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 char *ad_select;
 102 static char *xmit_hash_policy;
 103 static int arp_interval = BOND_LINK_ARP_INTERV;
 104 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 105 static char *arp_validate;
 106 static char *fail_over_mac;
 107 static int all_slaves_active = 0;
 108 static struct bond_params bonding_defaults;
 109 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 110
 111 module_param(max_bonds, int, 0);
 112 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 113 module_param(tx_queues, int, 0);
 114 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 115 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 116 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 117                                "failover event (alias of num_unsol_na)");
 118 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 119 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 120                                "failover event (alias of num_grat_arp)");
 121 module_param(miimon, int, 0);
 122 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 123 module_param(updelay, int, 0);
 124 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 125 module_param(downdelay, int, 0);
 126 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 127                             "in milliseconds");
 128 module_param(use_carrier, int, 0);
 129 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 130                               "0 for off, 1 for on (default)");
 131 module_param(mode, charp, 0);
 132 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 133                        "1 for active-backup, 2 for balance-xor, "
 134                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 135                        "6 for balance-alb");
 136 module_param(primary, charp, 0);
 137 MODULE_PARM_DESC(primary, "Primary network device to use");
 138 module_param(primary_reselect, charp, 0);
 139 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 140                                    "once it comes up; "
 141                                    "0 for always (default), "
 142                                    "1 for only if speed of primary is "
 143                                    "better, "
 144                                    "2 for only on active slave "
 145                                    "failure");
 146 module_param(lacp_rate, charp, 0);
 147 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 148                             "0 for slow, 1 for fast");
 149 module_param(ad_select, charp, 0);
 150 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 151                             "0 for stable (default), 1 for bandwidth, "
 152                             "2 for count");
 153 module_param(xmit_hash_policy, charp, 0);
 154 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 155                                    "0 for layer 2 (default), 1 for layer 3+4, "
 156                                    "2 for layer 2+3");
 157 module_param(arp_interval, int, 0);
 158 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 159 module_param_array(arp_ip_target, charp, NULL, 0);
 160 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 161 module_param(arp_validate, charp, 0);
 162 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 163                                "0 for none (default), 1 for active, "
 164                                "2 for backup, 3 for all");
 165 module_param(fail_over_mac, charp, 0);
 166 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 167                                 "the same MAC; 0 for none (default), "
 168                                 "1 for active, 2 for follow");
 169 module_param(all_slaves_active, int, 0);
 170 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 171                                      "by setting active flag for all slaves; "
 172                                      "0 for never (default), 1 for always.");
 173 module_param(resend_igmp, int, 0);
 174 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 175                               "link failure");
 176
 177 /*----------------------------- Global variables ----------------------------*/
 178
 179 #ifdef CONFIG_NET_POLL_CONTROLLER
 180 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 181 #endif
 182
 183 int bond_net_id __read_mostly;
 184
 185 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 186 static int arp_ip_count;
 187 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 188 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 189 static int lacp_fast;
 190
 191 const struct bond_parm_tbl bond_lacp_tbl[] = {
 192 {       "slow",         AD_LACP_SLOW},
 193 {       "fast",         AD_LACP_FAST},
 194 {       NULL,           -1},
 195 };
 196
 197 const struct bond_parm_tbl bond_mode_tbl[] = {
 198 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 199 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 200 {       "balance-xor",          BOND_MODE_XOR},
 201 {       "broadcast",            BOND_MODE_BROADCAST},
 202 {       "802.3ad",              BOND_MODE_8023AD},
 203 {       "balance-tlb",          BOND_MODE_TLB},
 204 {       "balance-alb",          BOND_MODE_ALB},
 205 {       NULL,                   -1},
 206 };
 207
 208 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 209 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 210 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 211 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 212 {       NULL,                   -1},
 213 };
 214
 215 const struct bond_parm_tbl arp_validate_tbl[] = {
 216 {       "none",                 BOND_ARP_VALIDATE_NONE},
 217 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 218 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 219 {       "all",                  BOND_ARP_VALIDATE_ALL},
 220 {       NULL,                   -1},
 221 };
 222
 223 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 224 {       "none",                 BOND_FOM_NONE},
 225 {       "active",               BOND_FOM_ACTIVE},
 226 {       "follow",               BOND_FOM_FOLLOW},
 227 {       NULL,                   -1},
 228 };
 229
 230 const struct bond_parm_tbl pri_reselect_tbl[] = {
 231 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 232 {       "better",               BOND_PRI_RESELECT_BETTER},
 233 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 234 {       NULL,                   -1},
 235 };
 236
 237 struct bond_parm_tbl ad_select_tbl[] = {
 238 {       "stable",       BOND_AD_STABLE},
 239 {       "bandwidth",    BOND_AD_BANDWIDTH},
 240 {       "count",        BOND_AD_COUNT},
 241 {       NULL,           -1},
 242 };
 243
 244 /*-------------------------- Forward declarations ---------------------------*/
 245
 246 static int bond_init(struct net_device *bond_dev);
 247 static void bond_uninit(struct net_device *bond_dev);
 248
 249 /*---------------------------- General routines -----------------------------*/
 250
 251 const char *bond_mode_name(int mode)
 252 {
 253         static const char *names[] = {
 254                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 255                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 256                 [BOND_MODE_XOR] = "load balancing (xor)",
 257                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 258                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 259                 [BOND_MODE_TLB] = "transmit load balancing",
 260                 [BOND_MODE_ALB] = "adaptive load balancing",
 261         };
 262
 263         if (mode < 0 || mode > BOND_MODE_ALB)
 264                 return "unknown";
 265
 266         return names[mode];
 267 }
 268
 269 /*---------------------------------- VLAN -----------------------------------*/
 270
 271 /**
 272  * bond_add_vlan - add a new vlan id on bond
 273  * @bond: bond that got the notification
 274  * @vlan_id: the vlan id to add
 275  *
 276  * Returns -ENOMEM if allocation failed.
 277  */
 278 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 279 {
 280         struct vlan_entry *vlan;
 281
 282         pr_debug("bond: %s, vlan id %d\n",
 283                  (bond ? bond->dev->name : "None"), vlan_id);
 284
 285         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 286         if (!vlan)
 287                 return -ENOMEM;
 288
 289         INIT_LIST_HEAD(&vlan->vlan_list);
 290         vlan->vlan_id = vlan_id;
 291
 292         write_lock_bh(&bond->lock);
 293
 294         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 295
 296         write_unlock_bh(&bond->lock);
 297
 298         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 299
 300         return 0;
 301 }
 302
 303 /**
 304  * bond_del_vlan - delete a vlan id from bond
 305  * @bond: bond that got the notification
 306  * @vlan_id: the vlan id to delete
 307  *
 308  * returns -ENODEV if @vlan_id was not found in @bond.
 309  */
 310 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 311 {
 312         struct vlan_entry *vlan;
 313         int res = -ENODEV;
 314
 315         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 316
 317         block_netpoll_tx();
 318         write_lock_bh(&bond->lock);
 319
 320         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 321                 if (vlan->vlan_id == vlan_id) {
 322                         list_del(&vlan->vlan_list);
 323
 324                         if (bond_is_lb(bond))
 325                                 bond_alb_clear_vlan(bond, vlan_id);
 326
 327                         pr_debug("removed VLAN ID %d from bond %s\n",
 328                                  vlan_id, bond->dev->name);
 329
 330                         kfree(vlan);
 331
 332                         if (list_empty(&bond->vlan_list) &&
 333                             (bond->slave_cnt == 0)) {
 334                                 /* Last VLAN removed and no slaves, so
 335                                  * restore block on adding VLANs. This will
 336                                  * be removed once new slaves that are not
 337                                  * VLAN challenged will be added.
 338                                  */
 339                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 340                         }
 341
 342                         res = 0;
 343                         goto out;
 344                 }
 345         }
 346
 347         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 348                  vlan_id, bond->dev->name);
 349
 350 out:
 351         write_unlock_bh(&bond->lock);
 352         unblock_netpoll_tx();
 353         return res;
 354 }
 355
 356 /**
 357  * bond_next_vlan - safely skip to the next item in the vlans list.
 358  * @bond: the bond we're working on
 359  * @curr: item we're advancing from
 360  *
 361  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 362  * or @curr->next otherwise (even if it is @curr itself again).
 363  *
 364  * Caller must hold bond->lock
 365  */
 366 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 367 {
 368         struct vlan_entry *next, *last;
 369
 370         if (list_empty(&bond->vlan_list))
 371                 return NULL;
 372
 373         if (!curr) {
 374                 next = list_entry(bond->vlan_list.next,
 375                                   struct vlan_entry, vlan_list);
 376         } else {
 377                 last = list_entry(bond->vlan_list.prev,
 378                                   struct vlan_entry, vlan_list);
 379                 if (last == curr) {
 380                         next = list_entry(bond->vlan_list.next,
 381                                           struct vlan_entry, vlan_list);
 382                 } else {
 383                         next = list_entry(curr->vlan_list.next,
 384                                           struct vlan_entry, vlan_list);
 385                 }
 386         }
 387
 388         return next;
 389 }
 390
 391 /**
 392  * bond_dev_queue_xmit - Prepare skb for xmit.
 393  *
 394  * @bond: bond device that got this skb for tx.
 395  * @skb: hw accel VLAN tagged skb to transmit
 396  * @slave_dev: slave that is supposed to xmit this skbuff
 397  */
 398 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 399                         struct net_device *slave_dev)
 400 {
 401         skb->dev = slave_dev;
 402         skb->priority = 1;
 403         if (unlikely(netpoll_tx_running(slave_dev)))
 404                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 405         else
 406                 dev_queue_xmit(skb);
 407
 408         return 0;
 409 }
 410
 411 /*
 412  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 413  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 414  * lock because:
 415  * a. This operation is performed in IOCTL context,
 416  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 417  * c. Holding a lock with BH disabled while directly calling a base driver
 418  *    entry point is generally a BAD idea.
 419  *
 420  * The design of synchronization/protection for this operation in the 8021q
 421  * module is good for one or more VLAN devices over a single physical device
 422  * and cannot be extended for a teaming solution like bonding, so there is a
 423  * potential race condition here where a net device from the vlan group might
 424  * be referenced (either by a base driver or the 8021q code) while it is being
 425  * removed from the system. However, it turns out we're not making matters
 426  * worse, and if it works for regular VLAN usage it will work here too.
 427 */
 428
 429 /**
 430  * bond_vlan_rx_register - Propagates registration to slaves
 431  * @bond_dev: bonding net device that got called
 432  * @grp: vlan group being registered
 433  */
 434 static void bond_vlan_rx_register(struct net_device *bond_dev,
 435                                   struct vlan_group *grp)
 436 {
 437         struct bonding *bond = netdev_priv(bond_dev);
 438         struct slave *slave;
 439         int i;
 440
 441         write_lock_bh(&bond->lock);
 442         bond->vlgrp = grp;
 443         write_unlock_bh(&bond->lock);
 444
 445         bond_for_each_slave(bond, slave, i) {
 446                 struct net_device *slave_dev = slave->dev;
 447                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 448
 449                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 450                     slave_ops->ndo_vlan_rx_register) {
 451                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 452                 }
 453         }
 454 }
 455
 456 /**
 457  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 458  * @bond_dev: bonding net device that got called
 459  * @vid: vlan id being added
 460  */
 461 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 462 {
 463         struct bonding *bond = netdev_priv(bond_dev);
 464         struct slave *slave;
 465         int i, res;
 466
 467         bond_for_each_slave(bond, slave, i) {
 468                 struct net_device *slave_dev = slave->dev;
 469                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 470
 471                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 472                     slave_ops->ndo_vlan_rx_add_vid) {
 473                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 474                 }
 475         }
 476
 477         res = bond_add_vlan(bond, vid);
 478         if (res) {
 479                 pr_err("%s: Error: Failed to add vlan id %d\n",
 480                        bond_dev->name, vid);
 481         }
 482 }
 483
 484 /**
 485  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 486  * @bond_dev: bonding net device that got called
 487  * @vid: vlan id being removed
 488  */
 489 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 490 {
 491         struct bonding *bond = netdev_priv(bond_dev);
 492         struct slave *slave;
 493         struct net_device *vlan_dev;
 494         int i, res;
 495
 496         bond_for_each_slave(bond, slave, i) {
 497                 struct net_device *slave_dev = slave->dev;
 498                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 499
 500                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 501                     slave_ops->ndo_vlan_rx_kill_vid) {
 502                         /* Save and then restore vlan_dev in the grp array,
 503                          * since the slave's driver might clear it.
 504                          */
 505                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 506                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 507                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 508                 }
 509         }
 510
 511         res = bond_del_vlan(bond, vid);
 512         if (res) {
 513                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 514                        bond_dev->name, vid);
 515         }
 516 }
 517
 518 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 519 {
 520         struct vlan_entry *vlan;
 521         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 522
 523         if (!bond->vlgrp)
 524                 return;
 525
 526         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 527             slave_ops->ndo_vlan_rx_register)
 528                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 529
 530         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 531             !(slave_ops->ndo_vlan_rx_add_vid))
 532                 return;
 533
 534         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 535                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 536 }
 537
 538 static void bond_del_vlans_from_slave(struct bonding *bond,
 539                                       struct net_device *slave_dev)
 540 {
 541         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 542         struct vlan_entry *vlan;
 543         struct net_device *vlan_dev;
 544
 545         if (!bond->vlgrp)
 546                 return;
 547
 548         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 549             !(slave_ops->ndo_vlan_rx_kill_vid))
 550                 goto unreg;
 551
 552         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 553                 if (!vlan->vlan_id)
 554                         continue;
 555                 /* Save and then restore vlan_dev in the grp array,
 556                  * since the slave's driver might clear it.
 557                  */
 558                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 559                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 560                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 561         }
 562
 563 unreg:
 564         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 565             slave_ops->ndo_vlan_rx_register)
 566                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 567 }
 568
 569 /*------------------------------- Link status -------------------------------*/
 570
 571 /*
 572  * Set the carrier state for the master according to the state of its
 573  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 574  * do special 802.3ad magic.
 575  *
 576  * Returns zero if carrier state does not change, nonzero if it does.
 577  */
 578 static int bond_set_carrier(struct bonding *bond)
 579 {
 580         struct slave *slave;
 581         int i;
 582
 583         if (bond->slave_cnt == 0)
 584                 goto down;
 585
 586         if (bond->params.mode == BOND_MODE_8023AD)
 587                 return bond_3ad_set_carrier(bond);
 588
 589         bond_for_each_slave(bond, slave, i) {
 590                 if (slave->link == BOND_LINK_UP) {
 591                         if (!netif_carrier_ok(bond->dev)) {
 592                                 netif_carrier_on(bond->dev);
 593                                 return 1;
 594                         }
 595                         return 0;
 596                 }
 597         }
 598
 599 down:
 600         if (netif_carrier_ok(bond->dev)) {
 601                 netif_carrier_off(bond->dev);
 602                 return 1;
 603         }
 604         return 0;
 605 }
 606
 607 /*
 608  * Get link speed and duplex from the slave's base driver
 609  * using ethtool. If for some reason the call fails or the
 610  * values are invalid, fake speed and duplex to 100/Full
 611  * and return error.
 612  */
 613 static int bond_update_speed_duplex(struct slave *slave)
 614 {
 615         struct net_device *slave_dev = slave->dev;
 616         struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
 617         u32 slave_speed;
 618         int res;
 619
 620         /* Fake speed and duplex */
 621         slave->speed = SPEED_100;
 622         slave->duplex = DUPLEX_FULL;
 623
 624         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 625                 return -1;
 626
 627         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 628         if (res < 0)
 629                 return -1;
 630
 631         slave_speed = ethtool_cmd_speed(&etool);
 632         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 633                 return -1;
 634
 635         switch (etool.duplex) {
 636         case DUPLEX_FULL:
 637         case DUPLEX_HALF:
 638                 break;
 639         default:
 640                 return -1;
 641         }
 642
 643         slave->speed = slave_speed;
 644         slave->duplex = etool.duplex;
 645
 646         return 0;
 647 }
 648
 649 /*
 650  * if <dev> supports MII link status reporting, check its link status.
 651  *
 652  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 653  * depending upon the setting of the use_carrier parameter.
 654  *
 655  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 656  * can't tell and just pretend it is), or 0, meaning that the link is
 657  * down.
 658  *
 659  * If reporting is non-zero, instead of faking link up, return -1 if
 660  * both ETHTOOL and MII ioctls fail (meaning the device does not
 661  * support them).  If use_carrier is set, return whatever it says.
 662  * It'd be nice if there was a good way to tell if a driver supports
 663  * netif_carrier, but there really isn't.
 664  */
 665 static int bond_check_dev_link(struct bonding *bond,
 666                                struct net_device *slave_dev, int reporting)
 667 {
 668         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 669         int (*ioctl)(struct net_device *, struct ifreq *, int);
 670         struct ifreq ifr;
 671         struct mii_ioctl_data *mii;
 672
 673         if (!reporting && !netif_running(slave_dev))
 674                 return 0;
 675
 676         if (bond->params.use_carrier)
 677                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 678
 679         /* Try to get link status using Ethtool first. */
 680         if (slave_dev->ethtool_ops) {
 681                 if (slave_dev->ethtool_ops->get_link) {
 682                         u32 link;
 683
 684                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 685
 686                         return link ? BMSR_LSTATUS : 0;
 687                 }
 688         }
 689
 690         /* Ethtool can't be used, fallback to MII ioctls. */
 691         ioctl = slave_ops->ndo_do_ioctl;
 692         if (ioctl) {
 693                 /* TODO: set pointer to correct ioctl on a per team member */
 694                 /*       bases to make this more efficient. that is, once  */
 695                 /*       we determine the correct ioctl, we will always    */
 696                 /*       call it and not the others for that team          */
 697                 /*       member.                                           */
 698
 699                 /*
 700                  * We cannot assume that SIOCGMIIPHY will also read a
 701                  * register; not all network drivers (e.g., e100)
 702                  * support that.
 703                  */
 704
 705                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 706                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 707                 mii = if_mii(&ifr);
 708                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 709                         mii->reg_num = MII_BMSR;
 710                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 711                                 return mii->val_out & BMSR_LSTATUS;
 712                 }
 713         }
 714
 715         /*
 716          * If reporting, report that either there's no dev->do_ioctl,
 717          * or both SIOCGMIIREG and get_link failed (meaning that we
 718          * cannot report link status).  If not reporting, pretend
 719          * we're ok.
 720          */
 721         return reporting ? -1 : BMSR_LSTATUS;
 722 }
 723
 724 /*----------------------------- Multicast list ------------------------------*/
 725
 726 /*
 727  * Push the promiscuity flag down to appropriate slaves
 728  */
 729 static int bond_set_promiscuity(struct bonding *bond, int inc)
 730 {
 731         int err = 0;
 732         if (USES_PRIMARY(bond->params.mode)) {
 733                 /* write lock already acquired */
 734                 if (bond->curr_active_slave) {
 735                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 736                                                   inc);
 737                 }
 738         } else {
 739                 struct slave *slave;
 740                 int i;
 741                 bond_for_each_slave(bond, slave, i) {
 742                         err = dev_set_promiscuity(slave->dev, inc);
 743                         if (err)
 744                                 return err;
 745                 }
 746         }
 747         return err;
 748 }
 749
 750 /*
 751  * Push the allmulti flag down to all slaves
 752  */
 753 static int bond_set_allmulti(struct bonding *bond, int inc)
 754 {
 755         int err = 0;
 756         if (USES_PRIMARY(bond->params.mode)) {
 757                 /* write lock already acquired */
 758                 if (bond->curr_active_slave) {
 759                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 760                                                inc);
 761                 }
 762         } else {
 763                 struct slave *slave;
 764                 int i;
 765                 bond_for_each_slave(bond, slave, i) {
 766                         err = dev_set_allmulti(slave->dev, inc);
 767                         if (err)
 768                                 return err;
 769                 }
 770         }
 771         return err;
 772 }
 773
 774 /*
 775  * Add a Multicast address to slaves
 776  * according to mode
 777  */
 778 static void bond_mc_add(struct bonding *bond, void *addr)
 779 {
 780         if (USES_PRIMARY(bond->params.mode)) {
 781                 /* write lock already acquired */
 782                 if (bond->curr_active_slave)
 783                         dev_mc_add(bond->curr_active_slave->dev, addr);
 784         } else {
 785                 struct slave *slave;
 786                 int i;
 787
 788                 bond_for_each_slave(bond, slave, i)
 789                         dev_mc_add(slave->dev, addr);
 790         }
 791 }
 792
 793 /*
 794  * Remove a multicast address from slave
 795  * according to mode
 796  */
 797 static void bond_mc_del(struct bonding *bond, void *addr)
 798 {
 799         if (USES_PRIMARY(bond->params.mode)) {
 800                 /* write lock already acquired */
 801                 if (bond->curr_active_slave)
 802                         dev_mc_del(bond->curr_active_slave->dev, addr);
 803         } else {
 804                 struct slave *slave;
 805                 int i;
 806                 bond_for_each_slave(bond, slave, i) {
 807                         dev_mc_del(slave->dev, addr);
 808                 }
 809         }
 810 }
 811
 812
 813 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 814 {
 815         struct in_device *in_dev;
 816
 817         rcu_read_lock();
 818         in_dev = __in_dev_get_rcu(dev);
 819         if (in_dev)
 820                 ip_mc_rejoin_groups(in_dev);
 821         rcu_read_unlock();
 822 }
 823
 824 /*
 825  * Retrieve the list of registered multicast addresses for the bonding
 826  * device and retransmit an IGMP JOIN request to the current active
 827  * slave.
 828  */
 829 static void bond_resend_igmp_join_requests(struct bonding *bond)
 830 {
 831         struct net_device *vlan_dev;
 832         struct vlan_entry *vlan;
 833
 834         read_lock(&bond->lock);
 835
 836         /* rejoin all groups on bond device */
 837         __bond_resend_igmp_join_requests(bond->dev);
 838
 839         /* rejoin all groups on vlan devices */
 840         if (bond->vlgrp) {
 841                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 842                         vlan_dev = vlan_group_get_device(bond->vlgrp,
 843                                                          vlan->vlan_id);
 844                         if (vlan_dev)
 845                                 __bond_resend_igmp_join_requests(vlan_dev);
 846                 }
 847         }
 848
 849         if (--bond->igmp_retrans > 0)
 850                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 851
 852         read_unlock(&bond->lock);
 853 }
 854
 855 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 856 {
 857         struct bonding *bond = container_of(work, struct bonding,
 858                                             mcast_work.work);
 859         bond_resend_igmp_join_requests(bond);
 860 }
 861
 862 /*
 863  * flush all members of flush->mc_list from device dev->mc_list
 864  */
 865 static void bond_mc_list_flush(struct net_device *bond_dev,
 866                                struct net_device *slave_dev)
 867 {
 868         struct bonding *bond = netdev_priv(bond_dev);
 869         struct netdev_hw_addr *ha;
 870
 871         netdev_for_each_mc_addr(ha, bond_dev)
 872                 dev_mc_del(slave_dev, ha->addr);
 873
 874         if (bond->params.mode == BOND_MODE_8023AD) {
 875                 /* del lacpdu mc addr from mc list */
 876                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 877
 878                 dev_mc_del(slave_dev, lacpdu_multicast);
 879         }
 880 }
 881
 882 /*--------------------------- Active slave change ---------------------------*/
 883
 884 /*
 885  * Update the mc list and multicast-related flags for the new and
 886  * old active slaves (if any) according to the multicast mode, and
 887  * promiscuous flags unconditionally.
 888  */
 889 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 890                          struct slave *old_active)
 891 {
 892         struct netdev_hw_addr *ha;
 893
 894         if (!USES_PRIMARY(bond->params.mode))
 895                 /* nothing to do -  mc list is already up-to-date on
 896                  * all slaves
 897                  */
 898                 return;
 899
 900         if (old_active) {
 901                 if (bond->dev->flags & IFF_PROMISC)
 902                         dev_set_promiscuity(old_active->dev, -1);
 903
 904                 if (bond->dev->flags & IFF_ALLMULTI)
 905                         dev_set_allmulti(old_active->dev, -1);
 906
 907                 netdev_for_each_mc_addr(ha, bond->dev)
 908                         dev_mc_del(old_active->dev, ha->addr);
 909         }
 910
 911         if (new_active) {
 912                 /* FIXME: Signal errors upstream. */
 913                 if (bond->dev->flags & IFF_PROMISC)
 914                         dev_set_promiscuity(new_active->dev, 1);
 915
 916                 if (bond->dev->flags & IFF_ALLMULTI)
 917                         dev_set_allmulti(new_active->dev, 1);
 918
 919                 netdev_for_each_mc_addr(ha, bond->dev)
 920                         dev_mc_add(new_active->dev, ha->addr);
 921         }
 922 }
 923
 924 /*
 925  * bond_do_fail_over_mac
 926  *
 927  * Perform special MAC address swapping for fail_over_mac settings
 928  *
 929  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 930  */
 931 static void bond_do_fail_over_mac(struct bonding *bond,
 932                                   struct slave *new_active,
 933                                   struct slave *old_active)
 934         __releases(&bond->curr_slave_lock)
 935         __releases(&bond->lock)
 936         __acquires(&bond->lock)
 937         __acquires(&bond->curr_slave_lock)
 938 {
 939         u8 tmp_mac[ETH_ALEN];
 940         struct sockaddr saddr;
 941         int rv;
 942
 943         switch (bond->params.fail_over_mac) {
 944         case BOND_FOM_ACTIVE:
 945                 if (new_active)
 946                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 947                                new_active->dev->addr_len);
 948                 break;
 949         case BOND_FOM_FOLLOW:
 950                 /*
 951                  * if new_active && old_active, swap them
 952                  * if just old_active, do nothing (going to no active slave)
 953                  * if just new_active, set new_active to bond's MAC
 954                  */
 955                 if (!new_active)
 956                         return;
 957
 958                 write_unlock_bh(&bond->curr_slave_lock);
 959                 read_unlock(&bond->lock);
 960
 961                 if (old_active) {
 962                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 963                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 964                                ETH_ALEN);
 965                         saddr.sa_family = new_active->dev->type;
 966                 } else {
 967                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 968                         saddr.sa_family = bond->dev->type;
 969                 }
 970
 971                 rv = dev_set_mac_address(new_active->dev, &saddr);
 972                 if (rv) {
 973                         pr_err("%s: Error %d setting MAC of slave %s\n",
 974                                bond->dev->name, -rv, new_active->dev->name);
 975                         goto out;
 976                 }
 977
 978                 if (!old_active)
 979                         goto out;
 980
 981                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 982                 saddr.sa_family = old_active->dev->type;
 983
 984                 rv = dev_set_mac_address(old_active->dev, &saddr);
 985                 if (rv)
 986                         pr_err("%s: Error %d setting MAC of slave %s\n",
 987                                bond->dev->name, -rv, new_active->dev->name);
 988 out:
 989                 read_lock(&bond->lock);
 990                 write_lock_bh(&bond->curr_slave_lock);
 991                 break;
 992         default:
 993                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 994                        bond->dev->name, bond->params.fail_over_mac);
 995                 break;
 996         }
 997
 998 }
 999
1000 static bool bond_should_change_active(struct bonding *bond)
1001 {
1002         struct slave *prim = bond->primary_slave;
1003         struct slave *curr = bond->curr_active_slave;
1004
1005         if (!prim || !curr || curr->link != BOND_LINK_UP)
1006                 return true;
1007         if (bond->force_primary) {
1008                 bond->force_primary = false;
1009                 return true;
1010         }
1011         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1012             (prim->speed < curr->speed ||
1013              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1014                 return false;
1015         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1016                 return false;
1017         return true;
1018 }
1019
1020 /**
1021  * find_best_interface - select the best available slave to be the active one
1022  * @bond: our bonding struct
1023  *
1024  * Warning: Caller must hold curr_slave_lock for writing.
1025  */
1026 static struct slave *bond_find_best_slave(struct bonding *bond)
1027 {
1028         struct slave *new_active, *old_active;
1029         struct slave *bestslave = NULL;
1030         int mintime = bond->params.updelay;
1031         int i;
1032
1033         new_active = bond->curr_active_slave;
1034
1035         if (!new_active) { /* there were no active slaves left */
1036                 if (bond->slave_cnt > 0)   /* found one slave */
1037                         new_active = bond->first_slave;
1038                 else
1039                         return NULL; /* still no slave, return NULL */
1040         }
1041
1042         if ((bond->primary_slave) &&
1043             bond->primary_slave->link == BOND_LINK_UP &&
1044             bond_should_change_active(bond)) {
1045                 new_active = bond->primary_slave;
1046         }
1047
1048         /* remember where to stop iterating over the slaves */
1049         old_active = new_active;
1050
1051         bond_for_each_slave_from(bond, new_active, i, old_active) {
1052                 if (new_active->link == BOND_LINK_UP) {
1053                         return new_active;
1054                 } else if (new_active->link == BOND_LINK_BACK &&
1055                            IS_UP(new_active->dev)) {
1056                         /* link up, but waiting for stabilization */
1057                         if (new_active->delay < mintime) {
1058                                 mintime = new_active->delay;
1059                                 bestslave = new_active;
1060                         }
1061                 }
1062         }
1063
1064         return bestslave;
1065 }
1066
1067 static bool bond_should_notify_peers(struct bonding *bond)
1068 {
1069         struct slave *slave = bond->curr_active_slave;
1070
1071         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1072                  bond->dev->name, slave ? slave->dev->name : "NULL");
1073
1074         if (!slave || !bond->send_peer_notif ||
1075             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1076                 return false;
1077
1078         bond->send_peer_notif--;
1079         return true;
1080 }
1081
1082 /**
1083  * change_active_interface - change the active slave into the specified one
1084  * @bond: our bonding struct
1085  * @new: the new slave to make the active one
1086  *
1087  * Set the new slave to the bond's settings and unset them on the old
1088  * curr_active_slave.
1089  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1090  *
1091  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1092  * because it is apparently the best available slave we have, even though its
1093  * updelay hasn't timed out yet.
1094  *
1095  * If new_active is not NULL, caller must hold bond->lock for read and
1096  * curr_slave_lock for write_bh.
1097  */
1098 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1099 {
1100         struct slave *old_active = bond->curr_active_slave;
1101
1102         if (old_active == new_active)
1103                 return;
1104
1105         if (new_active) {
1106                 new_active->jiffies = jiffies;
1107
1108                 if (new_active->link == BOND_LINK_BACK) {
1109                         if (USES_PRIMARY(bond->params.mode)) {
1110                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1111                                         bond->dev->name, new_active->dev->name,
1112                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1113                         }
1114
1115                         new_active->delay = 0;
1116                         new_active->link = BOND_LINK_UP;
1117
1118                         if (bond->params.mode == BOND_MODE_8023AD)
1119                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1120
1121                         if (bond_is_lb(bond))
1122                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1123                 } else {
1124                         if (USES_PRIMARY(bond->params.mode)) {
1125                                 pr_info("%s: making interface %s the new active one.\n",
1126                                         bond->dev->name, new_active->dev->name);
1127                         }
1128                 }
1129         }
1130
1131         if (USES_PRIMARY(bond->params.mode))
1132                 bond_mc_swap(bond, new_active, old_active);
1133
1134         if (bond_is_lb(bond)) {
1135                 bond_alb_handle_active_change(bond, new_active);
1136                 if (old_active)
1137                         bond_set_slave_inactive_flags(old_active);
1138                 if (new_active)
1139                         bond_set_slave_active_flags(new_active);
1140         } else {
1141                 bond->curr_active_slave = new_active;
1142         }
1143
1144         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1145                 if (old_active)
1146                         bond_set_slave_inactive_flags(old_active);
1147
1148                 if (new_active) {
1149                         bool should_notify_peers = false;
1150
1151                         bond_set_slave_active_flags(new_active);
1152
1153                         if (bond->params.fail_over_mac)
1154                                 bond_do_fail_over_mac(bond, new_active,
1155                                                       old_active);
1156
1157                         if (netif_running(bond->dev)) {
1158                                 bond->send_peer_notif =
1159                                         bond->params.num_peer_notif;
1160                                 should_notify_peers =
1161                                         bond_should_notify_peers(bond);
1162                         }
1163
1164                         write_unlock_bh(&bond->curr_slave_lock);
1165                         read_unlock(&bond->lock);
1166
1167                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1168                         if (should_notify_peers)
1169                                 netdev_bonding_change(bond->dev,
1170                                                       NETDEV_NOTIFY_PEERS);
1171
1172                         read_lock(&bond->lock);
1173                         write_lock_bh(&bond->curr_slave_lock);
1174                 }
1175         }
1176
1177         /* resend IGMP joins since active slave has changed or
1178          * all were sent on curr_active_slave.
1179          * resend only if bond is brought up with the affected
1180          * bonding modes and the retransmission is enabled */
1181         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1182             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1183              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1184                 bond->igmp_retrans = bond->params.resend_igmp;
1185                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1186         }
1187 }
1188
1189 /**
1190  * bond_select_active_slave - select a new active slave, if needed
1191  * @bond: our bonding struct
1192  *
1193  * This functions should be called when one of the following occurs:
1194  * - The old curr_active_slave has been released or lost its link.
1195  * - The primary_slave has got its link back.
1196  * - A slave has got its link back and there's no old curr_active_slave.
1197  *
1198  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1199  */
1200 void bond_select_active_slave(struct bonding *bond)
1201 {
1202         struct slave *best_slave;
1203         int rv;
1204
1205         best_slave = bond_find_best_slave(bond);
1206         if (best_slave != bond->curr_active_slave) {
1207                 bond_change_active_slave(bond, best_slave);
1208                 rv = bond_set_carrier(bond);
1209                 if (!rv)
1210                         return;
1211
1212                 if (netif_carrier_ok(bond->dev)) {
1213                         pr_info("%s: first active interface up!\n",
1214                                 bond->dev->name);
1215                 } else {
1216                         pr_info("%s: now running without any active interface !\n",
1217                                 bond->dev->name);
1218                 }
1219         }
1220 }
1221
1222 /*--------------------------- slave list handling ---------------------------*/
1223
1224 /*
1225  * This function attaches the slave to the end of list.
1226  *
1227  * bond->lock held for writing by caller.
1228  */
1229 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1230 {
1231         if (bond->first_slave == NULL) { /* attaching the first slave */
1232                 new_slave->next = new_slave;
1233                 new_slave->prev = new_slave;
1234                 bond->first_slave = new_slave;
1235         } else {
1236                 new_slave->next = bond->first_slave;
1237                 new_slave->prev = bond->first_slave->prev;
1238                 new_slave->next->prev = new_slave;
1239                 new_slave->prev->next = new_slave;
1240         }
1241
1242         bond->slave_cnt++;
1243 }
1244
1245 /*
1246  * This function detaches the slave from the list.
1247  * WARNING: no check is made to verify if the slave effectively
1248  * belongs to <bond>.
1249  * Nothing is freed on return, structures are just unchained.
1250  * If any slave pointer in bond was pointing to <slave>,
1251  * it should be changed by the calling function.
1252  *
1253  * bond->lock held for writing by caller.
1254  */
1255 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1256 {
1257         if (slave->next)
1258                 slave->next->prev = slave->prev;
1259
1260         if (slave->prev)
1261                 slave->prev->next = slave->next;
1262
1263         if (bond->first_slave == slave) { /* slave is the first slave */
1264                 if (bond->slave_cnt > 1) { /* there are more slave */
1265                         bond->first_slave = slave->next;
1266                 } else {
1267                         bond->first_slave = NULL; /* slave was the last one */
1268                 }
1269         }
1270
1271         slave->next = NULL;
1272         slave->prev = NULL;
1273         bond->slave_cnt--;
1274 }
1275
1276 #ifdef CONFIG_NET_POLL_CONTROLLER
1277 static inline int slave_enable_netpoll(struct slave *slave)
1278 {
1279         struct netpoll *np;
1280         int err = 0;
1281
1282         np = kzalloc(sizeof(*np), GFP_KERNEL);
1283         err = -ENOMEM;
1284         if (!np)
1285                 goto out;
1286
1287         np->dev = slave->dev;
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_TX_OFFLOADS | \
1419                                  NETIF_F_SOFT_FEATURES | \
1420                                  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                                             bond->params.lacp_fast);
1848                 } else {
1849                         SLAVE_AD_INFO(new_slave).id =
1850                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1851                 }
1852
1853                 bond_3ad_bind_slave(new_slave);
1854                 break;
1855         case BOND_MODE_TLB:
1856         case BOND_MODE_ALB:
1857                 bond_set_active_slave(new_slave);
1858                 bond_set_slave_inactive_flags(new_slave);
1859                 bond_select_active_slave(bond);
1860                 break;
1861         default:
1862                 pr_debug("This slave is always active in trunk mode\n");
1863
1864                 /* always active in trunk mode */
1865                 bond_set_active_slave(new_slave);
1866
1867                 /* In trunking mode there is little meaning to curr_active_slave
1868                  * anyway (it holds no special properties of the bond device),
1869                  * so we can change it without calling change_active_interface()
1870                  */
1871                 if (!bond->curr_active_slave)
1872                         bond->curr_active_slave = new_slave;
1873
1874                 break;
1875         } /* switch(bond_mode) */
1876
1877         write_unlock_bh(&bond->curr_slave_lock);
1878
1879         bond_set_carrier(bond);
1880
1881 #ifdef CONFIG_NET_POLL_CONTROLLER
1882         slave_dev->npinfo = bond_netpoll_info(bond);
1883         if (slave_dev->npinfo) {
1884                 if (slave_enable_netpoll(new_slave)) {
1885                         read_unlock(&bond->lock);
1886                         pr_info("Error, %s: master_dev is using netpoll, "
1887                                  "but new slave device does not support netpoll.\n",
1888                                  bond_dev->name);
1889                         res = -EBUSY;
1890                         goto err_close;
1891                 }
1892         }
1893 #endif
1894
1895         read_unlock(&bond->lock);
1896
1897         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1898         if (res)
1899                 goto err_close;
1900
1901         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1902                                          new_slave);
1903         if (res) {
1904                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1905                 goto err_dest_symlinks;
1906         }
1907
1908         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1909                 bond_dev->name, slave_dev->name,
1910                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1911                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1912
1913         /* enslave is successful */
1914         return 0;
1915
1916 /* Undo stages on error */
1917 err_dest_symlinks:
1918         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1919
1920 err_close:
1921         dev_close(slave_dev);
1922
1923 err_unset_master:
1924         netdev_set_bond_master(slave_dev, NULL);
1925
1926 err_restore_mac:
1927         if (!bond->params.fail_over_mac) {
1928                 /* XXX TODO - fom follow mode needs to change master's
1929                  * MAC if this slave's MAC is in use by the bond, or at
1930                  * least print a warning.
1931                  */
1932                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1933                 addr.sa_family = slave_dev->type;
1934                 dev_set_mac_address(slave_dev, &addr);
1935         }
1936
1937 err_restore_mtu:
1938         dev_set_mtu(slave_dev, new_slave->original_mtu);
1939
1940 err_free:
1941         kfree(new_slave);
1942
1943 err_undo_flags:
1944         bond_compute_features(bond);
1945
1946         return res;
1947 }
1948
1949 /*
1950  * Try to release the slave device <slave> from the bond device <master>
1951  * It is legal to access curr_active_slave without a lock because all the function
1952  * is write-locked.
1953  *
1954  * The rules for slave state should be:
1955  *   for Active/Backup:
1956  *     Active stays on all backups go down
1957  *   for Bonded connections:
1958  *     The first up interface should be left on and all others downed.
1959  */
1960 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1961 {
1962         struct bonding *bond = netdev_priv(bond_dev);
1963         struct slave *slave, *oldcurrent;
1964         struct sockaddr addr;
1965         u32 old_features = bond_dev->features;
1966
1967         /* slave is not a slave or master is not master of this slave */
1968         if (!(slave_dev->flags & IFF_SLAVE) ||
1969             (slave_dev->master != bond_dev)) {
1970                 pr_err("%s: Error: cannot release %s.\n",
1971                        bond_dev->name, slave_dev->name);
1972                 return -EINVAL;
1973         }
1974
1975         block_netpoll_tx();
1976         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1977         write_lock_bh(&bond->lock);
1978
1979         slave = bond_get_slave_by_dev(bond, slave_dev);
1980         if (!slave) {
1981                 /* not a slave of this bond */
1982                 pr_info("%s: %s not enslaved\n",
1983                         bond_dev->name, slave_dev->name);
1984                 write_unlock_bh(&bond->lock);
1985                 unblock_netpoll_tx();
1986                 return -EINVAL;
1987         }
1988
1989         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1990          * for this slave anymore.
1991          */
1992         netdev_rx_handler_unregister(slave_dev);
1993         write_unlock_bh(&bond->lock);
1994         synchronize_net();
1995         write_lock_bh(&bond->lock);
1996
1997         if (!bond->params.fail_over_mac) {
1998                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1999                     bond->slave_cnt > 1)
2000                         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",
2001                                    bond_dev->name, slave_dev->name,
2002                                    slave->perm_hwaddr,
2003                                    bond_dev->name, slave_dev->name);
2004         }
2005
2006         /* Inform AD package of unbinding of slave. */
2007         if (bond->params.mode == BOND_MODE_8023AD) {
2008                 /* must be called before the slave is
2009                  * detached from the list
2010                  */
2011                 bond_3ad_unbind_slave(slave);
2012         }
2013
2014         pr_info("%s: releasing %s interface %s\n",
2015                 bond_dev->name,
2016                 bond_is_active_slave(slave) ? "active" : "backup",
2017                 slave_dev->name);
2018
2019         oldcurrent = bond->curr_active_slave;
2020
2021         bond->current_arp_slave = NULL;
2022
2023         /* release the slave from its bond */
2024         bond_detach_slave(bond, slave);
2025
2026         if (bond->primary_slave == slave)
2027                 bond->primary_slave = NULL;
2028
2029         if (oldcurrent == slave)
2030                 bond_change_active_slave(bond, NULL);
2031
2032         if (bond_is_lb(bond)) {
2033                 /* Must be called only after the slave has been
2034                  * detached from the list and the curr_active_slave
2035                  * has been cleared (if our_slave == old_current),
2036                  * but before a new active slave is selected.
2037                  */
2038                 write_unlock_bh(&bond->lock);
2039                 bond_alb_deinit_slave(bond, slave);
2040                 write_lock_bh(&bond->lock);
2041         }
2042
2043         if (oldcurrent == slave) {
2044                 /*
2045                  * Note that we hold RTNL over this sequence, so there
2046                  * is no concern that another slave add/remove event
2047                  * will interfere.
2048                  */
2049                 write_unlock_bh(&bond->lock);
2050                 read_lock(&bond->lock);
2051                 write_lock_bh(&bond->curr_slave_lock);
2052
2053                 bond_select_active_slave(bond);
2054
2055                 write_unlock_bh(&bond->curr_slave_lock);
2056                 read_unlock(&bond->lock);
2057                 write_lock_bh(&bond->lock);
2058         }
2059
2060         if (bond->slave_cnt == 0) {
2061                 bond_set_carrier(bond);
2062
2063                 /* if the last slave was removed, zero the mac address
2064                  * of the master so it will be set by the application
2065                  * to the mac address of the first slave
2066                  */
2067                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2068
2069                 if (bond->vlgrp) {
2070                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2071                                    bond_dev->name, bond_dev->name);
2072                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2073                                    bond_dev->name);
2074                 }
2075         }
2076
2077         write_unlock_bh(&bond->lock);
2078         unblock_netpoll_tx();
2079
2080         bond_compute_features(bond);
2081         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2082             (old_features & NETIF_F_VLAN_CHALLENGED))
2083                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2084                         bond_dev->name, slave_dev->name, bond_dev->name);
2085
2086         /* must do this from outside any spinlocks */
2087         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2088
2089         bond_del_vlans_from_slave(bond, slave_dev);
2090
2091         /* If the mode USES_PRIMARY, then we should only remove its
2092          * promisc and mc settings if it was the curr_active_slave, but that was
2093          * already taken care of above when we detached the slave
2094          */
2095         if (!USES_PRIMARY(bond->params.mode)) {
2096                 /* unset promiscuity level from slave */
2097                 if (bond_dev->flags & IFF_PROMISC)
2098                         dev_set_promiscuity(slave_dev, -1);
2099
2100                 /* unset allmulti level from slave */
2101                 if (bond_dev->flags & IFF_ALLMULTI)
2102                         dev_set_allmulti(slave_dev, -1);
2103
2104                 /* flush master's mc_list from slave */
2105                 netif_addr_lock_bh(bond_dev);
2106                 bond_mc_list_flush(bond_dev, slave_dev);
2107                 netif_addr_unlock_bh(bond_dev);
2108         }
2109
2110         netdev_set_bond_master(slave_dev, NULL);
2111
2112         slave_disable_netpoll(slave);
2113
2114         /* close slave before restoring its mac address */
2115         dev_close(slave_dev);
2116
2117         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2118                 /* restore original ("permanent") mac address */
2119                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2120                 addr.sa_family = slave_dev->type;
2121                 dev_set_mac_address(slave_dev, &addr);
2122         }
2123
2124         dev_set_mtu(slave_dev, slave->original_mtu);
2125
2126         slave_dev->priv_flags &= ~IFF_BONDING;
2127
2128         kfree(slave);
2129
2130         return 0;  /* deletion OK */
2131 }
2132
2133 /*
2134 * First release a slave and then destroy the bond if no more slaves are left.
2135 * Must be under rtnl_lock when this function is called.
2136 */
2137 static int  bond_release_and_destroy(struct net_device *bond_dev,
2138                                      struct net_device *slave_dev)
2139 {
2140         struct bonding *bond = netdev_priv(bond_dev);
2141         int ret;
2142
2143         ret = bond_release(bond_dev, slave_dev);
2144         if ((ret == 0) && (bond->slave_cnt == 0)) {
2145                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2146                 pr_info("%s: destroying bond %s.\n",
2147                         bond_dev->name, bond_dev->name);
2148                 unregister_netdevice(bond_dev);
2149         }
2150         return ret;
2151 }
2152
2153 /*
2154  * This function releases all slaves.
2155  */
2156 static int bond_release_all(struct net_device *bond_dev)
2157 {
2158         struct bonding *bond = netdev_priv(bond_dev);
2159         struct slave *slave;
2160         struct net_device *slave_dev;
2161         struct sockaddr addr;
2162
2163         write_lock_bh(&bond->lock);
2164
2165         netif_carrier_off(bond_dev);
2166
2167         if (bond->slave_cnt == 0)
2168                 goto out;
2169
2170         bond->current_arp_slave = NULL;
2171         bond->primary_slave = NULL;
2172         bond_change_active_slave(bond, NULL);
2173
2174         while ((slave = bond->first_slave) != NULL) {
2175                 /* Inform AD package of unbinding of slave
2176                  * before slave is detached from the list.
2177                  */
2178                 if (bond->params.mode == BOND_MODE_8023AD)
2179                         bond_3ad_unbind_slave(slave);
2180
2181                 slave_dev = slave->dev;
2182                 bond_detach_slave(bond, slave);
2183
2184                 /* now that the slave is detached, unlock and perform
2185                  * all the undo steps that should not be called from
2186                  * within a lock.
2187                  */
2188                 write_unlock_bh(&bond->lock);
2189
2190                 /* unregister rx_handler early so bond_handle_frame wouldn't
2191                  * be called for this slave anymore.
2192                  */
2193                 netdev_rx_handler_unregister(slave_dev);
2194                 synchronize_net();
2195
2196                 if (bond_is_lb(bond)) {
2197                         /* must be called only after the slave
2198                          * has been detached from the list
2199                          */
2200                         bond_alb_deinit_slave(bond, slave);
2201                 }
2202
2203                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2204                 bond_del_vlans_from_slave(bond, slave_dev);
2205
2206                 /* If the mode USES_PRIMARY, then we should only remove its
2207                  * promisc and mc settings if it was the curr_active_slave, but that was
2208                  * already taken care of above when we detached the slave
2209                  */
2210                 if (!USES_PRIMARY(bond->params.mode)) {
2211                         /* unset promiscuity level from slave */
2212                         if (bond_dev->flags & IFF_PROMISC)
2213                                 dev_set_promiscuity(slave_dev, -1);
2214
2215                         /* unset allmulti level from slave */
2216                         if (bond_dev->flags & IFF_ALLMULTI)
2217                                 dev_set_allmulti(slave_dev, -1);
2218
2219                         /* flush master's mc_list from slave */
2220                         netif_addr_lock_bh(bond_dev);
2221                         bond_mc_list_flush(bond_dev, slave_dev);
2222                         netif_addr_unlock_bh(bond_dev);
2223                 }
2224
2225                 netdev_set_bond_master(slave_dev, NULL);
2226
2227                 slave_disable_netpoll(slave);
2228
2229                 /* close slave before restoring its mac address */
2230                 dev_close(slave_dev);
2231
2232                 if (!bond->params.fail_over_mac) {
2233                         /* restore original ("permanent") mac address*/
2234                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2235                         addr.sa_family = slave_dev->type;
2236                         dev_set_mac_address(slave_dev, &addr);
2237                 }
2238
2239                 kfree(slave);
2240
2241                 /* re-acquire the lock before getting the next slave */
2242                 write_lock_bh(&bond->lock);
2243         }
2244
2245         /* zero the mac address of the master so it will be
2246          * set by the application to the mac address of the
2247          * first slave
2248          */
2249         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2250
2251         if (bond->vlgrp) {
2252                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2253                            bond_dev->name, bond_dev->name);
2254                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2255                            bond_dev->name);
2256         }
2257
2258         pr_info("%s: released all slaves\n", bond_dev->name);
2259
2260 out:
2261         write_unlock_bh(&bond->lock);
2262
2263         bond_compute_features(bond);
2264
2265         return 0;
2266 }
2267
2268 /*
2269  * This function changes the active slave to slave <slave_dev>.
2270  * It returns -EINVAL in the following cases.
2271  *  - <slave_dev> is not found in the list.
2272  *  - There is not active slave now.
2273  *  - <slave_dev> is already active.
2274  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2275  *  - <slave_dev> is not running.
2276  * In these cases, this function does nothing.
2277  * In the other cases, current_slave pointer is changed and 0 is returned.
2278  */
2279 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2280 {
2281         struct bonding *bond = netdev_priv(bond_dev);
2282         struct slave *old_active = NULL;
2283         struct slave *new_active = NULL;
2284         int res = 0;
2285
2286         if (!USES_PRIMARY(bond->params.mode))
2287                 return -EINVAL;
2288
2289         /* Verify that master_dev is indeed the master of slave_dev */
2290         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2291                 return -EINVAL;
2292
2293         read_lock(&bond->lock);
2294
2295         read_lock(&bond->curr_slave_lock);
2296         old_active = bond->curr_active_slave;
2297         read_unlock(&bond->curr_slave_lock);
2298
2299         new_active = bond_get_slave_by_dev(bond, slave_dev);
2300
2301         /*
2302          * Changing to the current active: do nothing; return success.
2303          */
2304         if (new_active && (new_active == old_active)) {
2305                 read_unlock(&bond->lock);
2306                 return 0;
2307         }
2308
2309         if ((new_active) &&
2310             (old_active) &&
2311             (new_active->link == BOND_LINK_UP) &&
2312             IS_UP(new_active->dev)) {
2313                 block_netpoll_tx();
2314                 write_lock_bh(&bond->curr_slave_lock);
2315                 bond_change_active_slave(bond, new_active);
2316                 write_unlock_bh(&bond->curr_slave_lock);
2317                 unblock_netpoll_tx();
2318         } else
2319                 res = -EINVAL;
2320
2321         read_unlock(&bond->lock);
2322
2323         return res;
2324 }
2325
2326 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2327 {
2328         struct bonding *bond = netdev_priv(bond_dev);
2329
2330         info->bond_mode = bond->params.mode;
2331         info->miimon = bond->params.miimon;
2332
2333         read_lock(&bond->lock);
2334         info->num_slaves = bond->slave_cnt;
2335         read_unlock(&bond->lock);
2336
2337         return 0;
2338 }
2339
2340 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2341 {
2342         struct bonding *bond = netdev_priv(bond_dev);
2343         struct slave *slave;
2344         int i, res = -ENODEV;
2345
2346         read_lock(&bond->lock);
2347
2348         bond_for_each_slave(bond, slave, i) {
2349                 if (i == (int)info->slave_id) {
2350                         res = 0;
2351                         strcpy(info->slave_name, slave->dev->name);
2352                         info->link = slave->link;
2353                         info->state = bond_slave_state(slave);
2354                         info->link_failure_count = slave->link_failure_count;
2355                         break;
2356                 }
2357         }
2358
2359         read_unlock(&bond->lock);
2360
2361         return res;
2362 }
2363
2364 /*-------------------------------- Monitoring -------------------------------*/
2365
2366
2367 static int bond_miimon_inspect(struct bonding *bond)
2368 {
2369         struct slave *slave;
2370         int i, link_state, commit = 0;
2371         bool ignore_updelay;
2372
2373         ignore_updelay = !bond->curr_active_slave ? true : false;
2374
2375         bond_for_each_slave(bond, slave, i) {
2376                 slave->new_link = BOND_LINK_NOCHANGE;
2377
2378                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2379
2380                 switch (slave->link) {
2381                 case BOND_LINK_UP:
2382                         if (link_state)
2383                                 continue;
2384
2385                         slave->link = BOND_LINK_FAIL;
2386                         slave->delay = bond->params.downdelay;
2387                         if (slave->delay) {
2388                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2389                                         bond->dev->name,
2390                                         (bond->params.mode ==
2391                                          BOND_MODE_ACTIVEBACKUP) ?
2392                                         (bond_is_active_slave(slave) ?
2393                                          "active " : "backup ") : "",
2394                                         slave->dev->name,
2395                                         bond->params.downdelay * bond->params.miimon);
2396                         }
2397                         /*FALLTHRU*/
2398                 case BOND_LINK_FAIL:
2399                         if (link_state) {
2400                                 /*
2401                                  * recovered before downdelay expired
2402                                  */
2403                                 slave->link = BOND_LINK_UP;
2404                                 slave->jiffies = jiffies;
2405                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2406                                         bond->dev->name,
2407                                         (bond->params.downdelay - slave->delay) *
2408                                         bond->params.miimon,
2409                                         slave->dev->name);
2410                                 continue;
2411                         }
2412
2413                         if (slave->delay <= 0) {
2414                                 slave->new_link = BOND_LINK_DOWN;
2415                                 commit++;
2416                                 continue;
2417                         }
2418
2419                         slave->delay--;
2420                         break;
2421
2422                 case BOND_LINK_DOWN:
2423                         if (!link_state)
2424                                 continue;
2425
2426                         slave->link = BOND_LINK_BACK;
2427                         slave->delay = bond->params.updelay;
2428
2429                         if (slave->delay) {
2430                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2431                                         bond->dev->name, slave->dev->name,
2432                                         ignore_updelay ? 0 :
2433                                         bond->params.updelay *
2434                                         bond->params.miimon);
2435                         }
2436                         /*FALLTHRU*/
2437                 case BOND_LINK_BACK:
2438                         if (!link_state) {
2439                                 slave->link = BOND_LINK_DOWN;
2440                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2441                                         bond->dev->name,
2442                                         (bond->params.updelay - slave->delay) *
2443                                         bond->params.miimon,
2444                                         slave->dev->name);
2445
2446                                 continue;
2447                         }
2448
2449                         if (ignore_updelay)
2450                                 slave->delay = 0;
2451
2452                         if (slave->delay <= 0) {
2453                                 slave->new_link = BOND_LINK_UP;
2454                                 commit++;
2455                                 ignore_updelay = false;
2456                                 continue;
2457                         }
2458
2459                         slave->delay--;
2460                         break;
2461                 }
2462         }
2463
2464         return commit;
2465 }
2466
2467 static void bond_miimon_commit(struct bonding *bond)
2468 {
2469         struct slave *slave;
2470         int i;
2471
2472         bond_for_each_slave(bond, slave, i) {
2473                 switch (slave->new_link) {
2474                 case BOND_LINK_NOCHANGE:
2475                         continue;
2476
2477                 case BOND_LINK_UP:
2478                         slave->link = BOND_LINK_UP;
2479                         slave->jiffies = jiffies;
2480
2481                         if (bond->params.mode == BOND_MODE_8023AD) {
2482                                 /* prevent it from being the active one */
2483                                 bond_set_backup_slave(slave);
2484                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2485                                 /* make it immediately active */
2486                                 bond_set_active_slave(slave);
2487                         } else if (slave != bond->primary_slave) {
2488                                 /* prevent it from being the active one */
2489                                 bond_set_backup_slave(slave);
2490                         }
2491
2492                         bond_update_speed_duplex(slave);
2493
2494                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2495                                 bond->dev->name, slave->dev->name,
2496                                 slave->speed, slave->duplex ? "full" : "half");
2497
2498                         /* notify ad that the link status has changed */
2499                         if (bond->params.mode == BOND_MODE_8023AD)
2500                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2501
2502                         if (bond_is_lb(bond))
2503                                 bond_alb_handle_link_change(bond, slave,
2504                                                             BOND_LINK_UP);
2505
2506                         if (!bond->curr_active_slave ||
2507                             (slave == bond->primary_slave))
2508                                 goto do_failover;
2509
2510                         continue;
2511
2512                 case BOND_LINK_DOWN:
2513                         if (slave->link_failure_count < UINT_MAX)
2514                                 slave->link_failure_count++;
2515
2516                         slave->link = BOND_LINK_DOWN;
2517
2518                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2519                             bond->params.mode == BOND_MODE_8023AD)
2520                                 bond_set_slave_inactive_flags(slave);
2521
2522                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2523                                 bond->dev->name, slave->dev->name);
2524
2525                         if (bond->params.mode == BOND_MODE_8023AD)
2526                                 bond_3ad_handle_link_change(slave,
2527                                                             BOND_LINK_DOWN);
2528
2529                         if (bond_is_lb(bond))
2530                                 bond_alb_handle_link_change(bond, slave,
2531                                                             BOND_LINK_DOWN);
2532
2533                         if (slave == bond->curr_active_slave)
2534                                 goto do_failover;
2535
2536                         continue;
2537
2538                 default:
2539                         pr_err("%s: invalid new link %d on slave %s\n",
2540                                bond->dev->name, slave->new_link,
2541                                slave->dev->name);
2542                         slave->new_link = BOND_LINK_NOCHANGE;
2543
2544                         continue;
2545                 }
2546
2547 do_failover:
2548                 ASSERT_RTNL();
2549                 block_netpoll_tx();
2550                 write_lock_bh(&bond->curr_slave_lock);
2551                 bond_select_active_slave(bond);
2552                 write_unlock_bh(&bond->curr_slave_lock);
2553                 unblock_netpoll_tx();
2554         }
2555
2556         bond_set_carrier(bond);
2557 }
2558
2559 /*
2560  * bond_mii_monitor
2561  *
2562  * Really a wrapper that splits the mii monitor into two phases: an
2563  * inspection, then (if inspection indicates something needs to be done)
2564  * an acquisition of appropriate locks followed by a commit phase to
2565  * implement whatever link state changes are indicated.
2566  */
2567 void bond_mii_monitor(struct work_struct *work)
2568 {
2569         struct bonding *bond = container_of(work, struct bonding,
2570                                             mii_work.work);
2571         bool should_notify_peers = false;
2572
2573         read_lock(&bond->lock);
2574         if (bond->kill_timers)
2575                 goto out;
2576
2577         if (bond->slave_cnt == 0)
2578                 goto re_arm;
2579
2580         should_notify_peers = bond_should_notify_peers(bond);
2581
2582         if (bond_miimon_inspect(bond)) {
2583                 read_unlock(&bond->lock);
2584                 rtnl_lock();
2585                 read_lock(&bond->lock);
2586
2587                 bond_miimon_commit(bond);
2588
2589                 read_unlock(&bond->lock);
2590                 rtnl_unlock();  /* might sleep, hold no other locks */
2591                 read_lock(&bond->lock);
2592         }
2593
2594 re_arm:
2595         if (bond->params.miimon)
2596                 queue_delayed_work(bond->wq, &bond->mii_work,
2597                                    msecs_to_jiffies(bond->params.miimon));
2598 out:
2599         read_unlock(&bond->lock);
2600
2601         if (should_notify_peers) {
2602                 rtnl_lock();
2603                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2604                 rtnl_unlock();
2605         }
2606 }
2607
2608 static __be32 bond_glean_dev_ip(struct net_device *dev)
2609 {
2610         struct in_device *idev;
2611         struct in_ifaddr *ifa;
2612         __be32 addr = 0;
2613
2614         if (!dev)
2615                 return 0;
2616
2617         rcu_read_lock();
2618         idev = __in_dev_get_rcu(dev);
2619         if (!idev)
2620                 goto out;
2621
2622         ifa = idev->ifa_list;
2623         if (!ifa)
2624                 goto out;
2625
2626         addr = ifa->ifa_local;
2627 out:
2628         rcu_read_unlock();
2629         return addr;
2630 }
2631
2632 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2633 {
2634         struct vlan_entry *vlan;
2635
2636         if (ip == bond->master_ip)
2637                 return 1;
2638
2639         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2640                 if (ip == vlan->vlan_ip)
2641                         return 1;
2642         }
2643
2644         return 0;
2645 }
2646
2647 /*
2648  * We go to the (large) trouble of VLAN tagging ARP frames because
2649  * switches in VLAN mode (especially if ports are configured as
2650  * "native" to a VLAN) might not pass non-tagged frames.
2651  */
2652 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2653 {
2654         struct sk_buff *skb;
2655
2656         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2657                  slave_dev->name, dest_ip, src_ip, vlan_id);
2658
2659         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2660                          NULL, slave_dev->dev_addr, NULL);
2661
2662         if (!skb) {
2663                 pr_err("ARP packet allocation failed\n");
2664                 return;
2665         }
2666         if (vlan_id) {
2667                 skb = vlan_put_tag(skb, vlan_id);
2668                 if (!skb) {
2669                         pr_err("failed to insert VLAN tag\n");
2670                         return;
2671                 }
2672         }
2673         arp_xmit(skb);
2674 }
2675
2676
2677 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2678 {
2679         int i, vlan_id;
2680         __be32 *targets = bond->params.arp_targets;
2681         struct vlan_entry *vlan;
2682         struct net_device *vlan_dev;
2683         struct rtable *rt;
2684
2685         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2686                 if (!targets[i])
2687                         break;
2688                 pr_debug("basa: target %x\n", targets[i]);
2689                 if (!bond->vlgrp) {
2690                         pr_debug("basa: empty vlan: arp_send\n");
2691                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2692                                       bond->master_ip, 0);
2693                         continue;
2694                 }
2695
2696                 /*
2697                  * If VLANs are configured, we do a route lookup to
2698                  * determine which VLAN interface would be used, so we
2699                  * can tag the ARP with the proper VLAN tag.
2700                  */
2701                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2702                                      RTO_ONLINK, 0);
2703                 if (IS_ERR(rt)) {
2704                         if (net_ratelimit()) {
2705                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2706                                            bond->dev->name, &targets[i]);
2707                         }
2708                         continue;
2709                 }
2710
2711                 /*
2712                  * This target is not on a VLAN
2713                  */
2714                 if (rt->dst.dev == bond->dev) {
2715                         ip_rt_put(rt);
2716                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2717                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2718                                       bond->master_ip, 0);
2719                         continue;
2720                 }
2721
2722                 vlan_id = 0;
2723                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2724                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2725                         if (vlan_dev == rt->dst.dev) {
2726                                 vlan_id = vlan->vlan_id;
2727                                 pr_debug("basa: vlan match on %s %d\n",
2728                                        vlan_dev->name, vlan_id);
2729                                 break;
2730                         }
2731                 }
2732
2733                 if (vlan_id) {
2734                         ip_rt_put(rt);
2735                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2736                                       vlan->vlan_ip, vlan_id);
2737                         continue;
2738                 }
2739
2740                 if (net_ratelimit()) {
2741                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2742                                    bond->dev->name, &targets[i],
2743                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2744                 }
2745                 ip_rt_put(rt);
2746         }
2747 }
2748
2749 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2750 {
2751         int i;
2752         __be32 *targets = bond->params.arp_targets;
2753
2754         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2755                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2756                          &sip, &tip, i, &targets[i],
2757                          bond_has_this_ip(bond, tip));
2758                 if (sip == targets[i]) {
2759                         if (bond_has_this_ip(bond, tip))
2760                                 slave->last_arp_rx = jiffies;
2761                         return;
2762                 }
2763         }
2764 }
2765
2766 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2767                          struct slave *slave)
2768 {
2769         struct arphdr *arp;
2770         unsigned char *arp_ptr;
2771         __be32 sip, tip;
2772
2773         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2774                 return;
2775
2776         read_lock(&bond->lock);
2777
2778         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2779                  bond->dev->name, skb->dev->name);
2780
2781         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2782                 goto out_unlock;
2783
2784         arp = arp_hdr(skb);
2785         if (arp->ar_hln != bond->dev->addr_len ||
2786             skb->pkt_type == PACKET_OTHERHOST ||
2787             skb->pkt_type == PACKET_LOOPBACK ||
2788             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2789             arp->ar_pro != htons(ETH_P_IP) ||
2790             arp->ar_pln != 4)
2791                 goto out_unlock;
2792
2793         arp_ptr = (unsigned char *)(arp + 1);
2794         arp_ptr += bond->dev->addr_len;
2795         memcpy(&sip, arp_ptr, 4);
2796         arp_ptr += 4 + bond->dev->addr_len;
2797         memcpy(&tip, arp_ptr, 4);
2798
2799         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2800                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2801                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2802                  &sip, &tip);
2803
2804         /*
2805          * Backup slaves won't see the ARP reply, but do come through
2806          * here for each ARP probe (so we swap the sip/tip to validate
2807          * the probe).  In a "redundant switch, common router" type of
2808          * configuration, the ARP probe will (hopefully) travel from
2809          * the active, through one switch, the router, then the other
2810          * switch before reaching the backup.
2811          */
2812         if (bond_is_active_slave(slave))
2813                 bond_validate_arp(bond, slave, sip, tip);
2814         else
2815                 bond_validate_arp(bond, slave, tip, sip);
2816
2817 out_unlock:
2818         read_unlock(&bond->lock);
2819 }
2820
2821 /*
2822  * this function is called regularly to monitor each slave's link
2823  * ensuring that traffic is being sent and received when arp monitoring
2824  * is used in load-balancing mode. if the adapter has been dormant, then an
2825  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2826  * arp monitoring in active backup mode.
2827  */
2828 void bond_loadbalance_arp_mon(struct work_struct *work)
2829 {
2830         struct bonding *bond = container_of(work, struct bonding,
2831                                             arp_work.work);
2832         struct slave *slave, *oldcurrent;
2833         int do_failover = 0;
2834         int delta_in_ticks;
2835         int i;
2836
2837         read_lock(&bond->lock);
2838
2839         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2840
2841         if (bond->kill_timers)
2842                 goto out;
2843
2844         if (bond->slave_cnt == 0)
2845                 goto re_arm;
2846
2847         read_lock(&bond->curr_slave_lock);
2848         oldcurrent = bond->curr_active_slave;
2849         read_unlock(&bond->curr_slave_lock);
2850
2851         /* see if any of the previous devices are up now (i.e. they have
2852          * xmt and rcv traffic). the curr_active_slave does not come into
2853          * the picture unless it is null. also, slave->jiffies is not needed
2854          * here because we send an arp on each slave and give a slave as
2855          * long as it needs to get the tx/rx within the delta.
2856          * TODO: what about up/down delay in arp mode? it wasn't here before
2857          *       so it can wait
2858          */
2859         bond_for_each_slave(bond, slave, i) {
2860                 unsigned long trans_start = dev_trans_start(slave->dev);
2861
2862                 if (slave->link != BOND_LINK_UP) {
2863                         if (time_in_range(jiffies,
2864                                 trans_start - delta_in_ticks,
2865                                 trans_start + delta_in_ticks) &&
2866                             time_in_range(jiffies,
2867                                 slave->dev->last_rx - delta_in_ticks,
2868                                 slave->dev->last_rx + delta_in_ticks)) {
2869
2870                                 slave->link  = BOND_LINK_UP;
2871                                 bond_set_active_slave(slave);
2872
2873                                 /* primary_slave has no meaning in round-robin
2874                                  * mode. the window of a slave being up and
2875                                  * curr_active_slave being null after enslaving
2876                                  * is closed.
2877                                  */
2878                                 if (!oldcurrent) {
2879                                         pr_info("%s: link status definitely up for interface %s, ",
2880                                                 bond->dev->name,
2881                                                 slave->dev->name);
2882                                         do_failover = 1;
2883                                 } else {
2884                                         pr_info("%s: interface %s is now up\n",
2885                                                 bond->dev->name,
2886                                                 slave->dev->name);
2887                                 }
2888                         }
2889                 } else {
2890                         /* slave->link == BOND_LINK_UP */
2891
2892                         /* not all switches will respond to an arp request
2893                          * when the source ip is 0, so don't take the link down
2894                          * if we don't know our ip yet
2895                          */
2896                         if (!time_in_range(jiffies,
2897                                 trans_start - delta_in_ticks,
2898                                 trans_start + 2 * delta_in_ticks) ||
2899                             !time_in_range(jiffies,
2900                                 slave->dev->last_rx - delta_in_ticks,
2901                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2902
2903                                 slave->link  = BOND_LINK_DOWN;
2904                                 bond_set_backup_slave(slave);
2905
2906                                 if (slave->link_failure_count < UINT_MAX)
2907                                         slave->link_failure_count++;
2908
2909                                 pr_info("%s: interface %s is now down.\n",
2910                                         bond->dev->name,
2911                                         slave->dev->name);
2912
2913                                 if (slave == oldcurrent)
2914                                         do_failover = 1;
2915                         }
2916                 }
2917
2918                 /* note: if switch is in round-robin mode, all links
2919                  * must tx arp to ensure all links rx an arp - otherwise
2920                  * links may oscillate or not come up at all; if switch is
2921                  * in something like xor mode, there is nothing we can
2922                  * do - all replies will be rx'ed on same link causing slaves
2923                  * to be unstable during low/no traffic periods
2924                  */
2925                 if (IS_UP(slave->dev))
2926                         bond_arp_send_all(bond, slave);
2927         }
2928
2929         if (do_failover) {
2930                 block_netpoll_tx();
2931                 write_lock_bh(&bond->curr_slave_lock);
2932
2933                 bond_select_active_slave(bond);
2934
2935                 write_unlock_bh(&bond->curr_slave_lock);
2936                 unblock_netpoll_tx();
2937         }
2938
2939 re_arm:
2940         if (bond->params.arp_interval)
2941                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2942 out:
2943         read_unlock(&bond->lock);
2944 }
2945
2946 /*
2947  * Called to inspect slaves for active-backup mode ARP monitor link state
2948  * changes.  Sets new_link in slaves to specify what action should take
2949  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2950  * to link states must be committed.
2951  *
2952  * Called with bond->lock held for read.
2953  */
2954 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2955 {
2956         struct slave *slave;
2957         int i, commit = 0;
2958         unsigned long trans_start;
2959
2960         bond_for_each_slave(bond, slave, i) {
2961                 slave->new_link = BOND_LINK_NOCHANGE;
2962
2963                 if (slave->link != BOND_LINK_UP) {
2964                         if (time_in_range(jiffies,
2965                                 slave_last_rx(bond, slave) - delta_in_ticks,
2966                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2967
2968                                 slave->new_link = BOND_LINK_UP;
2969                                 commit++;
2970                         }
2971
2972                         continue;
2973                 }
2974
2975                 /*
2976                  * Give slaves 2*delta after being enslaved or made
2977                  * active.  This avoids bouncing, as the last receive
2978                  * times need a full ARP monitor cycle to be updated.
2979                  */
2980                 if (time_in_range(jiffies,
2981                                   slave->jiffies - delta_in_ticks,
2982                                   slave->jiffies + 2 * delta_in_ticks))
2983                         continue;
2984
2985                 /*
2986                  * Backup slave is down if:
2987                  * - No current_arp_slave AND
2988                  * - more than 3*delta since last receive AND
2989                  * - the bond has an IP address
2990                  *
2991                  * Note: a non-null current_arp_slave indicates
2992                  * the curr_active_slave went down and we are
2993                  * searching for a new one; under this condition
2994                  * we only take the curr_active_slave down - this
2995                  * gives each slave a chance to tx/rx traffic
2996                  * before being taken out
2997                  */
2998                 if (!bond_is_active_slave(slave) &&
2999                     !bond->current_arp_slave &&
3000                     !time_in_range(jiffies,
3001                         slave_last_rx(bond, slave) - delta_in_ticks,
3002                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3003
3004                         slave->new_link = BOND_LINK_DOWN;
3005                         commit++;
3006                 }
3007
3008                 /*
3009                  * Active slave is down if:
3010                  * - more than 2*delta since transmitting OR
3011                  * - (more than 2*delta since receive AND
3012                  *    the bond has an IP address)
3013                  */
3014                 trans_start = dev_trans_start(slave->dev);
3015                 if (bond_is_active_slave(slave) &&
3016                     (!time_in_range(jiffies,
3017                         trans_start - delta_in_ticks,
3018                         trans_start + 2 * delta_in_ticks) ||
3019                      !time_in_range(jiffies,
3020                         slave_last_rx(bond, slave) - delta_in_ticks,
3021                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3022
3023                         slave->new_link = BOND_LINK_DOWN;
3024                         commit++;
3025                 }
3026         }
3027
3028         return commit;
3029 }
3030
3031 /*
3032  * Called to commit link state changes noted by inspection step of
3033  * active-backup mode ARP monitor.
3034  *
3035  * Called with RTNL and bond->lock for read.
3036  */
3037 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3038 {
3039         struct slave *slave;
3040         int i;
3041         unsigned long trans_start;
3042
3043         bond_for_each_slave(bond, slave, i) {
3044                 switch (slave->new_link) {
3045                 case BOND_LINK_NOCHANGE:
3046                         continue;
3047
3048                 case BOND_LINK_UP:
3049                         trans_start = dev_trans_start(slave->dev);
3050                         if ((!bond->curr_active_slave &&
3051                              time_in_range(jiffies,
3052                                            trans_start - delta_in_ticks,
3053                                            trans_start + delta_in_ticks)) ||
3054                             bond->curr_active_slave != slave) {
3055                                 slave->link = BOND_LINK_UP;
3056                                 bond->current_arp_slave = NULL;
3057
3058                                 pr_info("%s: link status definitely up for interface %s.\n",
3059                                         bond->dev->name, slave->dev->name);
3060
3061                                 if (!bond->curr_active_slave ||
3062                                     (slave == bond->primary_slave))
3063                                         goto do_failover;
3064
3065                         }
3066
3067                         continue;
3068
3069                 case BOND_LINK_DOWN:
3070                         if (slave->link_failure_count < UINT_MAX)
3071                                 slave->link_failure_count++;
3072
3073                         slave->link = BOND_LINK_DOWN;
3074                         bond_set_slave_inactive_flags(slave);
3075
3076                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3077                                 bond->dev->name, slave->dev->name);
3078
3079                         if (slave == bond->curr_active_slave) {
3080                                 bond->current_arp_slave = NULL;
3081                                 goto do_failover;
3082                         }
3083
3084                         continue;
3085
3086                 default:
3087                         pr_err("%s: impossible: new_link %d on slave %s\n",
3088                                bond->dev->name, slave->new_link,
3089                                slave->dev->name);
3090                         continue;
3091                 }
3092
3093 do_failover:
3094                 ASSERT_RTNL();
3095                 block_netpoll_tx();
3096                 write_lock_bh(&bond->curr_slave_lock);
3097                 bond_select_active_slave(bond);
3098                 write_unlock_bh(&bond->curr_slave_lock);
3099                 unblock_netpoll_tx();
3100         }
3101
3102         bond_set_carrier(bond);
3103 }
3104
3105 /*
3106  * Send ARP probes for active-backup mode ARP monitor.
3107  *
3108  * Called with bond->lock held for read.
3109  */
3110 static void bond_ab_arp_probe(struct bonding *bond)
3111 {
3112         struct slave *slave;
3113         int i;
3114
3115         read_lock(&bond->curr_slave_lock);
3116
3117         if (bond->current_arp_slave && bond->curr_active_slave)
3118                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3119                         bond->current_arp_slave->dev->name,
3120                         bond->curr_active_slave->dev->name);
3121
3122         if (bond->curr_active_slave) {
3123                 bond_arp_send_all(bond, bond->curr_active_slave);
3124                 read_unlock(&bond->curr_slave_lock);
3125                 return;
3126         }
3127
3128         read_unlock(&bond->curr_slave_lock);
3129
3130         /* if we don't have a curr_active_slave, search for the next available
3131          * backup slave from the current_arp_slave and make it the candidate
3132          * for becoming the curr_active_slave
3133          */
3134
3135         if (!bond->current_arp_slave) {
3136                 bond->current_arp_slave = bond->first_slave;
3137                 if (!bond->current_arp_slave)
3138                         return;
3139         }
3140
3141         bond_set_slave_inactive_flags(bond->current_arp_slave);
3142
3143         /* search for next candidate */
3144         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3145                 if (IS_UP(slave->dev)) {
3146                         slave->link = BOND_LINK_BACK;
3147                         bond_set_slave_active_flags(slave);
3148                         bond_arp_send_all(bond, slave);
3149                         slave->jiffies = jiffies;
3150                         bond->current_arp_slave = slave;
3151                         break;
3152                 }
3153
3154                 /* if the link state is up at this point, we
3155                  * mark it down - this can happen if we have
3156                  * simultaneous link failures and
3157                  * reselect_active_interface doesn't make this
3158                  * one the current slave so it is still marked
3159                  * up when it is actually down
3160                  */
3161                 if (slave->link == BOND_LINK_UP) {
3162                         slave->link = BOND_LINK_DOWN;
3163                         if (slave->link_failure_count < UINT_MAX)
3164                                 slave->link_failure_count++;
3165
3166                         bond_set_slave_inactive_flags(slave);
3167
3168                         pr_info("%s: backup interface %s is now down.\n",
3169                                 bond->dev->name, slave->dev->name);
3170                 }
3171         }
3172 }
3173
3174 void bond_activebackup_arp_mon(struct work_struct *work)
3175 {
3176         struct bonding *bond = container_of(work, struct bonding,
3177                                             arp_work.work);
3178         bool should_notify_peers = false;
3179         int delta_in_ticks;
3180
3181         read_lock(&bond->lock);
3182
3183         if (bond->kill_timers)
3184                 goto out;
3185
3186         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3187
3188         if (bond->slave_cnt == 0)
3189                 goto re_arm;
3190
3191         should_notify_peers = bond_should_notify_peers(bond);
3192
3193         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3194                 read_unlock(&bond->lock);
3195                 rtnl_lock();
3196                 read_lock(&bond->lock);
3197
3198                 bond_ab_arp_commit(bond, delta_in_ticks);
3199
3200                 read_unlock(&bond->lock);
3201                 rtnl_unlock();
3202                 read_lock(&bond->lock);
3203         }
3204
3205         bond_ab_arp_probe(bond);
3206
3207 re_arm:
3208         if (bond->params.arp_interval)
3209                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3210 out:
3211         read_unlock(&bond->lock);
3212
3213         if (should_notify_peers) {
3214                 rtnl_lock();
3215                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3216                 rtnl_unlock();
3217         }
3218 }
3219
3220 /*-------------------------- netdev event handling --------------------------*/
3221
3222 /*
3223  * Change device name
3224  */
3225 static int bond_event_changename(struct bonding *bond)
3226 {
3227         bond_remove_proc_entry(bond);
3228         bond_create_proc_entry(bond);
3229
3230         bond_debug_reregister(bond);
3231
3232         return NOTIFY_DONE;
3233 }
3234
3235 static int bond_master_netdev_event(unsigned long event,
3236                                     struct net_device *bond_dev)
3237 {
3238         struct bonding *event_bond = netdev_priv(bond_dev);
3239
3240         switch (event) {
3241         case NETDEV_CHANGENAME:
3242                 return bond_event_changename(event_bond);
3243         default:
3244                 break;
3245         }
3246
3247         return NOTIFY_DONE;
3248 }
3249
3250 static int bond_slave_netdev_event(unsigned long event,
3251                                    struct net_device *slave_dev)
3252 {
3253         struct net_device *bond_dev = slave_dev->master;
3254         struct bonding *bond = netdev_priv(bond_dev);
3255
3256         switch (event) {
3257         case NETDEV_UNREGISTER:
3258                 if (bond_dev) {
3259                         if (bond->setup_by_slave)
3260                                 bond_release_and_destroy(bond_dev, slave_dev);
3261                         else
3262                                 bond_release(bond_dev, slave_dev);
3263                 }
3264                 break;
3265         case NETDEV_CHANGE:
3266                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3267                         struct slave *slave;
3268
3269                         slave = bond_get_slave_by_dev(bond, slave_dev);
3270                         if (slave) {
3271                                 u32 old_speed = slave->speed;
3272                                 u8  old_duplex = slave->duplex;
3273
3274                                 bond_update_speed_duplex(slave);
3275
3276                                 if (bond_is_lb(bond))
3277                                         break;
3278
3279                                 if (old_speed != slave->speed)
3280                                         bond_3ad_adapter_speed_changed(slave);
3281                                 if (old_duplex != slave->duplex)
3282                                         bond_3ad_adapter_duplex_changed(slave);
3283                         }
3284                 }
3285
3286                 break;
3287         case NETDEV_DOWN:
3288                 /*
3289                  * ... Or is it this?
3290                  */
3291                 break;
3292         case NETDEV_CHANGEMTU:
3293                 /*
3294                  * TODO: Should slaves be allowed to
3295                  * independently alter their MTU?  For
3296                  * an active-backup bond, slaves need
3297                  * not be the same type of device, so
3298                  * MTUs may vary.  For other modes,
3299                  * slaves arguably should have the
3300                  * same MTUs. To do this, we'd need to
3301                  * take over the slave's change_mtu
3302                  * function for the duration of their
3303                  * servitude.
3304                  */
3305                 break;
3306         case NETDEV_CHANGENAME:
3307                 /*
3308                  * TODO: handle changing the primary's name
3309                  */
3310                 break;
3311         case NETDEV_FEAT_CHANGE:
3312                 bond_compute_features(bond);
3313                 break;
3314         default:
3315                 break;
3316         }
3317
3318         return NOTIFY_DONE;
3319 }
3320
3321 /*
3322  * bond_netdev_event: handle netdev notifier chain events.
3323  *
3324  * This function receives events for the netdev chain.  The caller (an
3325  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3326  * locks for us to safely manipulate the slave devices (RTNL lock,
3327  * dev_probe_lock).
3328  */
3329 static int bond_netdev_event(struct notifier_block *this,
3330                              unsigned long event, void *ptr)
3331 {
3332         struct net_device *event_dev = (struct net_device *)ptr;
3333
3334         pr_debug("event_dev: %s, event: %lx\n",
3335                  event_dev ? event_dev->name : "None",
3336                  event);
3337
3338         if (!(event_dev->priv_flags & IFF_BONDING))
3339                 return NOTIFY_DONE;
3340
3341         if (event_dev->flags & IFF_MASTER) {
3342                 pr_debug("IFF_MASTER\n");
3343                 return bond_master_netdev_event(event, event_dev);
3344         }
3345
3346         if (event_dev->flags & IFF_SLAVE) {
3347                 pr_debug("IFF_SLAVE\n");
3348                 return bond_slave_netdev_event(event, event_dev);
3349         }
3350
3351         return NOTIFY_DONE;
3352 }
3353
3354 /*
3355  * bond_inetaddr_event: handle inetaddr notifier chain events.
3356  *
3357  * We keep track of device IPs primarily to use as source addresses in
3358  * ARP monitor probes (rather than spewing out broadcasts all the time).
3359  *
3360  * We track one IP for the main device (if it has one), plus one per VLAN.
3361  */
3362 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3363 {
3364         struct in_ifaddr *ifa = ptr;
3365         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3366         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3367         struct bonding *bond;
3368         struct vlan_entry *vlan;
3369
3370         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3371                 if (bond->dev == event_dev) {
3372                         switch (event) {
3373                         case NETDEV_UP:
3374                                 bond->master_ip = ifa->ifa_local;
3375                                 return NOTIFY_OK;
3376                         case NETDEV_DOWN:
3377                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3378                                 return NOTIFY_OK;
3379                         default:
3380                                 return NOTIFY_DONE;
3381                         }
3382                 }
3383
3384                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3385                         if (!bond->vlgrp)
3386                                 continue;
3387                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3388                         if (vlan_dev == event_dev) {
3389                                 switch (event) {
3390                                 case NETDEV_UP:
3391                                         vlan->vlan_ip = ifa->ifa_local;
3392                                         return NOTIFY_OK;
3393                                 case NETDEV_DOWN:
3394                                         vlan->vlan_ip =
3395                                                 bond_glean_dev_ip(vlan_dev);
3396                                         return NOTIFY_OK;
3397                                 default:
3398                                         return NOTIFY_DONE;
3399                                 }
3400                         }
3401                 }
3402         }
3403         return NOTIFY_DONE;
3404 }
3405
3406 static struct notifier_block bond_netdev_notifier = {
3407         .notifier_call = bond_netdev_event,
3408 };
3409
3410 static struct notifier_block bond_inetaddr_notifier = {
3411         .notifier_call = bond_inetaddr_event,
3412 };
3413
3414 /*---------------------------- Hashing Policies -----------------------------*/
3415
3416 /*
3417  * Hash for the output device based upon layer 2 and layer 3 data. If
3418  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3419  */
3420 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3421 {
3422         struct ethhdr *data = (struct ethhdr *)skb->data;
3423         struct iphdr *iph = ip_hdr(skb);
3424
3425         if (skb->protocol == htons(ETH_P_IP)) {
3426                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3427                         (data->h_dest[5] ^ data->h_source[5])) % count;
3428         }
3429
3430         return (data->h_dest[5] ^ data->h_source[5]) % count;
3431 }
3432
3433 /*
3434  * Hash for the output device based upon layer 3 and layer 4 data. If
3435  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3436  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3437  */
3438 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3439 {
3440         struct ethhdr *data = (struct ethhdr *)skb->data;
3441         struct iphdr *iph = ip_hdr(skb);
3442         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3443         int layer4_xor = 0;
3444
3445         if (skb->protocol == htons(ETH_P_IP)) {
3446                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3447                     (iph->protocol == IPPROTO_TCP ||
3448                      iph->protocol == IPPROTO_UDP)) {
3449                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3450                 }
3451                 return (layer4_xor ^
3452                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3453
3454         }
3455
3456         return (data->h_dest[5] ^ data->h_source[5]) % count;
3457 }
3458
3459 /*
3460  * Hash for the output device based upon layer 2 data
3461  */
3462 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3463 {
3464         struct ethhdr *data = (struct ethhdr *)skb->data;
3465
3466         return (data->h_dest[5] ^ data->h_source[5]) % count;
3467 }
3468
3469 /*-------------------------- Device entry points ----------------------------*/
3470
3471 static int bond_open(struct net_device *bond_dev)
3472 {
3473         struct bonding *bond = netdev_priv(bond_dev);
3474
3475         bond->kill_timers = 0;
3476
3477         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3478
3479         if (bond_is_lb(bond)) {
3480                 /* bond_alb_initialize must be called before the timer
3481                  * is started.
3482                  */
3483                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3484                         /* something went wrong - fail the open operation */
3485                         return -ENOMEM;
3486                 }
3487
3488                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3489                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3490         }
3491
3492         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3493                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3494                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3495         }
3496
3497         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3498                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3499                         INIT_DELAYED_WORK(&bond->arp_work,
3500                                           bond_activebackup_arp_mon);
3501                 else
3502                         INIT_DELAYED_WORK(&bond->arp_work,
3503                                           bond_loadbalance_arp_mon);
3504
3505                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3506                 if (bond->params.arp_validate)
3507                         bond->recv_probe = bond_arp_rcv;
3508         }
3509
3510         if (bond->params.mode == BOND_MODE_8023AD) {
3511                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3512                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3513                 /* register to receive LACPDUs */
3514                 bond->recv_probe = bond_3ad_lacpdu_recv;
3515                 bond_3ad_initiate_agg_selection(bond, 1);
3516         }
3517
3518         return 0;
3519 }
3520
3521 static int bond_close(struct net_device *bond_dev)
3522 {
3523         struct bonding *bond = netdev_priv(bond_dev);
3524
3525         write_lock_bh(&bond->lock);
3526
3527         bond->send_peer_notif = 0;
3528
3529         /* signal timers not to re-arm */
3530         bond->kill_timers = 1;
3531
3532         write_unlock_bh(&bond->lock);
3533
3534         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3535                 cancel_delayed_work(&bond->mii_work);
3536         }
3537
3538         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3539                 cancel_delayed_work(&bond->arp_work);
3540         }
3541
3542         switch (bond->params.mode) {
3543         case BOND_MODE_8023AD:
3544                 cancel_delayed_work(&bond->ad_work);
3545                 break;
3546         case BOND_MODE_TLB:
3547         case BOND_MODE_ALB:
3548                 cancel_delayed_work(&bond->alb_work);
3549                 break;
3550         default:
3551                 break;
3552         }
3553
3554         if (delayed_work_pending(&bond->mcast_work))
3555                 cancel_delayed_work(&bond->mcast_work);
3556
3557         if (bond_is_lb(bond)) {
3558                 /* Must be called only after all
3559                  * slaves have been released
3560                  */
3561                 bond_alb_deinitialize(bond);
3562         }
3563         bond->recv_probe = NULL;
3564
3565         return 0;
3566 }
3567
3568 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3569                                                 struct rtnl_link_stats64 *stats)
3570 {
3571         struct bonding *bond = netdev_priv(bond_dev);
3572         struct rtnl_link_stats64 temp;
3573         struct slave *slave;
3574         int i;
3575
3576         memset(stats, 0, sizeof(*stats));
3577
3578         read_lock_bh(&bond->lock);
3579
3580         bond_for_each_slave(bond, slave, i) {
3581                 const struct rtnl_link_stats64 *sstats =
3582                         dev_get_stats(slave->dev, &temp);
3583
3584                 stats->rx_packets += sstats->rx_packets;
3585                 stats->rx_bytes += sstats->rx_bytes;
3586                 stats->rx_errors += sstats->rx_errors;
3587                 stats->rx_dropped += sstats->rx_dropped;
3588
3589                 stats->tx_packets += sstats->tx_packets;
3590                 stats->tx_bytes += sstats->tx_bytes;
3591                 stats->tx_errors += sstats->tx_errors;
3592                 stats->tx_dropped += sstats->tx_dropped;
3593
3594                 stats->multicast += sstats->multicast;
3595                 stats->collisions += sstats->collisions;
3596
3597                 stats->rx_length_errors += sstats->rx_length_errors;
3598                 stats->rx_over_errors += sstats->rx_over_errors;
3599                 stats->rx_crc_errors += sstats->rx_crc_errors;
3600                 stats->rx_frame_errors += sstats->rx_frame_errors;
3601                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3602                 stats->rx_missed_errors += sstats->rx_missed_errors;
3603
3604                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3605                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3606                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3607                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3608                 stats->tx_window_errors += sstats->tx_window_errors;
3609         }
3610
3611         read_unlock_bh(&bond->lock);
3612
3613         return stats;
3614 }
3615
3616 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3617 {
3618         struct net_device *slave_dev = NULL;
3619         struct ifbond k_binfo;
3620         struct ifbond __user *u_binfo = NULL;
3621         struct ifslave k_sinfo;
3622         struct ifslave __user *u_sinfo = NULL;
3623         struct mii_ioctl_data *mii = NULL;
3624         int res = 0;
3625
3626         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3627
3628         switch (cmd) {
3629         case SIOCGMIIPHY:
3630                 mii = if_mii(ifr);
3631                 if (!mii)
3632                         return -EINVAL;
3633
3634                 mii->phy_id = 0;
3635                 /* Fall Through */
3636         case SIOCGMIIREG:
3637                 /*
3638                  * We do this again just in case we were called by SIOCGMIIREG
3639                  * instead of SIOCGMIIPHY.
3640                  */
3641                 mii = if_mii(ifr);
3642                 if (!mii)
3643                         return -EINVAL;
3644
3645
3646                 if (mii->reg_num == 1) {
3647                         struct bonding *bond = netdev_priv(bond_dev);
3648                         mii->val_out = 0;
3649                         read_lock(&bond->lock);
3650                         read_lock(&bond->curr_slave_lock);
3651                         if (netif_carrier_ok(bond->dev))
3652                                 mii->val_out = BMSR_LSTATUS;
3653
3654                         read_unlock(&bond->curr_slave_lock);
3655                         read_unlock(&bond->lock);
3656                 }
3657
3658                 return 0;
3659         case BOND_INFO_QUERY_OLD:
3660         case SIOCBONDINFOQUERY:
3661                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3662
3663                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3664                         return -EFAULT;
3665
3666                 res = bond_info_query(bond_dev, &k_binfo);
3667                 if (res == 0 &&
3668                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3669                         return -EFAULT;
3670
3671                 return res;
3672         case BOND_SLAVE_INFO_QUERY_OLD:
3673         case SIOCBONDSLAVEINFOQUERY:
3674                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3675
3676                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3677                         return -EFAULT;
3678
3679                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3680                 if (res == 0 &&
3681                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3682                         return -EFAULT;
3683
3684                 return res;
3685         default:
3686                 /* Go on */
3687                 break;
3688         }
3689
3690         if (!capable(CAP_NET_ADMIN))
3691                 return -EPERM;
3692
3693         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3694
3695         pr_debug("slave_dev=%p:\n", slave_dev);
3696
3697         if (!slave_dev)
3698                 res = -ENODEV;
3699         else {
3700                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3701                 switch (cmd) {
3702                 case BOND_ENSLAVE_OLD:
3703                 case SIOCBONDENSLAVE:
3704                         res = bond_enslave(bond_dev, slave_dev);
3705                         break;
3706                 case BOND_RELEASE_OLD:
3707                 case SIOCBONDRELEASE:
3708                         res = bond_release(bond_dev, slave_dev);
3709                         break;
3710                 case BOND_SETHWADDR_OLD:
3711                 case SIOCBONDSETHWADDR:
3712                         res = bond_sethwaddr(bond_dev, slave_dev);
3713                         break;
3714                 case BOND_CHANGE_ACTIVE_OLD:
3715                 case SIOCBONDCHANGEACTIVE:
3716                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3717                         break;
3718                 default:
3719                         res = -EOPNOTSUPP;
3720                 }
3721
3722                 dev_put(slave_dev);
3723         }
3724
3725         return res;
3726 }
3727
3728 static bool bond_addr_in_mc_list(unsigned char *addr,
3729                                  struct netdev_hw_addr_list *list,
3730                                  int addrlen)
3731 {
3732         struct netdev_hw_addr *ha;
3733
3734         netdev_hw_addr_list_for_each(ha, list)
3735                 if (!memcmp(ha->addr, addr, addrlen))
3736                         return true;
3737
3738         return false;
3739 }
3740
3741 static void bond_set_multicast_list(struct net_device *bond_dev)
3742 {
3743         struct bonding *bond = netdev_priv(bond_dev);
3744         struct netdev_hw_addr *ha;
3745         bool found;
3746
3747         /*
3748          * Do promisc before checking multicast_mode
3749          */
3750         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3751                 /*
3752                  * FIXME: Need to handle the error when one of the multi-slaves
3753                  * encounters error.
3754                  */
3755                 bond_set_promiscuity(bond, 1);
3756
3757
3758         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3759                 bond_set_promiscuity(bond, -1);
3760
3761
3762         /* set allmulti flag to slaves */
3763         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3764                 /*
3765                  * FIXME: Need to handle the error when one of the multi-slaves
3766                  * encounters error.
3767                  */
3768                 bond_set_allmulti(bond, 1);
3769
3770
3771         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3772                 bond_set_allmulti(bond, -1);
3773
3774
3775         read_lock(&bond->lock);
3776
3777         bond->flags = bond_dev->flags;
3778
3779         /* looking for addresses to add to slaves' mc list */
3780         netdev_for_each_mc_addr(ha, bond_dev) {
3781                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3782                                              bond_dev->addr_len);
3783                 if (!found)
3784                         bond_mc_add(bond, ha->addr);
3785         }
3786
3787         /* looking for addresses to delete from slaves' list */
3788         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3789                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3790                                              bond_dev->addr_len);
3791                 if (!found)
3792                         bond_mc_del(bond, ha->addr);
3793         }
3794
3795         /* save master's multicast list */
3796         __hw_addr_flush(&bond->mc_list);
3797         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3798                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3799
3800         read_unlock(&bond->lock);
3801 }
3802
3803 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3804 {
3805         struct bonding *bond = netdev_priv(dev);
3806         struct slave *slave = bond->first_slave;
3807
3808         if (slave) {
3809                 const struct net_device_ops *slave_ops
3810                         = slave->dev->netdev_ops;
3811                 if (slave_ops->ndo_neigh_setup)
3812                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3813         }
3814         return 0;
3815 }
3816
3817 /*
3818  * Change the MTU of all of a master's slaves to match the master
3819  */
3820 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3821 {
3822         struct bonding *bond = netdev_priv(bond_dev);
3823         struct slave *slave, *stop_at;
3824         int res = 0;
3825         int i;
3826
3827         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3828                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3829
3830         /* Can't hold bond->lock with bh disabled here since
3831          * some base drivers panic. On the other hand we can't
3832          * hold bond->lock without bh disabled because we'll
3833          * deadlock. The only solution is to rely on the fact
3834          * that we're under rtnl_lock here, and the slaves
3835          * list won't change. This doesn't solve the problem
3836          * of setting the slave's MTU while it is
3837          * transmitting, but the assumption is that the base
3838          * driver can handle that.
3839          *
3840          * TODO: figure out a way to safely iterate the slaves
3841          * list, but without holding a lock around the actual
3842          * call to the base driver.
3843          */
3844
3845         bond_for_each_slave(bond, slave, i) {
3846                 pr_debug("s %p s->p %p c_m %p\n",
3847                          slave,
3848                          slave->prev,
3849                          slave->dev->netdev_ops->ndo_change_mtu);
3850
3851                 res = dev_set_mtu(slave->dev, new_mtu);
3852
3853                 if (res) {
3854                         /* If we failed to set the slave's mtu to the new value
3855                          * we must abort the operation even in ACTIVE_BACKUP
3856                          * mode, because if we allow the backup slaves to have
3857                          * different mtu values than the active slave we'll
3858                          * need to change their mtu when doing a failover. That
3859                          * means changing their mtu from timer context, which
3860                          * is probably not a good idea.
3861                          */
3862                         pr_debug("err %d %s\n", res, slave->dev->name);
3863                         goto unwind;
3864                 }
3865         }
3866
3867         bond_dev->mtu = new_mtu;
3868
3869         return 0;
3870
3871 unwind:
3872         /* unwind from head to the slave that failed */
3873         stop_at = slave;
3874         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3875                 int tmp_res;
3876
3877                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3878                 if (tmp_res) {
3879                         pr_debug("unwind err %d dev %s\n",
3880                                  tmp_res, slave->dev->name);
3881                 }
3882         }
3883
3884         return res;
3885 }
3886
3887 /*
3888  * Change HW address
3889  *
3890  * Note that many devices must be down to change the HW address, and
3891  * downing the master releases all slaves.  We can make bonds full of
3892  * bonding devices to test this, however.
3893  */
3894 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3895 {
3896         struct bonding *bond = netdev_priv(bond_dev);
3897         struct sockaddr *sa = addr, tmp_sa;
3898         struct slave *slave, *stop_at;
3899         int res = 0;
3900         int i;
3901
3902         if (bond->params.mode == BOND_MODE_ALB)
3903                 return bond_alb_set_mac_address(bond_dev, addr);
3904
3905
3906         pr_debug("bond=%p, name=%s\n",
3907                  bond, bond_dev ? bond_dev->name : "None");
3908
3909         /*
3910          * If fail_over_mac is set to active, do nothing and return
3911          * success.  Returning an error causes ifenslave to fail.
3912          */
3913         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3914                 return 0;
3915
3916         if (!is_valid_ether_addr(sa->sa_data))
3917                 return -EADDRNOTAVAIL;
3918
3919         /* Can't hold bond->lock with bh disabled here since
3920          * some base drivers panic. On the other hand we can't
3921          * hold bond->lock without bh disabled because we'll
3922          * deadlock. The only solution is to rely on the fact
3923          * that we're under rtnl_lock here, and the slaves
3924          * list won't change. This doesn't solve the problem
3925          * of setting the slave's hw address while it is
3926          * transmitting, but the assumption is that the base
3927          * driver can handle that.
3928          *
3929          * TODO: figure out a way to safely iterate the slaves
3930          * list, but without holding a lock around the actual
3931          * call to the base driver.
3932          */
3933
3934         bond_for_each_slave(bond, slave, i) {
3935                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3936                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3937
3938                 if (slave_ops->ndo_set_mac_address == NULL) {
3939                         res = -EOPNOTSUPP;
3940                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3941                         goto unwind;
3942                 }
3943
3944                 res = dev_set_mac_address(slave->dev, addr);
3945                 if (res) {
3946                         /* TODO: consider downing the slave
3947                          * and retry ?
3948                          * User should expect communications
3949                          * breakage anyway until ARP finish
3950                          * updating, so...
3951                          */
3952                         pr_debug("err %d %s\n", res, slave->dev->name);
3953                         goto unwind;
3954                 }
3955         }
3956
3957         /* success */
3958         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3959         return 0;
3960
3961 unwind:
3962         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3963         tmp_sa.sa_family = bond_dev->type;
3964
3965         /* unwind from head to the slave that failed */
3966         stop_at = slave;
3967         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3968                 int tmp_res;
3969
3970                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3971                 if (tmp_res) {
3972                         pr_debug("unwind err %d dev %s\n",
3973                                  tmp_res, slave->dev->name);
3974                 }
3975         }
3976
3977         return res;
3978 }
3979
3980 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3981 {
3982         struct bonding *bond = netdev_priv(bond_dev);
3983         struct slave *slave, *start_at;
3984         int i, slave_no, res = 1;
3985         struct iphdr *iph = ip_hdr(skb);
3986
3987         /*
3988          * Start with the curr_active_slave that joined the bond as the
3989          * default for sending IGMP traffic.  For failover purposes one
3990          * needs to maintain some consistency for the interface that will
3991          * send the join/membership reports.  The curr_active_slave found
3992          * will send all of this type of traffic.
3993          */
3994         if ((iph->protocol == IPPROTO_IGMP) &&
3995             (skb->protocol == htons(ETH_P_IP))) {
3996
3997                 read_lock(&bond->curr_slave_lock);
3998                 slave = bond->curr_active_slave;
3999                 read_unlock(&bond->curr_slave_lock);
4000
4001                 if (!slave)
4002                         goto out;
4003         } else {
4004                 /*
4005                  * Concurrent TX may collide on rr_tx_counter; we accept
4006                  * that as being rare enough not to justify using an
4007                  * atomic op here.
4008                  */
4009                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4010
4011                 bond_for_each_slave(bond, slave, i) {
4012                         slave_no--;
4013                         if (slave_no < 0)
4014                                 break;
4015                 }
4016         }
4017
4018         start_at = slave;
4019         bond_for_each_slave_from(bond, slave, i, start_at) {
4020                 if (IS_UP(slave->dev) &&
4021                     (slave->link == BOND_LINK_UP) &&
4022                     bond_is_active_slave(slave)) {
4023                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4024                         break;
4025                 }
4026         }
4027
4028 out:
4029         if (res) {
4030                 /* no suitable interface, frame not sent */
4031                 dev_kfree_skb(skb);
4032         }
4033
4034         return NETDEV_TX_OK;
4035 }
4036
4037
4038 /*
4039  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4040  * the bond has a usable interface.
4041  */
4042 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4043 {
4044         struct bonding *bond = netdev_priv(bond_dev);
4045         int res = 1;
4046
4047         read_lock(&bond->curr_slave_lock);
4048
4049         if (bond->curr_active_slave)
4050                 res = bond_dev_queue_xmit(bond, skb,
4051                         bond->curr_active_slave->dev);
4052
4053         if (res)
4054                 /* no suitable interface, frame not sent */
4055                 dev_kfree_skb(skb);
4056
4057         read_unlock(&bond->curr_slave_lock);
4058
4059         return NETDEV_TX_OK;
4060 }
4061
4062 /*
4063  * In bond_xmit_xor() , we determine the output device by using a pre-
4064  * determined xmit_hash_policy(), If the selected device is not enabled,
4065  * find the next active slave.
4066  */
4067 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4068 {
4069         struct bonding *bond = netdev_priv(bond_dev);
4070         struct slave *slave, *start_at;
4071         int slave_no;
4072         int i;
4073         int res = 1;
4074
4075         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4076
4077         bond_for_each_slave(bond, slave, i) {
4078                 slave_no--;
4079                 if (slave_no < 0)
4080                         break;
4081         }
4082
4083         start_at = slave;
4084
4085         bond_for_each_slave_from(bond, slave, i, start_at) {
4086                 if (IS_UP(slave->dev) &&
4087                     (slave->link == BOND_LINK_UP) &&
4088                     bond_is_active_slave(slave)) {
4089                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4090                         break;
4091                 }
4092         }
4093
4094         if (res) {
4095                 /* no suitable interface, frame not sent */
4096                 dev_kfree_skb(skb);
4097         }
4098
4099         return NETDEV_TX_OK;
4100 }
4101
4102 /*
4103  * in broadcast mode, we send everything to all usable interfaces.
4104  */
4105 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4106 {
4107         struct bonding *bond = netdev_priv(bond_dev);
4108         struct slave *slave, *start_at;
4109         struct net_device *tx_dev = NULL;
4110         int i;
4111         int res = 1;
4112
4113         read_lock(&bond->curr_slave_lock);
4114         start_at = bond->curr_active_slave;
4115         read_unlock(&bond->curr_slave_lock);
4116
4117         if (!start_at)
4118                 goto out;
4119
4120         bond_for_each_slave_from(bond, slave, i, start_at) {
4121                 if (IS_UP(slave->dev) &&
4122                     (slave->link == BOND_LINK_UP) &&
4123                     bond_is_active_slave(slave)) {
4124                         if (tx_dev) {
4125                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4126                                 if (!skb2) {
4127                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4128                                                bond_dev->name);
4129                                         continue;
4130                                 }
4131
4132                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4133                                 if (res) {
4134                                         dev_kfree_skb(skb2);
4135                                         continue;
4136                                 }
4137                         }
4138                         tx_dev = slave->dev;
4139                 }
4140         }
4141
4142         if (tx_dev)
4143                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4144
4145 out:
4146         if (res)
4147                 /* no suitable interface, frame not sent */
4148                 dev_kfree_skb(skb);
4149
4150         /* frame sent to all suitable interfaces */
4151         return NETDEV_TX_OK;
4152 }
4153
4154 /*------------------------- Device initialization ---------------------------*/
4155
4156 static void bond_set_xmit_hash_policy(struct bonding *bond)
4157 {
4158         switch (bond->params.xmit_policy) {
4159         case BOND_XMIT_POLICY_LAYER23:
4160                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4161                 break;
4162         case BOND_XMIT_POLICY_LAYER34:
4163                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4164                 break;
4165         case BOND_XMIT_POLICY_LAYER2:
4166         default:
4167                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4168                 break;
4169         }
4170 }
4171
4172 /*
4173  * Lookup the slave that corresponds to a qid
4174  */
4175 static inline int bond_slave_override(struct bonding *bond,
4176                                       struct sk_buff *skb)
4177 {
4178         int i, res = 1;
4179         struct slave *slave = NULL;
4180         struct slave *check_slave;
4181
4182         if (!skb->queue_mapping)
4183                 return 1;
4184
4185         /* Find out if any slaves have the same mapping as this skb. */
4186         bond_for_each_slave(bond, check_slave, i) {
4187                 if (check_slave->queue_id == skb->queue_mapping) {
4188                         slave = check_slave;
4189                         break;
4190                 }
4191         }
4192
4193         /* If the slave isn't UP, use default transmit policy. */
4194         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4195             (slave->link == BOND_LINK_UP)) {
4196                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4197         }
4198
4199         return res;
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         if (unlikely(txq >= dev->real_num_tx_queues)) {
4213                 do {
4214                         txq -= dev->real_num_tx_queues;
4215                 } while (txq >= dev->real_num_tx_queues);
4216         }
4217         return txq;
4218 }
4219
4220 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4221 {
4222         struct bonding *bond = netdev_priv(dev);
4223
4224         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4225                 if (!bond_slave_override(bond, skb))
4226                         return NETDEV_TX_OK;
4227         }
4228
4229         switch (bond->params.mode) {
4230         case BOND_MODE_ROUNDROBIN:
4231                 return bond_xmit_roundrobin(skb, dev);
4232         case BOND_MODE_ACTIVEBACKUP:
4233                 return bond_xmit_activebackup(skb, dev);
4234         case BOND_MODE_XOR:
4235                 return bond_xmit_xor(skb, dev);
4236         case BOND_MODE_BROADCAST:
4237                 return bond_xmit_broadcast(skb, dev);
4238         case BOND_MODE_8023AD:
4239                 return bond_3ad_xmit_xor(skb, dev);
4240         case BOND_MODE_ALB:
4241         case BOND_MODE_TLB:
4242                 return bond_alb_xmit(skb, dev);
4243         default:
4244                 /* Should never happen, mode already checked */
4245                 pr_err("%s: Error: Unknown bonding mode %d\n",
4246                        dev->name, bond->params.mode);
4247                 WARN_ON_ONCE(1);
4248                 dev_kfree_skb(skb);
4249                 return NETDEV_TX_OK;
4250         }
4251 }
4252
4253 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4254 {
4255         struct bonding *bond = netdev_priv(dev);
4256         netdev_tx_t ret = NETDEV_TX_OK;
4257
4258         /*
4259          * If we risk deadlock from transmitting this in the
4260          * netpoll path, tell netpoll to queue the frame for later tx
4261          */
4262         if (is_netpoll_tx_blocked(dev))
4263                 return NETDEV_TX_BUSY;
4264
4265         read_lock(&bond->lock);
4266
4267         if (bond->slave_cnt)
4268                 ret = __bond_start_xmit(skb, dev);
4269         else
4270                 dev_kfree_skb(skb);
4271
4272         read_unlock(&bond->lock);
4273
4274         return ret;
4275 }
4276
4277 /*
4278  * set bond mode specific net device operations
4279  */
4280 void bond_set_mode_ops(struct bonding *bond, int mode)
4281 {
4282         struct net_device *bond_dev = bond->dev;
4283
4284         switch (mode) {
4285         case BOND_MODE_ROUNDROBIN:
4286                 break;
4287         case BOND_MODE_ACTIVEBACKUP:
4288                 break;
4289         case BOND_MODE_XOR:
4290                 bond_set_xmit_hash_policy(bond);
4291                 break;
4292         case BOND_MODE_BROADCAST:
4293                 break;
4294         case BOND_MODE_8023AD:
4295                 bond_set_xmit_hash_policy(bond);
4296                 break;
4297         case BOND_MODE_ALB:
4298                 /* FALLTHRU */
4299         case BOND_MODE_TLB:
4300                 break;
4301         default:
4302                 /* Should never happen, mode already checked */
4303                 pr_err("%s: Error: Unknown bonding mode %d\n",
4304                        bond_dev->name, mode);
4305                 break;
4306         }
4307 }
4308
4309 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4310                                     struct ethtool_drvinfo *drvinfo)
4311 {
4312         strncpy(drvinfo->driver, DRV_NAME, 32);
4313         strncpy(drvinfo->version, DRV_VERSION, 32);
4314         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4315 }
4316
4317 static const struct ethtool_ops bond_ethtool_ops = {
4318         .get_drvinfo            = bond_ethtool_get_drvinfo,
4319         .get_link               = ethtool_op_get_link,
4320 };
4321
4322 static const struct net_device_ops bond_netdev_ops = {
4323         .ndo_init               = bond_init,
4324         .ndo_uninit             = bond_uninit,
4325         .ndo_open               = bond_open,
4326         .ndo_stop               = bond_close,
4327         .ndo_start_xmit         = bond_start_xmit,
4328         .ndo_select_queue       = bond_select_queue,
4329         .ndo_get_stats64        = bond_get_stats,
4330         .ndo_do_ioctl           = bond_do_ioctl,
4331         .ndo_set_multicast_list = bond_set_multicast_list,
4332         .ndo_change_mtu         = bond_change_mtu,
4333         .ndo_set_mac_address    = bond_set_mac_address,
4334         .ndo_neigh_setup        = bond_neigh_setup,
4335         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4336         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4337         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4338 #ifdef CONFIG_NET_POLL_CONTROLLER
4339         .ndo_netpoll_setup      = bond_netpoll_setup,
4340         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4341         .ndo_poll_controller    = bond_poll_controller,
4342 #endif
4343         .ndo_add_slave          = bond_enslave,
4344         .ndo_del_slave          = bond_release,
4345         .ndo_fix_features       = bond_fix_features,
4346 };
4347
4348 static void bond_destructor(struct net_device *bond_dev)
4349 {
4350         struct bonding *bond = netdev_priv(bond_dev);
4351         if (bond->wq)
4352                 destroy_workqueue(bond->wq);
4353         free_netdev(bond_dev);
4354 }
4355
4356 static void bond_setup(struct net_device *bond_dev)
4357 {
4358         struct bonding *bond = netdev_priv(bond_dev);
4359
4360         /* initialize rwlocks */
4361         rwlock_init(&bond->lock);
4362         rwlock_init(&bond->curr_slave_lock);
4363
4364         bond->params = bonding_defaults;
4365
4366         /* Initialize pointers */
4367         bond->dev = bond_dev;
4368         INIT_LIST_HEAD(&bond->vlan_list);
4369
4370         /* Initialize the device entry points */
4371         ether_setup(bond_dev);
4372         bond_dev->netdev_ops = &bond_netdev_ops;
4373         bond_dev->ethtool_ops = &bond_ethtool_ops;
4374         bond_set_mode_ops(bond, bond->params.mode);
4375
4376         bond_dev->destructor = bond_destructor;
4377
4378         /* Initialize the device options */
4379         bond_dev->tx_queue_len = 0;
4380         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4381         bond_dev->priv_flags |= IFF_BONDING;
4382         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4383
4384         /* At first, we block adding VLANs. That's the only way to
4385          * prevent problems that occur when adding VLANs over an
4386          * empty bond. The block will be removed once non-challenged
4387          * slaves are enslaved.
4388          */
4389         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4390
4391         /* don't acquire bond device's netif_tx_lock when
4392          * transmitting */
4393         bond_dev->features |= NETIF_F_LLTX;
4394
4395         /* By default, we declare the bond to be fully
4396          * VLAN hardware accelerated capable. Special
4397          * care is taken in the various xmit functions
4398          * when there are slaves that are not hw accel
4399          * capable
4400          */
4401
4402         bond_dev->hw_features = BOND_VLAN_FEATURES |
4403                                 NETIF_F_HW_VLAN_TX |
4404                                 NETIF_F_HW_VLAN_RX |
4405                                 NETIF_F_HW_VLAN_FILTER;
4406
4407         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4408         bond_dev->features |= bond_dev->hw_features;
4409 }
4410
4411 static void bond_work_cancel_all(struct bonding *bond)
4412 {
4413         write_lock_bh(&bond->lock);
4414         bond->kill_timers = 1;
4415         write_unlock_bh(&bond->lock);
4416
4417         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4418                 cancel_delayed_work(&bond->mii_work);
4419
4420         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4421                 cancel_delayed_work(&bond->arp_work);
4422
4423         if (bond->params.mode == BOND_MODE_ALB &&
4424             delayed_work_pending(&bond->alb_work))
4425                 cancel_delayed_work(&bond->alb_work);
4426
4427         if (bond->params.mode == BOND_MODE_8023AD &&
4428             delayed_work_pending(&bond->ad_work))
4429                 cancel_delayed_work(&bond->ad_work);
4430
4431         if (delayed_work_pending(&bond->mcast_work))
4432                 cancel_delayed_work(&bond->mcast_work);
4433 }
4434
4435 /*
4436 * Destroy a bonding device.
4437 * Must be under rtnl_lock when this function is called.
4438 */
4439 static void bond_uninit(struct net_device *bond_dev)
4440 {
4441         struct bonding *bond = netdev_priv(bond_dev);
4442         struct vlan_entry *vlan, *tmp;
4443
4444         bond_netpoll_cleanup(bond_dev);
4445
4446         /* Release the bonded slaves */
4447         bond_release_all(bond_dev);
4448
4449         list_del(&bond->bond_list);
4450
4451         bond_work_cancel_all(bond);
4452
4453         bond_remove_proc_entry(bond);
4454
4455         bond_debug_unregister(bond);
4456
4457         __hw_addr_flush(&bond->mc_list);
4458
4459         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4460                 list_del(&vlan->vlan_list);
4461                 kfree(vlan);
4462         }
4463 }
4464
4465 /*------------------------- Module initialization ---------------------------*/
4466
4467 /*
4468  * Convert string input module parms.  Accept either the
4469  * number of the mode or its string name.  A bit complicated because
4470  * some mode names are substrings of other names, and calls from sysfs
4471  * may have whitespace in the name (trailing newlines, for example).
4472  */
4473 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4474 {
4475         int modeint = -1, i, rv;
4476         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4477
4478         for (p = (char *)buf; *p; p++)
4479                 if (!(isdigit(*p) || isspace(*p)))
4480                         break;
4481
4482         if (*p)
4483                 rv = sscanf(buf, "%20s", modestr);
4484         else
4485                 rv = sscanf(buf, "%d", &modeint);
4486
4487         if (!rv)
4488                 return -1;
4489
4490         for (i = 0; tbl[i].modename; i++) {
4491                 if (modeint == tbl[i].mode)
4492                         return tbl[i].mode;
4493                 if (strcmp(modestr, tbl[i].modename) == 0)
4494                         return tbl[i].mode;
4495         }
4496
4497         return -1;
4498 }
4499
4500 static int bond_check_params(struct bond_params *params)
4501 {
4502         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4503
4504         /*
4505          * Convert string parameters.
4506          */
4507         if (mode) {
4508                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4509                 if (bond_mode == -1) {
4510                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4511                                mode == NULL ? "NULL" : mode);
4512                         return -EINVAL;
4513                 }
4514         }
4515
4516         if (xmit_hash_policy) {
4517                 if ((bond_mode != BOND_MODE_XOR) &&
4518                     (bond_mode != BOND_MODE_8023AD)) {
4519                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4520                                bond_mode_name(bond_mode));
4521                 } else {
4522                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4523                                                         xmit_hashtype_tbl);
4524                         if (xmit_hashtype == -1) {
4525                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4526                                        xmit_hash_policy == NULL ? "NULL" :
4527                                        xmit_hash_policy);
4528                                 return -EINVAL;
4529                         }
4530                 }
4531         }
4532
4533         if (lacp_rate) {
4534                 if (bond_mode != BOND_MODE_8023AD) {
4535                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4536                                 bond_mode_name(bond_mode));
4537                 } else {
4538                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4539                         if (lacp_fast == -1) {
4540                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4541                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4542                                 return -EINVAL;
4543                         }
4544                 }
4545         }
4546
4547         if (ad_select) {
4548                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4549                 if (params->ad_select == -1) {
4550                         pr_err("Error: Invalid ad_select \"%s\"\n",
4551                                ad_select == NULL ? "NULL" : ad_select);
4552                         return -EINVAL;
4553                 }
4554
4555                 if (bond_mode != BOND_MODE_8023AD) {
4556                         pr_warning("ad_select param only affects 802.3ad mode\n");
4557                 }
4558         } else {
4559                 params->ad_select = BOND_AD_STABLE;
4560         }
4561
4562         if (max_bonds < 0) {
4563                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4564                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4565                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4566         }
4567
4568         if (miimon < 0) {
4569                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4570                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4571                 miimon = BOND_LINK_MON_INTERV;
4572         }
4573
4574         if (updelay < 0) {
4575                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4576                            updelay, INT_MAX);
4577                 updelay = 0;
4578         }
4579
4580         if (downdelay < 0) {
4581                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4582                            downdelay, INT_MAX);
4583                 downdelay = 0;
4584         }
4585
4586         if ((use_carrier != 0) && (use_carrier != 1)) {
4587                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4588                            use_carrier);
4589                 use_carrier = 1;
4590         }
4591
4592         if (num_peer_notif < 0 || num_peer_notif > 255) {
4593                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4594                            num_peer_notif);
4595                 num_peer_notif = 1;
4596         }
4597
4598         /* reset values for 802.3ad */
4599         if (bond_mode == BOND_MODE_8023AD) {
4600                 if (!miimon) {
4601                         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");
4602                         pr_warning("Forcing miimon to 100msec\n");
4603                         miimon = 100;
4604                 }
4605         }
4606
4607         if (tx_queues < 1 || tx_queues > 255) {
4608                 pr_warning("Warning: tx_queues (%d) should be between "
4609                            "1 and 255, resetting to %d\n",
4610                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4611                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4612         }
4613
4614         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4615                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4616                            "not of valid value (0/1), so it was set to "
4617                            "0\n", all_slaves_active);
4618                 all_slaves_active = 0;
4619         }
4620
4621         if (resend_igmp < 0 || resend_igmp > 255) {
4622                 pr_warning("Warning: resend_igmp (%d) should be between "
4623                            "0 and 255, resetting to %d\n",
4624                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4625                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4626         }
4627
4628         /* reset values for TLB/ALB */
4629         if ((bond_mode == BOND_MODE_TLB) ||
4630             (bond_mode == BOND_MODE_ALB)) {
4631                 if (!miimon) {
4632                         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");
4633                         pr_warning("Forcing miimon to 100msec\n");
4634                         miimon = 100;
4635                 }
4636         }
4637
4638         if (bond_mode == BOND_MODE_ALB) {
4639                 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",
4640                           updelay);
4641         }
4642
4643         if (!miimon) {
4644                 if (updelay || downdelay) {
4645                         /* just warn the user the up/down delay will have
4646                          * no effect since miimon is zero...
4647                          */
4648                         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",
4649                                    updelay, downdelay);
4650                 }
4651         } else {
4652                 /* don't allow arp monitoring */
4653                 if (arp_interval) {
4654                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4655                                    miimon, arp_interval);
4656                         arp_interval = 0;
4657                 }
4658
4659                 if ((updelay % miimon) != 0) {
4660                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4661                                    updelay, miimon,
4662                                    (updelay / miimon) * miimon);
4663                 }
4664
4665                 updelay /= miimon;
4666
4667                 if ((downdelay % miimon) != 0) {
4668                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4669                                    downdelay, miimon,
4670                                    (downdelay / miimon) * miimon);
4671                 }
4672
4673                 downdelay /= miimon;
4674         }
4675
4676         if (arp_interval < 0) {
4677                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4678                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4679                 arp_interval = BOND_LINK_ARP_INTERV;
4680         }
4681
4682         for (arp_ip_count = 0;
4683              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4684              arp_ip_count++) {
4685                 /* not complete check, but should be good enough to
4686                    catch mistakes */
4687                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4688                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4689                                    arp_ip_target[arp_ip_count]);
4690                         arp_interval = 0;
4691                 } else {
4692                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4693                         arp_target[arp_ip_count] = ip;
4694                 }
4695         }
4696
4697         if (arp_interval && !arp_ip_count) {
4698                 /* don't allow arping if no arp_ip_target given... */
4699                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4700                            arp_interval);
4701                 arp_interval = 0;
4702         }
4703
4704         if (arp_validate) {
4705                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4706                         pr_err("arp_validate only supported in active-backup mode\n");
4707                         return -EINVAL;
4708                 }
4709                 if (!arp_interval) {
4710                         pr_err("arp_validate requires arp_interval\n");
4711                         return -EINVAL;
4712                 }
4713
4714                 arp_validate_value = bond_parse_parm(arp_validate,
4715                                                      arp_validate_tbl);
4716                 if (arp_validate_value == -1) {
4717                         pr_err("Error: invalid arp_validate \"%s\"\n",
4718                                arp_validate == NULL ? "NULL" : arp_validate);
4719                         return -EINVAL;
4720                 }
4721         } else
4722                 arp_validate_value = 0;
4723
4724         if (miimon) {
4725                 pr_info("MII link monitoring set to %d ms\n", miimon);
4726         } else if (arp_interval) {
4727                 int i;
4728
4729                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4730                         arp_interval,
4731                         arp_validate_tbl[arp_validate_value].modename,
4732                         arp_ip_count);
4733
4734                 for (i = 0; i < arp_ip_count; i++)
4735                         pr_info(" %s", arp_ip_target[i]);
4736
4737                 pr_info("\n");
4738
4739         } else if (max_bonds) {
4740                 /* miimon and arp_interval not set, we need one so things
4741                  * work as expected, see bonding.txt for details
4742                  */
4743                 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");
4744         }
4745
4746         if (primary && !USES_PRIMARY(bond_mode)) {
4747                 /* currently, using a primary only makes sense
4748                  * in active backup, TLB or ALB modes
4749                  */
4750                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4751                            primary, bond_mode_name(bond_mode));
4752                 primary = NULL;
4753         }
4754
4755         if (primary && primary_reselect) {
4756                 primary_reselect_value = bond_parse_parm(primary_reselect,
4757                                                          pri_reselect_tbl);
4758                 if (primary_reselect_value == -1) {
4759                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4760                                primary_reselect ==
4761                                         NULL ? "NULL" : primary_reselect);
4762                         return -EINVAL;
4763                 }
4764         } else {
4765                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4766         }
4767
4768         if (fail_over_mac) {
4769                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4770                                                       fail_over_mac_tbl);
4771                 if (fail_over_mac_value == -1) {
4772                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4773                                arp_validate == NULL ? "NULL" : arp_validate);
4774                         return -EINVAL;
4775                 }
4776
4777                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4778                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4779         } else {
4780                 fail_over_mac_value = BOND_FOM_NONE;
4781         }
4782
4783         /* fill params struct with the proper values */
4784         params->mode = bond_mode;
4785         params->xmit_policy = xmit_hashtype;
4786         params->miimon = miimon;
4787         params->num_peer_notif = num_peer_notif;
4788         params->arp_interval = arp_interval;
4789         params->arp_validate = arp_validate_value;
4790         params->updelay = updelay;
4791         params->downdelay = downdelay;
4792         params->use_carrier = use_carrier;
4793         params->lacp_fast = lacp_fast;
4794         params->primary[0] = 0;
4795         params->primary_reselect = primary_reselect_value;
4796         params->fail_over_mac = fail_over_mac_value;
4797         params->tx_queues = tx_queues;
4798         params->all_slaves_active = all_slaves_active;
4799         params->resend_igmp = resend_igmp;
4800
4801         if (primary) {
4802                 strncpy(params->primary, primary, IFNAMSIZ);
4803                 params->primary[IFNAMSIZ - 1] = 0;
4804         }
4805
4806         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4807
4808         return 0;
4809 }
4810
4811 static struct lock_class_key bonding_netdev_xmit_lock_key;
4812 static struct lock_class_key bonding_netdev_addr_lock_key;
4813
4814 static void bond_set_lockdep_class_one(struct net_device *dev,
4815                                        struct netdev_queue *txq,
4816                                        void *_unused)
4817 {
4818         lockdep_set_class(&txq->_xmit_lock,
4819                           &bonding_netdev_xmit_lock_key);
4820 }
4821
4822 static void bond_set_lockdep_class(struct net_device *dev)
4823 {
4824         lockdep_set_class(&dev->addr_list_lock,
4825                           &bonding_netdev_addr_lock_key);
4826         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4827 }
4828
4829 /*
4830  * Called from registration process
4831  */
4832 static int bond_init(struct net_device *bond_dev)
4833 {
4834         struct bonding *bond = netdev_priv(bond_dev);
4835         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4836         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4837
4838         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4839
4840         /*
4841          * Initialize locks that may be required during
4842          * en/deslave operations.  All of the bond_open work
4843          * (of which this is part) should really be moved to
4844          * a phase prior to dev_open
4845          */
4846         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4847         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4848
4849         bond->wq = create_singlethread_workqueue(bond_dev->name);
4850         if (!bond->wq)
4851                 return -ENOMEM;
4852
4853         bond_set_lockdep_class(bond_dev);
4854
4855         bond_create_proc_entry(bond);
4856         list_add_tail(&bond->bond_list, &bn->dev_list);
4857
4858         bond_prepare_sysfs_group(bond);
4859
4860         bond_debug_register(bond);
4861
4862         __hw_addr_init(&bond->mc_list);
4863         return 0;
4864 }
4865
4866 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4867 {
4868         if (tb[IFLA_ADDRESS]) {
4869                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4870                         return -EINVAL;
4871                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4872                         return -EADDRNOTAVAIL;
4873         }
4874         return 0;
4875 }
4876
4877 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4878         .kind           = "bond",
4879         .priv_size      = sizeof(struct bonding),
4880         .setup          = bond_setup,
4881         .validate       = bond_validate,
4882 };
4883
4884 /* Create a new bond based on the specified name and bonding parameters.
4885  * If name is NULL, obtain a suitable "bond%d" name for us.
4886  * Caller must NOT hold rtnl_lock; we need to release it here before we
4887  * set up our sysfs entries.
4888  */
4889 int bond_create(struct net *net, const char *name)
4890 {
4891         struct net_device *bond_dev;
4892         int res;
4893
4894         rtnl_lock();
4895
4896         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4897                                    name ? name : "bond%d",
4898                                    bond_setup, tx_queues);
4899         if (!bond_dev) {
4900                 pr_err("%s: eek! can't alloc netdev!\n", name);
4901                 rtnl_unlock();
4902                 return -ENOMEM;
4903         }
4904
4905         dev_net_set(bond_dev, net);
4906         bond_dev->rtnl_link_ops = &bond_link_ops;
4907
4908         res = register_netdevice(bond_dev);
4909
4910         netif_carrier_off(bond_dev);
4911
4912         rtnl_unlock();
4913         if (res < 0)
4914                 bond_destructor(bond_dev);
4915         return res;
4916 }
4917
4918 static int __net_init bond_net_init(struct net *net)
4919 {
4920         struct bond_net *bn = net_generic(net, bond_net_id);
4921
4922         bn->net = net;
4923         INIT_LIST_HEAD(&bn->dev_list);
4924
4925         bond_create_proc_dir(bn);
4926
4927         return 0;
4928 }
4929
4930 static void __net_exit bond_net_exit(struct net *net)
4931 {
4932         struct bond_net *bn = net_generic(net, bond_net_id);
4933
4934         bond_destroy_proc_dir(bn);
4935 }
4936
4937 static struct pernet_operations bond_net_ops = {
4938         .init = bond_net_init,
4939         .exit = bond_net_exit,
4940         .id   = &bond_net_id,
4941         .size = sizeof(struct bond_net),
4942 };
4943
4944 static int __init bonding_init(void)
4945 {
4946         int i;
4947         int res;
4948
4949         pr_info("%s", bond_version);
4950
4951         res = bond_check_params(&bonding_defaults);
4952         if (res)
4953                 goto out;
4954
4955         res = register_pernet_subsys(&bond_net_ops);
4956         if (res)
4957                 goto out;
4958
4959         res = rtnl_link_register(&bond_link_ops);
4960         if (res)
4961                 goto err_link;
4962
4963         bond_create_debugfs();
4964
4965         for (i = 0; i < max_bonds; i++) {
4966                 res = bond_create(&init_net, NULL);
4967                 if (res)
4968                         goto err;
4969         }
4970
4971         res = bond_create_sysfs();
4972         if (res)
4973                 goto err;
4974
4975         register_netdevice_notifier(&bond_netdev_notifier);
4976         register_inetaddr_notifier(&bond_inetaddr_notifier);
4977 out:
4978         return res;
4979 err:
4980         rtnl_link_unregister(&bond_link_ops);
4981 err_link:
4982         unregister_pernet_subsys(&bond_net_ops);
4983         goto out;
4984
4985 }
4986
4987 static void __exit bonding_exit(void)
4988 {
4989         unregister_netdevice_notifier(&bond_netdev_notifier);
4990         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4991
4992         bond_destroy_sysfs();
4993         bond_destroy_debugfs();
4994
4995         rtnl_link_unregister(&bond_link_ops);
4996         unregister_pernet_subsys(&bond_net_ops);
4997
4998 #ifdef CONFIG_NET_POLL_CONTROLLER
4999         /*
5000          * Make sure we don't have an imbalance on our netpoll blocking
5001          */
5002         WARN_ON(atomic_read(&netpoll_block_tx));
5003 #endif
5004 }
5005
5006 module_init(bonding_init);
5007 module_exit(bonding_exit);
5008 MODULE_LICENSE("GPL");
5009 MODULE_VERSION(DRV_VERSION);
5010 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5011 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5012 MODULE_ALIAS_RTNL_LINK("bond");