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