dev->name, -rv, new_active->dev->name);
 944 out:
 945                 read_lock(&bond->lock);
 946                 write_lock_bh(&bond->curr_slave_lock);
 947                 break;
 948         default:
 949                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 950                        bond->dev->name, bond->params.fail_over_mac);
 951                 break;
 952         }
 953
 954 }
 955
 956 static bool bond_should_change_active(struct bonding *bond)
 957 {
 958         struct slave *prim = bond->primary_slave;
 959         struct slave *curr = bond->curr_active_slave;
 960
 961         if (!prim || !curr || curr->link != BOND_LINK_UP)
 962                 return true;
 963         if (bond->force_primary) {
 964                 bond->force_primary = false;
 965                 return true;
 966         }
 967         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 968             (prim->speed < curr->speed ||
 969              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 970                 return false;
 971         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 972                 return false;
 973         return true;
 974 }
 975
 976 /**
 977  * find_best_interface - select the best available slave to be the active one
 978  * @bond: our bonding struct
 979  *
 980  * Warning: Caller must hold curr_slave_lock for writing.
 981  */
 982 static struct slave *bond_find_best_slave(struct bonding *bond)
 983 {
 984         struct slave *new_active, *old_active;
 985         struct slave *bestslave = NULL;
 986         int mintime = bond->params.updelay;
 987         int i;
 988
 989         new_active = bond->curr_active_slave;
 990
 991         if (!new_active) { /* there were no active slaves left */
 992                 if (bond->slave_cnt > 0)   /* found one slave */
 993                         new_active = bond->first_slave;
 994                 else
 995                         return NULL; /* still no slave, return NULL */
 996         }
 997
 998         if ((bond->primary_slave) &&
 999             bond->primary_slave->link == BOND_LINK_UP &&
1000             bond_should_change_active(bond)) {
1001                 new_active = bond->primary_slave;
1002         }
1003
1004         /* remember where to stop iterating over the slaves */
1005         old_active = new_active;
1006
1007         bond_for_each_slave_from(bond, new_active, i, old_active) {
1008                 if (new_active->link == BOND_LINK_UP) {
1009                         return new_active;
1010                 } else if (new_active->link == BOND_LINK_BACK &&
1011                            IS_UP(new_active->dev)) {
1012                         /* link up, but waiting for stabilization */
1013                         if (new_active->delay < mintime) {
1014                                 mintime = new_active->delay;
1015                                 bestslave = new_active;
1016                         }
1017                 }
1018         }
1019
1020         return bestslave;
1021 }
1022
1023 static bool bond_should_notify_peers(struct bonding *bond)
1024 {
1025         struct slave *slave = bond->curr_active_slave;
1026
1027         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1028                  bond->dev->name, slave ? slave->dev->name : "NULL");
1029
1030         if (!slave || !bond->send_peer_notif ||
1031             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1032                 return false;
1033
1034         bond->send_peer_notif--;
1035         return true;
1036 }
1037
1038 /**
1039  * change_active_interface - change the active slave into the specified one
1040  * @bond: our bonding struct
1041  * @new: the new slave to make the active one
1042  *
1043  * Set the new slave to the bond's settings and unset them on the old
1044  * curr_active_slave.
1045  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1046  *
1047  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1048  * because it is apparently the best available slave we have, even though its
1049  * updelay hasn't timed out yet.
1050  *
1051  * If new_active is not NULL, caller must hold bond->lock for read and
1052  * curr_slave_lock for write_bh.
1053  */
1054 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1055 {
1056         struct slave *old_active = bond->curr_active_slave;
1057
1058         if (old_active == new_active)
1059                 return;
1060
1061         if (new_active) {
1062                 new_active->jiffies = jiffies;
1063
1064                 if (new_active->link == BOND_LINK_BACK) {
1065                         if (USES_PRIMARY(bond->params.mode)) {
1066                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1067                                         bond->dev->name, new_active->dev->name,
1068                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1069                         }
1070
1071                         new_active->delay = 0;
1072                         new_active->link = BOND_LINK_UP;
1073
1074                         if (bond->params.mode == BOND_MODE_8023AD)
1075                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1076
1077                         if (bond_is_lb(bond))
1078                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1079                 } else {
1080                         if (USES_PRIMARY(bond->params.mode)) {
1081                                 pr_info("%s: making interface %s the new active one.\n",
1082                                         bond->dev->name, new_active->dev->name);
1083                         }
1084                 }
1085         }
1086
1087         if (USES_PRIMARY(bond->params.mode))
1088                 bond_mc_swap(bond, new_active, old_active);
1089
1090         if (bond_is_lb(bond)) {
1091                 bond_alb_handle_active_change(bond, new_active);
1092                 if (old_active)
1093                         bond_set_slave_inactive_flags(old_active);
1094                 if (new_active)
1095                         bond_set_slave_active_flags(new_active);
1096         } else {
1097                 bond->curr_active_slave = new_active;
1098         }
1099
1100         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1101                 if (old_active)
1102                         bond_set_slave_inactive_flags(old_active);
1103
1104                 if (new_active) {
1105                         bool should_notify_peers = false;
1106
1107                         bond_set_slave_active_flags(new_active);
1108
1109                         if (bond->params.fail_over_mac)
1110                                 bond_do_fail_over_mac(bond, new_active,
1111                                                       old_active);
1112
1113                         if (netif_running(bond->dev)) {
1114                                 bond->send_peer_notif =
1115                                         bond->params.num_peer_notif;
1116                                 should_notify_peers =
1117                                         bond_should_notify_peers(bond);
1118                         }
1119
1120                         write_unlock_bh(&bond->curr_slave_lock);
1121                         read_unlock(&bond->lock);
1122
1123                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1124                         if (should_notify_peers)
1125                                 netdev_bonding_change(bond->dev,
1126                                                       NETDEV_NOTIFY_PEERS);
1127
1128                         read_lock(&bond->lock);
1129                         write_lock_bh(&bond->curr_slave_lock);
1130                 }
1131         }
1132
1133         /* resend IGMP joins since active slave has changed or
1134          * all were sent on curr_active_slave.
1135          * resend only if bond is brought up with the affected
1136          * bonding modes and the retransmission is enabled */
1137         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1138             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1139              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1140                 bond->igmp_retrans = bond->params.resend_igmp;
1141                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1142         }
1143 }
1144
1145 /**
1146  * bond_select_active_slave - select a new active slave, if needed
1147  * @bond: our bonding struct
1148  *
1149  * This functions should be called when one of the following occurs:
1150  * - The old curr_active_slave has been released or lost its link.
1151  * - The primary_slave has got its link back.
1152  * - A slave has got its link back and there's no old curr_active_slave.
1153  *
1154  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1155  */
1156 void bond_select_active_slave(struct bonding *bond)
1157 {
1158         struct slave *best_slave;
1159         int rv;
1160
1161         best_slave = bond_find_best_slave(bond);
1162         if (best_slave != bond->curr_active_slave) {
1163                 bond_change_active_slave(bond, best_slave);
1164                 rv = bond_set_carrier(bond);
1165                 if (!rv)
1166                         return;
1167
1168                 if (netif_carrier_ok(bond->dev)) {
1169                         pr_info("%s: first active interface up!\n",
1170                                 bond->dev->name);
1171                 } else {
1172                         pr_info("%s: now running without any active interface !\n",
1173                                 bond->dev->name);
1174                 }
1175         }
1176 }
1177
1178 /*--------------------------- slave list handling ---------------------------*/
1179
1180 /*
1181  * This function attaches the slave to the end of list.
1182  *
1183  * bond->lock held for writing by caller.
1184  */
1185 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1186 {
1187         if (bond->first_slave == NULL) { /* attaching the first slave */
1188                 new_slave->next = new_slave;
1189                 new_slave->prev = new_slave;
1190                 bond->first_slave = new_slave;
1191         } else {
1192                 new_slave->next = bond->first_slave;
1193                 new_slave->prev = bond->first_slave->prev;
1194                 new_slave->next->prev = new_slave;
1195                 new_slave->prev->next = new_slave;
1196         }
1197
1198         bond->slave_cnt++;
1199 }
1200
1201 /*
1202  * This function detaches the slave from the list.
1203  * WARNING: no check is made to verify if the slave effectively
1204  * belongs to <bond>.
1205  * Nothing is freed on return, structures are just unchained.
1206  * If any slave pointer in bond was pointing to <slave>,
1207  * it should be changed by the calling function.
1208  *
1209  * bond->lock held for writing by caller.
1210  */
1211 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1212 {
1213         if (slave->next)
1214                 slave->next->prev = slave->prev;
1215
1216         if (slave->prev)
1217                 slave->prev->next = slave->next;
1218
1219         if (bond->first_slave == slave) { /* slave is the first slave */
1220                 if (bond->slave_cnt > 1) { /* there are more slave */
1221                         bond->first_slave = slave->next;
1222                 } else {
1223                         bond->first_slave = NULL; /* slave was the last one */
1224                 }
1225         }
1226
1227         slave->next = NULL;
1228         slave->prev = NULL;
1229         bond->slave_cnt--;
1230 }
1231
1232 #ifdef CONFIG_NET_POLL_CONTROLLER
1233 static inline int slave_enable_netpoll(struct slave *slave)
1234 {
1235         struct netpoll *np;
1236         int err = 0;
1237
1238         np = kzalloc(sizeof(*np), GFP_KERNEL);
1239         err = -ENOMEM;
1240         if (!np)
1241                 goto out;
1242
1243         np->dev = slave->dev;
1244         strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1245         err = __netpoll_setup(np);
1246         if (err) {
1247                 kfree(np);
1248                 goto out;
1249         }
1250         slave->np = np;
1251 out:
1252         return err;
1253 }
1254 static inline void slave_disable_netpoll(struct slave *slave)
1255 {
1256         struct netpoll *np = slave->np;
1257
1258         if (!np)
1259                 return;
1260
1261         slave->np = NULL;
1262         synchronize_rcu_bh();
1263         __netpoll_cleanup(np);
1264         kfree(np);
1265 }
1266 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1267 {
1268         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1269                 return false;
1270         if (!slave_dev->netdev_ops->ndo_poll_controller)
1271                 return false;
1272         return true;
1273 }
1274
1275 static void bond_poll_controller(struct net_device *bond_dev)
1276 {
1277 }
1278
1279 static void __bond_netpoll_cleanup(struct bonding *bond)
1280 {
1281         struct slave *slave;
1282         int i;
1283
1284         bond_for_each_slave(bond, slave, i)
1285                 if (IS_UP(slave->dev))
1286                         slave_disable_netpoll(slave);
1287 }
1288 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1289 {
1290         struct bonding *bond = netdev_priv(bond_dev);
1291
1292         read_lock(&bond->lock);
1293         __bond_netpoll_cleanup(bond);
1294         read_unlock(&bond->lock);
1295 }
1296
1297 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1298 {
1299         struct bonding *bond = netdev_priv(dev);
1300         struct slave *slave;
1301         int i, err = 0;
1302
1303         read_lock(&bond->lock);
1304         bond_for_each_slave(bond, slave, i) {
1305                 err = slave_enable_netpoll(slave);
1306                 if (err) {
1307                         __bond_netpoll_cleanup(bond);
1308                         break;
1309                 }
1310         }
1311         read_unlock(&bond->lock);
1312         return err;
1313 }
1314
1315 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1316 {
1317         return bond->dev->npinfo;
1318 }
1319
1320 #else
1321 static inline int slave_enable_netpoll(struct slave *slave)
1322 {
1323         return 0;
1324 }
1325 static inline void slave_disable_netpoll(struct slave *slave)
1326 {
1327 }
1328 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1329 {
1330 }
1331 #endif
1332
1333 /*---------------------------------- IOCTL ----------------------------------*/
1334
1335 static int bond_sethwaddr(struct net_device *bond_dev,
1336                           struct net_device *slave_dev)
1337 {
1338         pr_debug("bond_dev=%p\n", bond_dev);
1339         pr_debug("slave_dev=%p\n", slave_dev);
1340         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1341         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1342         return 0;
1343 }
1344
1345 static netdev_features_t bond_fix_features(struct net_device *dev,
1346         netdev_features_t features)
1347 {
1348         struct slave *slave;
1349         struct bonding *bond = netdev_priv(dev);
1350         netdev_features_t mask;
1351         int i;
1352
1353         read_lock(&bond->lock);
1354
1355         if (!bond->first_slave) {
1356                 /* Disable adding VLANs to empty bond. But why? --mq */
1357                 features |= NETIF_F_VLAN_CHALLENGED;
1358                 goto out;
1359         }
1360
1361         mask = features;
1362         features &= ~NETIF_F_ONE_FOR_ALL;
1363         features |= NETIF_F_ALL_FOR_ALL;
1364
1365         bond_for_each_slave(bond, slave, i) {
1366                 features = netdev_increment_features(features,
1367                                                      slave->dev->features,
1368                                                      mask);
1369         }
1370
1371 out:
1372         read_unlock(&bond->lock);
1373         return features;
1374 }
1375
1376 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1377                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1378                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1379
1380 static void bond_compute_features(struct bonding *bond)
1381 {
1382         struct slave *slave;
1383         struct net_device *bond_dev = bond->dev;
1384         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1385         unsigned short max_hard_header_len = ETH_HLEN;
1386         int i;
1387
1388         read_lock(&bond->lock);
1389
1390         if (!bond->first_slave)
1391                 goto done;
1392
1393         bond_for_each_slave(bond, slave, i) {
1394                 vlan_features = netdev_increment_features(vlan_features,
1395                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1396
1397                 if (slave->dev->hard_header_len > max_hard_header_len)
1398                         max_hard_header_len = slave->dev->hard_header_len;
1399         }
1400
1401 done:
1402         bond_dev->vlan_features = vlan_features;
1403         bond_dev->hard_header_len = max_hard_header_len;
1404
1405         read_unlock(&bond->lock);
1406
1407         netdev_change_features(bond_dev);
1408 }
1409
1410 static void bond_setup_by_slave(struct net_device *bond_dev,
1411                                 struct net_device *slave_dev)
1412 {
1413         struct bonding *bond = netdev_priv(bond_dev);
1414
1415         bond_dev->header_ops        = slave_dev->header_ops;
1416
1417         bond_dev->type              = slave_dev->type;
1418         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1419         bond_dev->addr_len          = slave_dev->addr_len;
1420
1421         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1422                 slave_dev->addr_len);
1423         bond->setup_by_slave = 1;
1424 }
1425
1426 /* On bonding slaves other than the currently active slave, suppress
1427  * duplicates except for alb non-mcast/bcast.
1428  */
1429 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1430                                             struct slave *slave,
1431                                             struct bonding *bond)
1432 {
1433         if (bond_is_slave_inactive(slave)) {
1434                 if (bond->params.mode == BOND_MODE_ALB &&
1435                     skb->pkt_type != PACKET_BROADCAST &&
1436                     skb->pkt_type != PACKET_MULTICAST)
1437                         return false;
1438                 return true;
1439         }
1440         return false;
1441 }
1442
1443 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1444 {
1445         struct sk_buff *skb = *pskb;
1446         struct slave *slave;
1447         struct bonding *bond;
1448         int (*recv_probe)(struct sk_buff *, struct bonding *,
1449                                 struct slave *);
1450         int ret = RX_HANDLER_ANOTHER;
1451
1452         skb = skb_share_check(skb, GFP_ATOMIC);
1453         if (unlikely(!skb))
1454                 return RX_HANDLER_CONSUMED;
1455
1456         *pskb = skb;
1457
1458         slave = bond_slave_get_rcu(skb->dev);
1459         bond = slave->bond;
1460
1461         if (bond->params.arp_interval)
1462                 slave->dev->last_rx = jiffies;
1463
1464         recv_probe = ACCESS_ONCE(bond->recv_probe);
1465         if (recv_probe) {
1466                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1467
1468                 if (likely(nskb)) {
1469                         ret = recv_probe(nskb, bond, slave);
1470                         dev_kfree_skb(nskb);
1471                         if (ret == RX_HANDLER_CONSUMED) {
1472                                 consume_skb(skb);
1473                                 return ret;
1474                         }
1475                 }
1476         }
1477
1478         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1479                 return RX_HANDLER_EXACT;
1480         }
1481
1482         skb->dev = bond->dev;
1483
1484         if (bond->params.mode == BOND_MODE_ALB &&
1485             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1486             skb->pkt_type == PACKET_HOST) {
1487
1488                 if (unlikely(skb_cow_head(skb,
1489                                           skb->data - skb_mac_header(skb)))) {
1490                         kfree_skb(skb);
1491                         return RX_HANDLER_CONSUMED;
1492                 }
1493                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1494         }
1495
1496         return ret;
1497 }
1498
1499 /* enslave device <slave> to bond device <master> */
1500 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1501 {
1502         struct bonding *bond = netdev_priv(bond_dev);
1503         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1504         struct slave *new_slave = NULL;
1505         struct netdev_hw_addr *ha;
1506         struct sockaddr addr;
1507         int link_reporting;
1508         int res = 0;
1509
1510         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1511                 slave_ops->ndo_do_ioctl == NULL) {
1512                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1513                            bond_dev->name, slave_dev->name);
1514         }
1515
1516         /* already enslaved */
1517         if (slave_dev->flags & IFF_SLAVE) {
1518                 pr_debug("Error, Device was already enslaved\n");
1519                 return -EBUSY;
1520         }
1521
1522         /* vlan challenged mutual exclusion */
1523         /* no need to lock since we're protected by rtnl_lock */
1524         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1525                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1526                 if (bond_vlan_used(bond)) {
1527                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1528                                bond_dev->name, slave_dev->name, bond_dev->name);
1529                         return -EPERM;
1530                 } else {
1531                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1532                                    bond_dev->name, slave_dev->name,
1533                                    slave_dev->name, bond_dev->name);
1534                 }
1535         } else {
1536                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1537         }
1538
1539         /*
1540          * Old ifenslave binaries are no longer supported.  These can
1541          * be identified with moderate accuracy by the state of the slave:
1542          * the current ifenslave will set the interface down prior to
1543          * enslaving it; the old ifenslave will not.
1544          */
1545         if ((slave_dev->flags & IFF_UP)) {
1546                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1547                        slave_dev->name);
1548                 res = -EPERM;
1549                 goto err_undo_flags;
1550         }
1551
1552         /* set bonding device ether type by slave - bonding netdevices are
1553          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1554          * there is a need to override some of the type dependent attribs/funcs.
1555          *
1556          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1557          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1558          */
1559         if (bond->slave_cnt == 0) {
1560                 if (bond_dev->type != slave_dev->type) {
1561                         pr_debug("%s: change device type from %d to %d\n",
1562                                  bond_dev->name,
1563                                  bond_dev->type, slave_dev->type);
1564
1565                         res = netdev_bonding_change(bond_dev,
1566                                                     NETDEV_PRE_TYPE_CHANGE);
1567                         res = notifier_to_errno(res);
1568                         if (res) {
1569                                 pr_err("%s: refused to change device type\n",
1570                                        bond_dev->name);
1571                                 res = -EBUSY;
1572                                 goto err_undo_flags;
1573                         }
1574
1575                         /* Flush unicast and multicast addresses */
1576                         dev_uc_flush(bond_dev);
1577                         dev_mc_flush(bond_dev);
1578
1579                         if (slave_dev->type != ARPHRD_ETHER)
1580                                 bond_setup_by_slave(bond_dev, slave_dev);
1581                         else {
1582                                 ether_setup(bond_dev);
1583                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1584                         }
1585
1586                         netdev_bonding_change(bond_dev,
1587                                               NETDEV_POST_TYPE_CHANGE);
1588                 }
1589         } else if (bond_dev->type != slave_dev->type) {
1590                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1591                        slave_dev->name,
1592                        slave_dev->type, bond_dev->type);
1593                 res = -EINVAL;
1594                 goto err_undo_flags;
1595         }
1596
1597         if (slave_ops->ndo_set_mac_address == NULL) {
1598                 if (bond->slave_cnt == 0) {
1599                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1600                                    bond_dev->name);
1601                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1602                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1603                         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",
1604                                bond_dev->name);
1605                         res = -EOPNOTSUPP;
1606                         goto err_undo_flags;
1607                 }
1608         }
1609
1610         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1611
1612         /* If this is the first slave, then we need to set the master's hardware
1613          * address to be the same as the slave's. */
1614         if (is_zero_ether_addr(bond->dev->dev_addr))
1615                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1616                        slave_dev->addr_len);
1617
1618
1619         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1620         if (!new_slave) {
1621                 res = -ENOMEM;
1622                 goto err_undo_flags;
1623         }
1624
1625         /*
1626          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1627          * is set via sysfs or module option if desired.
1628          */
1629         new_slave->queue_id = 0;
1630
1631         /* Save slave's original mtu and then set it to match the bond */
1632         new_slave->original_mtu = slave_dev->mtu;
1633         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1634         if (res) {
1635                 pr_debug("Error %d calling dev_set_mtu\n", res);
1636                 goto err_free;
1637         }
1638
1639         /*
1640          * Save slave's original ("permanent") mac address for modes
1641          * that need it, and for restoring it upon release, and then
1642          * set it to the master's address
1643          */
1644         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1645
1646         if (!bond->params.fail_over_mac) {
1647                 /*
1648                  * Set slave to master's mac address.  The application already
1649                  * set the master's mac address to that of the first slave
1650                  */
1651                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1652                 addr.sa_family = slave_dev->type;
1653                 res = dev_set_mac_address(slave_dev, &addr);
1654                 if (res) {
1655                         pr_debug("Error %d calling set_mac_address\n", res);
1656                         goto err_restore_mtu;
1657                 }
1658         }
1659
1660         res = netdev_set_bond_master(slave_dev, bond_dev);
1661         if (res) {
1662                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1663                 goto err_restore_mac;
1664         }
1665
1666         /* open the slave since the application closed it */
1667         res = dev_open(slave_dev);
1668         if (res) {
1669                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1670                 goto err_unset_master;
1671         }
1672
1673         new_slave->bond = bond;
1674         new_slave->dev = slave_dev;
1675         slave_dev->priv_flags |= IFF_BONDING;
1676
1677         if (bond_is_lb(bond)) {
1678                 /* bond_alb_init_slave() must be called before all other stages since
1679                  * it might fail and we do not want to have to undo everything
1680                  */
1681                 res = bond_alb_init_slave(bond, new_slave);
1682                 if (res)
1683                         goto err_close;
1684         }
1685
1686         /* If the mode USES_PRIMARY, then the new slave gets the
1687          * master's promisc (and mc) settings only if it becomes the
1688          * curr_active_slave, and that is taken care of later when calling
1689          * bond_change_active()
1690          */
1691         if (!USES_PRIMARY(bond->params.mode)) {
1692                 /* set promiscuity level to new slave */
1693                 if (bond_dev->flags & IFF_PROMISC) {
1694                         res = dev_set_promiscuity(slave_dev, 1);
1695                         if (res)
1696                                 goto err_close;
1697                 }
1698
1699                 /* set allmulti level to new slave */
1700                 if (bond_dev->flags & IFF_ALLMULTI) {
1701                         res = dev_set_allmulti(slave_dev, 1);
1702                         if (res)
1703                                 goto err_close;
1704                 }
1705
1706                 netif_addr_lock_bh(bond_dev);
1707                 /* upload master's mc_list to new slave */
1708                 netdev_for_each_mc_addr(ha, bond_dev)
1709                         dev_mc_add(slave_dev, ha->addr);
1710                 netif_addr_unlock_bh(bond_dev);
1711         }
1712
1713         if (bond->params.mode == BOND_MODE_8023AD) {
1714                 /* add lacpdu mc addr to mc list */
1715                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1716
1717                 dev_mc_add(slave_dev, lacpdu_multicast);
1718         }
1719
1720         bond_add_vlans_on_slave(bond, slave_dev);
1721
1722         write_lock_bh(&bond->lock);
1723
1724         bond_attach_slave(bond, new_slave);
1725
1726         new_slave->delay = 0;
1727         new_slave->link_failure_count = 0;
1728
1729         write_unlock_bh(&bond->lock);
1730
1731         bond_compute_features(bond);
1732
1733         read_lock(&bond->lock);
1734
1735         new_slave->last_arp_rx = jiffies -
1736                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1737
1738         if (bond->params.miimon && !bond->params.use_carrier) {
1739                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1740
1741                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1742                         /*
1743                          * miimon is set but a bonded network driver
1744                          * does not support ETHTOOL/MII and
1745                          * arp_interval is not set.  Note: if
1746                          * use_carrier is enabled, we will never go
1747                          * here (because netif_carrier is always
1748                          * supported); thus, we don't need to change
1749                          * the messages for netif_carrier.
1750                          */
1751                         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",
1752                                bond_dev->name, slave_dev->name);
1753                 } else if (link_reporting == -1) {
1754                         /* unable get link status using mii/ethtool */
1755                         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",
1756                                    bond_dev->name, slave_dev->name);
1757                 }
1758         }
1759
1760         /* check for initial state */
1761         if (bond->params.miimon) {
1762                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1763                         if (bond->params.updelay) {
1764                                 new_slave->link = BOND_LINK_BACK;
1765                                 new_slave->delay = bond->params.updelay;
1766                         } else {
1767                                 new_slave->link = BOND_LINK_UP;
1768                         }
1769                 } else {
1770                         new_slave->link = BOND_LINK_DOWN;
1771                 }
1772         } else if (bond->params.arp_interval) {
1773                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1774                         BOND_LINK_UP : BOND_LINK_DOWN);
1775         } else {
1776                 new_slave->link = BOND_LINK_UP;
1777         }
1778
1779         if (new_slave->link != BOND_LINK_DOWN)
1780                 new_slave->jiffies = jiffies;
1781         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1782                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1783                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1784
1785         bond_update_speed_duplex(new_slave);
1786
1787         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1788                 /* if there is a primary slave, remember it */
1789                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1790                         bond->primary_slave = new_slave;
1791                         bond->force_primary = true;
1792                 }
1793         }
1794
1795         write_lock_bh(&bond->curr_slave_lock);
1796
1797         switch (bond->params.mode) {
1798         case BOND_MODE_ACTIVEBACKUP:
1799                 bond_set_slave_inactive_flags(new_slave);
1800                 bond_select_active_slave(bond);
1801                 break;
1802         case BOND_MODE_8023AD:
1803                 /* in 802.3ad mode, the internal mechanism
1804                  * will activate the slaves in the selected
1805                  * aggregator
1806                  */
1807                 bond_set_slave_inactive_flags(new_slave);
1808                 /* if this is the first slave */
1809                 if (bond->slave_cnt == 1) {
1810                         SLAVE_AD_INFO(new_slave).id = 1;
1811                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1812                          * can be called only after the mac address of the bond is set
1813                          */
1814                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1815                 } else {
1816                         SLAVE_AD_INFO(new_slave).id =
1817                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1818                 }
1819
1820                 bond_3ad_bind_slave(new_slave);
1821                 break;
1822         case BOND_MODE_TLB:
1823         case BOND_MODE_ALB:
1824                 bond_set_active_slave(new_slave);
1825                 bond_set_slave_inactive_flags(new_slave);
1826                 bond_select_active_slave(bond);
1827                 break;
1828         default:
1829                 pr_debug("This slave is always active in trunk mode\n");
1830
1831                 /* always active in trunk mode */
1832                 bond_set_active_slave(new_slave);
1833
1834                 /* In trunking mode there is little meaning to curr_active_slave
1835                  * anyway (it holds no special properties of the bond device),
1836                  * so we can change it without calling change_active_interface()
1837                  */
1838                 if (!bond->curr_active_slave)
1839                         bond->curr_active_slave = new_slave;
1840
1841                 break;
1842         } /* switch(bond_mode) */
1843
1844         write_unlock_bh(&bond->curr_slave_lock);
1845
1846         bond_set_carrier(bond);
1847
1848 #ifdef CONFIG_NET_POLL_CONTROLLER
1849         slave_dev->npinfo = bond_netpoll_info(bond);
1850         if (slave_dev->npinfo) {
1851                 if (slave_enable_netpoll(new_slave)) {
1852                         read_unlock(&bond->lock);
1853                         pr_info("Error, %s: master_dev is using netpoll, "
1854                                  "but new slave device does not support netpoll.\n",
1855                                  bond_dev->name);
1856                         res = -EBUSY;
1857                         goto err_detach;
1858                 }
1859         }
1860 #endif
1861
1862         read_unlock(&bond->lock);
1863
1864         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1865         if (res)
1866                 goto err_detach;
1867
1868         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1869                                          new_slave);
1870         if (res) {
1871                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1872                 goto err_dest_symlinks;
1873         }
1874
1875         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1876                 bond_dev->name, slave_dev->name,
1877                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1878                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1879
1880         /* enslave is successful */
1881         return 0;
1882
1883 /* Undo stages on error */
1884 err_dest_symlinks:
1885         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1886
1887 err_detach:
1888         write_lock_bh(&bond->lock);
1889         bond_detach_slave(bond, new_slave);
1890         write_unlock_bh(&bond->lock);
1891
1892 err_close:
1893         dev_close(slave_dev);
1894
1895 err_unset_master:
1896         netdev_set_bond_master(slave_dev, NULL);
1897
1898 err_restore_mac:
1899         if (!bond->params.fail_over_mac) {
1900                 /* XXX TODO - fom follow mode needs to change master's
1901                  * MAC if this slave's MAC is in use by the bond, or at
1902                  * least print a warning.
1903                  */
1904                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1905                 addr.sa_family = slave_dev->type;
1906                 dev_set_mac_address(slave_dev, &addr);
1907         }
1908
1909 err_restore_mtu:
1910         dev_set_mtu(slave_dev, new_slave->original_mtu);
1911
1912 err_free:
1913         kfree(new_slave);
1914
1915 err_undo_flags:
1916         bond_compute_features(bond);
1917
1918         return res;
1919 }
1920
1921 /*
1922  * Try to release the slave device <slave> from the bond device <master>
1923  * It is legal to access curr_active_slave without a lock because all the function
1924  * is write-locked.
1925  *
1926  * The rules for slave state should be:
1927  *   for Active/Backup:
1928  *     Active stays on all backups go down
1929  *   for Bonded connections:
1930  *     The first up interface should be left on and all others downed.
1931  */
1932 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1933 {
1934         struct bonding *bond = netdev_priv(bond_dev);
1935         struct slave *slave, *oldcurrent;
1936         struct sockaddr addr;
1937         netdev_features_t old_features = bond_dev->features;
1938
1939         /* slave is not a slave or master is not master of this slave */
1940         if (!(slave_dev->flags & IFF_SLAVE) ||
1941             (slave_dev->master != bond_dev)) {
1942                 pr_err("%s: Error: cannot release %s.\n",
1943                        bond_dev->name, slave_dev->name);
1944                 return -EINVAL;
1945         }
1946
1947         block_netpoll_tx();
1948         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1949         write_lock_bh(&bond->lock);
1950
1951         slave = bond_get_slave_by_dev(bond, slave_dev);
1952         if (!slave) {
1953                 /* not a slave of this bond */
1954                 pr_info("%s: %s not enslaved\n",
1955                         bond_dev->name, slave_dev->name);
1956                 write_unlock_bh(&bond->lock);
1957                 unblock_netpoll_tx();
1958                 return -EINVAL;
1959         }
1960
1961         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1962          * for this slave anymore.
1963          */
1964         netdev_rx_handler_unregister(slave_dev);
1965         write_unlock_bh(&bond->lock);
1966         synchronize_net();
1967         write_lock_bh(&bond->lock);
1968
1969         if (!bond->params.fail_over_mac) {
1970                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1971                     bond->slave_cnt > 1)
1972                         pr_warning("%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",
1973                                    bond_dev->name, slave_dev->name,
1974                                    slave->perm_hwaddr,
1975                                    bond_dev->name, slave_dev->name);
1976         }
1977
1978         /* Inform AD package of unbinding of slave. */
1979         if (bond->params.mode == BOND_MODE_8023AD) {
1980                 /* must be called before the slave is
1981                  * detached from the list
1982                  */
1983                 bond_3ad_unbind_slave(slave);
1984         }
1985
1986         pr_info("%s: releasing %s interface %s\n",
1987                 bond_dev->name,
1988                 bond_is_active_slave(slave) ? "active" : "backup",
1989                 slave_dev->name);
1990
1991         oldcurrent = bond->curr_active_slave;
1992
1993         bond->current_arp_slave = NULL;
1994
1995         /* release the slave from its bond */
1996         bond_detach_slave(bond, slave);
1997
1998         if (bond->primary_slave == slave)
1999                 bond->primary_slave = NULL;
2000
2001         if (oldcurrent == slave)
2002                 bond_change_active_slave(bond, NULL);
2003
2004         if (bond_is_lb(bond)) {
2005                 /* Must be called only after the slave has been
2006                  * detached from the list and the curr_active_slave
2007                  * has been cleared (if our_slave == old_current),
2008                  * but before a new active slave is selected.
2009                  */
2010                 write_unlock_bh(&bond->lock);
2011                 bond_alb_deinit_slave(bond, slave);
2012                 write_lock_bh(&bond->lock);
2013         }
2014
2015         if (oldcurrent == slave) {
2016                 /*
2017                  * Note that we hold RTNL over this sequence, so there
2018                  * is no concern that another slave add/remove event
2019                  * will interfere.
2020                  */
2021                 write_unlock_bh(&bond->lock);
2022                 read_lock(&bond->lock);
2023                 write_lock_bh(&bond->curr_slave_lock);
2024
2025                 bond_select_active_slave(bond);
2026
2027                 write_unlock_bh(&bond->curr_slave_lock);
2028                 read_unlock(&bond->lock);
2029                 write_lock_bh(&bond->lock);
2030         }
2031
2032         if (bond->slave_cnt == 0) {
2033                 bond_set_carrier(bond);
2034
2035                 /* if the last slave was removed, zero the mac address
2036                  * of the master so it will be set by the application
2037                  * to the mac address of the first slave
2038                  */
2039                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2040
2041                 if (bond_vlan_used(bond)) {
2042                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2043                                    bond_dev->name, bond_dev->name);
2044                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2045                                    bond_dev->name);
2046                 }
2047         }
2048
2049         write_unlock_bh(&bond->lock);
2050         unblock_netpoll_tx();
2051
2052         if (bond->slave_cnt == 0)
2053                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2054
2055         bond_compute_features(bond);
2056         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2057             (old_features & NETIF_F_VLAN_CHALLENGED))
2058                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2059                         bond_dev->name, slave_dev->name, bond_dev->name);
2060
2061         /* must do this from outside any spinlocks */
2062         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2063
2064         bond_del_vlans_from_slave(bond, slave_dev);
2065
2066         /* If the mode USES_PRIMARY, then we should only remove its
2067          * promisc and mc settings if it was the curr_active_slave, but that was
2068          * already taken care of above when we detached the slave
2069          */
2070         if (!USES_PRIMARY(bond->params.mode)) {
2071                 /* unset promiscuity level from slave */
2072                 if (bond_dev->flags & IFF_PROMISC)
2073                         dev_set_promiscuity(slave_dev, -1);
2074
2075                 /* unset allmulti level from slave */
2076                 if (bond_dev->flags & IFF_ALLMULTI)
2077                         dev_set_allmulti(slave_dev, -1);
2078
2079                 /* flush master's mc_list from slave */
2080                 netif_addr_lock_bh(bond_dev);
2081                 bond_mc_list_flush(bond_dev, slave_dev);
2082                 netif_addr_unlock_bh(bond_dev);
2083         }
2084
2085         netdev_set_bond_master(slave_dev, NULL);
2086
2087         slave_disable_netpoll(slave);
2088
2089         /* close slave before restoring its mac address */
2090         dev_close(slave_dev);
2091
2092         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2093                 /* restore original ("permanent") mac address */
2094                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2095                 addr.sa_family = slave_dev->type;
2096                 dev_set_mac_address(slave_dev, &addr);
2097         }
2098
2099         dev_set_mtu(slave_dev, slave->original_mtu);
2100
2101         slave_dev->priv_flags &= ~IFF_BONDING;
2102
2103         kfree(slave);
2104
2105         return 0;  /* deletion OK */
2106 }
2107
2108 /*
2109 * First release a slave and then destroy the bond if no more slaves are left.
2110 * Must be under rtnl_lock when this function is called.
2111 */
2112 static int  bond_release_and_destroy(struct net_device *bond_dev,
2113                                      struct net_device *slave_dev)
2114 {
2115         struct bonding *bond = netdev_priv(bond_dev);
2116         int ret;
2117
2118         ret = bond_release(bond_dev, slave_dev);
2119         if ((ret == 0) && (bond->slave_cnt == 0)) {
2120                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2121                 pr_info("%s: destroying bond %s.\n",
2122                         bond_dev->name, bond_dev->name);
2123                 unregister_netdevice(bond_dev);
2124         }
2125         return ret;
2126 }
2127
2128 /*
2129  * This function releases all slaves.
2130  */
2131 static int bond_release_all(struct net_device *bond_dev)
2132 {
2133         struct bonding *bond = netdev_priv(bond_dev);
2134         struct slave *slave;
2135         struct net_device *slave_dev;
2136         struct sockaddr addr;
2137
2138         write_lock_bh(&bond->lock);
2139
2140         netif_carrier_off(bond_dev);
2141
2142         if (bond->slave_cnt == 0)
2143                 goto out;
2144
2145         bond->current_arp_slave = NULL;
2146         bond->primary_slave = NULL;
2147         bond_change_active_slave(bond, NULL);
2148
2149         while ((slave = bond->first_slave) != NULL) {
2150                 /* Inform AD package of unbinding of slave
2151                  * before slave is detached from the list.
2152                  */
2153                 if (bond->params.mode == BOND_MODE_8023AD)
2154                         bond_3ad_unbind_slave(slave);
2155
2156                 slave_dev = slave->dev;
2157                 bond_detach_slave(bond, slave);
2158
2159                 /* now that the slave is detached, unlock and perform
2160                  * all the undo steps that should not be called from
2161                  * within a lock.
2162                  */
2163                 write_unlock_bh(&bond->lock);
2164
2165                 /* unregister rx_handler early so bond_handle_frame wouldn't
2166                  * be called for this slave anymore.
2167                  */
2168                 netdev_rx_handler_unregister(slave_dev);
2169                 synchronize_net();
2170
2171                 if (bond_is_lb(bond)) {
2172                         /* must be called only after the slave
2173                          * has been detached from the list
2174                          */
2175                         bond_alb_deinit_slave(bond, slave);
2176                 }
2177
2178                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2179                 bond_del_vlans_from_slave(bond, slave_dev);
2180
2181                 /* If the mode USES_PRIMARY, then we should only remove its
2182                  * promisc and mc settings if it was the curr_active_slave, but that was
2183                  * already taken care of above when we detached the slave
2184                  */
2185                 if (!USES_PRIMARY(bond->params.mode)) {
2186                         /* unset promiscuity level from slave */
2187                         if (bond_dev->flags & IFF_PROMISC)
2188                                 dev_set_promiscuity(slave_dev, -1);
2189
2190                         /* unset allmulti level from slave */
2191                         if (bond_dev->flags & IFF_ALLMULTI)
2192                                 dev_set_allmulti(slave_dev, -1);
2193
2194                         /* flush master's mc_list from slave */
2195                         netif_addr_lock_bh(bond_dev);
2196                         bond_mc_list_flush(bond_dev, slave_dev);
2197                         netif_addr_unlock_bh(bond_dev);
2198                 }
2199
2200                 netdev_set_bond_master(slave_dev, NULL);
2201
2202                 slave_disable_netpoll(slave);
2203
2204                 /* close slave before restoring its mac address */
2205                 dev_close(slave_dev);
2206
2207                 if (!bond->params.fail_over_mac) {
2208                         /* restore original ("permanent") mac address*/
2209                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2210                         addr.sa_family = slave_dev->type;
2211                         dev_set_mac_address(slave_dev, &addr);
2212                 }
2213
2214                 kfree(slave);
2215
2216                 /* re-acquire the lock before getting the next slave */
2217                 write_lock_bh(&bond->lock);
2218         }
2219
2220         /* zero the mac address of the master so it will be
2221          * set by the application to the mac address of the
2222          * first slave
2223          */
2224         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2225
2226         if (bond_vlan_used(bond)) {
2227                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2228                            bond_dev->name, bond_dev->name);
2229                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2230                            bond_dev->name);
2231         }
2232
2233         pr_info("%s: released all slaves\n", bond_dev->name);
2234
2235 out:
2236         write_unlock_bh(&bond->lock);
2237
2238         bond_compute_features(bond);
2239
2240         return 0;
2241 }
2242
2243 /*
2244  * This function changes the active slave to slave <slave_dev>.
2245  * It returns -EINVAL in the following cases.
2246  *  - <slave_dev> is not found in the list.
2247  *  - There is not active slave now.
2248  *  - <slave_dev> is already active.
2249  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2250  *  - <slave_dev> is not running.
2251  * In these cases, this function does nothing.
2252  * In the other cases, current_slave pointer is changed and 0 is returned.
2253  */
2254 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2255 {
2256         struct bonding *bond = netdev_priv(bond_dev);
2257         struct slave *old_active = NULL;
2258         struct slave *new_active = NULL;
2259         int res = 0;
2260
2261         if (!USES_PRIMARY(bond->params.mode))
2262                 return -EINVAL;
2263
2264         /* Verify that master_dev is indeed the master of slave_dev */
2265         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2266                 return -EINVAL;
2267
2268         read_lock(&bond->lock);
2269
2270         read_lock(&bond->curr_slave_lock);
2271         old_active = bond->curr_active_slave;
2272         read_unlock(&bond->curr_slave_lock);
2273
2274         new_active = bond_get_slave_by_dev(bond, slave_dev);
2275
2276         /*
2277          * Changing to the current active: do nothing; return success.
2278          */
2279         if (new_active && (new_active == old_active)) {
2280                 read_unlock(&bond->lock);
2281                 return 0;
2282         }
2283
2284         if ((new_active) &&
2285             (old_active) &&
2286             (new_active->link == BOND_LINK_UP) &&
2287             IS_UP(new_active->dev)) {
2288                 block_netpoll_tx();
2289                 write_lock_bh(&bond->curr_slave_lock);
2290                 bond_change_active_slave(bond, new_active);
2291                 write_unlock_bh(&bond->curr_slave_lock);
2292                 unblock_netpoll_tx();
2293         } else
2294                 res = -EINVAL;
2295
2296         read_unlock(&bond->lock);
2297
2298         return res;
2299 }
2300
2301 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2302 {
2303         struct bonding *bond = netdev_priv(bond_dev);
2304
2305         info->bond_mode = bond->params.mode;
2306         info->miimon = bond->params.miimon;
2307
2308         read_lock(&bond->lock);
2309         info->num_slaves = bond->slave_cnt;
2310         read_unlock(&bond->lock);
2311
2312         return 0;
2313 }
2314
2315 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2316 {
2317         struct bonding *bond = netdev_priv(bond_dev);
2318         struct slave *slave;
2319         int i, res = -ENODEV;
2320
2321         read_lock(&bond->lock);
2322
2323         bond_for_each_slave(bond, slave, i) {
2324                 if (i == (int)info->slave_id) {
2325                         res = 0;
2326                         strcpy(info->slave_name, slave->dev->name);
2327                         info->link = slave->link;
2328                         info->state = bond_slave_state(slave);
2329                         info->link_failure_count = slave->link_failure_count;
2330                         break;
2331                 }
2332         }
2333
2334         read_unlock(&bond->lock);
2335
2336         return res;
2337 }
2338
2339 /*-------------------------------- Monitoring -------------------------------*/
2340
2341
2342 static int bond_miimon_inspect(struct bonding *bond)
2343 {
2344         struct slave *slave;
2345         int i, link_state, commit = 0;
2346         bool ignore_updelay;
2347
2348         ignore_updelay = !bond->curr_active_slave ? true : false;
2349
2350         bond_for_each_slave(bond, slave, i) {
2351                 slave->new_link = BOND_LINK_NOCHANGE;
2352
2353                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2354
2355                 switch (slave->link) {
2356                 case BOND_LINK_UP:
2357                         if (link_state)
2358                                 continue;
2359
2360                         slave->link = BOND_LINK_FAIL;
2361                         slave->delay = bond->params.downdelay;
2362                         if (slave->delay) {
2363                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2364                                         bond->dev->name,
2365                                         (bond->params.mode ==
2366                                          BOND_MODE_ACTIVEBACKUP) ?
2367                                         (bond_is_active_slave(slave) ?
2368                                          "active " : "backup ") : "",
2369                                         slave->dev->name,
2370                                         bond->params.downdelay * bond->params.miimon);
2371                         }
2372                         /*FALLTHRU*/
2373                 case BOND_LINK_FAIL:
2374                         if (link_state) {
2375                                 /*
2376                                  * recovered before downdelay expired
2377                                  */
2378                                 slave->link = BOND_LINK_UP;
2379                                 slave->jiffies = jiffies;
2380                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2381                                         bond->dev->name,
2382                                         (bond->params.downdelay - slave->delay) *
2383                                         bond->params.miimon,
2384                                         slave->dev->name);
2385                                 continue;
2386                         }
2387
2388                         if (slave->delay <= 0) {
2389                                 slave->new_link = BOND_LINK_DOWN;
2390                                 commit++;
2391                                 continue;
2392                         }
2393
2394                         slave->delay--;
2395                         break;
2396
2397                 case BOND_LINK_DOWN:
2398                         if (!link_state)
2399                                 continue;
2400
2401                         slave->link = BOND_LINK_BACK;
2402                         slave->delay = bond->params.updelay;
2403
2404                         if (slave->delay) {
2405                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2406                                         bond->dev->name, slave->dev->name,
2407                                         ignore_updelay ? 0 :
2408                                         bond->params.updelay *
2409                                         bond->params.miimon);
2410                         }
2411                         /*FALLTHRU*/
2412                 case BOND_LINK_BACK:
2413                         if (!link_state) {
2414                                 slave->link = BOND_LINK_DOWN;
2415                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2416                                         bond->dev->name,
2417                                         (bond->params.updelay - slave->delay) *
2418                                         bond->params.miimon,
2419                                         slave->dev->name);
2420
2421                                 continue;
2422                         }
2423
2424                         if (ignore_updelay)
2425                                 slave->delay = 0;
2426
2427                         if (slave->delay <= 0) {
2428                                 slave->new_link = BOND_LINK_UP;
2429                                 commit++;
2430                                 ignore_updelay = false;
2431                                 continue;
2432                         }
2433
2434                         slave->delay--;
2435                         break;
2436                 }
2437         }
2438
2439         return commit;
2440 }
2441
2442 static void bond_miimon_commit(struct bonding *bond)
2443 {
2444         struct slave *slave;
2445         int i;
2446
2447         bond_for_each_slave(bond, slave, i) {
2448                 switch (slave->new_link) {
2449                 case BOND_LINK_NOCHANGE:
2450                         continue;
2451
2452                 case BOND_LINK_UP:
2453                         slave->link = BOND_LINK_UP;
2454                         slave->jiffies = jiffies;
2455
2456                         if (bond->params.mode == BOND_MODE_8023AD) {
2457                                 /* prevent it from being the active one */
2458                                 bond_set_backup_slave(slave);
2459                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2460                                 /* make it immediately active */
2461                                 bond_set_active_slave(slave);
2462                         } else if (slave != bond->primary_slave) {
2463                                 /* prevent it from being the active one */
2464                                 bond_set_backup_slave(slave);
2465                         }
2466
2467                         bond_update_speed_duplex(slave);
2468
2469                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2470                                 bond->dev->name, slave->dev->name,
2471                                 slave->speed, slave->duplex ? "full" : "half");
2472
2473                         /* notify ad that the link status has changed */
2474                         if (bond->params.mode == BOND_MODE_8023AD)
2475                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2476
2477                         if (bond_is_lb(bond))
2478                                 bond_alb_handle_link_change(bond, slave,
2479                                                             BOND_LINK_UP);
2480
2481                         if (!bond->curr_active_slave ||
2482                             (slave == bond->primary_slave))
2483                                 goto do_failover;
2484
2485                         continue;
2486
2487                 case BOND_LINK_DOWN:
2488                         if (slave->link_failure_count < UINT_MAX)
2489                                 slave->link_failure_count++;
2490
2491                         slave->link = BOND_LINK_DOWN;
2492
2493                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2494                             bond->params.mode == BOND_MODE_8023AD)
2495                                 bond_set_slave_inactive_flags(slave);
2496
2497                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2498                                 bond->dev->name, slave->dev->name);
2499
2500                         if (bond->params.mode == BOND_MODE_8023AD)
2501                                 bond_3ad_handle_link_change(slave,
2502                                                             BOND_LINK_DOWN);
2503
2504                         if (bond_is_lb(bond))
2505                                 bond_alb_handle_link_change(bond, slave,
2506                                                             BOND_LINK_DOWN);
2507
2508                         if (slave == bond->curr_active_slave)
2509                                 goto do_failover;
2510
2511                         continue;
2512
2513                 default:
2514                         pr_err("%s: invalid new link %d on slave %s\n",
2515                                bond->dev->name, slave->new_link,
2516                                slave->dev->name);
2517                         slave->new_link = BOND_LINK_NOCHANGE;
2518
2519                         continue;
2520                 }
2521
2522 do_failover:
2523                 ASSERT_RTNL();
2524                 block_netpoll_tx();
2525                 write_lock_bh(&bond->curr_slave_lock);
2526                 bond_select_active_slave(bond);
2527                 write_unlock_bh(&bond->curr_slave_lock);
2528                 unblock_netpoll_tx();
2529         }
2530
2531         bond_set_carrier(bond);
2532 }
2533
2534 /*
2535  * bond_mii_monitor
2536  *
2537  * Really a wrapper that splits the mii monitor into two phases: an
2538  * inspection, then (if inspection indicates something needs to be done)
2539  * an acquisition of appropriate locks followed by a commit phase to
2540  * implement whatever link state changes are indicated.
2541  */
2542 void bond_mii_monitor(struct work_struct *work)
2543 {
2544         struct bonding *bond = container_of(work, struct bonding,
2545                                             mii_work.work);
2546         bool should_notify_peers = false;
2547         unsigned long delay;
2548
2549         read_lock(&bond->lock);
2550
2551         delay = msecs_to_jiffies(bond->params.miimon);
2552
2553         if (bond->slave_cnt == 0)
2554                 goto re_arm;
2555
2556         should_notify_peers = bond_should_notify_peers(bond);
2557
2558         if (bond_miimon_inspect(bond)) {
2559                 read_unlock(&bond->lock);
2560
2561                 /* Race avoidance with bond_close cancel of workqueue */
2562                 if (!rtnl_trylock()) {
2563                         read_lock(&bond->lock);
2564                         delay = 1;
2565                         should_notify_peers = false;
2566                         goto re_arm;
2567                 }
2568
2569                 read_lock(&bond->lock);
2570
2571                 bond_miimon_commit(bond);
2572
2573                 read_unlock(&bond->lock);
2574                 rtnl_unlock();  /* might sleep, hold no other locks */
2575                 read_lock(&bond->lock);
2576         }
2577
2578 re_arm:
2579         if (bond->params.miimon)
2580                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2581
2582         read_unlock(&bond->lock);
2583
2584         if (should_notify_peers) {
2585                 if (!rtnl_trylock()) {
2586                         read_lock(&bond->lock);
2587                         bond->send_peer_notif++;
2588                         read_unlock(&bond->lock);
2589                         return;
2590                 }
2591                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2592                 rtnl_unlock();
2593         }
2594 }
2595
2596 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2597 {
2598         struct vlan_entry *vlan;
2599         struct net_device *vlan_dev;
2600
2601         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2602                 return 1;
2603
2604         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2605                 rcu_read_lock();
2606                 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2607                 rcu_read_unlock();
2608                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2609                         return 1;
2610         }
2611
2612         return 0;
2613 }
2614
2615 /*
2616  * We go to the (large) trouble of VLAN tagging ARP frames because
2617  * switches in VLAN mode (especially if ports are configured as
2618  * "native" to a VLAN) might not pass non-tagged frames.
2619  */
2620 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2621 {
2622         struct sk_buff *skb;
2623
2624         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2625                  slave_dev->name, dest_ip, src_ip, vlan_id);
2626
2627         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2628                          NULL, slave_dev->dev_addr, NULL);
2629
2630         if (!skb) {
2631                 pr_err("ARP packet allocation failed\n");
2632                 return;
2633         }
2634         if (vlan_id) {
2635                 skb = vlan_put_tag(skb, vlan_id);
2636                 if (!skb) {
2637                         pr_err("failed to insert VLAN tag\n");
2638                         return;
2639                 }
2640         }
2641         arp_xmit(skb);
2642 }
2643
2644
2645 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2646 {
2647         int i, vlan_id;
2648         __be32 *targets = bond->params.arp_targets;
2649         struct vlan_entry *vlan;
2650         struct net_device *vlan_dev = NULL;
2651         struct rtable *rt;
2652
2653         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2654                 __be32 addr;
2655                 if (!targets[i])
2656                         break;
2657                 pr_debug("basa: target %x\n", targets[i]);
2658                 if (!bond_vlan_used(bond)) {
2659                         pr_debug("basa: empty vlan: arp_send\n");
2660                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2661                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2662                                       addr, 0);
2663                         continue;
2664                 }
2665
2666                 /*
2667                  * If VLANs are configured, we do a route lookup to
2668                  * determine which VLAN interface would be used, so we
2669                  * can tag the ARP with the proper VLAN tag.
2670                  */
2671                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2672                                      RTO_ONLINK, 0);
2673                 if (IS_ERR(rt)) {
2674                         if (net_ratelimit()) {
2675                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2676                                            bond->dev->name, &targets[i]);
2677                         }
2678                         continue;
2679                 }
2680
2681                 /*
2682                  * This target is not on a VLAN
2683                  */
2684                 if (rt->dst.dev == bond->dev) {
2685                         ip_rt_put(rt);
2686                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2687                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2688                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2689                                       addr, 0);
2690                         continue;
2691                 }
2692
2693                 vlan_id = 0;
2694                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2695                         rcu_read_lock();
2696                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2697                                                         vlan->vlan_id);
2698                         rcu_read_unlock();
2699                         if (vlan_dev == rt->dst.dev) {
2700                                 vlan_id = vlan->vlan_id;
2701                                 pr_debug("basa: vlan match on %s %d\n",
2702                                        vlan_dev->name, vlan_id);
2703                                 break;
2704                         }
2705                 }
2706
2707                 if (vlan_id && vlan_dev) {
2708                         ip_rt_put(rt);
2709                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2710                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2711                                       addr, vlan_id);
2712                         continue;
2713                 }
2714
2715                 if (net_ratelimit()) {
2716                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2717                                    bond->dev->name, &targets[i],
2718                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2719                 }
2720                 ip_rt_put(rt);
2721         }
2722 }
2723
2724 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2725 {
2726         int i;
2727         __be32 *targets = bond->params.arp_targets;
2728
2729         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2730                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2731                          &sip, &tip, i, &targets[i],
2732                          bond_has_this_ip(bond, tip));
2733                 if (sip == targets[i]) {
2734                         if (bond_has_this_ip(bond, tip))
2735                                 slave->last_arp_rx = jiffies;
2736                         return;
2737                 }
2738         }
2739 }
2740
2741 static int bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2742                          struct slave *slave)
2743 {
2744         struct arphdr *arp;
2745         unsigned char *arp_ptr;
2746         __be32 sip, tip;
2747
2748         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2749                 return RX_HANDLER_ANOTHER;
2750
2751         read_lock(&bond->lock);
2752
2753         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2754                  bond->dev->name, skb->dev->name);
2755
2756         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2757                 goto out_unlock;
2758
2759         arp = arp_hdr(skb);
2760         if (arp->ar_hln != bond->dev->addr_len ||
2761             skb->pkt_type == PACKET_OTHERHOST ||
2762             skb->pkt_type == PACKET_LOOPBACK ||
2763             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2764             arp->ar_pro != htons(ETH_P_IP) ||
2765             arp->ar_pln != 4)
2766                 goto out_unlock;
2767
2768         arp_ptr = (unsigned char *)(arp + 1);
2769         arp_ptr += bond->dev->addr_len;
2770         memcpy(&sip, arp_ptr, 4);
2771         arp_ptr += 4 + bond->dev->addr_len;
2772         memcpy(&tip, arp_ptr, 4);
2773
2774         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2775                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2776                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2777                  &sip, &tip);
2778
2779         /*
2780          * Backup slaves won't see the ARP reply, but do come through
2781          * here for each ARP probe (so we swap the sip/tip to validate
2782          * the probe).  In a "redundant switch, common router" type of
2783          * configuration, the ARP probe will (hopefully) travel from
2784          * the active, through one switch, the router, then the other
2785          * switch before reaching the backup.
2786          */
2787         if (bond_is_active_slave(slave))
2788                 bond_validate_arp(bond, slave, sip, tip);
2789         else
2790                 bond_validate_arp(bond, slave, tip, sip);
2791
2792 out_unlock:
2793         read_unlock(&bond->lock);
2794         return RX_HANDLER_ANOTHER;
2795 }
2796
2797 /*
2798  * this function is called regularly to monitor each slave's link
2799  * ensuring that traffic is being sent and received when arp monitoring
2800  * is used in load-balancing mode. if the adapter has been dormant, then an
2801  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2802  * arp monitoring in active backup mode.
2803  */
2804 void bond_loadbalance_arp_mon(struct work_struct *work)
2805 {
2806         struct bonding *bond = container_of(work, struct bonding,
2807                                             arp_work.work);
2808         struct slave *slave, *oldcurrent;
2809         int do_failover = 0;
2810         int delta_in_ticks;
2811         int i;
2812
2813         read_lock(&bond->lock);
2814
2815         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2816
2817         if (bond->slave_cnt == 0)
2818                 goto re_arm;
2819
2820         read_lock(&bond->curr_slave_lock);
2821         oldcurrent = bond->curr_active_slave;
2822         read_unlock(&bond->curr_slave_lock);
2823
2824         /* see if any of the previous devices are up now (i.e. they have
2825          * xmt and rcv traffic). the curr_active_slave does not come into
2826          * the picture unless it is null. also, slave->jiffies is not needed
2827          * here because we send an arp on each slave and give a slave as
2828          * long as it needs to get the tx/rx within the delta.
2829          * TODO: what about up/down delay in arp mode? it wasn't here before
2830          *       so it can wait
2831          */
2832         bond_for_each_slave(bond, slave, i) {
2833                 unsigned long trans_start = dev_trans_start(slave->dev);
2834
2835                 if (slave->link != BOND_LINK_UP) {
2836                         if (time_in_range(jiffies,
2837                                 trans_start - delta_in_ticks,
2838                                 trans_start + delta_in_ticks) &&
2839                             time_in_range(jiffies,
2840                                 slave->dev->last_rx - delta_in_ticks,
2841                                 slave->dev->last_rx + delta_in_ticks)) {
2842
2843                                 slave->link  = BOND_LINK_UP;
2844                                 bond_set_active_slave(slave);
2845
2846                                 /* primary_slave has no meaning in round-robin
2847                                  * mode. the window of a slave being up and
2848                                  * curr_active_slave being null after enslaving
2849                                  * is closed.
2850                                  */
2851                                 if (!oldcurrent) {
2852                                         pr_info("%s: link status definitely up for interface %s, ",
2853                                                 bond->dev->name,
2854                                                 slave->dev->name);
2855                                         do_failover = 1;
2856                                 } else {
2857                                         pr_info("%s: interface %s is now up\n",
2858                                                 bond->dev->name,
2859                                                 slave->dev->name);
2860                                 }
2861                         }
2862                 } else {
2863                         /* slave->link == BOND_LINK_UP */
2864
2865                         /* not all switches will respond to an arp request
2866                          * when the source ip is 0, so don't take the link down
2867                          * if we don't know our ip yet
2868                          */
2869                         if (!time_in_range(jiffies,
2870                                 trans_start - delta_in_ticks,
2871                                 trans_start + 2 * delta_in_ticks) ||
2872                             !time_in_range(jiffies,
2873                                 slave->dev->last_rx - delta_in_ticks,
2874                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2875
2876                                 slave->link  = BOND_LINK_DOWN;
2877                                 bond_set_backup_slave(slave);
2878
2879                                 if (slave->link_failure_count < UINT_MAX)
2880                                         slave->link_failure_count++;
2881
2882                                 pr_info("%s: interface %s is now down.\n",
2883                                         bond->dev->name,
2884                                         slave->dev->name);
2885
2886                                 if (slave == oldcurrent)
2887                                         do_failover = 1;
2888                         }
2889                 }
2890
2891                 /* note: if switch is in round-robin mode, all links
2892                  * must tx arp to ensure all links rx an arp - otherwise
2893                  * links may oscillate or not come up at all; if switch is
2894                  * in something like xor mode, there is nothing we can
2895                  * do - all replies will be rx'ed on same link causing slaves
2896                  * to be unstable during low/no traffic periods
2897                  */
2898                 if (IS_UP(slave->dev))
2899                         bond_arp_send_all(bond, slave);
2900         }
2901
2902         if (do_failover) {
2903                 block_netpoll_tx();
2904                 write_lock_bh(&bond->curr_slave_lock);
2905
2906                 bond_select_active_slave(bond);
2907
2908                 write_unlock_bh(&bond->curr_slave_lock);
2909                 unblock_netpoll_tx();
2910         }
2911
2912 re_arm:
2913         if (bond->params.arp_interval)
2914                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2915
2916         read_unlock(&bond->lock);
2917 }
2918
2919 /*
2920  * Called to inspect slaves for active-backup mode ARP monitor link state
2921  * changes.  Sets new_link in slaves to specify what action should take
2922  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2923  * to link states must be committed.
2924  *
2925  * Called with bond->lock held for read.
2926  */
2927 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2928 {
2929         struct slave *slave;
2930         int i, commit = 0;
2931         unsigned long trans_start;
2932
2933         bond_for_each_slave(bond, slave, i) {
2934                 slave->new_link = BOND_LINK_NOCHANGE;
2935
2936                 if (slave->link != BOND_LINK_UP) {
2937                         if (time_in_range(jiffies,
2938                                 slave_last_rx(bond, slave) - delta_in_ticks,
2939                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2940
2941                                 slave->new_link = BOND_LINK_UP;
2942                                 commit++;
2943                         }
2944
2945                         continue;
2946                 }
2947
2948                 /*
2949                  * Give slaves 2*delta after being enslaved or made
2950                  * active.  This avoids bouncing, as the last receive
2951                  * times need a full ARP monitor cycle to be updated.
2952                  */
2953                 if (time_in_range(jiffies,
2954                                   slave->jiffies - delta_in_ticks,
2955                                   slave->jiffies + 2 * delta_in_ticks))
2956                         continue;
2957
2958                 /*
2959                  * Backup slave is down if:
2960                  * - No current_arp_slave AND
2961                  * - more than 3*delta since last receive AND
2962                  * - the bond has an IP address
2963                  *
2964                  * Note: a non-null current_arp_slave indicates
2965                  * the curr_active_slave went down and we are
2966                  * searching for a new one; under this condition
2967                  * we only take the curr_active_slave down - this
2968                  * gives each slave a chance to tx/rx traffic
2969                  * before being taken out
2970                  */
2971                 if (!bond_is_active_slave(slave) &&
2972                     !bond->current_arp_slave &&
2973                     !time_in_range(jiffies,
2974                         slave_last_rx(bond, slave) - delta_in_ticks,
2975                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2976
2977                         slave->new_link = BOND_LINK_DOWN;
2978                         commit++;
2979                 }
2980
2981                 /*
2982                  * Active slave is down if:
2983                  * - more than 2*delta since transmitting OR
2984                  * - (more than 2*delta since receive AND
2985                  *    the bond has an IP address)
2986                  */
2987                 trans_start = dev_trans_start(slave->dev);
2988                 if (bond_is_active_slave(slave) &&
2989                     (!time_in_range(jiffies,
2990                         trans_start - delta_in_ticks,
2991                         trans_start + 2 * delta_in_ticks) ||
2992                      !time_in_range(jiffies,
2993                         slave_last_rx(bond, slave) - delta_in_ticks,
2994                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2995
2996                         slave->new_link = BOND_LINK_DOWN;
2997                         commit++;
2998                 }
2999         }
3000
3001         return commit;
3002 }
3003
3004 /*
3005  * Called to commit link state changes noted by inspection step of
3006  * active-backup mode ARP monitor.
3007  *
3008  * Called with RTNL and bond->lock for read.
3009  */
3010 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3011 {
3012         struct slave *slave;
3013         int i;
3014         unsigned long trans_start;
3015
3016         bond_for_each_slave(bond, slave, i) {
3017                 switch (slave->new_link) {
3018                 case BOND_LINK_NOCHANGE:
3019                         continue;
3020
3021                 case BOND_LINK_UP:
3022                         trans_start = dev_trans_start(slave->dev);
3023                         if ((!bond->curr_active_slave &&
3024                              time_in_range(jiffies,
3025                                            trans_start - delta_in_ticks,
3026                                            trans_start + delta_in_ticks)) ||
3027                             bond->curr_active_slave != slave) {
3028                                 slave->link = BOND_LINK_UP;
3029                                 if (bond->current_arp_slave) {
3030                                         bond_set_slave_inactive_flags(
3031                                                 bond->current_arp_slave);
3032                                         bond->current_arp_slave = NULL;
3033                                 }
3034
3035                                 pr_info("%s: link status definitely up for interface %s.\n",
3036                                         bond->dev->name, slave->dev->name);
3037
3038                                 if (!bond->curr_active_slave ||
3039                                     (slave == bond->primary_slave))
3040                                         goto do_failover;
3041
3042                         }
3043
3044                         continue;
3045
3046                 case BOND_LINK_DOWN:
3047                         if (slave->link_failure_count < UINT_MAX)
3048                                 slave->link_failure_count++;
3049
3050                         slave->link = BOND_LINK_DOWN;
3051                         bond_set_slave_inactive_flags(slave);
3052
3053                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3054                                 bond->dev->name, slave->dev->name);
3055
3056                         if (slave == bond->curr_active_slave) {
3057                                 bond->current_arp_slave = NULL;
3058                                 goto do_failover;
3059                         }
3060
3061                         continue;
3062
3063                 default:
3064                         pr_err("%s: impossible: new_link %d on slave %s\n",
3065                                bond->dev->name, slave->new_link,
3066                                slave->dev->name);
3067                         continue;
3068                 }
3069
3070 do_failover:
3071                 ASSERT_RTNL();
3072                 block_netpoll_tx();
3073                 write_lock_bh(&bond->curr_slave_lock);
3074                 bond_select_active_slave(bond);
3075                 write_unlock_bh(&bond->curr_slave_lock);
3076                 unblock_netpoll_tx();
3077         }
3078
3079         bond_set_carrier(bond);
3080 }
3081
3082 /*
3083  * Send ARP probes for active-backup mode ARP monitor.
3084  *
3085  * Called with bond->lock held for read.
3086  */
3087 static void bond_ab_arp_probe(struct bonding *bond)
3088 {
3089         struct slave *slave;
3090         int i;
3091
3092         read_lock(&bond->curr_slave_lock);
3093
3094         if (bond->current_arp_slave && bond->curr_active_slave)
3095                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3096                         bond->current_arp_slave->dev->name,
3097                         bond->curr_active_slave->dev->name);
3098
3099         if (bond->curr_active_slave) {
3100                 bond_arp_send_all(bond, bond->curr_active_slave);
3101                 read_unlock(&bond->curr_slave_lock);
3102                 return;
3103         }
3104
3105         read_unlock(&bond->curr_slave_lock);
3106
3107         /* if we don't have a curr_active_slave, search for the next available
3108          * backup slave from the current_arp_slave and make it the candidate
3109          * for becoming the curr_active_slave
3110          */
3111
3112         if (!bond->current_arp_slave) {
3113                 bond->current_arp_slave = bond->first_slave;
3114                 if (!bond->current_arp_slave)
3115                         return;
3116         }
3117
3118         bond_set_slave_inactive_flags(bond->current_arp_slave);
3119
3120         /* search for next candidate */
3121         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3122                 if (IS_UP(slave->dev)) {
3123                         slave->link = BOND_LINK_BACK;
3124                         bond_set_slave_active_flags(slave);
3125                         bond_arp_send_all(bond, slave);
3126                         slave->jiffies = jiffies;
3127                         bond->current_arp_slave = slave;
3128                         break;
3129                 }
3130
3131                 /* if the link state is up at this point, we
3132                  * mark it down - this can happen if we have
3133                  * simultaneous link failures and
3134                  * reselect_active_interface doesn't make this
3135                  * one the current slave so it is still marked
3136                  * up when it is actually down
3137                  */
3138                 if (slave->link == BOND_LINK_UP) {
3139                         slave->link = BOND_LINK_DOWN;
3140                         if (slave->link_failure_count < UINT_MAX)
3141                                 slave->link_failure_count++;
3142
3143                         bond_set_slave_inactive_flags(slave);
3144
3145                         pr_info("%s: backup interface %s is now down.\n",
3146                                 bond->dev->name, slave->dev->name);
3147                 }
3148         }
3149 }
3150
3151 void bond_activebackup_arp_mon(struct work_struct *work)
3152 {
3153         struct bonding *bond = container_of(work, struct bonding,
3154                                             arp_work.work);
3155         bool should_notify_peers = false;
3156         int delta_in_ticks;
3157
3158         read_lock(&bond->lock);
3159
3160         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3161
3162         if (bond->slave_cnt == 0)
3163                 goto re_arm;
3164
3165         should_notify_peers = bond_should_notify_peers(bond);
3166
3167         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3168                 read_unlock(&bond->lock);
3169
3170                 /* Race avoidance with bond_close flush of workqueue */
3171                 if (!rtnl_trylock()) {
3172                         read_lock(&bond->lock);
3173                         delta_in_ticks = 1;
3174                         should_notify_peers = false;
3175                         goto re_arm;
3176                 }
3177
3178                 read_lock(&bond->lock);
3179
3180                 bond_ab_arp_commit(bond, delta_in_ticks);
3181
3182                 read_unlock(&bond->lock);
3183                 rtnl_unlock();
3184                 read_lock(&bond->lock);
3185         }
3186
3187         bond_ab_arp_probe(bond);
3188
3189 re_arm:
3190         if (bond->params.arp_interval)
3191                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3192
3193         read_unlock(&bond->lock);
3194
3195         if (should_notify_peers) {
3196                 if (!rtnl_trylock()) {
3197                         read_lock(&bond->lock);
3198                         bond->send_peer_notif++;
3199                         read_unlock(&bond->lock);
3200                         return;
3201                 }
3202                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3203                 rtnl_unlock();
3204         }
3205 }
3206
3207 /*-------------------------- netdev event handling --------------------------*/
3208
3209 /*
3210  * Change device name
3211  */
3212 static int bond_event_changename(struct bonding *bond)
3213 {
3214         bond_remove_proc_entry(bond);
3215         bond_create_proc_entry(bond);
3216
3217         bond_debug_reregister(bond);
3218
3219         return NOTIFY_DONE;
3220 }
3221
3222 static int bond_master_netdev_event(unsigned long event,
3223                                     struct net_device *bond_dev)
3224 {
3225         struct bonding *event_bond = netdev_priv(bond_dev);
3226
3227         switch (event) {
3228         case NETDEV_CHANGENAME:
3229                 return bond_event_changename(event_bond);
3230         case NETDEV_UNREGISTER:
3231                 bond_remove_proc_entry(event_bond);
3232                 break;
3233         case NETDEV_REGISTER:
3234                 bond_create_proc_entry(event_bond);
3235                 break;
3236         default:
3237                 break;
3238         }
3239
3240         return NOTIFY_DONE;
3241 }
3242
3243 static int bond_slave_netdev_event(unsigned long event,
3244                                    struct net_device *slave_dev)
3245 {
3246         struct net_device *bond_dev = slave_dev->master;
3247         struct bonding *bond = netdev_priv(bond_dev);
3248         struct slave *slave = NULL;
3249
3250         switch (event) {
3251         case NETDEV_UNREGISTER:
3252                 if (bond_dev) {
3253                         if (bond->setup_by_slave)
3254                                 bond_release_and_destroy(bond_dev, slave_dev);
3255                         else
3256                                 bond_release(bond_dev, slave_dev);
3257                 }
3258                 break;
3259         case NETDEV_UP:
3260         case NETDEV_CHANGE:
3261                 slave = bond_get_slave_by_dev(bond, slave_dev);
3262                 if (slave) {
3263                         u32 old_speed = slave->speed;
3264                         u8  old_duplex = slave->duplex;
3265
3266                         bond_update_speed_duplex(slave);
3267
3268                         if (bond->params.mode == BOND_MODE_8023AD) {
3269                                 if (old_speed != slave->speed)
3270                                         bond_3ad_adapter_speed_changed(slave);
3271                                 if (old_duplex != slave->duplex)
3272                                         bond_3ad_adapter_duplex_changed(slave);
3273                         }
3274                 }
3275
3276                 break;
3277         case NETDEV_DOWN:
3278                 /*
3279                  * ... Or is it this?
3280                  */
3281                 break;
3282         case NETDEV_CHANGEMTU:
3283                 /*
3284                  * TODO: Should slaves be allowed to
3285                  * independently alter their MTU?  For
3286                  * an active-backup bond, slaves need
3287                  * not be the same type of device, so
3288                  * MTUs may vary.  For other modes,
3289                  * slaves arguably should have the
3290                  * same MTUs. To do this, we'd need to
3291                  * take over the slave's change_mtu
3292                  * function for the duration of their
3293                  * servitude.
3294                  */
3295                 break;
3296         case NETDEV_CHANGENAME:
3297                 /*
3298                  * TODO: handle changing the primary's name
3299                  */
3300                 break;
3301         case NETDEV_FEAT_CHANGE:
3302                 bond_compute_features(bond);
3303                 break;
3304         default:
3305                 break;
3306         }
3307
3308         return NOTIFY_DONE;
3309 }
3310
3311 /*
3312  * bond_netdev_event: handle netdev notifier chain events.
3313  *
3314  * This function receives events for the netdev chain.  The caller (an
3315  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3316  * locks for us to safely manipulate the slave devices (RTNL lock,
3317  * dev_probe_lock).
3318  */
3319 static int bond_netdev_event(struct notifier_block *this,
3320                              unsigned long event, void *ptr)
3321 {
3322         struct net_device *event_dev = (struct net_device *)ptr;
3323
3324         pr_debug("event_dev: %s, event: %lx\n",
3325                  event_dev ? event_dev->name : "None",
3326                  event);
3327
3328         if (!(event_dev->priv_flags & IFF_BONDING))
3329                 return NOTIFY_DONE;
3330
3331         if (event_dev->flags & IFF_MASTER) {
3332                 pr_debug("IFF_MASTER\n");
3333                 return bond_master_netdev_event(event, event_dev);
3334         }
3335
3336         if (event_dev->flags & IFF_SLAVE) {
3337                 pr_debug("IFF_SLAVE\n");
3338                 return bond_slave_netdev_event(event, event_dev);
3339         }
3340
3341         return NOTIFY_DONE;
3342 }
3343
3344 static struct notifier_block bond_netdev_notifier = {
3345         .notifier_call = bond_netdev_event,
3346 };
3347
3348 /*---------------------------- Hashing Policies -----------------------------*/
3349
3350 /*
3351  * Hash for the output device based upon layer 2 and layer 3 data. If
3352  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3353  */
3354 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3355 {
3356         struct ethhdr *data = (struct ethhdr *)skb->data;
3357         struct iphdr *iph = ip_hdr(skb);
3358
3359         if (skb->protocol == htons(ETH_P_IP)) {
3360                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3361                         (data->h_dest[5] ^ data->h_source[5])) % count;
3362         }
3363
3364         return (data->h_dest[5] ^ data->h_source[5]) % count;
3365 }
3366
3367 /*
3368  * Hash for the output device based upon layer 3 and layer 4 data. If
3369  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3370  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3371  */
3372 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3373 {
3374         struct ethhdr *data = (struct ethhdr *)skb->data;
3375         struct iphdr *iph = ip_hdr(skb);
3376         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3377         int layer4_xor = 0;
3378
3379         if (skb->protocol == htons(ETH_P_IP)) {
3380                 if (!ip_is_fragment(iph) &&
3381                     (iph->protocol == IPPROTO_TCP ||
3382                      iph->protocol == IPPROTO_UDP)) {
3383                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3384                 }
3385                 return (layer4_xor ^
3386                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3387
3388         }
3389
3390         return (data->h_dest[5] ^ data->h_source[5]) % count;
3391 }
3392
3393 /*
3394  * Hash for the output device based upon layer 2 data
3395  */
3396 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3397 {
3398         struct ethhdr *data = (struct ethhdr *)skb->data;
3399
3400         return (data->h_dest[5] ^ data->h_source[5]) % count;
3401 }
3402
3403 /*-------------------------- Device entry points ----------------------------*/
3404
3405 static int bond_open(struct net_device *bond_dev)
3406 {
3407         struct bonding *bond = netdev_priv(bond_dev);
3408         struct slave *slave;
3409         int i;
3410
3411         /* reset slave->backup and slave->inactive */
3412         read_lock(&bond->lock);
3413         if (bond->slave_cnt > 0) {
3414                 read_lock(&bond->curr_slave_lock);
3415                 bond_for_each_slave(bond, slave, i) {
3416                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3417                                 && (slave != bond->curr_active_slave)) {
3418                                 bond_set_slave_inactive_flags(slave);
3419                         } else {
3420                                 bond_set_slave_active_flags(slave);
3421                         }
3422                 }
3423                 read_unlock(&bond->curr_slave_lock);
3424         }
3425         read_unlock(&bond->lock);
3426
3427         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3428
3429         if (bond_is_lb(bond)) {
3430                 /* bond_alb_initialize must be called before the timer
3431                  * is started.
3432                  */
3433                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3434                         /* something went wrong - fail the open operation */
3435                         return -ENOMEM;
3436                 }
3437
3438                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3439                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3440         }
3441
3442         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3443                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3444                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3445         }
3446
3447         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3448                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3449                         INIT_DELAYED_WORK(&bond->arp_work,
3450                                           bond_activebackup_arp_mon);
3451                 else
3452                         INIT_DELAYED_WORK(&bond->arp_work,
3453                                           bond_loadbalance_arp_mon);
3454
3455                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3456                 if (bond->params.arp_validate)
3457                         bond->recv_probe = bond_arp_rcv;
3458         }
3459
3460         if (bond->params.mode == BOND_MODE_8023AD) {
3461                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3462                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3463                 /* register to receive LACPDUs */
3464                 bond->recv_probe = bond_3ad_lacpdu_recv;
3465                 bond_3ad_initiate_agg_selection(bond, 1);
3466         }
3467
3468         return 0;
3469 }
3470
3471 static int bond_close(struct net_device *bond_dev)
3472 {
3473         struct bonding *bond = netdev_priv(bond_dev);
3474
3475         write_lock_bh(&bond->lock);
3476
3477         bond->send_peer_notif = 0;
3478
3479         write_unlock_bh(&bond->lock);
3480
3481         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3482                 cancel_delayed_work_sync(&bond->mii_work);
3483         }
3484
3485         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3486                 cancel_delayed_work_sync(&bond->arp_work);
3487         }
3488
3489         switch (bond->params.mode) {
3490         case BOND_MODE_8023AD:
3491                 cancel_delayed_work_sync(&bond->ad_work);
3492                 break;
3493         case BOND_MODE_TLB:
3494         case BOND_MODE_ALB:
3495                 cancel_delayed_work_sync(&bond->alb_work);
3496                 break;
3497         default:
3498                 break;
3499         }
3500
3501         if (delayed_work_pending(&bond->mcast_work))
3502                 cancel_delayed_work_sync(&bond->mcast_work);
3503
3504         if (bond_is_lb(bond)) {
3505                 /* Must be called only after all
3506                  * slaves have been released
3507                  */
3508                 bond_alb_deinitialize(bond);
3509         }
3510         bond->recv_probe = NULL;
3511
3512         return 0;
3513 }
3514
3515 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3516                                                 struct rtnl_link_stats64 *stats)
3517 {
3518         struct bonding *bond = netdev_priv(bond_dev);
3519         struct rtnl_link_stats64 temp;
3520         struct slave *slave;
3521         int i;
3522
3523         memset(stats, 0, sizeof(*stats));
3524
3525         read_lock_bh(&bond->lock);
3526
3527         bond_for_each_slave(bond, slave, i) {
3528                 const struct rtnl_link_stats64 *sstats =
3529                         dev_get_stats(slave->dev, &temp);
3530
3531                 stats->rx_packets += sstats->rx_packets;
3532                 stats->rx_bytes += sstats->rx_bytes;
3533                 stats->rx_errors += sstats->rx_errors;
3534                 stats->rx_dropped += sstats->rx_dropped;
3535
3536                 stats->tx_packets += sstats->tx_packets;
3537                 stats->tx_bytes += sstats->tx_bytes;
3538                 stats->tx_errors += sstats->tx_errors;
3539                 stats->tx_dropped += sstats->tx_dropped;
3540
3541                 stats->multicast += sstats->multicast;
3542                 stats->collisions += sstats->collisions;
3543
3544                 stats->rx_length_errors += sstats->rx_length_errors;
3545                 stats->rx_over_errors += sstats->rx_over_errors;
3546                 stats->rx_crc_errors += sstats->rx_crc_errors;
3547                 stats->rx_frame_errors += sstats->rx_frame_errors;
3548                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3549                 stats->rx_missed_errors += sstats->rx_missed_errors;
3550
3551                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3552                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3553                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3554                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3555                 stats->tx_window_errors += sstats->tx_window_errors;
3556         }
3557
3558         read_unlock_bh(&bond->lock);
3559
3560         return stats;
3561 }
3562
3563 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3564 {
3565         struct net_device *slave_dev = NULL;
3566         struct ifbond k_binfo;
3567         struct ifbond __user *u_binfo = NULL;
3568         struct ifslave k_sinfo;
3569         struct ifslave __user *u_sinfo = NULL;
3570         struct mii_ioctl_data *mii = NULL;
3571         int res = 0;
3572
3573         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3574
3575         switch (cmd) {
3576         case SIOCGMIIPHY:
3577                 mii = if_mii(ifr);
3578                 if (!mii)
3579                         return -EINVAL;
3580
3581                 mii->phy_id = 0;
3582                 /* Fall Through */
3583         case SIOCGMIIREG:
3584                 /*
3585                  * We do this again just in case we were called by SIOCGMIIREG
3586                  * instead of SIOCGMIIPHY.
3587                  */
3588                 mii = if_mii(ifr);
3589                 if (!mii)
3590                         return -EINVAL;
3591
3592
3593                 if (mii->reg_num == 1) {
3594                         struct bonding *bond = netdev_priv(bond_dev);
3595                         mii->val_out = 0;
3596                         read_lock(&bond->lock);
3597                         read_lock(&bond->curr_slave_lock);
3598                         if (netif_carrier_ok(bond->dev))
3599                                 mii->val_out = BMSR_LSTATUS;
3600
3601                         read_unlock(&bond->curr_slave_lock);
3602                         read_unlock(&bond->lock);
3603                 }
3604
3605                 return 0;
3606         case BOND_INFO_QUERY_OLD:
3607         case SIOCBONDINFOQUERY:
3608                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3609
3610                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3611                         return -EFAULT;
3612
3613                 res = bond_info_query(bond_dev, &k_binfo);
3614                 if (res == 0 &&
3615                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3616                         return -EFAULT;
3617
3618                 return res;
3619         case BOND_SLAVE_INFO_QUERY_OLD:
3620         case SIOCBONDSLAVEINFOQUERY:
3621                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3622
3623                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3624                         return -EFAULT;
3625
3626                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3627                 if (res == 0 &&
3628                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3629                         return -EFAULT;
3630
3631                 return res;
3632         default:
3633                 /* Go on */
3634                 break;
3635         }
3636
3637         if (!capable(CAP_NET_ADMIN))
3638                 return -EPERM;
3639
3640         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3641
3642         pr_debug("slave_dev=%p:\n", slave_dev);
3643
3644         if (!slave_dev)
3645                 res = -ENODEV;
3646         else {
3647                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3648                 switch (cmd) {
3649                 case BOND_ENSLAVE_OLD:
3650                 case SIOCBONDENSLAVE:
3651                         res = bond_enslave(bond_dev, slave_dev);
3652                         break;
3653                 case BOND_RELEASE_OLD:
3654                 case SIOCBONDRELEASE:
3655                         res = bond_release(bond_dev, slave_dev);
3656                         break;
3657                 case BOND_SETHWADDR_OLD:
3658                 case SIOCBONDSETHWADDR:
3659                         res = bond_sethwaddr(bond_dev, slave_dev);
3660                         break;
3661                 case BOND_CHANGE_ACTIVE_OLD:
3662                 case SIOCBONDCHANGEACTIVE:
3663                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3664                         break;
3665                 default:
3666                         res = -EOPNOTSUPP;
3667                 }
3668
3669                 dev_put(slave_dev);
3670         }
3671
3672         return res;
3673 }
3674
3675 static bool bond_addr_in_mc_list(unsigned char *addr,
3676                                  struct netdev_hw_addr_list *list,
3677                                  int addrlen)
3678 {
3679         struct netdev_hw_addr *ha;
3680
3681         netdev_hw_addr_list_for_each(ha, list)
3682                 if (!memcmp(ha->addr, addr, addrlen))
3683                         return true;
3684
3685         return false;
3686 }
3687
3688 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3689 {
3690         struct bonding *bond = netdev_priv(bond_dev);
3691
3692         if (change & IFF_PROMISC)
3693                 bond_set_promiscuity(bond,
3694                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3695
3696         if (change & IFF_ALLMULTI)
3697                 bond_set_allmulti(bond,
3698                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3699 }
3700
3701 static void bond_set_multicast_list(struct net_device *bond_dev)
3702 {
3703         struct bonding *bond = netdev_priv(bond_dev);
3704         struct netdev_hw_addr *ha;
3705         bool found;
3706
3707         read_lock(&bond->lock);
3708
3709         /* looking for addresses to add to slaves' mc list */
3710         netdev_for_each_mc_addr(ha, bond_dev) {
3711                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3712                                              bond_dev->addr_len);
3713                 if (!found)
3714                         bond_mc_add(bond, ha->addr);
3715         }
3716
3717         /* looking for addresses to delete from slaves' list */
3718         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3719                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3720                                              bond_dev->addr_len);
3721                 if (!found)
3722                         bond_mc_del(bond, ha->addr);
3723         }
3724
3725         /* save master's multicast list */
3726         __hw_addr_flush(&bond->mc_list);
3727         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3728                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3729
3730         read_unlock(&bond->lock);
3731 }
3732
3733 static int bond_neigh_init(struct neighbour *n)
3734 {
3735         struct bonding *bond = netdev_priv(n->dev);
3736         struct slave *slave = bond->first_slave;
3737         const struct net_device_ops *slave_ops;
3738         struct neigh_parms parms;
3739         int ret;
3740
3741         if (!slave)
3742                 return 0;
3743
3744         slave_ops = slave->dev->netdev_ops;
3745
3746         if (!slave_ops->ndo_neigh_setup)
3747                 return 0;
3748
3749         parms.neigh_setup = NULL;
3750         parms.neigh_cleanup = NULL;
3751         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3752         if (ret)
3753                 return ret;
3754
3755         /*
3756          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3757          * after the last slave has been detached.  Assumes that all slaves
3758          * utilize the same neigh_cleanup (true at this writing as only user
3759          * is ipoib).
3760          */
3761         n->parms->neigh_cleanup = parms.neigh_cleanup;
3762
3763         if (!parms.neigh_setup)
3764                 return 0;
3765
3766         return parms.neigh_setup(n);
3767 }
3768
3769 /*
3770  * The bonding ndo_neigh_setup is called at init time beofre any
3771  * slave exists. So we must declare proxy setup function which will
3772  * be used at run time to resolve the actual slave neigh param setup.
3773  */
3774 static int bond_neigh_setup(struct net_device *dev,
3775                             struct neigh_parms *parms)
3776 {
3777         parms->neigh_setup   = bond_neigh_init;
3778
3779         return 0;
3780 }
3781
3782 /*
3783  * Change the MTU of all of a master's slaves to match the master
3784  */
3785 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3786 {
3787         struct bonding *bond = netdev_priv(bond_dev);
3788         struct slave *slave, *stop_at;
3789         int res = 0;
3790         int i;
3791
3792         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3793                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3794
3795         /* Can't hold bond->lock with bh disabled here since
3796          * some base drivers panic. On the other hand we can't
3797          * hold bond->lock without bh disabled because we'll
3798          * deadlock. The only solution is to rely on the fact
3799          * that we're under rtnl_lock here, and the slaves
3800          * list won't change. This doesn't solve the problem
3801          * of setting the slave's MTU while it is
3802          * transmitting, but the assumption is that the base
3803          * driver can handle that.
3804          *
3805          * TODO: figure out a way to safely iterate the slaves
3806          * list, but without holding a lock around the actual
3807          * call to the base driver.
3808          */
3809
3810         bond_for_each_slave(bond, slave, i) {
3811                 pr_debug("s %p s->p %p c_m %p\n",
3812                          slave,
3813                          slave->prev,
3814                          slave->dev->netdev_ops->ndo_change_mtu);
3815
3816                 res = dev_set_mtu(slave->dev, new_mtu);
3817
3818                 if (res) {
3819                         /* If we failed to set the slave's mtu to the new value
3820                          * we must abort the operation even in ACTIVE_BACKUP
3821                          * mode, because if we allow the backup slaves to have
3822                          * different mtu values than the active slave we'll
3823                          * need to change their mtu when doing a failover. That
3824                          * means changing their mtu from timer context, which
3825                          * is probably not a good idea.
3826                          */
3827                         pr_debug("err %d %s\n", res, slave->dev->name);
3828                         goto unwind;
3829                 }
3830         }
3831
3832         bond_dev->mtu = new_mtu;
3833
3834         return 0;
3835
3836 unwind:
3837         /* unwind from head to the slave that failed */
3838         stop_at = slave;
3839         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3840                 int tmp_res;
3841
3842                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3843                 if (tmp_res) {
3844                         pr_debug("unwind err %d dev %s\n",
3845                                  tmp_res, slave->dev->name);
3846                 }
3847         }
3848
3849         return res;
3850 }
3851
3852 /*
3853  * Change HW address
3854  *
3855  * Note that many devices must be down to change the HW address, and
3856  * downing the master releases all slaves.  We can make bonds full of
3857  * bonding devices to test this, however.
3858  */
3859 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3860 {
3861         struct bonding *bond = netdev_priv(bond_dev);
3862         struct sockaddr *sa = addr, tmp_sa;
3863         struct slave *slave, *stop_at;
3864         int res = 0;
3865         int i;
3866
3867         if (bond->params.mode == BOND_MODE_ALB)
3868                 return bond_alb_set_mac_address(bond_dev, addr);
3869
3870
3871         pr_debug("bond=%p, name=%s\n",
3872                  bond, bond_dev ? bond_dev->name : "None");
3873
3874         /*
3875          * If fail_over_mac is set to active, do nothing and return
3876          * success.  Returning an error causes ifenslave to fail.
3877          */
3878         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3879                 return 0;
3880
3881         if (!is_valid_ether_addr(sa->sa_data))
3882                 return -EADDRNOTAVAIL;
3883
3884         /* Can't hold bond->lock with bh disabled here since
3885          * some base drivers panic. On the other hand we can't
3886          * hold bond->lock without bh disabled because we'll
3887          * deadlock. The only solution is to rely on the fact
3888          * that we're under rtnl_lock here, and the slaves
3889          * list won't change. This doesn't solve the problem
3890          * of setting the slave's hw address while it is
3891          * transmitting, but the assumption is that the base
3892          * driver can handle that.
3893          *
3894          * TODO: figure out a way to safely iterate the slaves
3895          * list, but without holding a lock around the actual
3896          * call to the base driver.
3897          */
3898
3899         bond_for_each_slave(bond, slave, i) {
3900                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3901                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3902
3903                 if (slave_ops->ndo_set_mac_address == NULL) {
3904                         res = -EOPNOTSUPP;
3905                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3906                         goto unwind;
3907                 }
3908
3909                 res = dev_set_mac_address(slave->dev, addr);
3910                 if (res) {
3911                         /* TODO: consider downing the slave
3912                          * and retry ?
3913                          * User should expect communications
3914                          * breakage anyway until ARP finish
3915                          * updating, so...
3916                          */
3917                         pr_debug("err %d %s\n", res, slave->dev->name);
3918                         goto unwind;
3919                 }
3920         }
3921
3922         /* success */
3923         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3924         return 0;
3925
3926 unwind:
3927         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3928         tmp_sa.sa_family = bond_dev->type;
3929
3930         /* unwind from head to the slave that failed */
3931         stop_at = slave;
3932         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3933                 int tmp_res;
3934
3935                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3936                 if (tmp_res) {
3937                         pr_debug("unwind err %d dev %s\n",
3938                                  tmp_res, slave->dev->name);
3939                 }
3940         }
3941
3942         return res;
3943 }
3944
3945 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3946 {
3947         struct bonding *bond = netdev_priv(bond_dev);
3948         struct slave *slave, *start_at;
3949         int i, slave_no, res = 1;
3950         struct iphdr *iph = ip_hdr(skb);
3951
3952         /*
3953          * Start with the curr_active_slave that joined the bond as the
3954          * default for sending IGMP traffic.  For failover purposes one
3955          * needs to maintain some consistency for the interface that will
3956          * send the join/membership reports.  The curr_active_slave found
3957          * will send all of this type of traffic.
3958          */
3959         if ((iph->protocol == IPPROTO_IGMP) &&
3960             (skb->protocol == htons(ETH_P_IP))) {
3961
3962                 read_lock(&bond->curr_slave_lock);
3963                 slave = bond->curr_active_slave;
3964                 read_unlock(&bond->curr_slave_lock);
3965
3966                 if (!slave)
3967                         goto out;
3968         } else {
3969                 /*
3970                  * Concurrent TX may collide on rr_tx_counter; we accept
3971                  * that as being rare enough not to justify using an
3972                  * atomic op here.
3973                  */
3974                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3975
3976                 bond_for_each_slave(bond, slave, i) {
3977                         slave_no--;
3978                         if (slave_no < 0)
3979                                 break;
3980                 }
3981         }
3982
3983         start_at = slave;
3984         bond_for_each_slave_from(bond, slave, i, start_at) {
3985                 if (IS_UP(slave->dev) &&
3986                     (slave->link == BOND_LINK_UP) &&
3987                     bond_is_active_slave(slave)) {
3988                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3989                         break;
3990                 }
3991         }
3992
3993 out:
3994         if (res) {
3995                 /* no suitable interface, frame not sent */
3996                 dev_kfree_skb(skb);
3997         }
3998
3999         return NETDEV_TX_OK;
4000 }
4001
4002
4003 /*
4004  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4005  * the bond has a usable interface.
4006  */
4007 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4008 {
4009         struct bonding *bond = netdev_priv(bond_dev);
4010         int res = 1;
4011
4012         read_lock(&bond->curr_slave_lock);
4013
4014         if (bond->curr_active_slave)
4015                 res = bond_dev_queue_xmit(bond, skb,
4016                         bond->curr_active_slave->dev);
4017
4018         if (res)
4019                 /* no suitable interface, frame not sent */
4020                 dev_kfree_skb(skb);
4021
4022         read_unlock(&bond->curr_slave_lock);
4023
4024         return NETDEV_TX_OK;
4025 }
4026
4027 /*
4028  * In bond_xmit_xor() , we determine the output device by using a pre-
4029  * determined xmit_hash_policy(), If the selected device is not enabled,
4030  * find the next active slave.
4031  */
4032 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4033 {
4034         struct bonding *bond = netdev_priv(bond_dev);
4035         struct slave *slave, *start_at;
4036         int slave_no;
4037         int i;
4038         int res = 1;
4039
4040         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4041
4042         bond_for_each_slave(bond, slave, i) {
4043                 slave_no--;
4044                 if (slave_no < 0)
4045                         break;
4046         }
4047
4048         start_at = slave;
4049
4050         bond_for_each_slave_from(bond, slave, i, start_at) {
4051                 if (IS_UP(slave->dev) &&
4052                     (slave->link == BOND_LINK_UP) &&
4053                     bond_is_active_slave(slave)) {
4054                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4055                         break;
4056                 }
4057         }
4058
4059         if (res) {
4060                 /* no suitable interface, frame not sent */
4061                 dev_kfree_skb(skb);
4062         }
4063
4064         return NETDEV_TX_OK;
4065 }
4066
4067 /*
4068  * in broadcast mode, we send everything to all usable interfaces.
4069  */
4070 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4071 {
4072         struct bonding *bond = netdev_priv(bond_dev);
4073         struct slave *slave, *start_at;
4074         struct net_device *tx_dev = NULL;
4075         int i;
4076         int res = 1;
4077
4078         read_lock(&bond->curr_slave_lock);
4079         start_at = bond->curr_active_slave;
4080         read_unlock(&bond->curr_slave_lock);
4081
4082         if (!start_at)
4083                 goto out;
4084
4085         bond_for_each_slave_from(bond, slave, i, start_at) {
4086                 if (IS_UP(slave->dev) &&
4087                     (slave->link == BOND_LINK_UP) &&
4088                     bond_is_active_slave(slave)) {
4089                         if (tx_dev) {
4090                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4091                                 if (!skb2) {
4092                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4093                                                bond_dev->name);
4094                                         continue;
4095                                 }
4096
4097                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4098                                 if (res) {
4099                                         dev_kfree_skb(skb2);
4100                                         continue;
4101                                 }
4102                         }
4103                         tx_dev = slave->dev;
4104                 }
4105         }
4106
4107         if (tx_dev)
4108                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4109
4110 out:
4111         if (res)
4112                 /* no suitable interface, frame not sent */
4113                 dev_kfree_skb(skb);
4114
4115         /* frame sent to all suitable interfaces */
4116         return NETDEV_TX_OK;
4117 }
4118
4119 /*------------------------- Device initialization ---------------------------*/
4120
4121 static void bond_set_xmit_hash_policy(struct bonding *bond)
4122 {
4123         switch (bond->params.xmit_policy) {
4124         case BOND_XMIT_POLICY_LAYER23:
4125                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4126                 break;
4127         case BOND_XMIT_POLICY_LAYER34:
4128                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4129                 break;
4130         case BOND_XMIT_POLICY_LAYER2:
4131         default:
4132                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4133                 break;
4134         }
4135 }
4136
4137 /*
4138  * Lookup the slave that corresponds to a qid
4139  */
4140 static inline int bond_slave_override(struct bonding *bond,
4141                                       struct sk_buff *skb)
4142 {
4143         int i, res = 1;
4144         struct slave *slave = NULL;
4145         struct slave *check_slave;
4146
4147         if (!skb->queue_mapping)
4148                 return 1;
4149
4150         /* Find out if any slaves have the same mapping as this skb. */
4151         bond_for_each_slave(bond, check_slave, i) {
4152                 if (check_slave->queue_id == skb->queue_mapping) {
4153                         slave = check_slave;
4154                         break;
4155                 }
4156         }
4157
4158         /* If the slave isn't UP, use default transmit policy. */
4159         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4160             (slave->link == BOND_LINK_UP)) {
4161                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4162         }
4163
4164         return res;
4165 }
4166
4167
4168 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4169 {
4170         /*
4171          * This helper function exists to help dev_pick_tx get the correct
4172          * destination queue.  Using a helper function skips a call to
4173          * skb_tx_hash and will put the skbs in the queue we expect on their
4174          * way down to the bonding driver.
4175          */
4176         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4177
4178         /*
4179          * Save the original txq to restore before passing to the driver
4180          */
4181         qdisc_skb_cb(skb)->bond_queue_mapping = skb->queue_mapping;
4182
4183         if (unlikely(txq >= dev->real_num_tx_queues)) {
4184                 do {
4185                         txq -= dev->real_num_tx_queues;
4186                 } while (txq >= dev->real_num_tx_queues);
4187         }
4188         return txq;
4189 }
4190
4191 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4192 {
4193         struct bonding *bond = netdev_priv(dev);
4194
4195         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4196                 if (!bond_slave_override(bond, skb))
4197                         return NETDEV_TX_OK;
4198         }
4199
4200         switch (bond->params.mode) {
4201         case BOND_MODE_ROUNDROBIN:
4202                 return bond_xmit_roundrobin(skb, dev);
4203         case BOND_MODE_ACTIVEBACKUP:
4204                 return bond_xmit_activebackup(skb, dev);
4205         case BOND_MODE_XOR:
4206                 return bond_xmit_xor(skb, dev);
4207         case BOND_MODE_BROADCAST:
4208                 return bond_xmit_broadcast(skb, dev);
4209         case BOND_MODE_8023AD:
4210                 return bond_3ad_xmit_xor(skb, dev);
4211         case BOND_MODE_ALB:
4212         case BOND_MODE_TLB:
4213                 return bond_alb_xmit(skb, dev);
4214         default:
4215                 /* Should never happen, mode already checked */
4216                 pr_err("%s: Error: Unknown bonding mode %d\n",
4217                        dev->name, bond->params.mode);
4218                 WARN_ON_ONCE(1);
4219                 dev_kfree_skb(skb);
4220                 return NETDEV_TX_OK;
4221         }
4222 }
4223
4224 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4225 {
4226         struct bonding *bond = netdev_priv(dev);
4227         netdev_tx_t ret = NETDEV_TX_OK;
4228
4229         /*
4230          * If we risk deadlock from transmitting this in the
4231          * netpoll path, tell netpoll to queue the frame for later tx
4232          */
4233         if (is_netpoll_tx_blocked(dev))
4234                 return NETDEV_TX_BUSY;
4235
4236         read_lock(&bond->lock);
4237
4238         if (bond->slave_cnt)
4239                 ret = __bond_start_xmit(skb, dev);
4240         else
4241                 dev_kfree_skb(skb);
4242
4243         read_unlock(&bond->lock);
4244
4245         return ret;
4246 }
4247
4248 /*
4249  * set bond mode specific net device operations
4250  */
4251 void bond_set_mode_ops(struct bonding *bond, int mode)
4252 {
4253         struct net_device *bond_dev = bond->dev;
4254
4255         switch (mode) {
4256         case BOND_MODE_ROUNDROBIN:
4257                 break;
4258         case BOND_MODE_ACTIVEBACKUP:
4259                 break;
4260         case BOND_MODE_XOR:
4261                 bond_set_xmit_hash_policy(bond);
4262                 break;
4263         case BOND_MODE_BROADCAST:
4264                 break;
4265         case BOND_MODE_8023AD:
4266                 bond_set_xmit_hash_policy(bond);
4267                 break;
4268         case BOND_MODE_ALB:
4269                 /* FALLTHRU */
4270         case BOND_MODE_TLB:
4271                 break;
4272         default:
4273                 /* Should never happen, mode already checked */
4274                 pr_err("%s: Error: Unknown bonding mode %d\n",
4275                        bond_dev->name, mode);
4276                 break;
4277         }
4278 }
4279
4280 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4281                                     struct ethtool_drvinfo *drvinfo)
4282 {
4283         strncpy(drvinfo->driver, DRV_NAME, 32);
4284         strncpy(drvinfo->version, DRV_VERSION, 32);
4285         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4286 }
4287
4288 static const struct ethtool_ops bond_ethtool_ops = {
4289         .get_drvinfo            = bond_ethtool_get_drvinfo,
4290         .get_link               = ethtool_op_get_link,
4291 };
4292
4293 static const struct net_device_ops bond_netdev_ops = {
4294         .ndo_init               = bond_init,
4295         .ndo_uninit             = bond_uninit,
4296         .ndo_open               = bond_open,
4297         .ndo_stop               = bond_close,
4298         .ndo_start_xmit         = bond_start_xmit,
4299         .ndo_select_queue       = bond_select_queue,
4300         .ndo_get_stats64        = bond_get_stats,
4301         .ndo_do_ioctl           = bond_do_ioctl,
4302         .ndo_change_rx_flags    = bond_change_rx_flags,
4303         .ndo_set_rx_mode        = bond_set_multicast_list,
4304         .ndo_change_mtu         = bond_change_mtu,
4305         .ndo_set_mac_address    = bond_set_mac_address,
4306         .ndo_neigh_setup        = bond_neigh_setup,
4307         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4308         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4309 #ifdef CONFIG_NET_POLL_CONTROLLER
4310         .ndo_netpoll_setup      = bond_netpoll_setup,
4311         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4312         .ndo_poll_controller    = bond_poll_controller,
4313 #endif
4314         .ndo_add_slave          = bond_enslave,
4315         .ndo_del_slave          = bond_release,
4316         .ndo_fix_features       = bond_fix_features,
4317 };
4318
4319 static void bond_destructor(struct net_device *bond_dev)
4320 {
4321         struct bonding *bond = netdev_priv(bond_dev);
4322         if (bond->wq)
4323                 destroy_workqueue(bond->wq);
4324         free_netdev(bond_dev);
4325 }
4326
4327 static void bond_setup(struct net_device *bond_dev)
4328 {
4329         struct bonding *bond = netdev_priv(bond_dev);
4330
4331         /* initialize rwlocks */
4332         rwlock_init(&bond->lock);
4333         rwlock_init(&bond->curr_slave_lock);
4334
4335         bond->params = bonding_defaults;
4336
4337         /* Initialize pointers */
4338         bond->dev = bond_dev;
4339         INIT_LIST_HEAD(&bond->vlan_list);
4340
4341         /* Initialize the device entry points */
4342         ether_setup(bond_dev);
4343         bond_dev->netdev_ops = &bond_netdev_ops;
4344         bond_dev->ethtool_ops = &bond_ethtool_ops;
4345         bond_set_mode_ops(bond, bond->params.mode);
4346
4347         bond_dev->destructor = bond_destructor;
4348
4349         /* Initialize the device options */
4350         bond_dev->tx_queue_len = 0;
4351         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4352         bond_dev->priv_flags |= IFF_BONDING;
4353         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4354
4355         /* At first, we block adding VLANs. That's the only way to
4356          * prevent problems that occur when adding VLANs over an
4357          * empty bond. The block will be removed once non-challenged
4358          * slaves are enslaved.
4359          */
4360         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4361
4362         /* don't acquire bond device's netif_tx_lock when
4363          * transmitting */
4364         bond_dev->features |= NETIF_F_LLTX;
4365
4366         /* By default, we declare the bond to be fully
4367          * VLAN hardware accelerated capable. Special
4368          * care is taken in the various xmit functions
4369          * when there are slaves that are not hw accel
4370          * capable
4371          */
4372
4373         bond_dev->hw_features = BOND_VLAN_FEATURES |
4374                                 NETIF_F_HW_VLAN_TX |
4375                                 NETIF_F_HW_VLAN_RX |
4376                                 NETIF_F_HW_VLAN_FILTER;
4377
4378         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4379         bond_dev->features |= bond_dev->hw_features;
4380 }
4381
4382 static void bond_work_cancel_all(struct bonding *bond)
4383 {
4384         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4385                 cancel_delayed_work_sync(&bond->mii_work);
4386
4387         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4388                 cancel_delayed_work_sync(&bond->arp_work);
4389
4390         if (bond->params.mode == BOND_MODE_ALB &&
4391             delayed_work_pending(&bond->alb_work))
4392                 cancel_delayed_work_sync(&bond->alb_work);
4393
4394         if (bond->params.mode == BOND_MODE_8023AD &&
4395             delayed_work_pending(&bond->ad_work))
4396                 cancel_delayed_work_sync(&bond->ad_work);
4397
4398         if (delayed_work_pending(&bond->mcast_work))
4399                 cancel_delayed_work_sync(&bond->mcast_work);
4400 }
4401
4402 /*
4403 * Destroy a bonding device.
4404 * Must be under rtnl_lock when this function is called.
4405 */
4406 static void bond_uninit(struct net_device *bond_dev)
4407 {
4408         struct bonding *bond = netdev_priv(bond_dev);
4409         struct vlan_entry *vlan, *tmp;
4410
4411         bond_netpoll_cleanup(bond_dev);
4412
4413         /* Release the bonded slaves */
4414         bond_release_all(bond_dev);
4415
4416         list_del(&bond->bond_list);
4417
4418         bond_work_cancel_all(bond);
4419
4420         bond_debug_unregister(bond);
4421
4422         __hw_addr_flush(&bond->mc_list);
4423
4424         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4425                 list_del(&vlan->vlan_list);
4426                 kfree(vlan);
4427         }
4428 }
4429
4430 /*------------------------- Module initialization ---------------------------*/
4431
4432 /*
4433  * Convert string input module parms.  Accept either the
4434  * number of the mode or its string name.  A bit complicated because
4435  * some mode names are substrings of other names, and calls from sysfs
4436  * may have whitespace in the name (trailing newlines, for example).
4437  */
4438 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4439 {
4440         int modeint = -1, i, rv;
4441         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4442
4443         for (p = (char *)buf; *p; p++)
4444                 if (!(isdigit(*p) || isspace(*p)))
4445                         break;
4446
4447         if (*p)
4448                 rv = sscanf(buf, "%20s", modestr);
4449         else
4450                 rv = sscanf(buf, "%d", &modeint);
4451
4452         if (!rv)
4453                 return -1;
4454
4455         for (i = 0; tbl[i].modename; i++) {
4456                 if (modeint == tbl[i].mode)
4457                         return tbl[i].mode;
4458                 if (strcmp(modestr, tbl[i].modename) == 0)
4459                         return tbl[i].mode;
4460         }
4461
4462         return -1;
4463 }
4464
4465 static int bond_check_params(struct bond_params *params)
4466 {
4467         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4468
4469         /*
4470          * Convert string parameters.
4471          */
4472         if (mode) {
4473                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4474                 if (bond_mode == -1) {
4475                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4476                                mode == NULL ? "NULL" : mode);
4477                         return -EINVAL;
4478                 }
4479         }
4480
4481         if (xmit_hash_policy) {
4482                 if ((bond_mode != BOND_MODE_XOR) &&
4483                     (bond_mode != BOND_MODE_8023AD)) {
4484                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4485                                bond_mode_name(bond_mode));
4486                 } else {
4487                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4488                                                         xmit_hashtype_tbl);
4489                         if (xmit_hashtype == -1) {
4490                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4491                                        xmit_hash_policy == NULL ? "NULL" :
4492                                        xmit_hash_policy);
4493                                 return -EINVAL;
4494                         }
4495                 }
4496         }
4497
4498         if (lacp_rate) {
4499                 if (bond_mode != BOND_MODE_8023AD) {
4500                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4501                                 bond_mode_name(bond_mode));
4502                 } else {
4503                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4504                         if (lacp_fast == -1) {
4505                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4506                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4507                                 return -EINVAL;
4508                         }
4509                 }
4510         }
4511
4512         if (ad_select) {
4513                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4514                 if (params->ad_select == -1) {
4515                         pr_err("Error: Invalid ad_select \"%s\"\n",
4516                                ad_select == NULL ? "NULL" : ad_select);
4517                         return -EINVAL;
4518                 }
4519
4520                 if (bond_mode != BOND_MODE_8023AD) {
4521                         pr_warning("ad_select param only affects 802.3ad mode\n");
4522                 }
4523         } else {
4524                 params->ad_select = BOND_AD_STABLE;
4525         }
4526
4527         if (max_bonds < 0) {
4528                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4529                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4530                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4531         }
4532
4533         if (miimon < 0) {
4534                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4535                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4536                 miimon = BOND_LINK_MON_INTERV;
4537         }
4538
4539         if (updelay < 0) {
4540                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4541                            updelay, INT_MAX);
4542                 updelay = 0;
4543         }
4544
4545         if (downdelay < 0) {
4546                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4547                            downdelay, INT_MAX);
4548                 downdelay = 0;
4549         }
4550
4551         if ((use_carrier != 0) && (use_carrier != 1)) {
4552                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4553                            use_carrier);
4554                 use_carrier = 1;
4555         }
4556
4557         if (num_peer_notif < 0 || num_peer_notif > 255) {
4558                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4559                            num_peer_notif);
4560                 num_peer_notif = 1;
4561         }
4562
4563         /* reset values for 802.3ad */
4564         if (bond_mode == BOND_MODE_8023AD) {
4565                 if (!miimon) {
4566                         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");
4567                         pr_warning("Forcing miimon to 100msec\n");
4568                         miimon = 100;
4569                 }
4570         }
4571
4572         if (tx_queues < 1 || tx_queues > 255) {
4573                 pr_warning("Warning: tx_queues (%d) should be between "
4574                            "1 and 255, resetting to %d\n",
4575                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4576                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4577         }
4578
4579         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4580                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4581                            "not of valid value (0/1), so it was set to "
4582                            "0\n", all_slaves_active);
4583                 all_slaves_active = 0;
4584         }
4585
4586         if (resend_igmp < 0 || resend_igmp > 255) {
4587                 pr_warning("Warning: resend_igmp (%d) should be between "
4588                            "0 and 255, resetting to %d\n",
4589                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4590                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4591         }
4592
4593         /* reset values for TLB/ALB */
4594         if ((bond_mode == BOND_MODE_TLB) ||
4595             (bond_mode == BOND_MODE_ALB)) {
4596                 if (!miimon) {
4597                         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");
4598                         pr_warning("Forcing miimon to 100msec\n");
4599                         miimon = 100;
4600                 }
4601         }
4602
4603         if (bond_mode == BOND_MODE_ALB) {
4604                 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",
4605                           updelay);
4606         }
4607
4608         if (!miimon) {
4609                 if (updelay || downdelay) {
4610                         /* just warn the user the up/down delay will have
4611                          * no effect since miimon is zero...
4612                          */
4613                         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",
4614                                    updelay, downdelay);
4615                 }
4616         } else {
4617                 /* don't allow arp monitoring */
4618                 if (arp_interval) {
4619                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4620                                    miimon, arp_interval);
4621                         arp_interval = 0;
4622                 }
4623
4624                 if ((updelay % miimon) != 0) {
4625                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4626                                    updelay, miimon,
4627                                    (updelay / miimon) * miimon);
4628                 }
4629
4630                 updelay /= miimon;
4631
4632                 if ((downdelay % miimon) != 0) {
4633                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4634                                    downdelay, miimon,
4635                                    (downdelay / miimon) * miimon);
4636                 }
4637
4638                 downdelay /= miimon;
4639         }
4640
4641         if (arp_interval < 0) {
4642                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4643                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4644                 arp_interval = BOND_LINK_ARP_INTERV;
4645         }
4646
4647         for (arp_ip_count = 0;
4648              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4649              arp_ip_count++) {
4650                 /* not complete check, but should be good enough to
4651                    catch mistakes */
4652                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4653                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4654                                    arp_ip_target[arp_ip_count]);
4655                         arp_interval = 0;
4656                 } else {
4657                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4658                         arp_target[arp_ip_count] = ip;
4659                 }
4660         }
4661
4662         if (arp_interval && !arp_ip_count) {
4663                 /* don't allow arping if no arp_ip_target given... */
4664                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4665                            arp_interval);
4666                 arp_interval = 0;
4667         }
4668
4669         if (arp_validate) {
4670                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4671                         pr_err("arp_validate only supported in active-backup mode\n");
4672                         return -EINVAL;
4673                 }
4674                 if (!arp_interval) {
4675                         pr_err("arp_validate requires arp_interval\n");
4676                         return -EINVAL;
4677                 }
4678
4679                 arp_validate_value = bond_parse_parm(arp_validate,
4680                                                      arp_validate_tbl);
4681                 if (arp_validate_value == -1) {
4682                         pr_err("Error: invalid arp_validate \"%s\"\n",
4683                                arp_validate == NULL ? "NULL" : arp_validate);
4684                         return -EINVAL;
4685                 }
4686         } else
4687                 arp_validate_value = 0;
4688
4689         if (miimon) {
4690                 pr_info("MII link monitoring set to %d ms\n", miimon);
4691         } else if (arp_interval) {
4692                 int i;
4693
4694                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4695                         arp_interval,
4696                         arp_validate_tbl[arp_validate_value].modename,
4697                         arp_ip_count);
4698
4699                 for (i = 0; i < arp_ip_count; i++)
4700                         pr_info(" %s", arp_ip_target[i]);
4701
4702                 pr_info("\n");
4703
4704         } else if (max_bonds) {
4705                 /* miimon and arp_interval not set, we need one so things
4706                  * work as expected, see bonding.txt for details
4707                  */
4708                 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");
4709         }
4710
4711         if (primary && !USES_PRIMARY(bond_mode)) {
4712                 /* currently, using a primary only makes sense
4713                  * in active backup, TLB or ALB modes
4714                  */
4715                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4716                            primary, bond_mode_name(bond_mode));
4717                 primary = NULL;
4718         }
4719
4720         if (primary && primary_reselect) {
4721                 primary_reselect_value = bond_parse_parm(primary_reselect,
4722                                                          pri_reselect_tbl);
4723                 if (primary_reselect_value == -1) {
4724                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4725                                primary_reselect ==
4726                                         NULL ? "NULL" : primary_reselect);
4727                         return -EINVAL;
4728                 }
4729         } else {
4730                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4731         }
4732
4733         if (fail_over_mac) {
4734                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4735                                                       fail_over_mac_tbl);
4736                 if (fail_over_mac_value == -1) {
4737                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4738                                arp_validate == NULL ? "NULL" : arp_validate);
4739                         return -EINVAL;
4740                 }
4741
4742                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4743                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4744         } else {
4745                 fail_over_mac_value = BOND_FOM_NONE;
4746         }
4747
4748         /* fill params struct with the proper values */
4749         params->mode = bond_mode;
4750         params->xmit_policy = xmit_hashtype;
4751         params->miimon = miimon;
4752         params->num_peer_notif = num_peer_notif;
4753         params->arp_interval = arp_interval;
4754         params->arp_validate = arp_validate_value;
4755         params->updelay = updelay;
4756         params->downdelay = downdelay;
4757         params->use_carrier = use_carrier;
4758         params->lacp_fast = lacp_fast;
4759         params->primary[0] = 0;
4760         params->primary_reselect = primary_reselect_value;
4761         params->fail_over_mac = fail_over_mac_value;
4762         params->tx_queues = tx_queues;
4763         params->all_slaves_active = all_slaves_active;
4764         params->resend_igmp = resend_igmp;
4765         params->min_links = min_links;
4766
4767         if (primary) {
4768                 strncpy(params->primary, primary, IFNAMSIZ);
4769                 params->primary[IFNAMSIZ - 1] = 0;
4770         }
4771
4772         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4773
4774         return 0;
4775 }
4776
4777 static struct lock_class_key bonding_netdev_xmit_lock_key;
4778 static struct lock_class_key bonding_netdev_addr_lock_key;
4779
4780 static void bond_set_lockdep_class_one(struct net_device *dev,
4781                                        struct netdev_queue *txq,
4782                                        void *_unused)
4783 {
4784         lockdep_set_class(&txq->_xmit_lock,
4785                           &bonding_netdev_xmit_lock_key);
4786 }
4787
4788 static void bond_set_lockdep_class(struct net_device *dev)
4789 {
4790         lockdep_set_class(&dev->addr_list_lock,
4791                           &bonding_netdev_addr_lock_key);
4792         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4793 }
4794
4795 /*
4796  * Called from registration process
4797  */
4798 static int bond_init(struct net_device *bond_dev)
4799 {
4800         struct bonding *bond = netdev_priv(bond_dev);
4801         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4802         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4803
4804         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4805
4806         /*
4807          * Initialize locks that may be required during
4808          * en/deslave operations.  All of the bond_open work
4809          * (of which this is part) should really be moved to
4810          * a phase prior to dev_open
4811          */
4812         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4813         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4814
4815         bond->wq = create_singlethread_workqueue(bond_dev->name);
4816         if (!bond->wq)
4817                 return -ENOMEM;
4818
4819         bond_set_lockdep_class(bond_dev);
4820
4821         list_add_tail(&bond->bond_list, &bn->dev_list);
4822
4823         bond_prepare_sysfs_group(bond);
4824
4825         bond_debug_register(bond);
4826
4827         __hw_addr_init(&bond->mc_list);
4828         return 0;
4829 }
4830
4831 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4832 {
4833         if (tb[IFLA_ADDRESS]) {
4834                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4835                         return -EINVAL;
4836                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4837                         return -EADDRNOTAVAIL;
4838         }
4839         return 0;
4840 }
4841
4842 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[])
4843 {
4844         return tx_queues;
4845 }
4846
4847 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4848         .kind           = "bond",
4849         .priv_size      = sizeof(struct bonding),
4850         .setup          = bond_setup,
4851         .validate       = bond_validate,
4852         .get_tx_queues  = bond_get_tx_queues,
4853 };
4854
4855 /* Create a new bond based on the specified name and bonding parameters.
4856  * If name is NULL, obtain a suitable "bond%d" name for us.
4857  * Caller must NOT hold rtnl_lock; we need to release it here before we
4858  * set up our sysfs entries.
4859  */
4860 int bond_create(struct net *net, const char *name)
4861 {
4862         struct net_device *bond_dev;
4863         int res;
4864
4865         rtnl_lock();
4866
4867         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4868                                    name ? name : "bond%d",
4869                                    bond_setup, tx_queues);
4870         if (!bond_dev) {
4871                 pr_err("%s: eek! can't alloc netdev!\n", name);
4872                 rtnl_unlock();
4873                 return -ENOMEM;
4874         }
4875
4876         dev_net_set(bond_dev, net);
4877         bond_dev->rtnl_link_ops = &bond_link_ops;
4878
4879         res = register_netdevice(bond_dev);
4880
4881         netif_carrier_off(bond_dev);
4882
4883         rtnl_unlock();
4884         if (res < 0)
4885                 bond_destructor(bond_dev);
4886         return res;
4887 }
4888
4889 static int __net_init bond_net_init(struct net *net)
4890 {
4891         struct bond_net *bn = net_generic(net, bond_net_id);
4892
4893         bn->net = net;
4894         INIT_LIST_HEAD(&bn->dev_list);
4895
4896         bond_create_proc_dir(bn);
4897         bond_create_sysfs(bn);
4898
4899         return 0;
4900 }
4901
4902 static void __net_exit bond_net_exit(struct net *net)
4903 {
4904         struct bond_net *bn = net_generic(net, bond_net_id);
4905
4906         bond_destroy_sysfs(bn);
4907         bond_destroy_proc_dir(bn);
4908 }
4909
4910 static struct pernet_operations bond_net_ops = {
4911         .init = bond_net_init,
4912         .exit = bond_net_exit,
4913         .id   = &bond_net_id,
4914         .size = sizeof(struct bond_net),
4915 };
4916
4917 static int __init bonding_init(void)
4918 {
4919         int i;
4920         int res;
4921
4922         pr_info("%s", bond_version);
4923
4924         res = bond_check_params(&bonding_defaults);
4925         if (res)
4926                 goto out;
4927
4928         res = register_pernet_subsys(&bond_net_ops);
4929         if (res)
4930                 goto out;
4931
4932         res = rtnl_link_register(&bond_link_ops);
4933         if (res)
4934                 goto err_link;
4935
4936         bond_create_debugfs();
4937
4938         for (i = 0; i < max_bonds; i++) {
4939                 res = bond_create(&init_net, NULL);
4940                 if (res)
4941                         goto err;
4942         }
4943
4944         register_netdevice_notifier(&bond_netdev_notifier);
4945 out:
4946         return res;
4947 err:
4948         rtnl_link_unregister(&bond_link_ops);
4949 err_link:
4950         unregister_pernet_subsys(&bond_net_ops);
4951         goto out;
4952
4953 }
4954
4955 static void __exit bonding_exit(void)
4956 {
4957         unregister_netdevice_notifier(&bond_netdev_notifier);
4958
4959         bond_destroy_debugfs();
4960
4961         rtnl_link_unregister(&bond_link_ops);
4962         unregister_pernet_subsys(&bond_net_ops);
4963
4964 #ifdef CONFIG_NET_POLL_CONTROLLER
4965         /*
4966          * Make sure we don't have an imbalance on our netpoll blocking
4967          */
4968         WARN_ON(atomic_read(&netpoll_block_tx));
4969 #endif
4970 }
4971
4972 module_init(bonding_init);
4973 module_exit(bonding_exit);
4974 MODULE_LICENSE("GPL");
4975 MODULE_VERSION(DRV_VERSION);
4976 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4977 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4978 MODULE_ALIAS_RTNL_LINK("bond");