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 "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon       = BOND_LINK_MON_INTERV;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static int all_slaves_active = 0;
 109 static struct bond_params bonding_defaults;
 110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 111
 112 module_param(max_bonds, int, 0);
 113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 114 module_param(tx_queues, int, 0);
 115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 118                                "failover event (alias of num_unsol_na)");
 119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 121                                "failover event (alias of num_grat_arp)");
 122 module_param(miimon, int, 0);
 123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 124 module_param(updelay, int, 0);
 125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 126 module_param(downdelay, int, 0);
 127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 128                             "in milliseconds");
 129 module_param(use_carrier, int, 0);
 130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 131                               "0 for off, 1 for on (default)");
 132 module_param(mode, charp, 0);
 133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 134                        "1 for active-backup, 2 for balance-xor, "
 135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 136                        "6 for balance-alb");
 137 module_param(primary, charp, 0);
 138 MODULE_PARM_DESC(primary, "Primary network device to use");
 139 module_param(primary_reselect, charp, 0);
 140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 141                                    "once it comes up; "
 142                                    "0 for always (default), "
 143                                    "1 for only if speed of primary is "
 144                                    "better, "
 145                                    "2 for only on active slave "
 146                                    "failure");
 147 module_param(lacp_rate, charp, 0);
 148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 149                             "0 for slow, 1 for fast");
 150 module_param(ad_select, charp, 0);
 151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 152                             "0 for stable (default), 1 for bandwidth, "
 153                             "2 for count");
 154 module_param(min_links, int, 0);
 155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 156
 157 module_param(xmit_hash_policy, charp, 0);
 158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 159                                    "0 for layer 2 (default), 1 for layer 3+4, "
 160                                    "2 for layer 2+3");
 161 module_param(arp_interval, int, 0);
 162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 163 module_param_array(arp_ip_target, charp, NULL, 0);
 164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 165 module_param(arp_validate, charp, 0);
 166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 167                                "0 for none (default), 1 for active, "
 168                                "2 for backup, 3 for all");
 169 module_param(fail_over_mac, charp, 0);
 170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 171                                 "the same MAC; 0 for none (default), "
 172                                 "1 for active, 2 for follow");
 173 module_param(all_slaves_active, int, 0);
 174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 175                                      "by setting active flag for all slaves; "
 176                                      "0 for never (default), 1 for always.");
 177 module_param(resend_igmp, int, 0);
 178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 179                               "link failure");
 180
 181 /*----------------------------- Global variables ----------------------------*/
 182
 183 #ifdef CONFIG_NET_POLL_CONTROLLER
 184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 185 #endif
 186
 187 int bond_net_id __read_mostly;
 188
 189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 190 static int arp_ip_count;
 191 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 193 static int lacp_fast;
 194
 195 const struct bond_parm_tbl bond_lacp_tbl[] = {
 196 {       "slow",         AD_LACP_SLOW},
 197 {       "fast",         AD_LACP_FAST},
 198 {       NULL,           -1},
 199 };
 200
 201 const struct bond_parm_tbl bond_mode_tbl[] = {
 202 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 203 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 204 {       "balance-xor",          BOND_MODE_XOR},
 205 {       "broadcast",            BOND_MODE_BROADCAST},
 206 {       "802.3ad",              BOND_MODE_8023AD},
 207 {       "balance-tlb",          BOND_MODE_TLB},
 208 {       "balance-alb",          BOND_MODE_ALB},
 209 {       NULL,                   -1},
 210 };
 211
 212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 213 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 214 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 215 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 216 {       NULL,                   -1},
 217 };
 218
 219 const struct bond_parm_tbl arp_validate_tbl[] = {
 220 {       "none",                 BOND_ARP_VALIDATE_NONE},
 221 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 222 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 223 {       "all",                  BOND_ARP_VALIDATE_ALL},
 224 {       NULL,                   -1},
 225 };
 226
 227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 228 {       "none",                 BOND_FOM_NONE},
 229 {       "active",               BOND_FOM_ACTIVE},
 230 {       "follow",               BOND_FOM_FOLLOW},
 231 {       NULL,                   -1},
 232 };
 233
 234 const struct bond_parm_tbl pri_reselect_tbl[] = {
 235 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 236 {       "better",               BOND_PRI_RESELECT_BETTER},
 237 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 238 {       NULL,                   -1},
 239 };
 240
 241 struct bond_parm_tbl ad_select_tbl[] = {
 242 {       "stable",       BOND_AD_STABLE},
 243 {       "bandwidth",    BOND_AD_BANDWIDTH},
 244 {       "count",        BOND_AD_COUNT},
 245 {       NULL,           -1},
 246 };
 247
 248 /*-------------------------- Forward declarations ---------------------------*/
 249
 250 static int bond_init(struct net_device *bond_dev);
 251 static void bond_uninit(struct net_device *bond_dev);
 252
 253 /*---------------------------- General routines -----------------------------*/
 254
 255 const char *bond_mode_name(int mode)
 256 {
 257         static const char *names[] = {
 258                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 259                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 260                 [BOND_MODE_XOR] = "load balancing (xor)",
 261                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 262                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 263                 [BOND_MODE_TLB] = "transmit load balancing",
 264                 [BOND_MODE_ALB] = "adaptive load balancing",
 265         };
 266
 267         if (mode < 0 || mode > BOND_MODE_ALB)
 268                 return "unknown";
 269
 270         return names[mode];
 271 }
 272
 273 /*---------------------------------- VLAN -----------------------------------*/
 274
 275 /**
 276  * bond_add_vlan - add a new vlan id on bond
 277  * @bond: bond that got the notification
 278  * @vlan_id: the vlan id to add
 279  *
 280  * Returns -ENOMEM if allocation failed.
 281  */
 282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 283 {
 284         struct vlan_entry *vlan;
 285
 286         pr_debug("bond: %s, vlan id %d\n",
 287                  (bond ? bond->dev->name : "None"), vlan_id);
 288
 289         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 290         if (!vlan)
 291                 return -ENOMEM;
 292
 293         INIT_LIST_HEAD(&vlan->vlan_list);
 294         vlan->vlan_id = vlan_id;
 295
 296         write_lock_bh(&bond->lock);
 297
 298         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 299
 300         write_unlock_bh(&bond->lock);
 301
 302         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 303
 304         return 0;
 305 }
 306
 307 /**
 308  * bond_del_vlan - delete a vlan id from bond
 309  * @bond: bond that got the notification
 310  * @vlan_id: the vlan id to delete
 311  *
 312  * returns -ENODEV if @vlan_id was not found in @bond.
 313  */
 314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 315 {
 316         struct vlan_entry *vlan;
 317         int res = -ENODEV;
 318
 319         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 320
 321         block_netpoll_tx();
 322         write_lock_bh(&bond->lock);
 323
 324         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 325                 if (vlan->vlan_id == vlan_id) {
 326                         list_del(&vlan->vlan_list);
 327
 328                         if (bond_is_lb(bond))
 329                                 bond_alb_clear_vlan(bond, vlan_id);
 330
 331                         pr_debug("removed VLAN ID %d from bond %s\n",
 332                                  vlan_id, bond->dev->name);
 333
 334                         kfree(vlan);
 335
 336                         res = 0;
 337                         goto out;
 338                 }
 339         }
 340
 341         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 342                  vlan_id, bond->dev->name);
 343
 344 out:
 345         write_unlock_bh(&bond->lock);
 346         unblock_netpoll_tx();
 347         return res;
 348 }
 349
 350 /**
 351  * bond_next_vlan - safely skip to the next item in the vlans list.
 352  * @bond: the bond we're working on
 353  * @curr: item we're advancing from
 354  *
 355  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 356  * or @curr->next otherwise (even if it is @curr itself again).
 357  *
 358  * Caller must hold bond->lock
 359  */
 360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 361 {
 362         struct vlan_entry *next, *last;
 363
 364         if (list_empty(&bond->vlan_list))
 365                 return NULL;
 366
 367         if (!curr) {
 368                 next = list_entry(bond->vlan_list.next,
 369                                   struct vlan_entry, vlan_list);
 370         } else {
 371                 last = list_entry(bond->vlan_list.prev,
 372                                   struct vlan_entry, vlan_list);
 373                 if (last == curr) {
 374                         next = list_entry(bond->vlan_list.next,
 375                                           struct vlan_entry, vlan_list);
 376                 } else {
 377                         next = list_entry(curr->vlan_list.next,
 378                                           struct vlan_entry, vlan_list);
 379                 }
 380         }
 381
 382         return next;
 383 }
 384
 385 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
 386
 387 /**
 388  * bond_dev_queue_xmit - Prepare skb for xmit.
 389  *
 390  * @bond: bond device that got this skb for tx.
 391  * @skb: hw accel VLAN tagged skb to transmit
 392  * @slave_dev: slave that is supposed to xmit this skbuff
 393  */
 394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 395                         struct net_device *slave_dev)
 396 {
 397         skb->dev = slave_dev;
 398         skb->priority = 1;
 399
 400         skb->queue_mapping = bond_queue_mapping(skb);
 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 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 412  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 413  * lock because:
 414  * a. This operation is performed in IOCTL context,
 415  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 416  * c. Holding a lock with BH disabled while directly calling a base driver
 417  *    entry point is generally a BAD idea.
 418  *
 419  * The design of synchronization/protection for this operation in the 8021q
 420  * module is good for one or more VLAN devices over a single physical device
 421  * and cannot be extended for a teaming solution like bonding, so there is a
 422  * potential race condition here where a net device from the vlan group might
 423  * be referenced (either by a base driver or the 8021q code) while it is being
 424  * removed from the system. However, it turns out we're not making matters
 425  * worse, and if it works for regular VLAN usage it will work here too.
 426 */
 427
 428 /**
 429  * bond_vlan_rx_register - Propagates registration to slaves
 430  * @bond_dev: bonding net device that got called
 431  * @grp: vlan group being registered
 432  */
 433 static void bond_vlan_rx_register(struct net_device *bond_dev,
 434                                   struct vlan_group *grp)
 435 {
 436         struct bonding *bond = netdev_priv(bond_dev);
 437         struct slave *slave;
 438         int i;
 439
 440         write_lock_bh(&bond->lock);
 441         bond->vlgrp = grp;
 442         write_unlock_bh(&bond->lock);
 443
 444         bond_for_each_slave(bond, slave, i) {
 445                 struct net_device *slave_dev = slave->dev;
 446                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 447
 448                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 449                     slave_ops->ndo_vlan_rx_register) {
 450                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 451                 }
 452         }
 453 }
 454
 455 /**
 456  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 457  * @bond_dev: bonding net device that got called
 458  * @vid: vlan id being added
 459  */
 460 static void bond_vlan_rx_add_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_add_vid) {
 472                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 473                 }
 474         }
 475
 476         res = bond_add_vlan(bond, vid);
 477         if (res) {
 478                 pr_err("%s: Error: Failed to add vlan id %d\n",
 479                        bond_dev->name, vid);
 480         }
 481 }
 482
 483 /**
 484  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 485  * @bond_dev: bonding net device that got called
 486  * @vid: vlan id being removed
 487  */
 488 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 489 {
 490         struct bonding *bond = netdev_priv(bond_dev);
 491         struct slave *slave;
 492         struct net_device *vlan_dev;
 493         int i, res;
 494
 495         bond_for_each_slave(bond, slave, i) {
 496                 struct net_device *slave_dev = slave->dev;
 497                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 498
 499                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 500                     slave_ops->ndo_vlan_rx_kill_vid) {
 501                         /* Save and then restore vlan_dev in the grp array,
 502                          * since the slave's driver might clear it.
 503                          */
 504                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 505                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 506                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 507                 }
 508         }
 509
 510         res = bond_del_vlan(bond, vid);
 511         if (res) {
 512                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 513                        bond_dev->name, vid);
 514         }
 515 }
 516
 517 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 518 {
 519         struct vlan_entry *vlan;
 520         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 521
 522         if (!bond->vlgrp)
 523                 return;
 524
 525         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 526             slave_ops->ndo_vlan_rx_register)
 527                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 528
 529         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 530             !(slave_ops->ndo_vlan_rx_add_vid))
 531                 return;
 532
 533         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 534                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 535 }
 536
 537 static void bond_del_vlans_from_slave(struct bonding *bond,
 538                                       struct net_device *slave_dev)
 539 {
 540         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 541         struct vlan_entry *vlan;
 542         struct net_device *vlan_dev;
 543
 544         if (!bond->vlgrp)
 545                 return;
 546
 547         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 548             !(slave_ops->ndo_vlan_rx_kill_vid))
 549                 goto unreg;
 550
 551         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 552                 if (!vlan->vlan_id)
 553                         continue;
 554                 /* Save and then restore vlan_dev in the grp array,
 555                  * since the slave's driver might clear it.
 556                  */
 557                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 558                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 559                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 560         }
 561
 562 unreg:
 563         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 564             slave_ops->ndo_vlan_rx_register)
 565                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 566 }
 567
 568 /*------------------------------- Link status -------------------------------*/
 569
 570 /*
 571  * Set the carrier state for the master according to the state of its
 572  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 573  * do special 802.3ad magic.
 574  *
 575  * Returns zero if carrier state does not change, nonzero if it does.
 576  */
 577 static int bond_set_carrier(struct bonding *bond)
 578 {
 579         struct slave *slave;
 580         int i;
 581
 582         if (bond->slave_cnt == 0)
 583                 goto down;
 584
 585         if (bond->params.mode == BOND_MODE_8023AD)
 586                 return bond_3ad_set_carrier(bond);
 587
 588         bond_for_each_slave(bond, slave, i) {
 589                 if (slave->link == BOND_LINK_UP) {
 590                         if (!netif_carrier_ok(bond->dev)) {
 591                                 netif_carrier_on(bond->dev);
 592                                 return 1;
 593                         }
 594                         return 0;
 595                 }
 596         }
 597
 598 down:
 599         if (netif_carrier_ok(bond->dev)) {
 600                 netif_carrier_off(bond->dev);
 601                 return 1;
 602         }
 603         return 0;
 604 }
 605
 606 /*
 607  * Get link speed and duplex from the slave's base driver
 608  * using ethtool. If for some reason the call fails or the
 609  * values are invalid, fake speed and duplex to 100/Full
 610  * and return error.
 611  */
 612 static int bond_update_speed_duplex(struct slave *slave)
 613 {
 614         struct net_device *slave_dev = slave->dev;
 615         struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
 616         u32 slave_speed;
 617         int res;
 618
 619         /* Fake speed and duplex */
 620         slave->speed = SPEED_100;
 621         slave->duplex = DUPLEX_FULL;
 622
 623         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 624                 return -1;
 625
 626         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 627         if (res < 0)
 628                 return -1;
 629
 630         slave_speed = ethtool_cmd_speed(&etool);
 631         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 632                 return -1;
 633
 634         switch (etool.duplex) {
 635         case DUPLEX_FULL:
 636         case DUPLEX_HALF:
 637                 break;
 638         default:
 639                 return -1;
 640         }
 641
 642         slave->speed = slave_speed;
 643         slave->duplex = etool.duplex;
 644
 645         return 0;
 646 }
 647
 648 /*
 649  * if <dev> supports MII link status reporting, check its link status.
 650  *
 651  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 652  * depending upon the setting of the use_carrier parameter.
 653  *
 654  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 655  * can't tell and just pretend it is), or 0, meaning that the link is
 656  * down.
 657  *
 658  * If reporting is non-zero, instead of faking link up, return -1 if
 659  * both ETHTOOL and MII ioctls fail (meaning the device does not
 660  * support them).  If use_carrier is set, return whatever it says.
 661  * It'd be nice if there was a good way to tell if a driver supports
 662  * netif_carrier, but there really isn't.
 663  */
 664 static int bond_check_dev_link(struct bonding *bond,
 665                                struct net_device *slave_dev, int reporting)
 666 {
 667         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 668         int (*ioctl)(struct net_device *, struct ifreq *, int);
 669         struct ifreq ifr;
 670         struct mii_ioctl_data *mii;
 671
 672         if (!reporting && !netif_running(slave_dev))
 673                 return 0;
 674
 675         if (bond->params.use_carrier)
 676                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 677
 678         /* Try to get link status using Ethtool first. */
 679         if (slave_dev->ethtool_ops) {
 680                 if (slave_dev->ethtool_ops->get_link) {
 681                         u32 link;
 682
 683                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 684
 685                         return link ? BMSR_LSTATUS : 0;
 686                 }
 687         }
 688
 689         /* Ethtool can't be used, fallback to MII ioctls. */
 690         ioctl = slave_ops->ndo_do_ioctl;
 691         if (ioctl) {
 692                 /* TODO: set pointer to correct ioctl on a per team member */
 693                 /*       bases to make this more efficient. that is, once  */
 694                 /*       we determine the correct ioctl, we will always    */
 695                 /*       call it and not the others for that team          */
 696                 /*       member.                                           */
 697
 698                 /*
 699                  * We cannot assume that SIOCGMIIPHY will also read a
 700                  * register; not all network drivers (e.g., e100)
 701                  * support that.
 702                  */
 703
 704                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 705                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 706                 mii = if_mii(&ifr);
 707                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 708                         mii->reg_num = MII_BMSR;
 709                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 710                                 return mii->val_out & BMSR_LSTATUS;
 711                 }
 712         }
 713
 714         /*
 715          * If reporting, report that either there's no dev->do_ioctl,
 716          * or both SIOCGMIIREG and get_link failed (meaning that we
 717          * cannot report link status).  If not reporting, pretend
 718          * we're ok.
 719          */
 720         return reporting ? -1 : BMSR_LSTATUS;
 721 }
 722
 723 /*----------------------------- Multicast list ------------------------------*/
 724
 725 /*
 726  * Push the promiscuity flag down to appropriate slaves
 727  */
 728 static int bond_set_promiscuity(struct bonding *bond, int inc)
 729 {
 730         int err = 0;
 731         if (USES_PRIMARY(bond->params.mode)) {
 732                 /* write lock already acquired */
 733                 if (bond->curr_active_slave) {
 734                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 735                                                   inc);
 736                 }
 737         } else {
 738                 struct slave *slave;
 739                 int i;
 740                 bond_for_each_slave(bond, slave, i) {
 741                         err = dev_set_promiscuity(slave->dev, inc);
 742                         if (err)
 743                                 return err;
 744                 }
 745         }
 746         return err;
 747 }
 748
 749 /*
 750  * Push the allmulti flag down to all slaves
 751  */
 752 static int bond_set_allmulti(struct bonding *bond, int inc)
 753 {
 754         int err = 0;
 755         if (USES_PRIMARY(bond->params.mode)) {
 756                 /* write lock already acquired */
 757                 if (bond->curr_active_slave) {
 758                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 759                                                inc);
 760                 }
 761         } else {
 762                 struct slave *slave;
 763                 int i;
 764                 bond_for_each_slave(bond, slave, i) {
 765                         err = dev_set_allmulti(slave->dev, inc);
 766                         if (err)
 767                                 return err;
 768                 }
 769         }
 770         return err;
 771 }
 772
 773 /*
 774  * Add a Multicast address to slaves
 775  * according to mode
 776  */
 777 static void bond_mc_add(struct bonding *bond, void *addr)
 778 {
 779         if (USES_PRIMARY(bond->params.mode)) {
 780                 /* write lock already acquired */
 781                 if (bond->curr_active_slave)
 782                         dev_mc_add(bond->curr_active_slave->dev, addr);
 783         } else {
 784                 struct slave *slave;
 785                 int i;
 786
 787                 bond_for_each_slave(bond, slave, i)
 788                         dev_mc_add(slave->dev, addr);
 789         }
 790 }
 791
 792 /*
 793  * Remove a multicast address from slave
 794  * according to mode
 795  */
 796 static void bond_mc_del(struct bonding *bond, void *addr)
 797 {
 798         if (USES_PRIMARY(bond->params.mode)) {
 799                 /* write lock already acquired */
 800                 if (bond->curr_active_slave)
 801                         dev_mc_del(bond->curr_active_slave->dev, addr);
 802         } else {
 803                 struct slave *slave;
 804                 int i;
 805                 bond_for_each_slave(bond, slave, i) {
 806                         dev_mc_del(slave->dev, addr);
 807                 }
 808         }
 809 }
 810
 811
 812 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 813 {
 814         struct in_device *in_dev;
 815
 816         rcu_read_lock();
 817         in_dev = __in_dev_get_rcu(dev);
 818         if (in_dev)
 819                 ip_mc_rejoin_groups(in_dev);
 820         rcu_read_unlock();
 821 }
 822
 823 /*
 824  * Retrieve the list of registered multicast addresses for the bonding
 825  * device and retransmit an IGMP JOIN request to the current active
 826  * slave.
 827  */
 828 static void bond_resend_igmp_join_requests(struct bonding *bond)
 829 {
 830         struct net_device *vlan_dev;
 831         struct vlan_entry *vlan;
 832
 833         read_lock(&bond->lock);
 834
 835         /* rejoin all groups on bond device */
 836         __bond_resend_igmp_join_requests(bond->dev);
 837
 838         /* rejoin all groups on vlan devices */
 839         if (bond->vlgrp) {
 840                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 841                         vlan_dev = vlan_group_get_device(bond->vlgrp,
 842                                                          vlan->vlan_id);
 843                         if (vlan_dev)
 844                                 __bond_resend_igmp_join_requests(vlan_dev);
 845                 }
 846         }
 847
 848         if (--bond->igmp_retrans > 0)
 849                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 850
 851         read_unlock(&bond->lock);
 852 }
 853
 854 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 855 {
 856         struct bonding *bond = container_of(work, struct bonding,
 857                                             mcast_work.work);
 858         bond_resend_igmp_join_requests(bond);
 859 }
 860
 861 /*
 862  * flush all members of flush->mc_list from device dev->mc_list
 863  */
 864 static void bond_mc_list_flush(struct net_device *bond_dev,
 865                                struct net_device *slave_dev)
 866 {
 867         struct bonding *bond = netdev_priv(bond_dev);
 868         struct netdev_hw_addr *ha;
 869
 870         netdev_for_each_mc_addr(ha, bond_dev)
 871                 dev_mc_del(slave_dev, ha->addr);
 872
 873         if (bond->params.mode == BOND_MODE_8023AD) {
 874                 /* del lacpdu mc addr from mc list */
 875                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 876
 877                 dev_mc_del(slave_dev, lacpdu_multicast);
 878         }
 879 }
 880
 881 /*--------------------------- Active slave change ---------------------------*/
 882
 883 /*
 884  * Update the mc list and multicast-related flags for the new and
 885  * old active slaves (if any) according to the multicast mode, and
 886  * promiscuous flags unconditionally.
 887  */
 888 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 889                          struct slave *old_active)
 890 {
 891         struct netdev_hw_addr *ha;
 892
 893         if (!USES_PRIMARY(bond->params.mode))
 894                 /* nothing to do -  mc list is already up-to-date on
 895                  * all slaves
 896                  */
 897                 return;
 898
 899         if (old_active) {
 900                 if (bond->dev->flags & IFF_PROMISC)
 901                         dev_set_promiscuity(old_active->dev, -1);
 902
 903                 if (bond->dev->flags & IFF_ALLMULTI)
 904                         dev_set_allmulti(old_active->dev, -1);
 905
 906                 netdev_for_each_mc_addr(ha, bond->dev)
 907                         dev_mc_del(old_active->dev, ha->addr);
 908         }
 909
 910         if (new_active) {
 911                 /* FIXME: Signal errors upstream. */
 912                 if (bond->dev->flags & IFF_PROMISC)
 913                         dev_set_promiscuity(new_active->dev, 1);
 914
 915                 if (bond->dev->flags & IFF_ALLMULTI)
 916                         dev_set_allmulti(new_active->dev, 1);
 917
 918                 netdev_for_each_mc_addr(ha, bond->dev)
 919                         dev_mc_add(new_active->dev, ha->addr);
 920         }
 921 }
 922
 923 /*
 924  * bond_do_fail_over_mac
 925  *
 926  * Perform special MAC address swapping for fail_over_mac settings
 927  *
 928  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 929  */
 930 static void bond_do_fail_over_mac(struct bonding *bond,
 931                                   struct slave *new_active,
 932                                   struct slave *old_active)
 933         __releases(&bond->curr_slave_lock)
 934         __releases(&bond->lock)
 935         __acquires(&bond->lock)
 936         __acquires(&bond->curr_slave_lock)
 937 {
 938         u8 tmp_mac[ETH_ALEN];
 939         struct sockaddr saddr;
 940         int rv;
 941
 942         switch (bond->params.fail_over_mac) {
 943         case BOND_FOM_ACTIVE:
 944                 if (new_active)
 945                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 946                                new_active->dev->addr_len);
 947                 break;
 948         case BOND_FOM_FOLLOW:
 949                 /*
 950                  * if new_active && old_active, swap them
 951                  * if just old_active, do nothing (going to no active slave)
 952                  * if just new_active, set new_active to bond's MAC
 953                  */
 954                 if (!new_active)
 955                         return;
 956
 957                 write_unlock_bh(&bond->curr_slave_lock);
 958                 read_unlock(&bond->lock);
 959
 960                 if (old_active) {
 961                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 962                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 963                                ETH_ALEN);
 964                         saddr.sa_family = new_active->dev->type;
 965                 } else {
 966                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 967                         saddr.sa_family = bond->dev->type;
 968                 }
 969
 970                 rv = dev_set_mac_address(new_active->dev, &saddr);
 971                 if (rv) {
 972                         pr_err("%s: Error %d setting MAC of slave %s\n",
 973                                bond->dev->name, -rv, new_active->dev->name);
 974                         goto out;
 975                 }
 976
 977                 if (!old_active)
 978                         goto out;
 979
 980                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 981                 saddr.sa_family = old_active->dev->type;
 982
 983                 rv = dev_set_mac_address(old_active->dev, &saddr);
 984                 if (rv)
 985                         pr_err("%s: Error %d setting MAC of slave %s\n",
 986                                bond->dev->name, -rv, new_active->dev->name);
 987 out:
 988                 read_lock(&bond->lock);
 989                 write_lock_bh(&bond->curr_slave_lock);
 990                 break;
 991         default:
 992                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 993                        bond->dev->name, bond->params.fail_over_mac);
 994                 break;
 995         }
 996
 997 }
 998
 999 static bool bond_should_change_active(struct bonding *bond)
1000 {
1001         struct slave *prim = bond->primary_slave;
1002         struct slave *curr = bond->curr_active_slave;
1003
1004         if (!prim || !curr || curr->link != BOND_LINK_UP)
1005                 return true;
1006         if (bond->force_primary) {
1007                 bond->force_primary = false;
1008                 return true;
1009         }
1010         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1011             (prim->speed < curr->speed ||
1012              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1013                 return false;
1014         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1015                 return false;
1016         return true;
1017 }
1018
1019 /**
1020  * find_best_interface - select the best available slave to be the active one
1021  * @bond: our bonding struct
1022  *
1023  * Warning: Caller must hold curr_slave_lock for writing.
1024  */
1025 static struct slave *bond_find_best_slave(struct bonding *bond)
1026 {
1027         struct slave *new_active, *old_active;
1028         struct slave *bestslave = NULL;
1029         int mintime = bond->params.updelay;
1030         int i;
1031
1032         new_active = bond->curr_active_slave;
1033
1034         if (!new_active) { /* there were no active slaves left */
1035                 if (bond->slave_cnt > 0)   /* found one slave */
1036                         new_active = bond->first_slave;
1037                 else
1038                         return NULL; /* still no slave, return NULL */
1039         }
1040
1041         if ((bond->primary_slave) &&
1042             bond->primary_slave->link == BOND_LINK_UP &&
1043             bond_should_change_active(bond)) {
1044                 new_active = bond->primary_slave;
1045         }
1046
1047         /* remember where to stop iterating over the slaves */
1048         old_active = new_active;
1049
1050         bond_for_each_slave_from(bond, new_active, i, old_active) {
1051                 if (new_active->link == BOND_LINK_UP) {
1052                         return new_active;
1053                 } else if (new_active->link == BOND_LINK_BACK &&
1054                            IS_UP(new_active->dev)) {
1055                         /* link up, but waiting for stabilization */
1056                         if (new_active->delay < mintime) {
1057                                 mintime = new_active->delay;
1058                                 bestslave = new_active;
1059                         }
1060                 }
1061         }
1062
1063         return bestslave;
1064 }
1065
1066 static bool bond_should_notify_peers(struct bonding *bond)
1067 {
1068         struct slave *slave = bond->curr_active_slave;
1069
1070         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1071                  bond->dev->name, slave ? slave->dev->name : "NULL");
1072
1073         if (!slave || !bond->send_peer_notif ||
1074             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1075                 return false;
1076
1077         bond->send_peer_notif--;
1078         return true;
1079 }
1080
1081 /**
1082  * change_active_interface - change the active slave into the specified one
1083  * @bond: our bonding struct
1084  * @new: the new slave to make the active one
1085  *
1086  * Set the new slave to the bond's settings and unset them on the old
1087  * curr_active_slave.
1088  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1089  *
1090  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1091  * because it is apparently the best available slave we have, even though its
1092  * updelay hasn't timed out yet.
1093  *
1094  * If new_active is not NULL, caller must hold bond->lock for read and
1095  * curr_slave_lock for write_bh.
1096  */
1097 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1098 {
1099         struct slave *old_active = bond->curr_active_slave;
1100
1101         if (old_active == new_active)
1102                 return;
1103
1104         if (new_active) {
1105                 new_active->jiffies = jiffies;
1106
1107                 if (new_active->link == BOND_LINK_BACK) {
1108                         if (USES_PRIMARY(bond->params.mode)) {
1109                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1110                                         bond->dev->name, new_active->dev->name,
1111                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1112                         }
1113
1114                         new_active->delay = 0;
1115                         new_active->link = BOND_LINK_UP;
1116
1117                         if (bond->params.mode == BOND_MODE_8023AD)
1118                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1119
1120                         if (bond_is_lb(bond))
1121                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1122                 } else {
1123                         if (USES_PRIMARY(bond->params.mode)) {
1124                                 pr_info("%s: making interface %s the new active one.\n",
1125                                         bond->dev->name, new_active->dev->name);
1126                         }
1127                 }
1128         }
1129
1130         if (USES_PRIMARY(bond->params.mode))
1131                 bond_mc_swap(bond, new_active, old_active);
1132
1133         if (bond_is_lb(bond)) {
1134                 bond_alb_handle_active_change(bond, new_active);
1135                 if (old_active)
1136                         bond_set_slave_inactive_flags(old_active);
1137                 if (new_active)
1138                         bond_set_slave_active_flags(new_active);
1139         } else {
1140                 bond->curr_active_slave = new_active;
1141         }
1142
1143         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1144                 if (old_active)
1145                         bond_set_slave_inactive_flags(old_active);
1146
1147                 if (new_active) {
1148                         bool should_notify_peers = false;
1149
1150                         bond_set_slave_active_flags(new_active);
1151
1152                         if (bond->params.fail_over_mac)
1153                                 bond_do_fail_over_mac(bond, new_active,
1154                                                       old_active);
1155
1156                         if (netif_running(bond->dev)) {
1157                                 bond->send_peer_notif =
1158                                         bond->params.num_peer_notif;
1159                                 should_notify_peers =
1160                                         bond_should_notify_peers(bond);
1161                         }
1162
1163                         write_unlock_bh(&bond->curr_slave_lock);
1164                         read_unlock(&bond->lock);
1165
1166                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1167                         if (should_notify_peers)
1168                                 netdev_bonding_change(bond->dev,
1169                                                       NETDEV_NOTIFY_PEERS);
1170
1171                         read_lock(&bond->lock);
1172                         write_lock_bh(&bond->curr_slave_lock);
1173                 }
1174         }
1175
1176         /* resend IGMP joins since active slave has changed or
1177          * all were sent on curr_active_slave.
1178          * resend only if bond is brought up with the affected
1179          * bonding modes and the retransmission is enabled */
1180         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1181             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1182              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1183                 bond->igmp_retrans = bond->params.resend_igmp;
1184                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1185         }
1186 }
1187
1188 /**
1189  * bond_select_active_slave - select a new active slave, if needed
1190  * @bond: our bonding struct
1191  *
1192  * This functions should be called when one of the following occurs:
1193  * - The old curr_active_slave has been released or lost its link.
1194  * - The primary_slave has got its link back.
1195  * - A slave has got its link back and there's no old curr_active_slave.
1196  *
1197  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1198  */
1199 void bond_select_active_slave(struct bonding *bond)
1200 {
1201         struct slave *best_slave;
1202         int rv;
1203
1204         best_slave = bond_find_best_slave(bond);
1205         if (best_slave != bond->curr_active_slave) {
1206                 bond_change_active_slave(bond, best_slave);
1207                 rv = bond_set_carrier(bond);
1208                 if (!rv)
1209                         return;
1210
1211                 if (netif_carrier_ok(bond->dev)) {
1212                         pr_info("%s: first active interface up!\n",
1213                                 bond->dev->name);
1214                 } else {
1215                         pr_info("%s: now running without any active interface !\n",
1216                                 bond->dev->name);
1217                 }
1218         }
1219 }
1220
1221 /*--------------------------- slave list handling ---------------------------*/
1222
1223 /*
1224  * This function attaches the slave to the end of list.
1225  *
1226  * bond->lock held for writing by caller.
1227  */
1228 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1229 {
1230         if (bond->first_slave == NULL) { /* attaching the first slave */
1231                 new_slave->next = new_slave;
1232                 new_slave->prev = new_slave;
1233                 bond->first_slave = new_slave;
1234         } else {
1235                 new_slave->next = bond->first_slave;
1236                 new_slave->prev = bond->first_slave->prev;
1237                 new_slave->next->prev = new_slave;
1238                 new_slave->prev->next = new_slave;
1239         }
1240
1241         bond->slave_cnt++;
1242 }
1243
1244 /*
1245  * This function detaches the slave from the list.
1246  * WARNING: no check is made to verify if the slave effectively
1247  * belongs to <bond>.
1248  * Nothing is freed on return, structures are just unchained.
1249  * If any slave pointer in bond was pointing to <slave>,
1250  * it should be changed by the calling function.
1251  *
1252  * bond->lock held for writing by caller.
1253  */
1254 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1255 {
1256         if (slave->next)
1257                 slave->next->prev = slave->prev;
1258
1259         if (slave->prev)
1260                 slave->prev->next = slave->next;
1261
1262         if (bond->first_slave == slave) { /* slave is the first slave */
1263                 if (bond->slave_cnt > 1) { /* there are more slave */
1264                         bond->first_slave = slave->next;
1265                 } else {
1266                         bond->first_slave = NULL; /* slave was the last one */
1267                 }
1268         }
1269
1270         slave->next = NULL;
1271         slave->prev = NULL;
1272         bond->slave_cnt--;
1273 }
1274
1275 #ifdef CONFIG_NET_POLL_CONTROLLER
1276 static inline int slave_enable_netpoll(struct slave *slave)
1277 {
1278         struct netpoll *np;
1279         int err = 0;
1280
1281         np = kzalloc(sizeof(*np), GFP_KERNEL);
1282         err = -ENOMEM;
1283         if (!np)
1284                 goto out;
1285
1286         np->dev = slave->dev;
1287         strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1288         err = __netpoll_setup(np);
1289         if (err) {
1290                 kfree(np);
1291                 goto out;
1292         }
1293         slave->np = np;
1294 out:
1295         return err;
1296 }
1297 static inline void slave_disable_netpoll(struct slave *slave)
1298 {
1299         struct netpoll *np = slave->np;
1300
1301         if (!np)
1302                 return;
1303
1304         slave->np = NULL;
1305         synchronize_rcu_bh();
1306         __netpoll_cleanup(np);
1307         kfree(np);
1308 }
1309 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1310 {
1311         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1312                 return false;
1313         if (!slave_dev->netdev_ops->ndo_poll_controller)
1314                 return false;
1315         return true;
1316 }
1317
1318 static void bond_poll_controller(struct net_device *bond_dev)
1319 {
1320 }
1321
1322 static void __bond_netpoll_cleanup(struct bonding *bond)
1323 {
1324         struct slave *slave;
1325         int i;
1326
1327         bond_for_each_slave(bond, slave, i)
1328                 if (IS_UP(slave->dev))
1329                         slave_disable_netpoll(slave);
1330 }
1331 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1332 {
1333         struct bonding *bond = netdev_priv(bond_dev);
1334
1335         read_lock(&bond->lock);
1336         __bond_netpoll_cleanup(bond);
1337         read_unlock(&bond->lock);
1338 }
1339
1340 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1341 {
1342         struct bonding *bond = netdev_priv(dev);
1343         struct slave *slave;
1344         int i, err = 0;
1345
1346         read_lock(&bond->lock);
1347         bond_for_each_slave(bond, slave, i) {
1348                 err = slave_enable_netpoll(slave);
1349                 if (err) {
1350                         __bond_netpoll_cleanup(bond);
1351                         break;
1352                 }
1353         }
1354         read_unlock(&bond->lock);
1355         return err;
1356 }
1357
1358 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1359 {
1360         return bond->dev->npinfo;
1361 }
1362
1363 #else
1364 static inline int slave_enable_netpoll(struct slave *slave)
1365 {
1366         return 0;
1367 }
1368 static inline void slave_disable_netpoll(struct slave *slave)
1369 {
1370 }
1371 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1372 {
1373 }
1374 #endif
1375
1376 /*---------------------------------- IOCTL ----------------------------------*/
1377
1378 static int bond_sethwaddr(struct net_device *bond_dev,
1379                           struct net_device *slave_dev)
1380 {
1381         pr_debug("bond_dev=%p\n", bond_dev);
1382         pr_debug("slave_dev=%p\n", slave_dev);
1383         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1384         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1385         return 0;
1386 }
1387
1388 static u32 bond_fix_features(struct net_device *dev, u32 features)
1389 {
1390         struct slave *slave;
1391         struct bonding *bond = netdev_priv(dev);
1392         u32 mask;
1393         int i;
1394
1395         read_lock(&bond->lock);
1396
1397         if (!bond->first_slave) {
1398                 /* Disable adding VLANs to empty bond. But why? --mq */
1399                 features |= NETIF_F_VLAN_CHALLENGED;
1400                 goto out;
1401         }
1402
1403         mask = features;
1404         features &= ~NETIF_F_ONE_FOR_ALL;
1405         features |= NETIF_F_ALL_FOR_ALL;
1406
1407         bond_for_each_slave(bond, slave, i) {
1408                 features = netdev_increment_features(features,
1409                                                      slave->dev->features,
1410                                                      mask);
1411         }
1412
1413 out:
1414         read_unlock(&bond->lock);
1415         return features;
1416 }
1417
1418 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1419                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1420                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1421
1422 static void bond_compute_features(struct bonding *bond)
1423 {
1424         struct slave *slave;
1425         struct net_device *bond_dev = bond->dev;
1426         u32 vlan_features = BOND_VLAN_FEATURES;
1427         unsigned short max_hard_header_len = ETH_HLEN;
1428         int i;
1429
1430         read_lock(&bond->lock);
1431
1432         if (!bond->first_slave)
1433                 goto done;
1434
1435         bond_for_each_slave(bond, slave, i) {
1436                 vlan_features = netdev_increment_features(vlan_features,
1437                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1438
1439                 if (slave->dev->hard_header_len > max_hard_header_len)
1440                         max_hard_header_len = slave->dev->hard_header_len;
1441         }
1442
1443 done:
1444         bond_dev->vlan_features = vlan_features;
1445         bond_dev->hard_header_len = max_hard_header_len;
1446
1447         read_unlock(&bond->lock);
1448
1449         netdev_change_features(bond_dev);
1450 }
1451
1452 static void bond_setup_by_slave(struct net_device *bond_dev,
1453                                 struct net_device *slave_dev)
1454 {
1455         struct bonding *bond = netdev_priv(bond_dev);
1456
1457         bond_dev->header_ops        = slave_dev->header_ops;
1458
1459         bond_dev->type              = slave_dev->type;
1460         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1461         bond_dev->addr_len          = slave_dev->addr_len;
1462
1463         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1464                 slave_dev->addr_len);
1465         bond->setup_by_slave = 1;
1466 }
1467
1468 /* On bonding slaves other than the currently active slave, suppress
1469  * duplicates except for alb non-mcast/bcast.
1470  */
1471 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1472                                             struct slave *slave,
1473                                             struct bonding *bond)
1474 {
1475         if (bond_is_slave_inactive(slave)) {
1476                 if (bond->params.mode == BOND_MODE_ALB &&
1477                     skb->pkt_type != PACKET_BROADCAST &&
1478                     skb->pkt_type != PACKET_MULTICAST)
1479                         return false;
1480                 return true;
1481         }
1482         return false;
1483 }
1484
1485 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1486 {
1487         struct sk_buff *skb = *pskb;
1488         struct slave *slave;
1489         struct bonding *bond;
1490
1491         skb = skb_share_check(skb, GFP_ATOMIC);
1492         if (unlikely(!skb))
1493                 return RX_HANDLER_CONSUMED;
1494
1495         *pskb = skb;
1496
1497         slave = bond_slave_get_rcu(skb->dev);
1498         bond = slave->bond;
1499
1500         if (bond->params.arp_interval)
1501                 slave->dev->last_rx = jiffies;
1502
1503         if (bond->recv_probe) {
1504                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1505
1506                 if (likely(nskb)) {
1507                         bond->recv_probe(nskb, bond, slave);
1508                         dev_kfree_skb(nskb);
1509                 }
1510         }
1511
1512         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1513                 return RX_HANDLER_EXACT;
1514         }
1515
1516         skb->dev = bond->dev;
1517
1518         if (bond->params.mode == BOND_MODE_ALB &&
1519             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1520             skb->pkt_type == PACKET_HOST) {
1521
1522                 if (unlikely(skb_cow_head(skb,
1523                                           skb->data - skb_mac_header(skb)))) {
1524                         kfree_skb(skb);
1525                         return RX_HANDLER_CONSUMED;
1526                 }
1527                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1528         }
1529
1530         return RX_HANDLER_ANOTHER;
1531 }
1532
1533 /* enslave device <slave> to bond device <master> */
1534 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1535 {
1536         struct bonding *bond = netdev_priv(bond_dev);
1537         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1538         struct slave *new_slave = NULL;
1539         struct netdev_hw_addr *ha;
1540         struct sockaddr addr;
1541         int link_reporting;
1542         int res = 0;
1543
1544         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1545                 slave_ops->ndo_do_ioctl == NULL) {
1546                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1547                            bond_dev->name, slave_dev->name);
1548         }
1549
1550         /* already enslaved */
1551         if (slave_dev->flags & IFF_SLAVE) {
1552                 pr_debug("Error, Device was already enslaved\n");
1553                 return -EBUSY;
1554         }
1555
1556         /* vlan challenged mutual exclusion */
1557         /* no need to lock since we're protected by rtnl_lock */
1558         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1559                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1560                 if (bond->vlgrp) {
1561                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1562                                bond_dev->name, slave_dev->name, bond_dev->name);
1563                         return -EPERM;
1564                 } else {
1565                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1566                                    bond_dev->name, slave_dev->name,
1567                                    slave_dev->name, bond_dev->name);
1568                 }
1569         } else {
1570                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1571         }
1572
1573         /*
1574          * Old ifenslave binaries are no longer supported.  These can
1575          * be identified with moderate accuracy by the state of the slave:
1576          * the current ifenslave will set the interface down prior to
1577          * enslaving it; the old ifenslave will not.
1578          */
1579         if ((slave_dev->flags & IFF_UP)) {
1580                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1581                        slave_dev->name);
1582                 res = -EPERM;
1583                 goto err_undo_flags;
1584         }
1585
1586         /* set bonding device ether type by slave - bonding netdevices are
1587          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1588          * there is a need to override some of the type dependent attribs/funcs.
1589          *
1590          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1591          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1592          */
1593         if (bond->slave_cnt == 0) {
1594                 if (bond_dev->type != slave_dev->type) {
1595                         pr_debug("%s: change device type from %d to %d\n",
1596                                  bond_dev->name,
1597                                  bond_dev->type, slave_dev->type);
1598
1599                         res = netdev_bonding_change(bond_dev,
1600                                                     NETDEV_PRE_TYPE_CHANGE);
1601                         res = notifier_to_errno(res);
1602                         if (res) {
1603                                 pr_err("%s: refused to change device type\n",
1604                                        bond_dev->name);
1605                                 res = -EBUSY;
1606                                 goto err_undo_flags;
1607                         }
1608
1609                         /* Flush unicast and multicast addresses */
1610                         dev_uc_flush(bond_dev);
1611                         dev_mc_flush(bond_dev);
1612
1613                         if (slave_dev->type != ARPHRD_ETHER)
1614                                 bond_setup_by_slave(bond_dev, slave_dev);
1615                         else
1616                                 ether_setup(bond_dev);
1617
1618                         netdev_bonding_change(bond_dev,
1619                                               NETDEV_POST_TYPE_CHANGE);
1620                 }
1621         } else if (bond_dev->type != slave_dev->type) {
1622                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1623                        slave_dev->name,
1624                        slave_dev->type, bond_dev->type);
1625                 res = -EINVAL;
1626                 goto err_undo_flags;
1627         }
1628
1629         if (slave_ops->ndo_set_mac_address == NULL) {
1630                 if (bond->slave_cnt == 0) {
1631                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1632                                    bond_dev->name);
1633                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1634                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1635                         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",
1636                                bond_dev->name);
1637                         res = -EOPNOTSUPP;
1638                         goto err_undo_flags;
1639                 }
1640         }
1641
1642         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1643
1644         /* If this is the first slave, then we need to set the master's hardware
1645          * address to be the same as the slave's. */
1646         if (is_zero_ether_addr(bond->dev->dev_addr))
1647                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1648                        slave_dev->addr_len);
1649
1650
1651         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1652         if (!new_slave) {
1653                 res = -ENOMEM;
1654                 goto err_undo_flags;
1655         }
1656
1657         /*
1658          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1659          * is set via sysfs or module option if desired.
1660          */
1661         new_slave->queue_id = 0;
1662
1663         /* Save slave's original mtu and then set it to match the bond */
1664         new_slave->original_mtu = slave_dev->mtu;
1665         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1666         if (res) {
1667                 pr_debug("Error %d calling dev_set_mtu\n", res);
1668                 goto err_free;
1669         }
1670
1671         /*
1672          * Save slave's original ("permanent") mac address for modes
1673          * that need it, and for restoring it upon release, and then
1674          * set it to the master's address
1675          */
1676         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1677
1678         if (!bond->params.fail_over_mac) {
1679                 /*
1680                  * Set slave to master's mac address.  The application already
1681                  * set the master's mac address to that of the first slave
1682                  */
1683                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1684                 addr.sa_family = slave_dev->type;
1685                 res = dev_set_mac_address(slave_dev, &addr);
1686                 if (res) {
1687                         pr_debug("Error %d calling set_mac_address\n", res);
1688                         goto err_restore_mtu;
1689                 }
1690         }
1691
1692         res = netdev_set_bond_master(slave_dev, bond_dev);
1693         if (res) {
1694                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1695                 goto err_restore_mac;
1696         }
1697
1698         /* open the slave since the application closed it */
1699         res = dev_open(slave_dev);
1700         if (res) {
1701                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1702                 goto err_unset_master;
1703         }
1704
1705         new_slave->bond = bond;
1706         new_slave->dev = slave_dev;
1707         slave_dev->priv_flags |= IFF_BONDING;
1708
1709         if (bond_is_lb(bond)) {
1710                 /* bond_alb_init_slave() must be called before all other stages since
1711                  * it might fail and we do not want to have to undo everything
1712                  */
1713                 res = bond_alb_init_slave(bond, new_slave);
1714                 if (res)
1715                         goto err_close;
1716         }
1717
1718         /* If the mode USES_PRIMARY, then the new slave gets the
1719          * master's promisc (and mc) settings only if it becomes the
1720          * curr_active_slave, and that is taken care of later when calling
1721          * bond_change_active()
1722          */
1723         if (!USES_PRIMARY(bond->params.mode)) {
1724                 /* set promiscuity level to new slave */
1725                 if (bond_dev->flags & IFF_PROMISC) {
1726                         res = dev_set_promiscuity(slave_dev, 1);
1727                         if (res)
1728                                 goto err_close;
1729                 }
1730
1731                 /* set allmulti level to new slave */
1732                 if (bond_dev->flags & IFF_ALLMULTI) {
1733                         res = dev_set_allmulti(slave_dev, 1);
1734                         if (res)
1735                                 goto err_close;
1736                 }
1737
1738                 netif_addr_lock_bh(bond_dev);
1739                 /* upload master's mc_list to new slave */
1740                 netdev_for_each_mc_addr(ha, bond_dev)
1741                         dev_mc_add(slave_dev, ha->addr);
1742                 netif_addr_unlock_bh(bond_dev);
1743         }
1744
1745         if (bond->params.mode == BOND_MODE_8023AD) {
1746                 /* add lacpdu mc addr to mc list */
1747                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1748
1749                 dev_mc_add(slave_dev, lacpdu_multicast);
1750         }
1751
1752         bond_add_vlans_on_slave(bond, slave_dev);
1753
1754         write_lock_bh(&bond->lock);
1755
1756         bond_attach_slave(bond, new_slave);
1757
1758         new_slave->delay = 0;
1759         new_slave->link_failure_count = 0;
1760
1761         write_unlock_bh(&bond->lock);
1762
1763         bond_compute_features(bond);
1764
1765         read_lock(&bond->lock);
1766
1767         new_slave->last_arp_rx = jiffies;
1768
1769         if (bond->params.miimon && !bond->params.use_carrier) {
1770                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1771
1772                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1773                         /*
1774                          * miimon is set but a bonded network driver
1775                          * does not support ETHTOOL/MII and
1776                          * arp_interval is not set.  Note: if
1777                          * use_carrier is enabled, we will never go
1778                          * here (because netif_carrier is always
1779                          * supported); thus, we don't need to change
1780                          * the messages for netif_carrier.
1781                          */
1782                         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",
1783                                bond_dev->name, slave_dev->name);
1784                 } else if (link_reporting == -1) {
1785                         /* unable get link status using mii/ethtool */
1786                         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",
1787                                    bond_dev->name, slave_dev->name);
1788                 }
1789         }
1790
1791         /* check for initial state */
1792         if (!bond->params.miimon ||
1793             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1794                 if (bond->params.updelay) {
1795                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1796                         new_slave->link  = BOND_LINK_BACK;
1797                         new_slave->delay = bond->params.updelay;
1798                 } else {
1799                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1800                         new_slave->link  = BOND_LINK_UP;
1801                 }
1802                 new_slave->jiffies = jiffies;
1803         } else {
1804                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1805                 new_slave->link  = BOND_LINK_DOWN;
1806         }
1807
1808         if (bond_update_speed_duplex(new_slave) &&
1809             (new_slave->link != BOND_LINK_DOWN)) {
1810                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1811                            bond_dev->name, new_slave->dev->name);
1812
1813                 if (bond->params.mode == BOND_MODE_8023AD) {
1814                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1815                                    bond_dev->name);
1816                 }
1817         }
1818
1819         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1820                 /* if there is a primary slave, remember it */
1821                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1822                         bond->primary_slave = new_slave;
1823                         bond->force_primary = true;
1824                 }
1825         }
1826
1827         write_lock_bh(&bond->curr_slave_lock);
1828
1829         switch (bond->params.mode) {
1830         case BOND_MODE_ACTIVEBACKUP:
1831                 bond_set_slave_inactive_flags(new_slave);
1832                 bond_select_active_slave(bond);
1833                 break;
1834         case BOND_MODE_8023AD:
1835                 /* in 802.3ad mode, the internal mechanism
1836                  * will activate the slaves in the selected
1837                  * aggregator
1838                  */
1839                 bond_set_slave_inactive_flags(new_slave);
1840                 /* if this is the first slave */
1841                 if (bond->slave_cnt == 1) {
1842                         SLAVE_AD_INFO(new_slave).id = 1;
1843                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1844                          * can be called only after the mac address of the bond is set
1845                          */
1846                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1847                 } else {
1848                         SLAVE_AD_INFO(new_slave).id =
1849                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1850                 }
1851
1852                 bond_3ad_bind_slave(new_slave);
1853                 break;
1854         case BOND_MODE_TLB:
1855         case BOND_MODE_ALB:
1856                 bond_set_active_slave(new_slave);
1857                 bond_set_slave_inactive_flags(new_slave);
1858                 bond_select_active_slave(bond);
1859                 break;
1860         default:
1861                 pr_debug("This slave is always active in trunk mode\n");
1862
1863                 /* always active in trunk mode */
1864                 bond_set_active_slave(new_slave);
1865
1866                 /* In trunking mode there is little meaning to curr_active_slave
1867                  * anyway (it holds no special properties of the bond device),
1868                  * so we can change it without calling change_active_interface()
1869                  */
1870                 if (!bond->curr_active_slave)
1871                         bond->curr_active_slave = new_slave;
1872
1873                 break;
1874         } /* switch(bond_mode) */
1875
1876         write_unlock_bh(&bond->curr_slave_lock);
1877
1878         bond_set_carrier(bond);
1879
1880 #ifdef CONFIG_NET_POLL_CONTROLLER
1881         slave_dev->npinfo = bond_netpoll_info(bond);
1882         if (slave_dev->npinfo) {
1883                 if (slave_enable_netpoll(new_slave)) {
1884                         read_unlock(&bond->lock);
1885                         pr_info("Error, %s: master_dev is using netpoll, "
1886                                  "but new slave device does not support netpoll.\n",
1887                                  bond_dev->name);
1888                         res = -EBUSY;
1889                         goto err_close;
1890                 }
1891         }
1892 #endif
1893
1894         read_unlock(&bond->lock);
1895
1896         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1897         if (res)
1898                 goto err_close;
1899
1900         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1901                                          new_slave);
1902         if (res) {
1903                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1904                 goto err_dest_symlinks;
1905         }
1906
1907         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1908                 bond_dev->name, slave_dev->name,
1909                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1910                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1911
1912         /* enslave is successful */
1913         return 0;
1914
1915 /* Undo stages on error */
1916 err_dest_symlinks:
1917         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1918
1919 err_close:
1920         dev_close(slave_dev);
1921
1922 err_unset_master:
1923         netdev_set_bond_master(slave_dev, NULL);
1924
1925 err_restore_mac:
1926         if (!bond->params.fail_over_mac) {
1927                 /* XXX TODO - fom follow mode needs to change master's
1928                  * MAC if this slave's MAC is in use by the bond, or at
1929                  * least print a warning.
1930                  */
1931                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1932                 addr.sa_family = slave_dev->type;
1933                 dev_set_mac_address(slave_dev, &addr);
1934         }
1935
1936 err_restore_mtu:
1937         dev_set_mtu(slave_dev, new_slave->original_mtu);
1938
1939 err_free:
1940         kfree(new_slave);
1941
1942 err_undo_flags:
1943         bond_compute_features(bond);
1944
1945         return res;
1946 }
1947
1948 /*
1949  * Try to release the slave device <slave> from the bond device <master>
1950  * It is legal to access curr_active_slave without a lock because all the function
1951  * is write-locked.
1952  *
1953  * The rules for slave state should be:
1954  *   for Active/Backup:
1955  *     Active stays on all backups go down
1956  *   for Bonded connections:
1957  *     The first up interface should be left on and all others downed.
1958  */
1959 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1960 {
1961         struct bonding *bond = netdev_priv(bond_dev);
1962         struct slave *slave, *oldcurrent;
1963         struct sockaddr addr;
1964         u32 old_features = bond_dev->features;
1965
1966         /* slave is not a slave or master is not master of this slave */
1967         if (!(slave_dev->flags & IFF_SLAVE) ||
1968             (slave_dev->master != bond_dev)) {
1969                 pr_err("%s: Error: cannot release %s.\n",
1970                        bond_dev->name, slave_dev->name);
1971                 return -EINVAL;
1972         }
1973
1974         block_netpoll_tx();
1975         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1976         write_lock_bh(&bond->lock);
1977
1978         slave = bond_get_slave_by_dev(bond, slave_dev);
1979         if (!slave) {
1980                 /* not a slave of this bond */
1981                 pr_info("%s: %s not enslaved\n",
1982                         bond_dev->name, slave_dev->name);
1983                 write_unlock_bh(&bond->lock);
1984                 unblock_netpoll_tx();
1985                 return -EINVAL;
1986         }
1987
1988         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1989          * for this slave anymore.
1990          */
1991         netdev_rx_handler_unregister(slave_dev);
1992         write_unlock_bh(&bond->lock);
1993         synchronize_net();
1994         write_lock_bh(&bond->lock);
1995
1996         if (!bond->params.fail_over_mac) {
1997                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1998                     bond->slave_cnt > 1)
1999                         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",
2000                                    bond_dev->name, slave_dev->name,
2001                                    slave->perm_hwaddr,
2002                                    bond_dev->name, slave_dev->name);
2003         }
2004
2005         /* Inform AD package of unbinding of slave. */
2006         if (bond->params.mode == BOND_MODE_8023AD) {
2007                 /* must be called before the slave is
2008                  * detached from the list
2009                  */
2010                 bond_3ad_unbind_slave(slave);
2011         }
2012
2013         pr_info("%s: releasing %s interface %s\n",
2014                 bond_dev->name,
2015                 bond_is_active_slave(slave) ? "active" : "backup",
2016                 slave_dev->name);
2017
2018         oldcurrent = bond->curr_active_slave;
2019
2020         bond->current_arp_slave = NULL;
2021
2022         /* release the slave from its bond */
2023         bond_detach_slave(bond, slave);
2024
2025         if (bond->primary_slave == slave)
2026                 bond->primary_slave = NULL;
2027
2028         if (oldcurrent == slave)
2029                 bond_change_active_slave(bond, NULL);
2030
2031         if (bond_is_lb(bond)) {
2032                 /* Must be called only after the slave has been
2033                  * detached from the list and the curr_active_slave
2034                  * has been cleared (if our_slave == old_current),
2035                  * but before a new active slave is selected.
2036                  */
2037                 write_unlock_bh(&bond->lock);
2038                 bond_alb_deinit_slave(bond, slave);
2039                 write_lock_bh(&bond->lock);
2040         }
2041
2042         if (oldcurrent == slave) {
2043                 /*
2044                  * Note that we hold RTNL over this sequence, so there
2045                  * is no concern that another slave add/remove event
2046                  * will interfere.
2047                  */
2048                 write_unlock_bh(&bond->lock);
2049                 read_lock(&bond->lock);
2050                 write_lock_bh(&bond->curr_slave_lock);
2051
2052                 bond_select_active_slave(bond);
2053
2054                 write_unlock_bh(&bond->curr_slave_lock);
2055                 read_unlock(&bond->lock);
2056                 write_lock_bh(&bond->lock);
2057         }
2058
2059         if (bond->slave_cnt == 0) {
2060                 bond_set_carrier(bond);
2061
2062                 /* if the last slave was removed, zero the mac address
2063                  * of the master so it will be set by the application
2064                  * to the mac address of the first slave
2065                  */
2066                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2067
2068                 if (bond->vlgrp) {
2069                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2070                                    bond_dev->name, bond_dev->name);
2071                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2072                                    bond_dev->name);
2073                 }
2074         }
2075
2076         write_unlock_bh(&bond->lock);
2077         unblock_netpoll_tx();
2078
2079         bond_compute_features(bond);
2080         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2081             (old_features & NETIF_F_VLAN_CHALLENGED))
2082                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2083                         bond_dev->name, slave_dev->name, bond_dev->name);
2084
2085         /* must do this from outside any spinlocks */
2086         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2087
2088         bond_del_vlans_from_slave(bond, slave_dev);
2089
2090         /* If the mode USES_PRIMARY, then we should only remove its
2091          * promisc and mc settings if it was the curr_active_slave, but that was
2092          * already taken care of above when we detached the slave
2093          */
2094         if (!USES_PRIMARY(bond->params.mode)) {
2095                 /* unset promiscuity level from slave */
2096                 if (bond_dev->flags & IFF_PROMISC)
2097                         dev_set_promiscuity(slave_dev, -1);
2098
2099                 /* unset allmulti level from slave */
2100                 if (bond_dev->flags & IFF_ALLMULTI)
2101                         dev_set_allmulti(slave_dev, -1);
2102
2103                 /* flush master's mc_list from slave */
2104                 netif_addr_lock_bh(bond_dev);
2105                 bond_mc_list_flush(bond_dev, slave_dev);
2106                 netif_addr_unlock_bh(bond_dev);
2107         }
2108
2109         netdev_set_bond_master(slave_dev, NULL);
2110
2111         slave_disable_netpoll(slave);
2112
2113         /* close slave before restoring its mac address */
2114         dev_close(slave_dev);
2115
2116         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2117                 /* restore original ("permanent") mac address */
2118                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2119                 addr.sa_family = slave_dev->type;
2120                 dev_set_mac_address(slave_dev, &addr);
2121         }
2122
2123         dev_set_mtu(slave_dev, slave->original_mtu);
2124
2125         slave_dev->priv_flags &= ~IFF_BONDING;
2126
2127         kfree(slave);
2128
2129         return 0;  /* deletion OK */
2130 }
2131
2132 /*
2133 * First release a slave and then destroy the bond if no more slaves are left.
2134 * Must be under rtnl_lock when this function is called.
2135 */
2136 static int  bond_release_and_destroy(struct net_device *bond_dev,
2137                                      struct net_device *slave_dev)
2138 {
2139         struct bonding *bond = netdev_priv(bond_dev);
2140         int ret;
2141
2142         ret = bond_release(bond_dev, slave_dev);
2143         if ((ret == 0) && (bond->slave_cnt == 0)) {
2144                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2145                 pr_info("%s: destroying bond %s.\n",
2146                         bond_dev->name, bond_dev->name);
2147                 unregister_netdevice(bond_dev);
2148         }
2149         return ret;
2150 }
2151
2152 /*
2153  * This function releases all slaves.
2154  */
2155 static int bond_release_all(struct net_device *bond_dev)
2156 {
2157         struct bonding *bond = netdev_priv(bond_dev);
2158         struct slave *slave;
2159         struct net_device *slave_dev;
2160         struct sockaddr addr;
2161
2162         write_lock_bh(&bond->lock);
2163
2164         netif_carrier_off(bond_dev);
2165
2166         if (bond->slave_cnt == 0)
2167                 goto out;
2168
2169         bond->current_arp_slave = NULL;
2170         bond->primary_slave = NULL;
2171         bond_change_active_slave(bond, NULL);
2172
2173         while ((slave = bond->first_slave) != NULL) {
2174                 /* Inform AD package of unbinding of slave
2175                  * before slave is detached from the list.
2176                  */
2177                 if (bond->params.mode == BOND_MODE_8023AD)
2178                         bond_3ad_unbind_slave(slave);
2179
2180                 slave_dev = slave->dev;
2181                 bond_detach_slave(bond, slave);
2182
2183                 /* now that the slave is detached, unlock and perform
2184                  * all the undo steps that should not be called from
2185                  * within a lock.
2186                  */
2187                 write_unlock_bh(&bond->lock);
2188
2189                 /* unregister rx_handler early so bond_handle_frame wouldn't
2190                  * be called for this slave anymore.
2191                  */
2192                 netdev_rx_handler_unregister(slave_dev);
2193                 synchronize_net();
2194
2195                 if (bond_is_lb(bond)) {
2196                         /* must be called only after the slave
2197                          * has been detached from the list
2198                          */
2199                         bond_alb_deinit_slave(bond, slave);
2200                 }
2201
2202                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2203                 bond_del_vlans_from_slave(bond, slave_dev);
2204
2205                 /* If the mode USES_PRIMARY, then we should only remove its
2206                  * promisc and mc settings if it was the curr_active_slave, but that was
2207                  * already taken care of above when we detached the slave
2208                  */
2209                 if (!USES_PRIMARY(bond->params.mode)) {
2210                         /* unset promiscuity level from slave */
2211                         if (bond_dev->flags & IFF_PROMISC)
2212                                 dev_set_promiscuity(slave_dev, -1);
2213
2214                         /* unset allmulti level from slave */
2215                         if (bond_dev->flags & IFF_ALLMULTI)
2216                                 dev_set_allmulti(slave_dev, -1);
2217
2218                         /* flush master's mc_list from slave */
2219                         netif_addr_lock_bh(bond_dev);
2220                         bond_mc_list_flush(bond_dev, slave_dev);
2221                         netif_addr_unlock_bh(bond_dev);
2222                 }
2223
2224                 netdev_set_bond_master(slave_dev, NULL);
2225
2226                 slave_disable_netpoll(slave);
2227
2228                 /* close slave before restoring its mac address */
2229                 dev_close(slave_dev);
2230
2231                 if (!bond->params.fail_over_mac) {
2232                         /* restore original ("permanent") mac address*/
2233                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2234                         addr.sa_family = slave_dev->type;
2235                         dev_set_mac_address(slave_dev, &addr);
2236                 }
2237
2238                 kfree(slave);
2239
2240                 /* re-acquire the lock before getting the next slave */
2241                 write_lock_bh(&bond->lock);
2242         }
2243
2244         /* zero the mac address of the master so it will be
2245          * set by the application to the mac address of the
2246          * first slave
2247          */
2248         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2249
2250         if (bond->vlgrp) {
2251                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2252                            bond_dev->name, bond_dev->name);
2253                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2254                            bond_dev->name);
2255         }
2256
2257         pr_info("%s: released all slaves\n", bond_dev->name);
2258
2259 out:
2260         write_unlock_bh(&bond->lock);
2261
2262         bond_compute_features(bond);
2263
2264         return 0;
2265 }
2266
2267 /*
2268  * This function changes the active slave to slave <slave_dev>.
2269  * It returns -EINVAL in the following cases.
2270  *  - <slave_dev> is not found in the list.
2271  *  - There is not active slave now.
2272  *  - <slave_dev> is already active.
2273  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2274  *  - <slave_dev> is not running.
2275  * In these cases, this function does nothing.
2276  * In the other cases, current_slave pointer is changed and 0 is returned.
2277  */
2278 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2279 {
2280         struct bonding *bond = netdev_priv(bond_dev);
2281         struct slave *old_active = NULL;
2282         struct slave *new_active = NULL;
2283         int res = 0;
2284
2285         if (!USES_PRIMARY(bond->params.mode))
2286                 return -EINVAL;
2287
2288         /* Verify that master_dev is indeed the master of slave_dev */
2289         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2290                 return -EINVAL;
2291
2292         read_lock(&bond->lock);
2293
2294         read_lock(&bond->curr_slave_lock);
2295         old_active = bond->curr_active_slave;
2296         read_unlock(&bond->curr_slave_lock);
2297
2298         new_active = bond_get_slave_by_dev(bond, slave_dev);
2299
2300         /*
2301          * Changing to the current active: do nothing; return success.
2302          */
2303         if (new_active && (new_active == old_active)) {
2304                 read_unlock(&bond->lock);
2305                 return 0;
2306         }
2307
2308         if ((new_active) &&
2309             (old_active) &&
2310             (new_active->link == BOND_LINK_UP) &&
2311             IS_UP(new_active->dev)) {
2312                 block_netpoll_tx();
2313                 write_lock_bh(&bond->curr_slave_lock);
2314                 bond_change_active_slave(bond, new_active);
2315                 write_unlock_bh(&bond->curr_slave_lock);
2316                 unblock_netpoll_tx();
2317         } else
2318                 res = -EINVAL;
2319
2320         read_unlock(&bond->lock);
2321
2322         return res;
2323 }
2324
2325 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2326 {
2327         struct bonding *bond = netdev_priv(bond_dev);
2328
2329         info->bond_mode = bond->params.mode;
2330         info->miimon = bond->params.miimon;
2331
2332         read_lock(&bond->lock);
2333         info->num_slaves = bond->slave_cnt;
2334         read_unlock(&bond->lock);
2335
2336         return 0;
2337 }
2338
2339 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2340 {
2341         struct bonding *bond = netdev_priv(bond_dev);
2342         struct slave *slave;
2343         int i, res = -ENODEV;
2344
2345         read_lock(&bond->lock);
2346
2347         bond_for_each_slave(bond, slave, i) {
2348                 if (i == (int)info->slave_id) {
2349                         res = 0;
2350                         strcpy(info->slave_name, slave->dev->name);
2351                         info->link = slave->link;
2352                         info->state = bond_slave_state(slave);
2353                         info->link_failure_count = slave->link_failure_count;
2354                         break;
2355                 }
2356         }
2357
2358         read_unlock(&bond->lock);
2359
2360         return res;
2361 }
2362
2363 /*-------------------------------- Monitoring -------------------------------*/
2364
2365
2366 static int bond_miimon_inspect(struct bonding *bond)
2367 {
2368         struct slave *slave;
2369         int i, link_state, commit = 0;
2370         bool ignore_updelay;
2371
2372         ignore_updelay = !bond->curr_active_slave ? true : false;
2373
2374         bond_for_each_slave(bond, slave, i) {
2375                 slave->new_link = BOND_LINK_NOCHANGE;
2376
2377                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2378
2379                 switch (slave->link) {
2380                 case BOND_LINK_UP:
2381                         if (link_state)
2382                                 continue;
2383
2384                         slave->link = BOND_LINK_FAIL;
2385                         slave->delay = bond->params.downdelay;
2386                         if (slave->delay) {
2387                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2388                                         bond->dev->name,
2389                                         (bond->params.mode ==
2390                                          BOND_MODE_ACTIVEBACKUP) ?
2391                                         (bond_is_active_slave(slave) ?
2392                                          "active " : "backup ") : "",
2393                                         slave->dev->name,
2394                                         bond->params.downdelay * bond->params.miimon);
2395                         }
2396                         /*FALLTHRU*/
2397                 case BOND_LINK_FAIL:
2398                         if (link_state) {
2399                                 /*
2400                                  * recovered before downdelay expired
2401                                  */
2402                                 slave->link = BOND_LINK_UP;
2403                                 slave->jiffies = jiffies;
2404                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2405                                         bond->dev->name,
2406                                         (bond->params.downdelay - slave->delay) *
2407                                         bond->params.miimon,
2408                                         slave->dev->name);
2409                                 continue;
2410                         }
2411
2412                         if (slave->delay <= 0) {
2413                                 slave->new_link = BOND_LINK_DOWN;
2414                                 commit++;
2415                                 continue;
2416                         }
2417
2418                         slave->delay--;
2419                         break;
2420
2421                 case BOND_LINK_DOWN:
2422                         if (!link_state)
2423                                 continue;
2424
2425                         slave->link = BOND_LINK_BACK;
2426                         slave->delay = bond->params.updelay;
2427
2428                         if (slave->delay) {
2429                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2430                                         bond->dev->name, slave->dev->name,
2431                                         ignore_updelay ? 0 :
2432                                         bond->params.updelay *
2433                                         bond->params.miimon);
2434                         }
2435                         /*FALLTHRU*/
2436                 case BOND_LINK_BACK:
2437                         if (!link_state) {
2438                                 slave->link = BOND_LINK_DOWN;
2439                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2440                                         bond->dev->name,
2441                                         (bond->params.updelay - slave->delay) *
2442                                         bond->params.miimon,
2443                                         slave->dev->name);
2444
2445                                 continue;
2446                         }
2447
2448                         if (ignore_updelay)
2449                                 slave->delay = 0;
2450
2451                         if (slave->delay <= 0) {
2452                                 slave->new_link = BOND_LINK_UP;
2453                                 commit++;
2454                                 ignore_updelay = false;
2455                                 continue;
2456                         }
2457
2458                         slave->delay--;
2459                         break;
2460                 }
2461         }
2462
2463         return commit;
2464 }
2465
2466 static void bond_miimon_commit(struct bonding *bond)
2467 {
2468         struct slave *slave;
2469         int i;
2470
2471         bond_for_each_slave(bond, slave, i) {
2472                 switch (slave->new_link) {
2473                 case BOND_LINK_NOCHANGE:
2474                         continue;
2475
2476                 case BOND_LINK_UP:
2477                         slave->link = BOND_LINK_UP;
2478                         slave->jiffies = jiffies;
2479
2480                         if (bond->params.mode == BOND_MODE_8023AD) {
2481                                 /* prevent it from being the active one */
2482                                 bond_set_backup_slave(slave);
2483                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2484                                 /* make it immediately active */
2485                                 bond_set_active_slave(slave);
2486                         } else if (slave != bond->primary_slave) {
2487                                 /* prevent it from being the active one */
2488                                 bond_set_backup_slave(slave);
2489                         }
2490
2491                         bond_update_speed_duplex(slave);
2492
2493                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2494                                 bond->dev->name, slave->dev->name,
2495                                 slave->speed, slave->duplex ? "full" : "half");
2496
2497                         /* notify ad that the link status has changed */
2498                         if (bond->params.mode == BOND_MODE_8023AD)
2499                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2500
2501                         if (bond_is_lb(bond))
2502                                 bond_alb_handle_link_change(bond, slave,
2503                                                             BOND_LINK_UP);
2504
2505                         if (!bond->curr_active_slave ||
2506                             (slave == bond->primary_slave))
2507                                 goto do_failover;
2508
2509                         continue;
2510
2511                 case BOND_LINK_DOWN:
2512                         if (slave->link_failure_count < UINT_MAX)
2513                                 slave->link_failure_count++;
2514
2515                         slave->link = BOND_LINK_DOWN;
2516
2517                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2518                             bond->params.mode == BOND_MODE_8023AD)
2519                                 bond_set_slave_inactive_flags(slave);
2520
2521                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2522                                 bond->dev->name, slave->dev->name);
2523
2524                         if (bond->params.mode == BOND_MODE_8023AD)
2525                                 bond_3ad_handle_link_change(slave,
2526                                                             BOND_LINK_DOWN);
2527
2528                         if (bond_is_lb(bond))
2529                                 bond_alb_handle_link_change(bond, slave,
2530                                                             BOND_LINK_DOWN);
2531
2532                         if (slave == bond->curr_active_slave)
2533                                 goto do_failover;
2534
2535                         continue;
2536
2537                 default:
2538                         pr_err("%s: invalid new link %d on slave %s\n",
2539                                bond->dev->name, slave->new_link,
2540                                slave->dev->name);
2541                         slave->new_link = BOND_LINK_NOCHANGE;
2542
2543                         continue;
2544                 }
2545
2546 do_failover:
2547                 ASSERT_RTNL();
2548                 block_netpoll_tx();
2549                 write_lock_bh(&bond->curr_slave_lock);
2550                 bond_select_active_slave(bond);
2551                 write_unlock_bh(&bond->curr_slave_lock);
2552                 unblock_netpoll_tx();
2553         }
2554
2555         bond_set_carrier(bond);
2556 }
2557
2558 /*
2559  * bond_mii_monitor
2560  *
2561  * Really a wrapper that splits the mii monitor into two phases: an
2562  * inspection, then (if inspection indicates something needs to be done)
2563  * an acquisition of appropriate locks followed by a commit phase to
2564  * implement whatever link state changes are indicated.
2565  */
2566 void bond_mii_monitor(struct work_struct *work)
2567 {
2568         struct bonding *bond = container_of(work, struct bonding,
2569                                             mii_work.work);
2570         bool should_notify_peers = false;
2571
2572         read_lock(&bond->lock);
2573         if (bond->kill_timers)
2574                 goto out;
2575
2576         if (bond->slave_cnt == 0)
2577                 goto re_arm;
2578
2579         should_notify_peers = bond_should_notify_peers(bond);
2580
2581         if (bond_miimon_inspect(bond)) {
2582                 read_unlock(&bond->lock);
2583                 rtnl_lock();
2584                 read_lock(&bond->lock);
2585
2586                 bond_miimon_commit(bond);
2587
2588                 read_unlock(&bond->lock);
2589                 rtnl_unlock();  /* might sleep, hold no other locks */
2590                 read_lock(&bond->lock);
2591         }
2592
2593 re_arm:
2594         if (bond->params.miimon)
2595                 queue_delayed_work(bond->wq, &bond->mii_work,
2596                                    msecs_to_jiffies(bond->params.miimon));
2597 out:
2598         read_unlock(&bond->lock);
2599
2600         if (should_notify_peers) {
2601                 rtnl_lock();
2602                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2603                 rtnl_unlock();
2604         }
2605 }
2606
2607 static __be32 bond_glean_dev_ip(struct net_device *dev)
2608 {
2609         struct in_device *idev;
2610         struct in_ifaddr *ifa;
2611         __be32 addr = 0;
2612
2613         if (!dev)
2614                 return 0;
2615
2616         rcu_read_lock();
2617         idev = __in_dev_get_rcu(dev);
2618         if (!idev)
2619                 goto out;
2620
2621         ifa = idev->ifa_list;
2622         if (!ifa)
2623                 goto out;
2624
2625         addr = ifa->ifa_local;
2626 out:
2627         rcu_read_unlock();
2628         return addr;
2629 }
2630
2631 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2632 {
2633         struct vlan_entry *vlan;
2634
2635         if (ip == bond->master_ip)
2636                 return 1;
2637
2638         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2639                 if (ip == vlan->vlan_ip)
2640                         return 1;
2641         }
2642
2643         return 0;
2644 }
2645
2646 /*
2647  * We go to the (large) trouble of VLAN tagging ARP frames because
2648  * switches in VLAN mode (especially if ports are configured as
2649  * "native" to a VLAN) might not pass non-tagged frames.
2650  */
2651 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2652 {
2653         struct sk_buff *skb;
2654
2655         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2656                  slave_dev->name, dest_ip, src_ip, vlan_id);
2657
2658         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2659                          NULL, slave_dev->dev_addr, NULL);
2660
2661         if (!skb) {
2662                 pr_err("ARP packet allocation failed\n");
2663                 return;
2664         }
2665         if (vlan_id) {
2666                 skb = vlan_put_tag(skb, vlan_id);
2667                 if (!skb) {
2668                         pr_err("failed to insert VLAN tag\n");
2669                         return;
2670                 }
2671         }
2672         arp_xmit(skb);
2673 }
2674
2675
2676 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2677 {
2678         int i, vlan_id;
2679         __be32 *targets = bond->params.arp_targets;
2680         struct vlan_entry *vlan;
2681         struct net_device *vlan_dev;
2682         struct rtable *rt;
2683
2684         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2685                 if (!targets[i])
2686                         break;
2687                 pr_debug("basa: target %x\n", targets[i]);
2688                 if (!bond->vlgrp) {
2689                         pr_debug("basa: empty vlan: arp_send\n");
2690                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2691                                       bond->master_ip, 0);
2692                         continue;
2693                 }
2694
2695                 /*
2696                  * If VLANs are configured, we do a route lookup to
2697                  * determine which VLAN interface would be used, so we
2698                  * can tag the ARP with the proper VLAN tag.
2699                  */
2700                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2701                                      RTO_ONLINK, 0);
2702                 if (IS_ERR(rt)) {
2703                         if (net_ratelimit()) {
2704                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2705                                            bond->dev->name, &targets[i]);
2706                         }
2707                         continue;
2708                 }
2709
2710                 /*
2711                  * This target is not on a VLAN
2712                  */
2713                 if (rt->dst.dev == bond->dev) {
2714                         ip_rt_put(rt);
2715                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2716                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2717                                       bond->master_ip, 0);
2718                         continue;
2719                 }
2720
2721                 vlan_id = 0;
2722                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2723                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2724                         if (vlan_dev == rt->dst.dev) {
2725                                 vlan_id = vlan->vlan_id;
2726                                 pr_debug("basa: vlan match on %s %d\n",
2727                                        vlan_dev->name, vlan_id);
2728                                 break;
2729                         }
2730                 }
2731
2732                 if (vlan_id) {
2733                         ip_rt_put(rt);
2734                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2735                                       vlan->vlan_ip, vlan_id);
2736                         continue;
2737                 }
2738
2739                 if (net_ratelimit()) {
2740                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2741                                    bond->dev->name, &targets[i],
2742                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2743                 }
2744                 ip_rt_put(rt);
2745         }
2746 }
2747
2748 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2749 {
2750         int i;
2751         __be32 *targets = bond->params.arp_targets;
2752
2753         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2754                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2755                          &sip, &tip, i, &targets[i],
2756                          bond_has_this_ip(bond, tip));
2757                 if (sip == targets[i]) {
2758                         if (bond_has_this_ip(bond, tip))
2759                                 slave->last_arp_rx = jiffies;
2760                         return;
2761                 }
2762         }
2763 }
2764
2765 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2766                          struct slave *slave)
2767 {
2768         struct arphdr *arp;
2769         unsigned char *arp_ptr;
2770         __be32 sip, tip;
2771
2772         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2773                 return;
2774
2775         read_lock(&bond->lock);
2776
2777         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2778                  bond->dev->name, skb->dev->name);
2779
2780         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2781                 goto out_unlock;
2782
2783         arp = arp_hdr(skb);
2784         if (arp->ar_hln != bond->dev->addr_len ||
2785             skb->pkt_type == PACKET_OTHERHOST ||
2786             skb->pkt_type == PACKET_LOOPBACK ||
2787             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2788             arp->ar_pro != htons(ETH_P_IP) ||
2789             arp->ar_pln != 4)
2790                 goto out_unlock;
2791
2792         arp_ptr = (unsigned char *)(arp + 1);
2793         arp_ptr += bond->dev->addr_len;
2794         memcpy(&sip, arp_ptr, 4);
2795         arp_ptr += 4 + bond->dev->addr_len;
2796         memcpy(&tip, arp_ptr, 4);
2797
2798         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2799                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2800                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2801                  &sip, &tip);
2802
2803         /*
2804          * Backup slaves won't see the ARP reply, but do come through
2805          * here for each ARP probe (so we swap the sip/tip to validate
2806          * the probe).  In a "redundant switch, common router" type of
2807          * configuration, the ARP probe will (hopefully) travel from
2808          * the active, through one switch, the router, then the other
2809          * switch before reaching the backup.
2810          */
2811         if (bond_is_active_slave(slave))
2812                 bond_validate_arp(bond, slave, sip, tip);
2813         else
2814                 bond_validate_arp(bond, slave, tip, sip);
2815
2816 out_unlock:
2817         read_unlock(&bond->lock);
2818 }
2819
2820 /*
2821  * this function is called regularly to monitor each slave's link
2822  * ensuring that traffic is being sent and received when arp monitoring
2823  * is used in load-balancing mode. if the adapter has been dormant, then an
2824  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2825  * arp monitoring in active backup mode.
2826  */
2827 void bond_loadbalance_arp_mon(struct work_struct *work)
2828 {
2829         struct bonding *bond = container_of(work, struct bonding,
2830                                             arp_work.work);
2831         struct slave *slave, *oldcurrent;
2832         int do_failover = 0;
2833         int delta_in_ticks;
2834         int i;
2835
2836         read_lock(&bond->lock);
2837
2838         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2839
2840         if (bond->kill_timers)
2841                 goto out;
2842
2843         if (bond->slave_cnt == 0)
2844                 goto re_arm;
2845
2846         read_lock(&bond->curr_slave_lock);
2847         oldcurrent = bond->curr_active_slave;
2848         read_unlock(&bond->curr_slave_lock);
2849
2850         /* see if any of the previous devices are up now (i.e. they have
2851          * xmt and rcv traffic). the curr_active_slave does not come into
2852          * the picture unless it is null. also, slave->jiffies is not needed
2853          * here because we send an arp on each slave and give a slave as
2854          * long as it needs to get the tx/rx within the delta.
2855          * TODO: what about up/down delay in arp mode? it wasn't here before
2856          *       so it can wait
2857          */
2858         bond_for_each_slave(bond, slave, i) {
2859                 unsigned long trans_start = dev_trans_start(slave->dev);
2860
2861                 if (slave->link != BOND_LINK_UP) {
2862                         if (time_in_range(jiffies,
2863                                 trans_start - delta_in_ticks,
2864                                 trans_start + delta_in_ticks) &&
2865                             time_in_range(jiffies,
2866                                 slave->dev->last_rx - delta_in_ticks,
2867                                 slave->dev->last_rx + delta_in_ticks)) {
2868
2869                                 slave->link  = BOND_LINK_UP;
2870                                 bond_set_active_slave(slave);
2871
2872                                 /* primary_slave has no meaning in round-robin
2873                                  * mode. the window of a slave being up and
2874                                  * curr_active_slave being null after enslaving
2875                                  * is closed.
2876                                  */
2877                                 if (!oldcurrent) {
2878                                         pr_info("%s: link status definitely up for interface %s, ",
2879                                                 bond->dev->name,
2880                                                 slave->dev->name);
2881                                         do_failover = 1;
2882                                 } else {
2883                                         pr_info("%s: interface %s is now up\n",
2884                                                 bond->dev->name,
2885                                                 slave->dev->name);
2886                                 }
2887                         }
2888                 } else {
2889                         /* slave->link == BOND_LINK_UP */
2890
2891                         /* not all switches will respond to an arp request
2892                          * when the source ip is 0, so don't take the link down
2893                          * if we don't know our ip yet
2894                          */
2895                         if (!time_in_range(jiffies,
2896                                 trans_start - delta_in_ticks,
2897                                 trans_start + 2 * delta_in_ticks) ||
2898                             !time_in_range(jiffies,
2899                                 slave->dev->last_rx - delta_in_ticks,
2900                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2901
2902                                 slave->link  = BOND_LINK_DOWN;
2903                                 bond_set_backup_slave(slave);
2904
2905                                 if (slave->link_failure_count < UINT_MAX)
2906                                         slave->link_failure_count++;
2907
2908                                 pr_info("%s: interface %s is now down.\n",
2909                                         bond->dev->name,
2910                                         slave->dev->name);
2911
2912                                 if (slave == oldcurrent)
2913                                         do_failover = 1;
2914                         }
2915                 }
2916
2917                 /* note: if switch is in round-robin mode, all links
2918                  * must tx arp to ensure all links rx an arp - otherwise
2919                  * links may oscillate or not come up at all; if switch is
2920                  * in something like xor mode, there is nothing we can
2921                  * do - all replies will be rx'ed on same link causing slaves
2922                  * to be unstable during low/no traffic periods
2923                  */
2924                 if (IS_UP(slave->dev))
2925                         bond_arp_send_all(bond, slave);
2926         }
2927
2928         if (do_failover) {
2929                 block_netpoll_tx();
2930                 write_lock_bh(&bond->curr_slave_lock);
2931
2932                 bond_select_active_slave(bond);
2933
2934                 write_unlock_bh(&bond->curr_slave_lock);
2935                 unblock_netpoll_tx();
2936         }
2937
2938 re_arm:
2939         if (bond->params.arp_interval)
2940                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2941 out:
2942         read_unlock(&bond->lock);
2943 }
2944
2945 /*
2946  * Called to inspect slaves for active-backup mode ARP monitor link state
2947  * changes.  Sets new_link in slaves to specify what action should take
2948  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2949  * to link states must be committed.
2950  *
2951  * Called with bond->lock held for read.
2952  */
2953 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2954 {
2955         struct slave *slave;
2956         int i, commit = 0;
2957         unsigned long trans_start;
2958
2959         bond_for_each_slave(bond, slave, i) {
2960                 slave->new_link = BOND_LINK_NOCHANGE;
2961
2962                 if (slave->link != BOND_LINK_UP) {
2963                         if (time_in_range(jiffies,
2964                                 slave_last_rx(bond, slave) - delta_in_ticks,
2965                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2966
2967                                 slave->new_link = BOND_LINK_UP;
2968                                 commit++;
2969                         }
2970
2971                         continue;
2972                 }
2973
2974                 /*
2975                  * Give slaves 2*delta after being enslaved or made
2976                  * active.  This avoids bouncing, as the last receive
2977                  * times need a full ARP monitor cycle to be updated.
2978                  */
2979                 if (time_in_range(jiffies,
2980                                   slave->jiffies - delta_in_ticks,
2981                                   slave->jiffies + 2 * delta_in_ticks))
2982                         continue;
2983
2984                 /*
2985                  * Backup slave is down if:
2986                  * - No current_arp_slave AND
2987                  * - more than 3*delta since last receive AND
2988                  * - the bond has an IP address
2989                  *
2990                  * Note: a non-null current_arp_slave indicates
2991                  * the curr_active_slave went down and we are
2992                  * searching for a new one; under this condition
2993                  * we only take the curr_active_slave down - this
2994                  * gives each slave a chance to tx/rx traffic
2995                  * before being taken out
2996                  */
2997                 if (!bond_is_active_slave(slave) &&
2998                     !bond->current_arp_slave &&
2999                     !time_in_range(jiffies,
3000                         slave_last_rx(bond, slave) - delta_in_ticks,
3001                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3002
3003                         slave->new_link = BOND_LINK_DOWN;
3004                         commit++;
3005                 }
3006
3007                 /*
3008                  * Active slave is down if:
3009                  * - more than 2*delta since transmitting OR
3010                  * - (more than 2*delta since receive AND
3011                  *    the bond has an IP address)
3012                  */
3013                 trans_start = dev_trans_start(slave->dev);
3014                 if (bond_is_active_slave(slave) &&
3015                     (!time_in_range(jiffies,
3016                         trans_start - delta_in_ticks,
3017                         trans_start + 2 * delta_in_ticks) ||
3018                      !time_in_range(jiffies,
3019                         slave_last_rx(bond, slave) - delta_in_ticks,
3020                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3021
3022                         slave->new_link = BOND_LINK_DOWN;
3023                         commit++;
3024                 }
3025         }
3026
3027         return commit;
3028 }
3029
3030 /*
3031  * Called to commit link state changes noted by inspection step of
3032  * active-backup mode ARP monitor.
3033  *
3034  * Called with RTNL and bond->lock for read.
3035  */
3036 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3037 {
3038         struct slave *slave;
3039         int i;
3040         unsigned long trans_start;
3041
3042         bond_for_each_slave(bond, slave, i) {
3043                 switch (slave->new_link) {
3044                 case BOND_LINK_NOCHANGE:
3045                         continue;
3046
3047                 case BOND_LINK_UP:
3048                         trans_start = dev_trans_start(slave->dev);
3049                         if ((!bond->curr_active_slave &&
3050                              time_in_range(jiffies,
3051                                            trans_start - delta_in_ticks,
3052                                            trans_start + delta_in_ticks)) ||
3053                             bond->curr_active_slave != slave) {
3054                                 slave->link = BOND_LINK_UP;
3055                                 bond->current_arp_slave = NULL;
3056
3057                                 pr_info("%s: link status definitely up for interface %s.\n",
3058                                         bond->dev->name, slave->dev->name);
3059
3060                                 if (!bond->curr_active_slave ||
3061                                     (slave == bond->primary_slave))
3062                                         goto do_failover;
3063
3064                         }
3065
3066                         continue;
3067
3068                 case BOND_LINK_DOWN:
3069                         if (slave->link_failure_count < UINT_MAX)
3070                                 slave->link_failure_count++;
3071
3072                         slave->link = BOND_LINK_DOWN;
3073                         bond_set_slave_inactive_flags(slave);
3074
3075                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3076                                 bond->dev->name, slave->dev->name);
3077
3078                         if (slave == bond->curr_active_slave) {
3079                                 bond->current_arp_slave = NULL;
3080                                 goto do_failover;
3081                         }
3082
3083                         continue;
3084
3085                 default:
3086                         pr_err("%s: impossible: new_link %d on slave %s\n",
3087                                bond->dev->name, slave->new_link,
3088                                slave->dev->name);
3089                         continue;
3090                 }
3091
3092 do_failover:
3093                 ASSERT_RTNL();
3094                 block_netpoll_tx();
3095                 write_lock_bh(&bond->curr_slave_lock);
3096                 bond_select_active_slave(bond);
3097                 write_unlock_bh(&bond->curr_slave_lock);
3098                 unblock_netpoll_tx();
3099         }
3100
3101         bond_set_carrier(bond);
3102 }
3103
3104 /*
3105  * Send ARP probes for active-backup mode ARP monitor.
3106  *
3107  * Called with bond->lock held for read.
3108  */
3109 static void bond_ab_arp_probe(struct bonding *bond)
3110 {
3111         struct slave *slave;
3112         int i;
3113
3114         read_lock(&bond->curr_slave_lock);
3115
3116         if (bond->current_arp_slave && bond->curr_active_slave)
3117                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3118                         bond->current_arp_slave->dev->name,
3119                         bond->curr_active_slave->dev->name);
3120
3121         if (bond->curr_active_slave) {
3122                 bond_arp_send_all(bond, bond->curr_active_slave);
3123                 read_unlock(&bond->curr_slave_lock);
3124                 return;
3125         }
3126
3127         read_unlock(&bond->curr_slave_lock);
3128
3129         /* if we don't have a curr_active_slave, search for the next available
3130          * backup slave from the current_arp_slave and make it the candidate
3131          * for becoming the curr_active_slave
3132          */
3133
3134         if (!bond->current_arp_slave) {
3135                 bond->current_arp_slave = bond->first_slave;
3136                 if (!bond->current_arp_slave)
3137                         return;
3138         }
3139
3140         bond_set_slave_inactive_flags(bond->current_arp_slave);
3141
3142         /* search for next candidate */
3143         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3144                 if (IS_UP(slave->dev)) {
3145                         slave->link = BOND_LINK_BACK;
3146                         bond_set_slave_active_flags(slave);
3147                         bond_arp_send_all(bond, slave);
3148                         slave->jiffies = jiffies;
3149                         bond->current_arp_slave = slave;
3150                         break;
3151                 }
3152
3153                 /* if the link state is up at this point, we
3154                  * mark it down - this can happen if we have
3155                  * simultaneous link failures and
3156                  * reselect_active_interface doesn't make this
3157                  * one the current slave so it is still marked
3158                  * up when it is actually down
3159                  */
3160                 if (slave->link == BOND_LINK_UP) {
3161                         slave->link = BOND_LINK_DOWN;
3162                         if (slave->link_failure_count < UINT_MAX)
3163                                 slave->link_failure_count++;
3164
3165                         bond_set_slave_inactive_flags(slave);
3166
3167                         pr_info("%s: backup interface %s is now down.\n",
3168                                 bond->dev->name, slave->dev->name);
3169                 }
3170         }
3171 }
3172
3173 void bond_activebackup_arp_mon(struct work_struct *work)
3174 {
3175         struct bonding *bond = container_of(work, struct bonding,
3176                                             arp_work.work);
3177         bool should_notify_peers = false;
3178         int delta_in_ticks;
3179
3180         read_lock(&bond->lock);
3181
3182         if (bond->kill_timers)
3183                 goto out;
3184
3185         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3186
3187         if (bond->slave_cnt == 0)
3188                 goto re_arm;
3189
3190         should_notify_peers = bond_should_notify_peers(bond);
3191
3192         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3193                 read_unlock(&bond->lock);
3194                 rtnl_lock();
3195                 read_lock(&bond->lock);
3196
3197                 bond_ab_arp_commit(bond, delta_in_ticks);
3198
3199                 read_unlock(&bond->lock);
3200                 rtnl_unlock();
3201                 read_lock(&bond->lock);
3202         }
3203
3204         bond_ab_arp_probe(bond);
3205
3206 re_arm:
3207         if (bond->params.arp_interval)
3208                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3209 out:
3210         read_unlock(&bond->lock);
3211
3212         if (should_notify_peers) {
3213                 rtnl_lock();
3214                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3215                 rtnl_unlock();
3216         }
3217 }
3218
3219 /*-------------------------- netdev event handling --------------------------*/
3220
3221 /*
3222  * Change device name
3223  */
3224 static int bond_event_changename(struct bonding *bond)
3225 {
3226         bond_remove_proc_entry(bond);
3227         bond_create_proc_entry(bond);
3228
3229         bond_debug_reregister(bond);
3230
3231         return NOTIFY_DONE;
3232 }
3233
3234 static int bond_master_netdev_event(unsigned long event,
3235                                     struct net_device *bond_dev)
3236 {
3237         struct bonding *event_bond = netdev_priv(bond_dev);
3238
3239         switch (event) {
3240         case NETDEV_CHANGENAME:
3241                 return bond_event_changename(event_bond);
3242         default:
3243                 break;
3244         }
3245
3246         return NOTIFY_DONE;
3247 }
3248
3249 static int bond_slave_netdev_event(unsigned long event,
3250                                    struct net_device *slave_dev)
3251 {
3252         struct net_device *bond_dev = slave_dev->master;
3253         struct bonding *bond = netdev_priv(bond_dev);
3254
3255         switch (event) {
3256         case NETDEV_UNREGISTER:
3257                 if (bond_dev) {
3258                         if (bond->setup_by_slave)
3259                                 bond_release_and_destroy(bond_dev, slave_dev);
3260                         else
3261                                 bond_release(bond_dev, slave_dev);
3262                 }
3263                 break;
3264         case NETDEV_CHANGE:
3265                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3266                         struct slave *slave;
3267
3268                         slave = bond_get_slave_by_dev(bond, slave_dev);
3269                         if (slave) {
3270                                 u32 old_speed = slave->speed;
3271                                 u8  old_duplex = slave->duplex;
3272
3273                                 bond_update_speed_duplex(slave);
3274
3275                                 if (bond_is_lb(bond))
3276                                         break;
3277
3278                                 if (old_speed != slave->speed)
3279                                         bond_3ad_adapter_speed_changed(slave);
3280                                 if (old_duplex != slave->duplex)
3281                                         bond_3ad_adapter_duplex_changed(slave);
3282                         }
3283                 }
3284
3285                 break;
3286         case NETDEV_DOWN:
3287                 /*
3288                  * ... Or is it this?
3289                  */
3290                 break;
3291         case NETDEV_CHANGEMTU:
3292                 /*
3293                  * TODO: Should slaves be allowed to
3294                  * independently alter their MTU?  For
3295                  * an active-backup bond, slaves need
3296                  * not be the same type of device, so
3297                  * MTUs may vary.  For other modes,
3298                  * slaves arguably should have the
3299                  * same MTUs. To do this, we'd need to
3300                  * take over the slave's change_mtu
3301                  * function for the duration of their
3302                  * servitude.
3303                  */
3304                 break;
3305         case NETDEV_CHANGENAME:
3306                 /*
3307                  * TODO: handle changing the primary's name
3308                  */
3309                 break;
3310         case NETDEV_FEAT_CHANGE:
3311                 bond_compute_features(bond);
3312                 break;
3313         default:
3314                 break;
3315         }
3316
3317         return NOTIFY_DONE;
3318 }
3319
3320 /*
3321  * bond_netdev_event: handle netdev notifier chain events.
3322  *
3323  * This function receives events for the netdev chain.  The caller (an
3324  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3325  * locks for us to safely manipulate the slave devices (RTNL lock,
3326  * dev_probe_lock).
3327  */
3328 static int bond_netdev_event(struct notifier_block *this,
3329                              unsigned long event, void *ptr)
3330 {
3331         struct net_device *event_dev = (struct net_device *)ptr;
3332
3333         pr_debug("event_dev: %s, event: %lx\n",
3334                  event_dev ? event_dev->name : "None",
3335                  event);
3336
3337         if (!(event_dev->priv_flags & IFF_BONDING))
3338                 return NOTIFY_DONE;
3339
3340         if (event_dev->flags & IFF_MASTER) {
3341                 pr_debug("IFF_MASTER\n");
3342                 return bond_master_netdev_event(event, event_dev);
3343         }
3344
3345         if (event_dev->flags & IFF_SLAVE) {
3346                 pr_debug("IFF_SLAVE\n");
3347                 return bond_slave_netdev_event(event, event_dev);
3348         }
3349
3350         return NOTIFY_DONE;
3351 }
3352
3353 /*
3354  * bond_inetaddr_event: handle inetaddr notifier chain events.
3355  *
3356  * We keep track of device IPs primarily to use as source addresses in
3357  * ARP monitor probes (rather than spewing out broadcasts all the time).
3358  *
3359  * We track one IP for the main device (if it has one), plus one per VLAN.
3360  */
3361 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3362 {
3363         struct in_ifaddr *ifa = ptr;
3364         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3365         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3366         struct bonding *bond;
3367         struct vlan_entry *vlan;
3368
3369         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3370                 if (bond->dev == event_dev) {
3371                         switch (event) {
3372                         case NETDEV_UP:
3373                                 bond->master_ip = ifa->ifa_local;
3374                                 return NOTIFY_OK;
3375                         case NETDEV_DOWN:
3376                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3377                                 return NOTIFY_OK;
3378                         default:
3379                                 return NOTIFY_DONE;
3380                         }
3381                 }
3382
3383                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3384                         if (!bond->vlgrp)
3385                                 continue;
3386                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3387                         if (vlan_dev == event_dev) {
3388                                 switch (event) {
3389                                 case NETDEV_UP:
3390                                         vlan->vlan_ip = ifa->ifa_local;
3391                                         return NOTIFY_OK;
3392                                 case NETDEV_DOWN:
3393                                         vlan->vlan_ip =
3394                                                 bond_glean_dev_ip(vlan_dev);
3395                                         return NOTIFY_OK;
3396                                 default:
3397                                         return NOTIFY_DONE;
3398                                 }
3399                         }
3400                 }
3401         }
3402         return NOTIFY_DONE;
3403 }
3404
3405 static struct notifier_block bond_netdev_notifier = {
3406         .notifier_call = bond_netdev_event,
3407 };
3408
3409 static struct notifier_block bond_inetaddr_notifier = {
3410         .notifier_call = bond_inetaddr_event,
3411 };
3412
3413 /*---------------------------- Hashing Policies -----------------------------*/
3414
3415 /*
3416  * Hash for the output device based upon layer 2 and layer 3 data. If
3417  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3418  */
3419 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3420 {
3421         struct ethhdr *data = (struct ethhdr *)skb->data;
3422         struct iphdr *iph = ip_hdr(skb);
3423
3424         if (skb->protocol == htons(ETH_P_IP)) {
3425                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3426                         (data->h_dest[5] ^ data->h_source[5])) % count;
3427         }
3428
3429         return (data->h_dest[5] ^ data->h_source[5]) % count;
3430 }
3431
3432 /*
3433  * Hash for the output device based upon layer 3 and layer 4 data. If
3434  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3435  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3436  */
3437 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3438 {
3439         struct ethhdr *data = (struct ethhdr *)skb->data;
3440         struct iphdr *iph = ip_hdr(skb);
3441         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3442         int layer4_xor = 0;
3443
3444         if (skb->protocol == htons(ETH_P_IP)) {
3445                 if (!ip_is_fragment(iph) &&
3446                     (iph->protocol == IPPROTO_TCP ||
3447                      iph->protocol == IPPROTO_UDP)) {
3448                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3449                 }
3450                 return (layer4_xor ^
3451                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3452
3453         }
3454
3455         return (data->h_dest[5] ^ data->h_source[5]) % count;
3456 }
3457
3458 /*
3459  * Hash for the output device based upon layer 2 data
3460  */
3461 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3462 {
3463         struct ethhdr *data = (struct ethhdr *)skb->data;
3464
3465         return (data->h_dest[5] ^ data->h_source[5]) % count;
3466 }
3467
3468 /*-------------------------- Device entry points ----------------------------*/
3469
3470 static int bond_open(struct net_device *bond_dev)
3471 {
3472         struct bonding *bond = netdev_priv(bond_dev);
3473
3474         bond->kill_timers = 0;
3475
3476         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3477
3478         if (bond_is_lb(bond)) {
3479                 /* bond_alb_initialize must be called before the timer
3480                  * is started.
3481                  */
3482                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3483                         /* something went wrong - fail the open operation */
3484                         return -ENOMEM;
3485                 }
3486
3487                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3488                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3489         }
3490
3491         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3492                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3493                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3494         }
3495
3496         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3497                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3498                         INIT_DELAYED_WORK(&bond->arp_work,
3499                                           bond_activebackup_arp_mon);
3500                 else
3501                         INIT_DELAYED_WORK(&bond->arp_work,
3502                                           bond_loadbalance_arp_mon);
3503
3504                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3505                 if (bond->params.arp_validate)
3506                         bond->recv_probe = bond_arp_rcv;
3507         }
3508
3509         if (bond->params.mode == BOND_MODE_8023AD) {
3510                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3511                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3512                 /* register to receive LACPDUs */
3513                 bond->recv_probe = bond_3ad_lacpdu_recv;
3514                 bond_3ad_initiate_agg_selection(bond, 1);
3515         }
3516
3517         return 0;
3518 }
3519
3520 static int bond_close(struct net_device *bond_dev)
3521 {
3522         struct bonding *bond = netdev_priv(bond_dev);
3523
3524         write_lock_bh(&bond->lock);
3525
3526         bond->send_peer_notif = 0;
3527
3528         /* signal timers not to re-arm */
3529         bond->kill_timers = 1;
3530
3531         write_unlock_bh(&bond->lock);
3532
3533         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3534                 cancel_delayed_work(&bond->mii_work);
3535         }
3536
3537         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3538                 cancel_delayed_work(&bond->arp_work);
3539         }
3540
3541         switch (bond->params.mode) {
3542         case BOND_MODE_8023AD:
3543                 cancel_delayed_work(&bond->ad_work);
3544                 break;
3545         case BOND_MODE_TLB:
3546         case BOND_MODE_ALB:
3547                 cancel_delayed_work(&bond->alb_work);
3548                 break;
3549         default:
3550                 break;
3551         }
3552
3553         if (delayed_work_pending(&bond->mcast_work))
3554                 cancel_delayed_work(&bond->mcast_work);
3555
3556         if (bond_is_lb(bond)) {
3557                 /* Must be called only after all
3558                  * slaves have been released
3559                  */
3560                 bond_alb_deinitialize(bond);
3561         }
3562         bond->recv_probe = NULL;
3563
3564         return 0;
3565 }
3566
3567 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3568                                                 struct rtnl_link_stats64 *stats)
3569 {
3570         struct bonding *bond = netdev_priv(bond_dev);
3571         struct rtnl_link_stats64 temp;
3572         struct slave *slave;
3573         int i;
3574
3575         memset(stats, 0, sizeof(*stats));
3576
3577         read_lock_bh(&bond->lock);
3578
3579         bond_for_each_slave(bond, slave, i) {
3580                 const struct rtnl_link_stats64 *sstats =
3581                         dev_get_stats(slave->dev, &temp);
3582
3583                 stats->rx_packets += sstats->rx_packets;
3584                 stats->rx_bytes += sstats->rx_bytes;
3585                 stats->rx_errors += sstats->rx_errors;
3586                 stats->rx_dropped += sstats->rx_dropped;
3587
3588                 stats->tx_packets += sstats->tx_packets;
3589                 stats->tx_bytes += sstats->tx_bytes;
3590                 stats->tx_errors += sstats->tx_errors;
3591                 stats->tx_dropped += sstats->tx_dropped;
3592
3593                 stats->multicast += sstats->multicast;
3594                 stats->collisions += sstats->collisions;
3595
3596                 stats->rx_length_errors += sstats->rx_length_errors;
3597                 stats->rx_over_errors += sstats->rx_over_errors;
3598                 stats->rx_crc_errors += sstats->rx_crc_errors;
3599                 stats->rx_frame_errors += sstats->rx_frame_errors;
3600                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3601                 stats->rx_missed_errors += sstats->rx_missed_errors;
3602
3603                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3604                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3605                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3606                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3607                 stats->tx_window_errors += sstats->tx_window_errors;
3608         }
3609
3610         read_unlock_bh(&bond->lock);
3611
3612         return stats;
3613 }
3614
3615 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3616 {
3617         struct net_device *slave_dev = NULL;
3618         struct ifbond k_binfo;
3619         struct ifbond __user *u_binfo = NULL;
3620         struct ifslave k_sinfo;
3621         struct ifslave __user *u_sinfo = NULL;
3622         struct mii_ioctl_data *mii = NULL;
3623         int res = 0;
3624
3625         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3626
3627         switch (cmd) {
3628         case SIOCGMIIPHY:
3629                 mii = if_mii(ifr);
3630                 if (!mii)
3631                         return -EINVAL;
3632
3633                 mii->phy_id = 0;
3634                 /* Fall Through */
3635         case SIOCGMIIREG:
3636                 /*
3637                  * We do this again just in case we were called by SIOCGMIIREG
3638                  * instead of SIOCGMIIPHY.
3639                  */
3640                 mii = if_mii(ifr);
3641                 if (!mii)
3642                         return -EINVAL;
3643
3644
3645                 if (mii->reg_num == 1) {
3646                         struct bonding *bond = netdev_priv(bond_dev);
3647                         mii->val_out = 0;
3648                         read_lock(&bond->lock);
3649                         read_lock(&bond->curr_slave_lock);
3650                         if (netif_carrier_ok(bond->dev))
3651                                 mii->val_out = BMSR_LSTATUS;
3652
3653                         read_unlock(&bond->curr_slave_lock);
3654                         read_unlock(&bond->lock);
3655                 }
3656
3657                 return 0;
3658         case BOND_INFO_QUERY_OLD:
3659         case SIOCBONDINFOQUERY:
3660                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3661
3662                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3663                         return -EFAULT;
3664
3665                 res = bond_info_query(bond_dev, &k_binfo);
3666                 if (res == 0 &&
3667                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3668                         return -EFAULT;
3669
3670                 return res;
3671         case BOND_SLAVE_INFO_QUERY_OLD:
3672         case SIOCBONDSLAVEINFOQUERY:
3673                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3674
3675                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3676                         return -EFAULT;
3677
3678                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3679                 if (res == 0 &&
3680                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3681                         return -EFAULT;
3682
3683                 return res;
3684         default:
3685                 /* Go on */
3686                 break;
3687         }
3688
3689         if (!capable(CAP_NET_ADMIN))
3690                 return -EPERM;
3691
3692         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3693
3694         pr_debug("slave_dev=%p:\n", slave_dev);
3695
3696         if (!slave_dev)
3697                 res = -ENODEV;
3698         else {
3699                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3700                 switch (cmd) {
3701                 case BOND_ENSLAVE_OLD:
3702                 case SIOCBONDENSLAVE:
3703                         res = bond_enslave(bond_dev, slave_dev);
3704                         break;
3705                 case BOND_RELEASE_OLD:
3706                 case SIOCBONDRELEASE:
3707                         res = bond_release(bond_dev, slave_dev);
3708                         break;
3709                 case BOND_SETHWADDR_OLD:
3710                 case SIOCBONDSETHWADDR:
3711                         res = bond_sethwaddr(bond_dev, slave_dev);
3712                         break;
3713                 case BOND_CHANGE_ACTIVE_OLD:
3714                 case SIOCBONDCHANGEACTIVE:
3715                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3716                         break;
3717                 default:
3718                         res = -EOPNOTSUPP;
3719                 }
3720
3721                 dev_put(slave_dev);
3722         }
3723
3724         return res;
3725 }
3726
3727 static bool bond_addr_in_mc_list(unsigned char *addr,
3728                                  struct netdev_hw_addr_list *list,
3729                                  int addrlen)
3730 {
3731         struct netdev_hw_addr *ha;
3732
3733         netdev_hw_addr_list_for_each(ha, list)
3734                 if (!memcmp(ha->addr, addr, addrlen))
3735                         return true;
3736
3737         return false;
3738 }
3739
3740 static void bond_set_multicast_list(struct net_device *bond_dev)
3741 {
3742         struct bonding *bond = netdev_priv(bond_dev);
3743         struct netdev_hw_addr *ha;
3744         bool found;
3745
3746         /*
3747          * Do promisc before checking multicast_mode
3748          */
3749         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3750                 /*
3751                  * FIXME: Need to handle the error when one of the multi-slaves
3752                  * encounters error.
3753                  */
3754                 bond_set_promiscuity(bond, 1);
3755
3756
3757         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3758                 bond_set_promiscuity(bond, -1);
3759
3760
3761         /* set allmulti flag to slaves */
3762         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3763                 /*
3764                  * FIXME: Need to handle the error when one of the multi-slaves
3765                  * encounters error.
3766                  */
3767                 bond_set_allmulti(bond, 1);
3768
3769
3770         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3771                 bond_set_allmulti(bond, -1);
3772
3773
3774         read_lock(&bond->lock);
3775
3776         bond->flags = bond_dev->flags;
3777
3778         /* looking for addresses to add to slaves' mc list */
3779         netdev_for_each_mc_addr(ha, bond_dev) {
3780                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3781                                              bond_dev->addr_len);
3782                 if (!found)
3783                         bond_mc_add(bond, ha->addr);
3784         }
3785
3786         /* looking for addresses to delete from slaves' list */
3787         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3788                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3789                                              bond_dev->addr_len);
3790                 if (!found)
3791                         bond_mc_del(bond, ha->addr);
3792         }
3793
3794         /* save master's multicast list */
3795         __hw_addr_flush(&bond->mc_list);
3796         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3797                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3798
3799         read_unlock(&bond->lock);
3800 }
3801
3802 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3803 {
3804         struct bonding *bond = netdev_priv(dev);
3805         struct slave *slave = bond->first_slave;
3806
3807         if (slave) {
3808                 const struct net_device_ops *slave_ops
3809                         = slave->dev->netdev_ops;
3810                 if (slave_ops->ndo_neigh_setup)
3811                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3812         }
3813         return 0;
3814 }
3815
3816 /*
3817  * Change the MTU of all of a master's slaves to match the master
3818  */
3819 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3820 {
3821         struct bonding *bond = netdev_priv(bond_dev);
3822         struct slave *slave, *stop_at;
3823         int res = 0;
3824         int i;
3825
3826         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3827                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3828
3829         /* Can't hold bond->lock with bh disabled here since
3830          * some base drivers panic. On the other hand we can't
3831          * hold bond->lock without bh disabled because we'll
3832          * deadlock. The only solution is to rely on the fact
3833          * that we're under rtnl_lock here, and the slaves
3834          * list won't change. This doesn't solve the problem
3835          * of setting the slave's MTU while it is
3836          * transmitting, but the assumption is that the base
3837          * driver can handle that.
3838          *
3839          * TODO: figure out a way to safely iterate the slaves
3840          * list, but without holding a lock around the actual
3841          * call to the base driver.
3842          */
3843
3844         bond_for_each_slave(bond, slave, i) {
3845                 pr_debug("s %p s->p %p c_m %p\n",
3846                          slave,
3847                          slave->prev,
3848                          slave->dev->netdev_ops->ndo_change_mtu);
3849
3850                 res = dev_set_mtu(slave->dev, new_mtu);
3851
3852                 if (res) {
3853                         /* If we failed to set the slave's mtu to the new value
3854                          * we must abort the operation even in ACTIVE_BACKUP
3855                          * mode, because if we allow the backup slaves to have
3856                          * different mtu values than the active slave we'll
3857                          * need to change their mtu when doing a failover. That
3858                          * means changing their mtu from timer context, which
3859                          * is probably not a good idea.
3860                          */
3861                         pr_debug("err %d %s\n", res, slave->dev->name);
3862                         goto unwind;
3863                 }
3864         }
3865
3866         bond_dev->mtu = new_mtu;
3867
3868         return 0;
3869
3870 unwind:
3871         /* unwind from head to the slave that failed */
3872         stop_at = slave;
3873         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3874                 int tmp_res;
3875
3876                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3877                 if (tmp_res) {
3878                         pr_debug("unwind err %d dev %s\n",
3879                                  tmp_res, slave->dev->name);
3880                 }
3881         }
3882
3883         return res;
3884 }
3885
3886 /*
3887  * Change HW address
3888  *
3889  * Note that many devices must be down to change the HW address, and
3890  * downing the master releases all slaves.  We can make bonds full of
3891  * bonding devices to test this, however.
3892  */
3893 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3894 {
3895         struct bonding *bond = netdev_priv(bond_dev);
3896         struct sockaddr *sa = addr, tmp_sa;
3897         struct slave *slave, *stop_at;
3898         int res = 0;
3899         int i;
3900
3901         if (bond->params.mode == BOND_MODE_ALB)
3902                 return bond_alb_set_mac_address(bond_dev, addr);
3903
3904
3905         pr_debug("bond=%p, name=%s\n",
3906                  bond, bond_dev ? bond_dev->name : "None");
3907
3908         /*
3909          * If fail_over_mac is set to active, do nothing and return
3910          * success.  Returning an error causes ifenslave to fail.
3911          */
3912         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3913                 return 0;
3914
3915         if (!is_valid_ether_addr(sa->sa_data))
3916                 return -EADDRNOTAVAIL;
3917
3918         /* Can't hold bond->lock with bh disabled here since
3919          * some base drivers panic. On the other hand we can't
3920          * hold bond->lock without bh disabled because we'll
3921          * deadlock. The only solution is to rely on the fact
3922          * that we're under rtnl_lock here, and the slaves
3923          * list won't change. This doesn't solve the problem
3924          * of setting the slave's hw address while it is
3925          * transmitting, but the assumption is that the base
3926          * driver can handle that.
3927          *
3928          * TODO: figure out a way to safely iterate the slaves
3929          * list, but without holding a lock around the actual
3930          * call to the base driver.
3931          */
3932
3933         bond_for_each_slave(bond, slave, i) {
3934                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3935                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3936
3937                 if (slave_ops->ndo_set_mac_address == NULL) {
3938                         res = -EOPNOTSUPP;
3939                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3940                         goto unwind;
3941                 }
3942
3943                 res = dev_set_mac_address(slave->dev, addr);
3944                 if (res) {
3945                         /* TODO: consider downing the slave
3946                          * and retry ?
3947                          * User should expect communications
3948                          * breakage anyway until ARP finish
3949                          * updating, so...
3950                          */
3951                         pr_debug("err %d %s\n", res, slave->dev->name);
3952                         goto unwind;
3953                 }
3954         }
3955
3956         /* success */
3957         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3958         return 0;
3959
3960 unwind:
3961         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3962         tmp_sa.sa_family = bond_dev->type;
3963
3964         /* unwind from head to the slave that failed */
3965         stop_at = slave;
3966         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3967                 int tmp_res;
3968
3969                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3970                 if (tmp_res) {
3971                         pr_debug("unwind err %d dev %s\n",
3972                                  tmp_res, slave->dev->name);
3973                 }
3974         }
3975
3976         return res;
3977 }
3978
3979 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3980 {
3981         struct bonding *bond = netdev_priv(bond_dev);
3982         struct slave *slave, *start_at;
3983         int i, slave_no, res = 1;
3984         struct iphdr *iph = ip_hdr(skb);
3985
3986         /*
3987          * Start with the curr_active_slave that joined the bond as the
3988          * default for sending IGMP traffic.  For failover purposes one
3989          * needs to maintain some consistency for the interface that will
3990          * send the join/membership reports.  The curr_active_slave found
3991          * will send all of this type of traffic.
3992          */
3993         if ((iph->protocol == IPPROTO_IGMP) &&
3994             (skb->protocol == htons(ETH_P_IP))) {
3995
3996                 read_lock(&bond->curr_slave_lock);
3997                 slave = bond->curr_active_slave;
3998                 read_unlock(&bond->curr_slave_lock);
3999
4000                 if (!slave)
4001                         goto out;
4002         } else {
4003                 /*
4004                  * Concurrent TX may collide on rr_tx_counter; we accept
4005                  * that as being rare enough not to justify using an
4006                  * atomic op here.
4007                  */
4008                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4009
4010                 bond_for_each_slave(bond, slave, i) {
4011                         slave_no--;
4012                         if (slave_no < 0)
4013                                 break;
4014                 }
4015         }
4016
4017         start_at = slave;
4018         bond_for_each_slave_from(bond, slave, i, start_at) {
4019                 if (IS_UP(slave->dev) &&
4020                     (slave->link == BOND_LINK_UP) &&
4021                     bond_is_active_slave(slave)) {
4022                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4023                         break;
4024                 }
4025         }
4026
4027 out:
4028         if (res) {
4029                 /* no suitable interface, frame not sent */
4030                 dev_kfree_skb(skb);
4031         }
4032
4033         return NETDEV_TX_OK;
4034 }
4035
4036
4037 /*
4038  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4039  * the bond has a usable interface.
4040  */
4041 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4042 {
4043         struct bonding *bond = netdev_priv(bond_dev);
4044         int res = 1;
4045
4046         read_lock(&bond->curr_slave_lock);
4047
4048         if (bond->curr_active_slave)
4049                 res = bond_dev_queue_xmit(bond, skb,
4050                         bond->curr_active_slave->dev);
4051
4052         if (res)
4053                 /* no suitable interface, frame not sent */
4054                 dev_kfree_skb(skb);
4055
4056         read_unlock(&bond->curr_slave_lock);
4057
4058         return NETDEV_TX_OK;
4059 }
4060
4061 /*
4062  * In bond_xmit_xor() , we determine the output device by using a pre-
4063  * determined xmit_hash_policy(), If the selected device is not enabled,
4064  * find the next active slave.
4065  */
4066 static int bond_xmit_xor(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         int slave_no;
4071         int i;
4072         int res = 1;
4073
4074         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4075
4076         bond_for_each_slave(bond, slave, i) {
4077                 slave_no--;
4078                 if (slave_no < 0)
4079                         break;
4080         }
4081
4082         start_at = slave;
4083
4084         bond_for_each_slave_from(bond, slave, i, start_at) {
4085                 if (IS_UP(slave->dev) &&
4086                     (slave->link == BOND_LINK_UP) &&
4087                     bond_is_active_slave(slave)) {
4088                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4089                         break;
4090                 }
4091         }
4092
4093         if (res) {
4094                 /* no suitable interface, frame not sent */
4095                 dev_kfree_skb(skb);
4096         }
4097
4098         return NETDEV_TX_OK;
4099 }
4100
4101 /*
4102  * in broadcast mode, we send everything to all usable interfaces.
4103  */
4104 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4105 {
4106         struct bonding *bond = netdev_priv(bond_dev);
4107         struct slave *slave, *start_at;
4108         struct net_device *tx_dev = NULL;
4109         int i;
4110         int res = 1;
4111
4112         read_lock(&bond->curr_slave_lock);
4113         start_at = bond->curr_active_slave;
4114         read_unlock(&bond->curr_slave_lock);
4115
4116         if (!start_at)
4117                 goto out;
4118
4119         bond_for_each_slave_from(bond, slave, i, start_at) {
4120                 if (IS_UP(slave->dev) &&
4121                     (slave->link == BOND_LINK_UP) &&
4122                     bond_is_active_slave(slave)) {
4123                         if (tx_dev) {
4124                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4125                                 if (!skb2) {
4126                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4127                                                bond_dev->name);
4128                                         continue;
4129                                 }
4130
4131                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4132                                 if (res) {
4133                                         dev_kfree_skb(skb2);
4134                                         continue;
4135                                 }
4136                         }
4137                         tx_dev = slave->dev;
4138                 }
4139         }
4140
4141         if (tx_dev)
4142                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4143
4144 out:
4145         if (res)
4146                 /* no suitable interface, frame not sent */
4147                 dev_kfree_skb(skb);
4148
4149         /* frame sent to all suitable interfaces */
4150         return NETDEV_TX_OK;
4151 }
4152
4153 /*------------------------- Device initialization ---------------------------*/
4154
4155 static void bond_set_xmit_hash_policy(struct bonding *bond)
4156 {
4157         switch (bond->params.xmit_policy) {
4158         case BOND_XMIT_POLICY_LAYER23:
4159                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4160                 break;
4161         case BOND_XMIT_POLICY_LAYER34:
4162                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4163                 break;
4164         case BOND_XMIT_POLICY_LAYER2:
4165         default:
4166                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4167                 break;
4168         }
4169 }
4170
4171 /*
4172  * Lookup the slave that corresponds to a qid
4173  */
4174 static inline int bond_slave_override(struct bonding *bond,
4175                                       struct sk_buff *skb)
4176 {
4177         int i, res = 1;
4178         struct slave *slave = NULL;
4179         struct slave *check_slave;
4180
4181         if (!skb->queue_mapping)
4182                 return 1;
4183
4184         /* Find out if any slaves have the same mapping as this skb. */
4185         bond_for_each_slave(bond, check_slave, i) {
4186                 if (check_slave->queue_id == skb->queue_mapping) {
4187                         slave = check_slave;
4188                         break;
4189                 }
4190         }
4191
4192         /* If the slave isn't UP, use default transmit policy. */
4193         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4194             (slave->link == BOND_LINK_UP)) {
4195                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4196         }
4197
4198         return res;
4199 }
4200
4201
4202 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4203 {
4204         /*
4205          * This helper function exists to help dev_pick_tx get the correct
4206          * destination queue.  Using a helper function skips a call to
4207          * skb_tx_hash and will put the skbs in the queue we expect on their
4208          * way down to the bonding driver.
4209          */
4210         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4211
4212         /*
4213          * Save the original txq to restore before passing to the driver
4214          */
4215         bond_queue_mapping(skb) = skb->queue_mapping;
4216
4217         if (unlikely(txq >= dev->real_num_tx_queues)) {
4218                 do {
4219                         txq -= dev->real_num_tx_queues;
4220                 } while (txq >= dev->real_num_tx_queues);
4221         }
4222         return txq;
4223 }
4224
4225 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4226 {
4227         struct bonding *bond = netdev_priv(dev);
4228
4229         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4230                 if (!bond_slave_override(bond, skb))
4231                         return NETDEV_TX_OK;
4232         }
4233
4234         switch (bond->params.mode) {
4235         case BOND_MODE_ROUNDROBIN:
4236                 return bond_xmit_roundrobin(skb, dev);
4237         case BOND_MODE_ACTIVEBACKUP:
4238                 return bond_xmit_activebackup(skb, dev);
4239         case BOND_MODE_XOR:
4240                 return bond_xmit_xor(skb, dev);
4241         case BOND_MODE_BROADCAST:
4242                 return bond_xmit_broadcast(skb, dev);
4243         case BOND_MODE_8023AD:
4244                 return bond_3ad_xmit_xor(skb, dev);
4245         case BOND_MODE_ALB:
4246         case BOND_MODE_TLB:
4247                 return bond_alb_xmit(skb, dev);
4248         default:
4249                 /* Should never happen, mode already checked */
4250                 pr_err("%s: Error: Unknown bonding mode %d\n",
4251                        dev->name, bond->params.mode);
4252                 WARN_ON_ONCE(1);
4253                 dev_kfree_skb(skb);
4254                 return NETDEV_TX_OK;
4255         }
4256 }
4257
4258 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4259 {
4260         struct bonding *bond = netdev_priv(dev);
4261         netdev_tx_t ret = NETDEV_TX_OK;
4262
4263         /*
4264          * If we risk deadlock from transmitting this in the
4265          * netpoll path, tell netpoll to queue the frame for later tx
4266          */
4267         if (is_netpoll_tx_blocked(dev))
4268                 return NETDEV_TX_BUSY;
4269
4270         read_lock(&bond->lock);
4271
4272         if (bond->slave_cnt)
4273                 ret = __bond_start_xmit(skb, dev);
4274         else
4275                 dev_kfree_skb(skb);
4276
4277         read_unlock(&bond->lock);
4278
4279         return ret;
4280 }
4281
4282 /*
4283  * set bond mode specific net device operations
4284  */
4285 void bond_set_mode_ops(struct bonding *bond, int mode)
4286 {
4287         struct net_device *bond_dev = bond->dev;
4288
4289         switch (mode) {
4290         case BOND_MODE_ROUNDROBIN:
4291                 break;
4292         case BOND_MODE_ACTIVEBACKUP:
4293                 break;
4294         case BOND_MODE_XOR:
4295                 bond_set_xmit_hash_policy(bond);
4296                 break;
4297         case BOND_MODE_BROADCAST:
4298                 break;
4299         case BOND_MODE_8023AD:
4300                 bond_set_xmit_hash_policy(bond);
4301                 break;
4302         case BOND_MODE_ALB:
4303                 /* FALLTHRU */
4304         case BOND_MODE_TLB:
4305                 break;
4306         default:
4307                 /* Should never happen, mode already checked */
4308                 pr_err("%s: Error: Unknown bonding mode %d\n",
4309                        bond_dev->name, mode);
4310                 break;
4311         }
4312 }
4313
4314 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4315                                     struct ethtool_drvinfo *drvinfo)
4316 {
4317         strncpy(drvinfo->driver, DRV_NAME, 32);
4318         strncpy(drvinfo->version, DRV_VERSION, 32);
4319         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4320 }
4321
4322 static const struct ethtool_ops bond_ethtool_ops = {
4323         .get_drvinfo            = bond_ethtool_get_drvinfo,
4324         .get_link               = ethtool_op_get_link,
4325 };
4326
4327 static const struct net_device_ops bond_netdev_ops = {
4328         .ndo_init               = bond_init,
4329         .ndo_uninit             = bond_uninit,
4330         .ndo_open               = bond_open,
4331         .ndo_stop               = bond_close,
4332         .ndo_start_xmit         = bond_start_xmit,
4333         .ndo_select_queue       = bond_select_queue,
4334         .ndo_get_stats64        = bond_get_stats,
4335         .ndo_do_ioctl           = bond_do_ioctl,
4336         .ndo_set_multicast_list = bond_set_multicast_list,
4337         .ndo_change_mtu         = bond_change_mtu,
4338         .ndo_set_mac_address    = bond_set_mac_address,
4339         .ndo_neigh_setup        = bond_neigh_setup,
4340         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4341         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4342         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4343 #ifdef CONFIG_NET_POLL_CONTROLLER
4344         .ndo_netpoll_setup      = bond_netpoll_setup,
4345         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4346         .ndo_poll_controller    = bond_poll_controller,
4347 #endif
4348         .ndo_add_slave          = bond_enslave,
4349         .ndo_del_slave          = bond_release,
4350         .ndo_fix_features       = bond_fix_features,
4351 };
4352
4353 static void bond_destructor(struct net_device *bond_dev)
4354 {
4355         struct bonding *bond = netdev_priv(bond_dev);
4356         if (bond->wq)
4357                 destroy_workqueue(bond->wq);
4358         free_netdev(bond_dev);
4359 }
4360
4361 static void bond_setup(struct net_device *bond_dev)
4362 {
4363         struct bonding *bond = netdev_priv(bond_dev);
4364
4365         /* initialize rwlocks */
4366         rwlock_init(&bond->lock);
4367         rwlock_init(&bond->curr_slave_lock);
4368
4369         bond->params = bonding_defaults;
4370
4371         /* Initialize pointers */
4372         bond->dev = bond_dev;
4373         INIT_LIST_HEAD(&bond->vlan_list);
4374
4375         /* Initialize the device entry points */
4376         ether_setup(bond_dev);
4377         bond_dev->netdev_ops = &bond_netdev_ops;
4378         bond_dev->ethtool_ops = &bond_ethtool_ops;
4379         bond_set_mode_ops(bond, bond->params.mode);
4380
4381         bond_dev->destructor = bond_destructor;
4382
4383         /* Initialize the device options */
4384         bond_dev->tx_queue_len = 0;
4385         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4386         bond_dev->priv_flags |= IFF_BONDING;
4387         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4388
4389         /* At first, we block adding VLANs. That's the only way to
4390          * prevent problems that occur when adding VLANs over an
4391          * empty bond. The block will be removed once non-challenged
4392          * slaves are enslaved.
4393          */
4394         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4395
4396         /* don't acquire bond device's netif_tx_lock when
4397          * transmitting */
4398         bond_dev->features |= NETIF_F_LLTX;
4399
4400         /* By default, we declare the bond to be fully
4401          * VLAN hardware accelerated capable. Special
4402          * care is taken in the various xmit functions
4403          * when there are slaves that are not hw accel
4404          * capable
4405          */
4406
4407         bond_dev->hw_features = BOND_VLAN_FEATURES |
4408                                 NETIF_F_HW_VLAN_TX |
4409                                 NETIF_F_HW_VLAN_RX |
4410                                 NETIF_F_HW_VLAN_FILTER;
4411
4412         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4413         bond_dev->features |= bond_dev->hw_features;
4414 }
4415
4416 static void bond_work_cancel_all(struct bonding *bond)
4417 {
4418         write_lock_bh(&bond->lock);
4419         bond->kill_timers = 1;
4420         write_unlock_bh(&bond->lock);
4421
4422         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4423                 cancel_delayed_work(&bond->mii_work);
4424
4425         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4426                 cancel_delayed_work(&bond->arp_work);
4427
4428         if (bond->params.mode == BOND_MODE_ALB &&
4429             delayed_work_pending(&bond->alb_work))
4430                 cancel_delayed_work(&bond->alb_work);
4431
4432         if (bond->params.mode == BOND_MODE_8023AD &&
4433             delayed_work_pending(&bond->ad_work))
4434                 cancel_delayed_work(&bond->ad_work);
4435
4436         if (delayed_work_pending(&bond->mcast_work))
4437                 cancel_delayed_work(&bond->mcast_work);
4438 }
4439
4440 /*
4441 * Destroy a bonding device.
4442 * Must be under rtnl_lock when this function is called.
4443 */
4444 static void bond_uninit(struct net_device *bond_dev)
4445 {
4446         struct bonding *bond = netdev_priv(bond_dev);
4447         struct vlan_entry *vlan, *tmp;
4448
4449         bond_netpoll_cleanup(bond_dev);
4450
4451         /* Release the bonded slaves */
4452         bond_release_all(bond_dev);
4453
4454         list_del(&bond->bond_list);
4455
4456         bond_work_cancel_all(bond);
4457
4458         bond_remove_proc_entry(bond);
4459
4460         bond_debug_unregister(bond);
4461
4462         __hw_addr_flush(&bond->mc_list);
4463
4464         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4465                 list_del(&vlan->vlan_list);
4466                 kfree(vlan);
4467         }
4468 }
4469
4470 /*------------------------- Module initialization ---------------------------*/
4471
4472 /*
4473  * Convert string input module parms.  Accept either the
4474  * number of the mode or its string name.  A bit complicated because
4475  * some mode names are substrings of other names, and calls from sysfs
4476  * may have whitespace in the name (trailing newlines, for example).
4477  */
4478 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4479 {
4480         int modeint = -1, i, rv;
4481         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4482
4483         for (p = (char *)buf; *p; p++)
4484                 if (!(isdigit(*p) || isspace(*p)))
4485                         break;
4486
4487         if (*p)
4488                 rv = sscanf(buf, "%20s", modestr);
4489         else
4490                 rv = sscanf(buf, "%d", &modeint);
4491
4492         if (!rv)
4493                 return -1;
4494
4495         for (i = 0; tbl[i].modename; i++) {
4496                 if (modeint == tbl[i].mode)
4497                         return tbl[i].mode;
4498                 if (strcmp(modestr, tbl[i].modename) == 0)
4499                         return tbl[i].mode;
4500         }
4501
4502         return -1;
4503 }
4504
4505 static int bond_check_params(struct bond_params *params)
4506 {
4507         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4508
4509         /*
4510          * Convert string parameters.
4511          */
4512         if (mode) {
4513                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4514                 if (bond_mode == -1) {
4515                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4516                                mode == NULL ? "NULL" : mode);
4517                         return -EINVAL;
4518                 }
4519         }
4520
4521         if (xmit_hash_policy) {
4522                 if ((bond_mode != BOND_MODE_XOR) &&
4523                     (bond_mode != BOND_MODE_8023AD)) {
4524                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4525                                bond_mode_name(bond_mode));
4526                 } else {
4527                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4528                                                         xmit_hashtype_tbl);
4529                         if (xmit_hashtype == -1) {
4530                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4531                                        xmit_hash_policy == NULL ? "NULL" :
4532                                        xmit_hash_policy);
4533                                 return -EINVAL;
4534                         }
4535                 }
4536         }
4537
4538         if (lacp_rate) {
4539                 if (bond_mode != BOND_MODE_8023AD) {
4540                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4541                                 bond_mode_name(bond_mode));
4542                 } else {
4543                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4544                         if (lacp_fast == -1) {
4545                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4546                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4547                                 return -EINVAL;
4548                         }
4549                 }
4550         }
4551
4552         if (ad_select) {
4553                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4554                 if (params->ad_select == -1) {
4555                         pr_err("Error: Invalid ad_select \"%s\"\n",
4556                                ad_select == NULL ? "NULL" : ad_select);
4557                         return -EINVAL;
4558                 }
4559
4560                 if (bond_mode != BOND_MODE_8023AD) {
4561                         pr_warning("ad_select param only affects 802.3ad mode\n");
4562                 }
4563         } else {
4564                 params->ad_select = BOND_AD_STABLE;
4565         }
4566
4567         if (max_bonds < 0) {
4568                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4569                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4570                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4571         }
4572
4573         if (miimon < 0) {
4574                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4575                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4576                 miimon = BOND_LINK_MON_INTERV;
4577         }
4578
4579         if (updelay < 0) {
4580                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4581                            updelay, INT_MAX);
4582                 updelay = 0;
4583         }
4584
4585         if (downdelay < 0) {
4586                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4587                            downdelay, INT_MAX);
4588                 downdelay = 0;
4589         }
4590
4591         if ((use_carrier != 0) && (use_carrier != 1)) {
4592                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4593                            use_carrier);
4594                 use_carrier = 1;
4595         }
4596
4597         if (num_peer_notif < 0 || num_peer_notif > 255) {
4598                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4599                            num_peer_notif);
4600                 num_peer_notif = 1;
4601         }
4602
4603         /* reset values for 802.3ad */
4604         if (bond_mode == BOND_MODE_8023AD) {
4605                 if (!miimon) {
4606                         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");
4607                         pr_warning("Forcing miimon to 100msec\n");
4608                         miimon = 100;
4609                 }
4610         }
4611
4612         if (tx_queues < 1 || tx_queues > 255) {
4613                 pr_warning("Warning: tx_queues (%d) should be between "
4614                            "1 and 255, resetting to %d\n",
4615                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4616                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4617         }
4618
4619         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4620                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4621                            "not of valid value (0/1), so it was set to "
4622                            "0\n", all_slaves_active);
4623                 all_slaves_active = 0;
4624         }
4625
4626         if (resend_igmp < 0 || resend_igmp > 255) {
4627                 pr_warning("Warning: resend_igmp (%d) should be between "
4628                            "0 and 255, resetting to %d\n",
4629                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4630                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4631         }
4632
4633         /* reset values for TLB/ALB */
4634         if ((bond_mode == BOND_MODE_TLB) ||
4635             (bond_mode == BOND_MODE_ALB)) {
4636                 if (!miimon) {
4637                         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");
4638                         pr_warning("Forcing miimon to 100msec\n");
4639                         miimon = 100;
4640                 }
4641         }
4642
4643         if (bond_mode == BOND_MODE_ALB) {
4644                 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",
4645                           updelay);
4646         }
4647
4648         if (!miimon) {
4649                 if (updelay || downdelay) {
4650                         /* just warn the user the up/down delay will have
4651                          * no effect since miimon is zero...
4652                          */
4653                         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",
4654                                    updelay, downdelay);
4655                 }
4656         } else {
4657                 /* don't allow arp monitoring */
4658                 if (arp_interval) {
4659                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4660                                    miimon, arp_interval);
4661                         arp_interval = 0;
4662                 }
4663
4664                 if ((updelay % miimon) != 0) {
4665                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4666                                    updelay, miimon,
4667                                    (updelay / miimon) * miimon);
4668                 }
4669
4670                 updelay /= miimon;
4671
4672                 if ((downdelay % miimon) != 0) {
4673                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4674                                    downdelay, miimon,
4675                                    (downdelay / miimon) * miimon);
4676                 }
4677
4678                 downdelay /= miimon;
4679         }
4680
4681         if (arp_interval < 0) {
4682                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4683                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4684                 arp_interval = BOND_LINK_ARP_INTERV;
4685         }
4686
4687         for (arp_ip_count = 0;
4688              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4689              arp_ip_count++) {
4690                 /* not complete check, but should be good enough to
4691                    catch mistakes */
4692                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4693                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4694                                    arp_ip_target[arp_ip_count]);
4695                         arp_interval = 0;
4696                 } else {
4697                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4698                         arp_target[arp_ip_count] = ip;
4699                 }
4700         }
4701
4702         if (arp_interval && !arp_ip_count) {
4703                 /* don't allow arping if no arp_ip_target given... */
4704                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4705                            arp_interval);
4706                 arp_interval = 0;
4707         }
4708
4709         if (arp_validate) {
4710                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4711                         pr_err("arp_validate only supported in active-backup mode\n");
4712                         return -EINVAL;
4713                 }
4714                 if (!arp_interval) {
4715                         pr_err("arp_validate requires arp_interval\n");
4716                         return -EINVAL;
4717                 }
4718
4719                 arp_validate_value = bond_parse_parm(arp_validate,
4720                                                      arp_validate_tbl);
4721                 if (arp_validate_value == -1) {
4722                         pr_err("Error: invalid arp_validate \"%s\"\n",
4723                                arp_validate == NULL ? "NULL" : arp_validate);
4724                         return -EINVAL;
4725                 }
4726         } else
4727                 arp_validate_value = 0;
4728
4729         if (miimon) {
4730                 pr_info("MII link monitoring set to %d ms\n", miimon);
4731         } else if (arp_interval) {
4732                 int i;
4733
4734                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4735                         arp_interval,
4736                         arp_validate_tbl[arp_validate_value].modename,
4737                         arp_ip_count);
4738
4739                 for (i = 0; i < arp_ip_count; i++)
4740                         pr_info(" %s", arp_ip_target[i]);
4741
4742                 pr_info("\n");
4743
4744         } else if (max_bonds) {
4745                 /* miimon and arp_interval not set, we need one so things
4746                  * work as expected, see bonding.txt for details
4747                  */
4748                 pr_warning("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");
4749         }
4750
4751         if (primary && !USES_PRIMARY(bond_mode)) {
4752                 /* currently, using a primary only makes sense
4753                  * in active backup, TLB or ALB modes
4754                  */
4755                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4756                            primary, bond_mode_name(bond_mode));
4757                 primary = NULL;
4758         }
4759
4760         if (primary && primary_reselect) {
4761                 primary_reselect_value = bond_parse_parm(primary_reselect,
4762                                                          pri_reselect_tbl);
4763                 if (primary_reselect_value == -1) {
4764                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4765                                primary_reselect ==
4766                                         NULL ? "NULL" : primary_reselect);
4767                         return -EINVAL;
4768                 }
4769         } else {
4770                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4771         }
4772
4773         if (fail_over_mac) {
4774                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4775                                                       fail_over_mac_tbl);
4776                 if (fail_over_mac_value == -1) {
4777                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4778                                arp_validate == NULL ? "NULL" : arp_validate);
4779                         return -EINVAL;
4780                 }
4781
4782                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4783                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4784         } else {
4785                 fail_over_mac_value = BOND_FOM_NONE;
4786         }
4787
4788         /* fill params struct with the proper values */
4789         params->mode = bond_mode;
4790         params->xmit_policy = xmit_hashtype;
4791         params->miimon = miimon;
4792         params->num_peer_notif = num_peer_notif;
4793         params->arp_interval = arp_interval;
4794         params->arp_validate = arp_validate_value;
4795         params->updelay = updelay;
4796         params->downdelay = downdelay;
4797         params->use_carrier = use_carrier;
4798         params->lacp_fast = lacp_fast;
4799         params->primary[0] = 0;
4800         params->primary_reselect = primary_reselect_value;
4801         params->fail_over_mac = fail_over_mac_value;
4802         params->tx_queues = tx_queues;
4803         params->all_slaves_active = all_slaves_active;
4804         params->resend_igmp = resend_igmp;
4805         params->min_links = min_links;
4806
4807         if (primary) {
4808                 strncpy(params->primary, primary, IFNAMSIZ);
4809                 params->primary[IFNAMSIZ - 1] = 0;
4810         }
4811
4812         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4813
4814         return 0;
4815 }
4816
4817 static struct lock_class_key bonding_netdev_xmit_lock_key;
4818 static struct lock_class_key bonding_netdev_addr_lock_key;
4819
4820 static void bond_set_lockdep_class_one(struct net_device *dev,
4821                                        struct netdev_queue *txq,
4822                                        void *_unused)
4823 {
4824         lockdep_set_class(&txq->_xmit_lock,
4825                           &bonding_netdev_xmit_lock_key);
4826 }
4827
4828 static void bond_set_lockdep_class(struct net_device *dev)
4829 {
4830         lockdep_set_class(&dev->addr_list_lock,
4831                           &bonding_netdev_addr_lock_key);
4832         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4833 }
4834
4835 /*
4836  * Called from registration process
4837  */
4838 static int bond_init(struct net_device *bond_dev)
4839 {
4840         struct bonding *bond = netdev_priv(bond_dev);
4841         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4842         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4843
4844         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4845
4846         /*
4847          * Initialize locks that may be required during
4848          * en/deslave operations.  All of the bond_open work
4849          * (of which this is part) should really be moved to
4850          * a phase prior to dev_open
4851          */
4852         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4853         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4854
4855         bond->wq = create_singlethread_workqueue(bond_dev->name);
4856         if (!bond->wq)
4857                 return -ENOMEM;
4858
4859         bond_set_lockdep_class(bond_dev);
4860
4861         bond_create_proc_entry(bond);
4862         list_add_tail(&bond->bond_list, &bn->dev_list);
4863
4864         bond_prepare_sysfs_group(bond);
4865
4866         bond_debug_register(bond);
4867
4868         __hw_addr_init(&bond->mc_list);
4869         return 0;
4870 }
4871
4872 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4873 {
4874         if (tb[IFLA_ADDRESS]) {
4875                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4876                         return -EINVAL;
4877                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4878                         return -EADDRNOTAVAIL;
4879         }
4880         return 0;
4881 }
4882
4883 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4884         .kind           = "bond",
4885         .priv_size      = sizeof(struct bonding),
4886         .setup          = bond_setup,
4887         .validate       = bond_validate,
4888 };
4889
4890 /* Create a new bond based on the specified name and bonding parameters.
4891  * If name is NULL, obtain a suitable "bond%d" name for us.
4892  * Caller must NOT hold rtnl_lock; we need to release it here before we
4893  * set up our sysfs entries.
4894  */
4895 int bond_create(struct net *net, const char *name)
4896 {
4897         struct net_device *bond_dev;
4898         int res;
4899
4900         rtnl_lock();
4901
4902         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4903                                    name ? name : "bond%d",
4904                                    bond_setup, tx_queues);
4905         if (!bond_dev) {
4906                 pr_err("%s: eek! can't alloc netdev!\n", name);
4907                 rtnl_unlock();
4908                 return -ENOMEM;
4909         }
4910
4911         dev_net_set(bond_dev, net);
4912         bond_dev->rtnl_link_ops = &bond_link_ops;
4913
4914         res = register_netdevice(bond_dev);
4915
4916         netif_carrier_off(bond_dev);
4917
4918         rtnl_unlock();
4919         if (res < 0)
4920                 bond_destructor(bond_dev);
4921         return res;
4922 }
4923
4924 static int __net_init bond_net_init(struct net *net)
4925 {
4926         struct bond_net *bn = net_generic(net, bond_net_id);
4927
4928         bn->net = net;
4929         INIT_LIST_HEAD(&bn->dev_list);
4930
4931         bond_create_proc_dir(bn);
4932
4933         return 0;
4934 }
4935
4936 static void __net_exit bond_net_exit(struct net *net)
4937 {
4938         struct bond_net *bn = net_generic(net, bond_net_id);
4939
4940         bond_destroy_proc_dir(bn);
4941 }
4942
4943 static struct pernet_operations bond_net_ops = {
4944         .init = bond_net_init,
4945         .exit = bond_net_exit,
4946         .id   = &bond_net_id,
4947         .size = sizeof(struct bond_net),
4948 };
4949
4950 static int __init bonding_init(void)
4951 {
4952         int i;
4953         int res;
4954
4955         pr_info("%s", bond_version);
4956
4957         res = bond_check_params(&bonding_defaults);
4958         if (res)
4959                 goto out;
4960
4961         res = register_pernet_subsys(&bond_net_ops);
4962         if (res)
4963                 goto out;
4964
4965         res = rtnl_link_register(&bond_link_ops);
4966         if (res)
4967                 goto err_link;
4968
4969         bond_create_debugfs();
4970
4971         for (i = 0; i < max_bonds; i++) {
4972                 res = bond_create(&init_net, NULL);
4973                 if (res)
4974                         goto err;
4975         }
4976
4977         res = bond_create_sysfs();
4978         if (res)
4979                 goto err;
4980
4981         register_netdevice_notifier(&bond_netdev_notifier);
4982         register_inetaddr_notifier(&bond_inetaddr_notifier);
4983 out:
4984         return res;
4985 err:
4986         rtnl_link_unregister(&bond_link_ops);
4987 err_link:
4988         unregister_pernet_subsys(&bond_net_ops);
4989         goto out;
4990
4991 }
4992
4993 static void __exit bonding_exit(void)
4994 {
4995         unregister_netdevice_notifier(&bond_netdev_notifier);
4996         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4997
4998         bond_destroy_sysfs();
4999         bond_destroy_debugfs();
5000
5001         rtnl_link_unregister(&bond_link_ops);
5002         unregister_pernet_subsys(&bond_net_ops);
5003
5004 #ifdef CONFIG_NET_POLL_CONTROLLER
5005         /*
5006          * Make sure we don't have an imbalance on our netpoll blocking
5007          */
5008         WARN_ON(atomic_read(&netpoll_block_tx));
5009 #endif
5010 }
5011
5012 module_init(bonding_init);
5013 module_exit(bonding_exit);
5014 MODULE_LICENSE("GPL");
5015 MODULE_VERSION(DRV_VERSION);
5016 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5017 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5018 MODULE_ALIAS_RTNL_LINK("bond");