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         switch (slave_speed) {
 633         case SPEED_10:
 634         case SPEED_100:
 635         case SPEED_1000:
 636         case SPEED_10000:
 637                 break;
 638         default:
 639                 return -1;
 640         }
 641
 642         switch (etool.duplex) {
 643         case DUPLEX_FULL:
 644         case DUPLEX_HALF:
 645                 break;
 646         default:
 647                 return -1;
 648         }
 649
 650         slave->speed = slave_speed;
 651         slave->duplex = etool.duplex;
 652
 653         return 0;
 654 }
 655
 656 /*
 657  * if <dev> supports MII link status reporting, check its link status.
 658  *
 659  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 660  * depending upon the setting of the use_carrier parameter.
 661  *
 662  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 663  * can't tell and just pretend it is), or 0, meaning that the link is
 664  * down.
 665  *
 666  * If reporting is non-zero, instead of faking link up, return -1 if
 667  * both ETHTOOL and MII ioctls fail (meaning the device does not
 668  * support them).  If use_carrier is set, return whatever it says.
 669  * It'd be nice if there was a good way to tell if a driver supports
 670  * netif_carrier, but there really isn't.
 671  */
 672 static int bond_check_dev_link(struct bonding *bond,
 673                                struct net_device *slave_dev, int reporting)
 674 {
 675         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 676         int (*ioctl)(struct net_device *, struct ifreq *, int);
 677         struct ifreq ifr;
 678         struct mii_ioctl_data *mii;
 679
 680         if (!reporting && !netif_running(slave_dev))
 681                 return 0;
 682
 683         if (bond->params.use_carrier)
 684                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 685
 686         /* Try to get link status using Ethtool first. */
 687         if (slave_dev->ethtool_ops) {
 688                 if (slave_dev->ethtool_ops->get_link) {
 689                         u32 link;
 690
 691                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 692
 693                         return link ? BMSR_LSTATUS : 0;
 694                 }
 695         }
 696
 697         /* Ethtool can't be used, fallback to MII ioctls. */
 698         ioctl = slave_ops->ndo_do_ioctl;
 699         if (ioctl) {
 700                 /* TODO: set pointer to correct ioctl on a per team member */
 701                 /*       bases to make this more efficient. that is, once  */
 702                 /*       we determine the correct ioctl, we will always    */
 703                 /*       call it and not the others for that team          */
 704                 /*       member.                                           */
 705
 706                 /*
 707                  * We cannot assume that SIOCGMIIPHY will also read a
 708                  * register; not all network drivers (e.g., e100)
 709                  * support that.
 710                  */
 711
 712                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 713                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 714                 mii = if_mii(&ifr);
 715                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 716                         mii->reg_num = MII_BMSR;
 717                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 718                                 return mii->val_out & BMSR_LSTATUS;
 719                 }
 720         }
 721
 722         /*
 723          * If reporting, report that either there's no dev->do_ioctl,
 724          * or both SIOCGMIIREG and get_link failed (meaning that we
 725          * cannot report link status).  If not reporting, pretend
 726          * we're ok.
 727          */
 728         return reporting ? -1 : BMSR_LSTATUS;
 729 }
 730
 731 /*----------------------------- Multicast list ------------------------------*/
 732
 733 /*
 734  * Push the promiscuity flag down to appropriate slaves
 735  */
 736 static int bond_set_promiscuity(struct bonding *bond, int inc)
 737 {
 738         int err = 0;
 739         if (USES_PRIMARY(bond->params.mode)) {
 740                 /* write lock already acquired */
 741                 if (bond->curr_active_slave) {
 742                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 743                                                   inc);
 744                 }
 745         } else {
 746                 struct slave *slave;
 747                 int i;
 748                 bond_for_each_slave(bond, slave, i) {
 749                         err = dev_set_promiscuity(slave->dev, inc);
 750                         if (err)
 751                                 return err;
 752                 }
 753         }
 754         return err;
 755 }
 756
 757 /*
 758  * Push the allmulti flag down to all slaves
 759  */
 760 static int bond_set_allmulti(struct bonding *bond, int inc)
 761 {
 762         int err = 0;
 763         if (USES_PRIMARY(bond->params.mode)) {
 764                 /* write lock already acquired */
 765                 if (bond->curr_active_slave) {
 766                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 767                                                inc);
 768                 }
 769         } else {
 770                 struct slave *slave;
 771                 int i;
 772                 bond_for_each_slave(bond, slave, i) {
 773                         err = dev_set_allmulti(slave->dev, inc);
 774                         if (err)
 775                                 return err;
 776                 }
 777         }
 778         return err;
 779 }
 780
 781 /*
 782  * Add a Multicast address to slaves
 783  * according to mode
 784  */
 785 static void bond_mc_add(struct bonding *bond, void *addr)
 786 {
 787         if (USES_PRIMARY(bond->params.mode)) {
 788                 /* write lock already acquired */
 789                 if (bond->curr_active_slave)
 790                         dev_mc_add(bond->curr_active_slave->dev, addr);
 791         } else {
 792                 struct slave *slave;
 793                 int i;
 794
 795                 bond_for_each_slave(bond, slave, i)
 796                         dev_mc_add(slave->dev, addr);
 797         }
 798 }
 799
 800 /*
 801  * Remove a multicast address from slave
 802  * according to mode
 803  */
 804 static void bond_mc_del(struct bonding *bond, void *addr)
 805 {
 806         if (USES_PRIMARY(bond->params.mode)) {
 807                 /* write lock already acquired */
 808                 if (bond->curr_active_slave)
 809                         dev_mc_del(bond->curr_active_slave->dev, addr);
 810         } else {
 811                 struct slave *slave;
 812                 int i;
 813                 bond_for_each_slave(bond, slave, i) {
 814                         dev_mc_del(slave->dev, addr);
 815                 }
 816         }
 817 }
 818
 819
 820 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 821 {
 822         struct in_device *in_dev;
 823
 824         rcu_read_lock();
 825         in_dev = __in_dev_get_rcu(dev);
 826         if (in_dev)
 827                 ip_mc_rejoin_groups(in_dev);
 828         rcu_read_unlock();
 829 }
 830
 831 /*
 832  * Retrieve the list of registered multicast addresses for the bonding
 833  * device and retransmit an IGMP JOIN request to the current active
 834  * slave.
 835  */
 836 static void bond_resend_igmp_join_requests(struct bonding *bond)
 837 {
 838         struct net_device *vlan_dev;
 839         struct vlan_entry *vlan;
 840
 841         read_lock(&bond->lock);
 842
 843         /* rejoin all groups on bond device */
 844         __bond_resend_igmp_join_requests(bond->dev);
 845
 846         /* rejoin all groups on vlan devices */
 847         if (bond->vlgrp) {
 848                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 849                         vlan_dev = vlan_group_get_device(bond->vlgrp,
 850                                                          vlan->vlan_id);
 851                         if (vlan_dev)
 852                                 __bond_resend_igmp_join_requests(vlan_dev);
 853                 }
 854         }
 855
 856         if (--bond->igmp_retrans > 0)
 857                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 858
 859         read_unlock(&bond->lock);
 860 }
 861
 862 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 863 {
 864         struct bonding *bond = container_of(work, struct bonding,
 865                                             mcast_work.work);
 866         bond_resend_igmp_join_requests(bond);
 867 }
 868
 869 /*
 870  * flush all members of flush->mc_list from device dev->mc_list
 871  */
 872 static void bond_mc_list_flush(struct net_device *bond_dev,
 873                                struct net_device *slave_dev)
 874 {
 875         struct bonding *bond = netdev_priv(bond_dev);
 876         struct netdev_hw_addr *ha;
 877
 878         netdev_for_each_mc_addr(ha, bond_dev)
 879                 dev_mc_del(slave_dev, ha->addr);
 880
 881         if (bond->params.mode == BOND_MODE_8023AD) {
 882                 /* del lacpdu mc addr from mc list */
 883                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 884
 885                 dev_mc_del(slave_dev, lacpdu_multicast);
 886         }
 887 }
 888
 889 /*--------------------------- Active slave change ---------------------------*/
 890
 891 /*
 892  * Update the mc list and multicast-related flags for the new and
 893  * old active slaves (if any) according to the multicast mode, and
 894  * promiscuous flags unconditionally.
 895  */
 896 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 897                          struct slave *old_active)
 898 {
 899         struct netdev_hw_addr *ha;
 900
 901         if (!USES_PRIMARY(bond->params.mode))
 902                 /* nothing to do -  mc list is already up-to-date on
 903                  * all slaves
 904                  */
 905                 return;
 906
 907         if (old_active) {
 908                 if (bond->dev->flags & IFF_PROMISC)
 909                         dev_set_promiscuity(old_active->dev, -1);
 910
 911                 if (bond->dev->flags & IFF_ALLMULTI)
 912                         dev_set_allmulti(old_active->dev, -1);
 913
 914                 netdev_for_each_mc_addr(ha, bond->dev)
 915                         dev_mc_del(old_active->dev, ha->addr);
 916         }
 917
 918         if (new_active) {
 919                 /* FIXME: Signal errors upstream. */
 920                 if (bond->dev->flags & IFF_PROMISC)
 921                         dev_set_promiscuity(new_active->dev, 1);
 922
 923                 if (bond->dev->flags & IFF_ALLMULTI)
 924                         dev_set_allmulti(new_active->dev, 1);
 925
 926                 netdev_for_each_mc_addr(ha, bond->dev)
 927                         dev_mc_add(new_active->dev, ha->addr);
 928         }
 929 }
 930
 931 /*
 932  * bond_do_fail_over_mac
 933  *
 934  * Perform special MAC address swapping for fail_over_mac settings
 935  *
 936  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 937  */
 938 static void bond_do_fail_over_mac(struct bonding *bond,
 939                                   struct slave *new_active,
 940                                   struct slave *old_active)
 941         __releases(&bond->curr_slave_lock)
 942         __releases(&bond->lock)
 943         __acquires(&bond->lock)
 944         __acquires(&bond->curr_slave_lock)
 945 {
 946         u8 tmp_mac[ETH_ALEN];
 947         struct sockaddr saddr;
 948         int rv;
 949
 950         switch (bond->params.fail_over_mac) {
 951         case BOND_FOM_ACTIVE:
 952                 if (new_active)
 953                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 954                                new_active->dev->addr_len);
 955                 break;
 956         case BOND_FOM_FOLLOW:
 957                 /*
 958                  * if new_active && old_active, swap them
 959                  * if just old_active, do nothing (going to no active slave)
 960                  * if just new_active, set new_active to bond's MAC
 961                  */
 962                 if (!new_active)
 963                         return;
 964
 965                 write_unlock_bh(&bond->curr_slave_lock);
 966                 read_unlock(&bond->lock);
 967
 968                 if (old_active) {
 969                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 970                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 971                                ETH_ALEN);
 972                         saddr.sa_family = new_active->dev->type;
 973                 } else {
 974                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 975                         saddr.sa_family = bond->dev->type;
 976                 }
 977
 978                 rv = dev_set_mac_address(new_active->dev, &saddr);
 979                 if (rv) {
 980                         pr_err("%s: Error %d setting MAC of slave %s\n",
 981                                bond->dev->name, -rv, new_active->dev->name);
 982                         goto out;
 983                 }
 984
 985                 if (!old_active)
 986                         goto out;
 987
 988                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 989                 saddr.sa_family = old_active->dev->type;
 990
 991                 rv = dev_set_mac_address(old_active->dev, &saddr);
 992                 if (rv)
 993                         pr_err("%s: Error %d setting MAC of slave %s\n",
 994                                bond->dev->name, -rv, new_active->dev->name);
 995 out:
 996                 read_lock(&bond->lock);
 997                 write_lock_bh(&bond->curr_slave_lock);
 998                 break;
 999         default:
1000                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1001                        bond->dev->name, bond->params.fail_over_mac);
1002                 break;
1003         }
1004
1005 }
1006
1007 static bool bond_should_change_active(struct bonding *bond)
1008 {
1009         struct slave *prim = bond->primary_slave;
1010         struct slave *curr = bond->curr_active_slave;
1011
1012         if (!prim || !curr || curr->link != BOND_LINK_UP)
1013                 return true;
1014         if (bond->force_primary) {
1015                 bond->force_primary = false;
1016                 return true;
1017         }
1018         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1019             (prim->speed < curr->speed ||
1020              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1021                 return false;
1022         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1023                 return false;
1024         return true;
1025 }
1026
1027 /**
1028  * find_best_interface - select the best available slave to be the active one
1029  * @bond: our bonding struct
1030  *
1031  * Warning: Caller must hold curr_slave_lock for writing.
1032  */
1033 static struct slave *bond_find_best_slave(struct bonding *bond)
1034 {
1035         struct slave *new_active, *old_active;
1036         struct slave *bestslave = NULL;
1037         int mintime = bond->params.updelay;
1038         int i;
1039
1040         new_active = bond->curr_active_slave;
1041
1042         if (!new_active) { /* there were no active slaves left */
1043                 if (bond->slave_cnt > 0)   /* found one slave */
1044                         new_active = bond->first_slave;
1045                 else
1046                         return NULL; /* still no slave, return NULL */
1047         }
1048
1049         if ((bond->primary_slave) &&
1050             bond->primary_slave->link == BOND_LINK_UP &&
1051             bond_should_change_active(bond)) {
1052                 new_active = bond->primary_slave;
1053         }
1054
1055         /* remember where to stop iterating over the slaves */
1056         old_active = new_active;
1057
1058         bond_for_each_slave_from(bond, new_active, i, old_active) {
1059                 if (new_active->link == BOND_LINK_UP) {
1060                         return new_active;
1061                 } else if (new_active->link == BOND_LINK_BACK &&
1062                            IS_UP(new_active->dev)) {
1063                         /* link up, but waiting for stabilization */
1064                         if (new_active->delay < mintime) {
1065                                 mintime = new_active->delay;
1066                                 bestslave = new_active;
1067                         }
1068                 }
1069         }
1070
1071         return bestslave;
1072 }
1073
1074 static bool bond_should_notify_peers(struct bonding *bond)
1075 {
1076         struct slave *slave = bond->curr_active_slave;
1077
1078         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1079                  bond->dev->name, slave ? slave->dev->name : "NULL");
1080
1081         if (!slave || !bond->send_peer_notif ||
1082             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1083                 return false;
1084
1085         bond->send_peer_notif--;
1086         return true;
1087 }
1088
1089 /**
1090  * change_active_interface - change the active slave into the specified one
1091  * @bond: our bonding struct
1092  * @new: the new slave to make the active one
1093  *
1094  * Set the new slave to the bond's settings and unset them on the old
1095  * curr_active_slave.
1096  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1097  *
1098  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1099  * because it is apparently the best available slave we have, even though its
1100  * updelay hasn't timed out yet.
1101  *
1102  * If new_active is not NULL, caller must hold bond->lock for read and
1103  * curr_slave_lock for write_bh.
1104  */
1105 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1106 {
1107         struct slave *old_active = bond->curr_active_slave;
1108
1109         if (old_active == new_active)
1110                 return;
1111
1112         if (new_active) {
1113                 new_active->jiffies = jiffies;
1114
1115                 if (new_active->link == BOND_LINK_BACK) {
1116                         if (USES_PRIMARY(bond->params.mode)) {
1117                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1118                                         bond->dev->name, new_active->dev->name,
1119                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1120                         }
1121
1122                         new_active->delay = 0;
1123                         new_active->link = BOND_LINK_UP;
1124
1125                         if (bond->params.mode == BOND_MODE_8023AD)
1126                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1127
1128                         if (bond_is_lb(bond))
1129                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1130                 } else {
1131                         if (USES_PRIMARY(bond->params.mode)) {
1132                                 pr_info("%s: making interface %s the new active one.\n",
1133                                         bond->dev->name, new_active->dev->name);
1134                         }
1135                 }
1136         }
1137
1138         if (USES_PRIMARY(bond->params.mode))
1139                 bond_mc_swap(bond, new_active, old_active);
1140
1141         if (bond_is_lb(bond)) {
1142                 bond_alb_handle_active_change(bond, new_active);
1143                 if (old_active)
1144                         bond_set_slave_inactive_flags(old_active);
1145                 if (new_active)
1146                         bond_set_slave_active_flags(new_active);
1147         } else {
1148                 bond->curr_active_slave = new_active;
1149         }
1150
1151         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1152                 if (old_active)
1153                         bond_set_slave_inactive_flags(old_active);
1154
1155                 if (new_active) {
1156                         bool should_notify_peers = false;
1157
1158                         bond_set_slave_active_flags(new_active);
1159
1160                         if (bond->params.fail_over_mac)
1161                                 bond_do_fail_over_mac(bond, new_active,
1162                                                       old_active);
1163
1164                         if (netif_running(bond->dev)) {
1165                                 bond->send_peer_notif =
1166                                         bond->params.num_peer_notif;
1167                                 should_notify_peers =
1168                                         bond_should_notify_peers(bond);
1169                         }
1170
1171                         write_unlock_bh(&bond->curr_slave_lock);
1172                         read_unlock(&bond->lock);
1173
1174                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1175                         if (should_notify_peers)
1176                                 netdev_bonding_change(bond->dev,
1177                                                       NETDEV_NOTIFY_PEERS);
1178
1179                         read_lock(&bond->lock);
1180                         write_lock_bh(&bond->curr_slave_lock);
1181                 }
1182         }
1183
1184         /* resend IGMP joins since active slave has changed or
1185          * all were sent on curr_active_slave.
1186          * resend only if bond is brought up with the affected
1187          * bonding modes and the retransmission is enabled */
1188         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1189             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1190              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1191                 bond->igmp_retrans = bond->params.resend_igmp;
1192                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1193         }
1194 }
1195
1196 /**
1197  * bond_select_active_slave - select a new active slave, if needed
1198  * @bond: our bonding struct
1199  *
1200  * This functions should be called when one of the following occurs:
1201  * - The old curr_active_slave has been released or lost its link.
1202  * - The primary_slave has got its link back.
1203  * - A slave has got its link back and there's no old curr_active_slave.
1204  *
1205  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1206  */
1207 void bond_select_active_slave(struct bonding *bond)
1208 {
1209         struct slave *best_slave;
1210         int rv;
1211
1212         best_slave = bond_find_best_slave(bond);
1213         if (best_slave != bond->curr_active_slave) {
1214                 bond_change_active_slave(bond, best_slave);
1215                 rv = bond_set_carrier(bond);
1216                 if (!rv)
1217                         return;
1218
1219                 if (netif_carrier_ok(bond->dev)) {
1220                         pr_info("%s: first active interface up!\n",
1221                                 bond->dev->name);
1222                 } else {
1223                         pr_info("%s: now running without any active interface !\n",
1224                                 bond->dev->name);
1225                 }
1226         }
1227 }
1228
1229 /*--------------------------- slave list handling ---------------------------*/
1230
1231 /*
1232  * This function attaches the slave to the end of list.
1233  *
1234  * bond->lock held for writing by caller.
1235  */
1236 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1237 {
1238         if (bond->first_slave == NULL) { /* attaching the first slave */
1239                 new_slave->next = new_slave;
1240                 new_slave->prev = new_slave;
1241                 bond->first_slave = new_slave;
1242         } else {
1243                 new_slave->next = bond->first_slave;
1244                 new_slave->prev = bond->first_slave->prev;
1245                 new_slave->next->prev = new_slave;
1246                 new_slave->prev->next = new_slave;
1247         }
1248
1249         bond->slave_cnt++;
1250 }
1251
1252 /*
1253  * This function detaches the slave from the list.
1254  * WARNING: no check is made to verify if the slave effectively
1255  * belongs to <bond>.
1256  * Nothing is freed on return, structures are just unchained.
1257  * If any slave pointer in bond was pointing to <slave>,
1258  * it should be changed by the calling function.
1259  *
1260  * bond->lock held for writing by caller.
1261  */
1262 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1263 {
1264         if (slave->next)
1265                 slave->next->prev = slave->prev;
1266
1267         if (slave->prev)
1268                 slave->prev->next = slave->next;
1269
1270         if (bond->first_slave == slave) { /* slave is the first slave */
1271                 if (bond->slave_cnt > 1) { /* there are more slave */
1272                         bond->first_slave = slave->next;
1273                 } else {
1274                         bond->first_slave = NULL; /* slave was the last one */
1275                 }
1276         }
1277
1278         slave->next = NULL;
1279         slave->prev = NULL;
1280         bond->slave_cnt--;
1281 }
1282
1283 #ifdef CONFIG_NET_POLL_CONTROLLER
1284 static inline int slave_enable_netpoll(struct slave *slave)
1285 {
1286         struct netpoll *np;
1287         int err = 0;
1288
1289         np = kzalloc(sizeof(*np), GFP_KERNEL);
1290         err = -ENOMEM;
1291         if (!np)
1292                 goto out;
1293
1294         np->dev = slave->dev;
1295         err = __netpoll_setup(np);
1296         if (err) {
1297                 kfree(np);
1298                 goto out;
1299         }
1300         slave->np = np;
1301 out:
1302         return err;
1303 }
1304 static inline void slave_disable_netpoll(struct slave *slave)
1305 {
1306         struct netpoll *np = slave->np;
1307
1308         if (!np)
1309                 return;
1310
1311         slave->np = NULL;
1312         synchronize_rcu_bh();
1313         __netpoll_cleanup(np);
1314         kfree(np);
1315 }
1316 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1317 {
1318         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1319                 return false;
1320         if (!slave_dev->netdev_ops->ndo_poll_controller)
1321                 return false;
1322         return true;
1323 }
1324
1325 static void bond_poll_controller(struct net_device *bond_dev)
1326 {
1327 }
1328
1329 static void __bond_netpoll_cleanup(struct bonding *bond)
1330 {
1331         struct slave *slave;
1332         int i;
1333
1334         bond_for_each_slave(bond, slave, i)
1335                 if (IS_UP(slave->dev))
1336                         slave_disable_netpoll(slave);
1337 }
1338 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1339 {
1340         struct bonding *bond = netdev_priv(bond_dev);
1341
1342         read_lock(&bond->lock);
1343         __bond_netpoll_cleanup(bond);
1344         read_unlock(&bond->lock);
1345 }
1346
1347 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1348 {
1349         struct bonding *bond = netdev_priv(dev);
1350         struct slave *slave;
1351         int i, err = 0;
1352
1353         read_lock(&bond->lock);
1354         bond_for_each_slave(bond, slave, i) {
1355                 err = slave_enable_netpoll(slave);
1356                 if (err) {
1357                         __bond_netpoll_cleanup(bond);
1358                         break;
1359                 }
1360         }
1361         read_unlock(&bond->lock);
1362         return err;
1363 }
1364
1365 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1366 {
1367         return bond->dev->npinfo;
1368 }
1369
1370 #else
1371 static inline int slave_enable_netpoll(struct slave *slave)
1372 {
1373         return 0;
1374 }
1375 static inline void slave_disable_netpoll(struct slave *slave)
1376 {
1377 }
1378 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1379 {
1380 }
1381 #endif
1382
1383 /*---------------------------------- IOCTL ----------------------------------*/
1384
1385 static int bond_sethwaddr(struct net_device *bond_dev,
1386                           struct net_device *slave_dev)
1387 {
1388         pr_debug("bond_dev=%p\n", bond_dev);
1389         pr_debug("slave_dev=%p\n", slave_dev);
1390         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1391         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1392         return 0;
1393 }
1394
1395 static u32 bond_fix_features(struct net_device *dev, u32 features)
1396 {
1397         struct slave *slave;
1398         struct bonding *bond = netdev_priv(dev);
1399         u32 mask;
1400         int i;
1401
1402         read_lock(&bond->lock);
1403
1404         if (!bond->first_slave) {
1405                 /* Disable adding VLANs to empty bond. But why? --mq */
1406                 features |= NETIF_F_VLAN_CHALLENGED;
1407                 goto out;
1408         }
1409
1410         mask = features;
1411         features &= ~NETIF_F_ONE_FOR_ALL;
1412         features |= NETIF_F_ALL_FOR_ALL;
1413
1414         bond_for_each_slave(bond, slave, i) {
1415                 features = netdev_increment_features(features,
1416                                                      slave->dev->features,
1417                                                      mask);
1418         }
1419
1420 out:
1421         read_unlock(&bond->lock);
1422         return features;
1423 }
1424
1425 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_TX_OFFLOADS | \
1426                                  NETIF_F_SOFT_FEATURES | \
1427                                  NETIF_F_LRO)
1428
1429 static void bond_compute_features(struct bonding *bond)
1430 {
1431         struct slave *slave;
1432         struct net_device *bond_dev = bond->dev;
1433         u32 vlan_features = BOND_VLAN_FEATURES;
1434         unsigned short max_hard_header_len = ETH_HLEN;
1435         int i;
1436
1437         read_lock(&bond->lock);
1438
1439         if (!bond->first_slave)
1440                 goto done;
1441
1442         bond_for_each_slave(bond, slave, i) {
1443                 vlan_features = netdev_increment_features(vlan_features,
1444                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1445
1446                 if (slave->dev->hard_header_len > max_hard_header_len)
1447                         max_hard_header_len = slave->dev->hard_header_len;
1448         }
1449
1450 done:
1451         bond_dev->vlan_features = vlan_features;
1452         bond_dev->hard_header_len = max_hard_header_len;
1453
1454         read_unlock(&bond->lock);
1455
1456         netdev_change_features(bond_dev);
1457 }
1458
1459 static void bond_setup_by_slave(struct net_device *bond_dev,
1460                                 struct net_device *slave_dev)
1461 {
1462         struct bonding *bond = netdev_priv(bond_dev);
1463
1464         bond_dev->header_ops        = slave_dev->header_ops;
1465
1466         bond_dev->type              = slave_dev->type;
1467         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1468         bond_dev->addr_len          = slave_dev->addr_len;
1469
1470         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1471                 slave_dev->addr_len);
1472         bond->setup_by_slave = 1;
1473 }
1474
1475 /* On bonding slaves other than the currently active slave, suppress
1476  * duplicates except for alb non-mcast/bcast.
1477  */
1478 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1479                                             struct slave *slave,
1480                                             struct bonding *bond)
1481 {
1482         if (bond_is_slave_inactive(slave)) {
1483                 if (bond->params.mode == BOND_MODE_ALB &&
1484                     skb->pkt_type != PACKET_BROADCAST &&
1485                     skb->pkt_type != PACKET_MULTICAST)
1486                         return false;
1487                 return true;
1488         }
1489         return false;
1490 }
1491
1492 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1493 {
1494         struct sk_buff *skb = *pskb;
1495         struct slave *slave;
1496         struct bonding *bond;
1497
1498         skb = skb_share_check(skb, GFP_ATOMIC);
1499         if (unlikely(!skb))
1500                 return RX_HANDLER_CONSUMED;
1501
1502         *pskb = skb;
1503
1504         slave = bond_slave_get_rcu(skb->dev);
1505         bond = slave->bond;
1506
1507         if (bond->params.arp_interval)
1508                 slave->dev->last_rx = jiffies;
1509
1510         if (bond->recv_probe) {
1511                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1512
1513                 if (likely(nskb)) {
1514                         bond->recv_probe(nskb, bond, slave);
1515                         dev_kfree_skb(nskb);
1516                 }
1517         }
1518
1519         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1520                 return RX_HANDLER_EXACT;
1521         }
1522
1523         skb->dev = bond->dev;
1524
1525         if (bond->params.mode == BOND_MODE_ALB &&
1526             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1527             skb->pkt_type == PACKET_HOST) {
1528
1529                 if (unlikely(skb_cow_head(skb,
1530                                           skb->data - skb_mac_header(skb)))) {
1531                         kfree_skb(skb);
1532                         return RX_HANDLER_CONSUMED;
1533                 }
1534                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1535         }
1536
1537         return RX_HANDLER_ANOTHER;
1538 }
1539
1540 /* enslave device <slave> to bond device <master> */
1541 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1542 {
1543         struct bonding *bond = netdev_priv(bond_dev);
1544         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1545         struct slave *new_slave = NULL;
1546         struct netdev_hw_addr *ha;
1547         struct sockaddr addr;
1548         int link_reporting;
1549         int res = 0;
1550
1551         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1552                 slave_ops->ndo_do_ioctl == NULL) {
1553                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1554                            bond_dev->name, slave_dev->name);
1555         }
1556
1557         /* already enslaved */
1558         if (slave_dev->flags & IFF_SLAVE) {
1559                 pr_debug("Error, Device was already enslaved\n");
1560                 return -EBUSY;
1561         }
1562
1563         /* vlan challenged mutual exclusion */
1564         /* no need to lock since we're protected by rtnl_lock */
1565         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1566                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1567                 if (bond->vlgrp) {
1568                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1569                                bond_dev->name, slave_dev->name, bond_dev->name);
1570                         return -EPERM;
1571                 } else {
1572                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1573                                    bond_dev->name, slave_dev->name,
1574                                    slave_dev->name, bond_dev->name);
1575                 }
1576         } else {
1577                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1578         }
1579
1580         /*
1581          * Old ifenslave binaries are no longer supported.  These can
1582          * be identified with moderate accuracy by the state of the slave:
1583          * the current ifenslave will set the interface down prior to
1584          * enslaving it; the old ifenslave will not.
1585          */
1586         if ((slave_dev->flags & IFF_UP)) {
1587                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1588                        slave_dev->name);
1589                 res = -EPERM;
1590                 goto err_undo_flags;
1591         }
1592
1593         /* set bonding device ether type by slave - bonding netdevices are
1594          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1595          * there is a need to override some of the type dependent attribs/funcs.
1596          *
1597          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1598          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1599          */
1600         if (bond->slave_cnt == 0) {
1601                 if (bond_dev->type != slave_dev->type) {
1602                         pr_debug("%s: change device type from %d to %d\n",
1603                                  bond_dev->name,
1604                                  bond_dev->type, slave_dev->type);
1605
1606                         res = netdev_bonding_change(bond_dev,
1607                                                     NETDEV_PRE_TYPE_CHANGE);
1608                         res = notifier_to_errno(res);
1609                         if (res) {
1610                                 pr_err("%s: refused to change device type\n",
1611                                        bond_dev->name);
1612                                 res = -EBUSY;
1613                                 goto err_undo_flags;
1614                         }
1615
1616                         /* Flush unicast and multicast addresses */
1617                         dev_uc_flush(bond_dev);
1618                         dev_mc_flush(bond_dev);
1619
1620                         if (slave_dev->type != ARPHRD_ETHER)
1621                                 bond_setup_by_slave(bond_dev, slave_dev);
1622                         else
1623                                 ether_setup(bond_dev);
1624
1625                         netdev_bonding_change(bond_dev,
1626                                               NETDEV_POST_TYPE_CHANGE);
1627                 }
1628         } else if (bond_dev->type != slave_dev->type) {
1629                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1630                        slave_dev->name,
1631                        slave_dev->type, bond_dev->type);
1632                 res = -EINVAL;
1633                 goto err_undo_flags;
1634         }
1635
1636         if (slave_ops->ndo_set_mac_address == NULL) {
1637                 if (bond->slave_cnt == 0) {
1638                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1639                                    bond_dev->name);
1640                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1641                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1642                         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",
1643                                bond_dev->name);
1644                         res = -EOPNOTSUPP;
1645                         goto err_undo_flags;
1646                 }
1647         }
1648
1649         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1650
1651         /* If this is the first slave, then we need to set the master's hardware
1652          * address to be the same as the slave's. */
1653         if (is_zero_ether_addr(bond->dev->dev_addr))
1654                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1655                        slave_dev->addr_len);
1656
1657
1658         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1659         if (!new_slave) {
1660                 res = -ENOMEM;
1661                 goto err_undo_flags;
1662         }
1663
1664         /*
1665          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1666          * is set via sysfs or module option if desired.
1667          */
1668         new_slave->queue_id = 0;
1669
1670         /* Save slave's original mtu and then set it to match the bond */
1671         new_slave->original_mtu = slave_dev->mtu;
1672         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1673         if (res) {
1674                 pr_debug("Error %d calling dev_set_mtu\n", res);
1675                 goto err_free;
1676         }
1677
1678         /*
1679          * Save slave's original ("permanent") mac address for modes
1680          * that need it, and for restoring it upon release, and then
1681          * set it to the master's address
1682          */
1683         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1684
1685         if (!bond->params.fail_over_mac) {
1686                 /*
1687                  * Set slave to master's mac address.  The application already
1688                  * set the master's mac address to that of the first slave
1689                  */
1690                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1691                 addr.sa_family = slave_dev->type;
1692                 res = dev_set_mac_address(slave_dev, &addr);
1693                 if (res) {
1694                         pr_debug("Error %d calling set_mac_address\n", res);
1695                         goto err_restore_mtu;
1696                 }
1697         }
1698
1699         res = netdev_set_bond_master(slave_dev, bond_dev);
1700         if (res) {
1701                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1702                 goto err_restore_mac;
1703         }
1704
1705         /* open the slave since the application closed it */
1706         res = dev_open(slave_dev);
1707         if (res) {
1708                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1709                 goto err_unset_master;
1710         }
1711
1712         new_slave->bond = bond;
1713         new_slave->dev = slave_dev;
1714         slave_dev->priv_flags |= IFF_BONDING;
1715
1716         if (bond_is_lb(bond)) {
1717                 /* bond_alb_init_slave() must be called before all other stages since
1718                  * it might fail and we do not want to have to undo everything
1719                  */
1720                 res = bond_alb_init_slave(bond, new_slave);
1721                 if (res)
1722                         goto err_close;
1723         }
1724
1725         /* If the mode USES_PRIMARY, then the new slave gets the
1726          * master's promisc (and mc) settings only if it becomes the
1727          * curr_active_slave, and that is taken care of later when calling
1728          * bond_change_active()
1729          */
1730         if (!USES_PRIMARY(bond->params.mode)) {
1731                 /* set promiscuity level to new slave */
1732                 if (bond_dev->flags & IFF_PROMISC) {
1733                         res = dev_set_promiscuity(slave_dev, 1);
1734                         if (res)
1735                                 goto err_close;
1736                 }
1737
1738                 /* set allmulti level to new slave */
1739                 if (bond_dev->flags & IFF_ALLMULTI) {
1740                         res = dev_set_allmulti(slave_dev, 1);
1741                         if (res)
1742                                 goto err_close;
1743                 }
1744
1745                 netif_addr_lock_bh(bond_dev);
1746                 /* upload master's mc_list to new slave */
1747                 netdev_for_each_mc_addr(ha, bond_dev)
1748                         dev_mc_add(slave_dev, ha->addr);
1749                 netif_addr_unlock_bh(bond_dev);
1750         }
1751
1752         if (bond->params.mode == BOND_MODE_8023AD) {
1753                 /* add lacpdu mc addr to mc list */
1754                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1755
1756                 dev_mc_add(slave_dev, lacpdu_multicast);
1757         }
1758
1759         bond_add_vlans_on_slave(bond, slave_dev);
1760
1761         write_lock_bh(&bond->lock);
1762
1763         bond_attach_slave(bond, new_slave);
1764
1765         new_slave->delay = 0;
1766         new_slave->link_failure_count = 0;
1767
1768         write_unlock_bh(&bond->lock);
1769
1770         bond_compute_features(bond);
1771
1772         read_lock(&bond->lock);
1773
1774         new_slave->last_arp_rx = jiffies;
1775
1776         if (bond->params.miimon && !bond->params.use_carrier) {
1777                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1778
1779                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1780                         /*
1781                          * miimon is set but a bonded network driver
1782                          * does not support ETHTOOL/MII and
1783                          * arp_interval is not set.  Note: if
1784                          * use_carrier is enabled, we will never go
1785                          * here (because netif_carrier is always
1786                          * supported); thus, we don't need to change
1787                          * the messages for netif_carrier.
1788                          */
1789                         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",
1790                                bond_dev->name, slave_dev->name);
1791                 } else if (link_reporting == -1) {
1792                         /* unable get link status using mii/ethtool */
1793                         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",
1794                                    bond_dev->name, slave_dev->name);
1795                 }
1796         }
1797
1798         /* check for initial state */
1799         if (!bond->params.miimon ||
1800             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1801                 if (bond->params.updelay) {
1802                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1803                         new_slave->link  = BOND_LINK_BACK;
1804                         new_slave->delay = bond->params.updelay;
1805                 } else {
1806                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1807                         new_slave->link  = BOND_LINK_UP;
1808                 }
1809                 new_slave->jiffies = jiffies;
1810         } else {
1811                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1812                 new_slave->link  = BOND_LINK_DOWN;
1813         }
1814
1815         if (bond_update_speed_duplex(new_slave) &&
1816             (new_slave->link != BOND_LINK_DOWN)) {
1817                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1818                            bond_dev->name, new_slave->dev->name);
1819
1820                 if (bond->params.mode == BOND_MODE_8023AD) {
1821                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1822                                    bond_dev->name);
1823                 }
1824         }
1825
1826         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1827                 /* if there is a primary slave, remember it */
1828                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1829                         bond->primary_slave = new_slave;
1830                         bond->force_primary = true;
1831                 }
1832         }
1833
1834         write_lock_bh(&bond->curr_slave_lock);
1835
1836         switch (bond->params.mode) {
1837         case BOND_MODE_ACTIVEBACKUP:
1838                 bond_set_slave_inactive_flags(new_slave);
1839                 bond_select_active_slave(bond);
1840                 break;
1841         case BOND_MODE_8023AD:
1842                 /* in 802.3ad mode, the internal mechanism
1843                  * will activate the slaves in the selected
1844                  * aggregator
1845                  */
1846                 bond_set_slave_inactive_flags(new_slave);
1847                 /* if this is the first slave */
1848                 if (bond->slave_cnt == 1) {
1849                         SLAVE_AD_INFO(new_slave).id = 1;
1850                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1851                          * can be called only after the mac address of the bond is set
1852                          */
1853                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1854                                             bond->params.lacp_fast);
1855                 } else {
1856                         SLAVE_AD_INFO(new_slave).id =
1857                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1858                 }
1859
1860                 bond_3ad_bind_slave(new_slave);
1861                 break;
1862         case BOND_MODE_TLB:
1863         case BOND_MODE_ALB:
1864                 bond_set_active_slave(new_slave);
1865                 bond_set_slave_inactive_flags(new_slave);
1866                 bond_select_active_slave(bond);
1867                 break;
1868         default:
1869                 pr_debug("This slave is always active in trunk mode\n");
1870
1871                 /* always active in trunk mode */
1872                 bond_set_active_slave(new_slave);
1873
1874                 /* In trunking mode there is little meaning to curr_active_slave
1875                  * anyway (it holds no special properties of the bond device),
1876                  * so we can change it without calling change_active_interface()
1877                  */
1878                 if (!bond->curr_active_slave)
1879                         bond->curr_active_slave = new_slave;
1880
1881                 break;
1882         } /* switch(bond_mode) */
1883
1884         write_unlock_bh(&bond->curr_slave_lock);
1885
1886         bond_set_carrier(bond);
1887
1888 #ifdef CONFIG_NET_POLL_CONTROLLER
1889         slave_dev->npinfo = bond_netpoll_info(bond);
1890         if (slave_dev->npinfo) {
1891                 if (slave_enable_netpoll(new_slave)) {
1892                         read_unlock(&bond->lock);
1893                         pr_info("Error, %s: master_dev is using netpoll, "
1894                                  "but new slave device does not support netpoll.\n",
1895                                  bond_dev->name);
1896                         res = -EBUSY;
1897                         goto err_close;
1898                 }
1899         }
1900 #endif
1901
1902         read_unlock(&bond->lock);
1903
1904         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1905         if (res)
1906                 goto err_close;
1907
1908         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1909                                          new_slave);
1910         if (res) {
1911                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1912                 goto err_dest_symlinks;
1913         }
1914
1915         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1916                 bond_dev->name, slave_dev->name,
1917                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1918                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1919
1920         /* enslave is successful */
1921         return 0;
1922
1923 /* Undo stages on error */
1924 err_dest_symlinks:
1925         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1926
1927 err_close:
1928         dev_close(slave_dev);
1929
1930 err_unset_master:
1931         netdev_set_bond_master(slave_dev, NULL);
1932
1933 err_restore_mac:
1934         if (!bond->params.fail_over_mac) {
1935                 /* XXX TODO - fom follow mode needs to change master's
1936                  * MAC if this slave's MAC is in use by the bond, or at
1937                  * least print a warning.
1938                  */
1939                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1940                 addr.sa_family = slave_dev->type;
1941                 dev_set_mac_address(slave_dev, &addr);
1942         }
1943
1944 err_restore_mtu:
1945         dev_set_mtu(slave_dev, new_slave->original_mtu);
1946
1947 err_free:
1948         kfree(new_slave);
1949
1950 err_undo_flags:
1951         bond_compute_features(bond);
1952
1953         return res;
1954 }
1955
1956 /*
1957  * Try to release the slave device <slave> from the bond device <master>
1958  * It is legal to access curr_active_slave without a lock because all the function
1959  * is write-locked.
1960  *
1961  * The rules for slave state should be:
1962  *   for Active/Backup:
1963  *     Active stays on all backups go down
1964  *   for Bonded connections:
1965  *     The first up interface should be left on and all others downed.
1966  */
1967 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1968 {
1969         struct bonding *bond = netdev_priv(bond_dev);
1970         struct slave *slave, *oldcurrent;
1971         struct sockaddr addr;
1972         u32 old_features = bond_dev->features;
1973
1974         /* slave is not a slave or master is not master of this slave */
1975         if (!(slave_dev->flags & IFF_SLAVE) ||
1976             (slave_dev->master != bond_dev)) {
1977                 pr_err("%s: Error: cannot release %s.\n",
1978                        bond_dev->name, slave_dev->name);
1979                 return -EINVAL;
1980         }
1981
1982         block_netpoll_tx();
1983         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1984         write_lock_bh(&bond->lock);
1985
1986         slave = bond_get_slave_by_dev(bond, slave_dev);
1987         if (!slave) {
1988                 /* not a slave of this bond */
1989                 pr_info("%s: %s not enslaved\n",
1990                         bond_dev->name, slave_dev->name);
1991                 write_unlock_bh(&bond->lock);
1992                 unblock_netpoll_tx();
1993                 return -EINVAL;
1994         }
1995
1996         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1997          * for this slave anymore.
1998          */
1999         netdev_rx_handler_unregister(slave_dev);
2000         write_unlock_bh(&bond->lock);
2001         synchronize_net();
2002         write_lock_bh(&bond->lock);
2003
2004         if (!bond->params.fail_over_mac) {
2005                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
2006                     bond->slave_cnt > 1)
2007                         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",
2008                                    bond_dev->name, slave_dev->name,
2009                                    slave->perm_hwaddr,
2010                                    bond_dev->name, slave_dev->name);
2011         }
2012
2013         /* Inform AD package of unbinding of slave. */
2014         if (bond->params.mode == BOND_MODE_8023AD) {
2015                 /* must be called before the slave is
2016                  * detached from the list
2017                  */
2018                 bond_3ad_unbind_slave(slave);
2019         }
2020
2021         pr_info("%s: releasing %s interface %s\n",
2022                 bond_dev->name,
2023                 bond_is_active_slave(slave) ? "active" : "backup",
2024                 slave_dev->name);
2025
2026         oldcurrent = bond->curr_active_slave;
2027
2028         bond->current_arp_slave = NULL;
2029
2030         /* release the slave from its bond */
2031         bond_detach_slave(bond, slave);
2032
2033         if (bond->primary_slave == slave)
2034                 bond->primary_slave = NULL;
2035
2036         if (oldcurrent == slave)
2037                 bond_change_active_slave(bond, NULL);
2038
2039         if (bond_is_lb(bond)) {
2040                 /* Must be called only after the slave has been
2041                  * detached from the list and the curr_active_slave
2042                  * has been cleared (if our_slave == old_current),
2043                  * but before a new active slave is selected.
2044                  */
2045                 write_unlock_bh(&bond->lock);
2046                 bond_alb_deinit_slave(bond, slave);
2047                 write_lock_bh(&bond->lock);
2048         }
2049
2050         if (oldcurrent == slave) {
2051                 /*
2052                  * Note that we hold RTNL over this sequence, so there
2053                  * is no concern that another slave add/remove event
2054                  * will interfere.
2055                  */
2056                 write_unlock_bh(&bond->lock);
2057                 read_lock(&bond->lock);
2058                 write_lock_bh(&bond->curr_slave_lock);
2059
2060                 bond_select_active_slave(bond);
2061
2062                 write_unlock_bh(&bond->curr_slave_lock);
2063                 read_unlock(&bond->lock);
2064                 write_lock_bh(&bond->lock);
2065         }
2066
2067         if (bond->slave_cnt == 0) {
2068                 bond_set_carrier(bond);
2069
2070                 /* if the last slave was removed, zero the mac address
2071                  * of the master so it will be set by the application
2072                  * to the mac address of the first slave
2073                  */
2074                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2075
2076                 if (bond->vlgrp) {
2077                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2078                                    bond_dev->name, bond_dev->name);
2079                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2080                                    bond_dev->name);
2081                 }
2082         }
2083
2084         write_unlock_bh(&bond->lock);
2085         unblock_netpoll_tx();
2086
2087         bond_compute_features(bond);
2088         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2089             (old_features & NETIF_F_VLAN_CHALLENGED))
2090                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2091                         bond_dev->name, slave_dev->name, bond_dev->name);
2092
2093         /* must do this from outside any spinlocks */
2094         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2095
2096         bond_del_vlans_from_slave(bond, slave_dev);
2097
2098         /* If the mode USES_PRIMARY, then we should only remove its
2099          * promisc and mc settings if it was the curr_active_slave, but that was
2100          * already taken care of above when we detached the slave
2101          */
2102         if (!USES_PRIMARY(bond->params.mode)) {
2103                 /* unset promiscuity level from slave */
2104                 if (bond_dev->flags & IFF_PROMISC)
2105                         dev_set_promiscuity(slave_dev, -1);
2106
2107                 /* unset allmulti level from slave */
2108                 if (bond_dev->flags & IFF_ALLMULTI)
2109                         dev_set_allmulti(slave_dev, -1);
2110
2111                 /* flush master's mc_list from slave */
2112                 netif_addr_lock_bh(bond_dev);
2113                 bond_mc_list_flush(bond_dev, slave_dev);
2114                 netif_addr_unlock_bh(bond_dev);
2115         }
2116
2117         netdev_set_bond_master(slave_dev, NULL);
2118
2119         slave_disable_netpoll(slave);
2120
2121         /* close slave before restoring its mac address */
2122         dev_close(slave_dev);
2123
2124         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2125                 /* restore original ("permanent") mac address */
2126                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2127                 addr.sa_family = slave_dev->type;
2128                 dev_set_mac_address(slave_dev, &addr);
2129         }
2130
2131         dev_set_mtu(slave_dev, slave->original_mtu);
2132
2133         slave_dev->priv_flags &= ~IFF_BONDING;
2134
2135         kfree(slave);
2136
2137         return 0;  /* deletion OK */
2138 }
2139
2140 /*
2141 * First release a slave and then destroy the bond if no more slaves are left.
2142 * Must be under rtnl_lock when this function is called.
2143 */
2144 static int  bond_release_and_destroy(struct net_device *bond_dev,
2145                                      struct net_device *slave_dev)
2146 {
2147         struct bonding *bond = netdev_priv(bond_dev);
2148         int ret;
2149
2150         ret = bond_release(bond_dev, slave_dev);
2151         if ((ret == 0) && (bond->slave_cnt == 0)) {
2152                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2153                 pr_info("%s: destroying bond %s.\n",
2154                         bond_dev->name, bond_dev->name);
2155                 unregister_netdevice(bond_dev);
2156         }
2157         return ret;
2158 }
2159
2160 /*
2161  * This function releases all slaves.
2162  */
2163 static int bond_release_all(struct net_device *bond_dev)
2164 {
2165         struct bonding *bond = netdev_priv(bond_dev);
2166         struct slave *slave;
2167         struct net_device *slave_dev;
2168         struct sockaddr addr;
2169
2170         write_lock_bh(&bond->lock);
2171
2172         netif_carrier_off(bond_dev);
2173
2174         if (bond->slave_cnt == 0)
2175                 goto out;
2176
2177         bond->current_arp_slave = NULL;
2178         bond->primary_slave = NULL;
2179         bond_change_active_slave(bond, NULL);
2180
2181         while ((slave = bond->first_slave) != NULL) {
2182                 /* Inform AD package of unbinding of slave
2183                  * before slave is detached from the list.
2184                  */
2185                 if (bond->params.mode == BOND_MODE_8023AD)
2186                         bond_3ad_unbind_slave(slave);
2187
2188                 slave_dev = slave->dev;
2189                 bond_detach_slave(bond, slave);
2190
2191                 /* now that the slave is detached, unlock and perform
2192                  * all the undo steps that should not be called from
2193                  * within a lock.
2194                  */
2195                 write_unlock_bh(&bond->lock);
2196
2197                 /* unregister rx_handler early so bond_handle_frame wouldn't
2198                  * be called for this slave anymore.
2199                  */
2200                 netdev_rx_handler_unregister(slave_dev);
2201                 synchronize_net();
2202
2203                 if (bond_is_lb(bond)) {
2204                         /* must be called only after the slave
2205                          * has been detached from the list
2206                          */
2207                         bond_alb_deinit_slave(bond, slave);
2208                 }
2209
2210                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2211                 bond_del_vlans_from_slave(bond, slave_dev);
2212
2213                 /* If the mode USES_PRIMARY, then we should only remove its
2214                  * promisc and mc settings if it was the curr_active_slave, but that was
2215                  * already taken care of above when we detached the slave
2216                  */
2217                 if (!USES_PRIMARY(bond->params.mode)) {
2218                         /* unset promiscuity level from slave */
2219                         if (bond_dev->flags & IFF_PROMISC)
2220                                 dev_set_promiscuity(slave_dev, -1);
2221
2222                         /* unset allmulti level from slave */
2223                         if (bond_dev->flags & IFF_ALLMULTI)
2224                                 dev_set_allmulti(slave_dev, -1);
2225
2226                         /* flush master's mc_list from slave */
2227                         netif_addr_lock_bh(bond_dev);
2228                         bond_mc_list_flush(bond_dev, slave_dev);
2229                         netif_addr_unlock_bh(bond_dev);
2230                 }
2231
2232                 netdev_set_bond_master(slave_dev, NULL);
2233
2234                 slave_disable_netpoll(slave);
2235
2236                 /* close slave before restoring its mac address */
2237                 dev_close(slave_dev);
2238
2239                 if (!bond->params.fail_over_mac) {
2240                         /* restore original ("permanent") mac address*/
2241                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2242                         addr.sa_family = slave_dev->type;
2243                         dev_set_mac_address(slave_dev, &addr);
2244                 }
2245
2246                 kfree(slave);
2247
2248                 /* re-acquire the lock before getting the next slave */
2249                 write_lock_bh(&bond->lock);
2250         }
2251
2252         /* zero the mac address of the master so it will be
2253          * set by the application to the mac address of the
2254          * first slave
2255          */
2256         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2257
2258         if (bond->vlgrp) {
2259                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2260                            bond_dev->name, bond_dev->name);
2261                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2262                            bond_dev->name);
2263         }
2264
2265         pr_info("%s: released all slaves\n", bond_dev->name);
2266
2267 out:
2268         write_unlock_bh(&bond->lock);
2269
2270         bond_compute_features(bond);
2271
2272         return 0;
2273 }
2274
2275 /*
2276  * This function changes the active slave to slave <slave_dev>.
2277  * It returns -EINVAL in the following cases.
2278  *  - <slave_dev> is not found in the list.
2279  *  - There is not active slave now.
2280  *  - <slave_dev> is already active.
2281  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2282  *  - <slave_dev> is not running.
2283  * In these cases, this function does nothing.
2284  * In the other cases, current_slave pointer is changed and 0 is returned.
2285  */
2286 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2287 {
2288         struct bonding *bond = netdev_priv(bond_dev);
2289         struct slave *old_active = NULL;
2290         struct slave *new_active = NULL;
2291         int res = 0;
2292
2293         if (!USES_PRIMARY(bond->params.mode))
2294                 return -EINVAL;
2295
2296         /* Verify that master_dev is indeed the master of slave_dev */
2297         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2298                 return -EINVAL;
2299
2300         read_lock(&bond->lock);
2301
2302         read_lock(&bond->curr_slave_lock);
2303         old_active = bond->curr_active_slave;
2304         read_unlock(&bond->curr_slave_lock);
2305
2306         new_active = bond_get_slave_by_dev(bond, slave_dev);
2307
2308         /*
2309          * Changing to the current active: do nothing; return success.
2310          */
2311         if (new_active && (new_active == old_active)) {
2312                 read_unlock(&bond->lock);
2313                 return 0;
2314         }
2315
2316         if ((new_active) &&
2317             (old_active) &&
2318             (new_active->link == BOND_LINK_UP) &&
2319             IS_UP(new_active->dev)) {
2320                 block_netpoll_tx();
2321                 write_lock_bh(&bond->curr_slave_lock);
2322                 bond_change_active_slave(bond, new_active);
2323                 write_unlock_bh(&bond->curr_slave_lock);
2324                 unblock_netpoll_tx();
2325         } else
2326                 res = -EINVAL;
2327
2328         read_unlock(&bond->lock);
2329
2330         return res;
2331 }
2332
2333 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2334 {
2335         struct bonding *bond = netdev_priv(bond_dev);
2336
2337         info->bond_mode = bond->params.mode;
2338         info->miimon = bond->params.miimon;
2339
2340         read_lock(&bond->lock);
2341         info->num_slaves = bond->slave_cnt;
2342         read_unlock(&bond->lock);
2343
2344         return 0;
2345 }
2346
2347 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2348 {
2349         struct bonding *bond = netdev_priv(bond_dev);
2350         struct slave *slave;
2351         int i, res = -ENODEV;
2352
2353         read_lock(&bond->lock);
2354
2355         bond_for_each_slave(bond, slave, i) {
2356                 if (i == (int)info->slave_id) {
2357                         res = 0;
2358                         strcpy(info->slave_name, slave->dev->name);
2359                         info->link = slave->link;
2360                         info->state = bond_slave_state(slave);
2361                         info->link_failure_count = slave->link_failure_count;
2362                         break;
2363                 }
2364         }
2365
2366         read_unlock(&bond->lock);
2367
2368         return res;
2369 }
2370
2371 /*-------------------------------- Monitoring -------------------------------*/
2372
2373
2374 static int bond_miimon_inspect(struct bonding *bond)
2375 {
2376         struct slave *slave;
2377         int i, link_state, commit = 0;
2378         bool ignore_updelay;
2379
2380         ignore_updelay = !bond->curr_active_slave ? true : false;
2381
2382         bond_for_each_slave(bond, slave, i) {
2383                 slave->new_link = BOND_LINK_NOCHANGE;
2384
2385                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2386
2387                 switch (slave->link) {
2388                 case BOND_LINK_UP:
2389                         if (link_state)
2390                                 continue;
2391
2392                         slave->link = BOND_LINK_FAIL;
2393                         slave->delay = bond->params.downdelay;
2394                         if (slave->delay) {
2395                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2396                                         bond->dev->name,
2397                                         (bond->params.mode ==
2398                                          BOND_MODE_ACTIVEBACKUP) ?
2399                                         (bond_is_active_slave(slave) ?
2400                                          "active " : "backup ") : "",
2401                                         slave->dev->name,
2402                                         bond->params.downdelay * bond->params.miimon);
2403                         }
2404                         /*FALLTHRU*/
2405                 case BOND_LINK_FAIL:
2406                         if (link_state) {
2407                                 /*
2408                                  * recovered before downdelay expired
2409                                  */
2410                                 slave->link = BOND_LINK_UP;
2411                                 slave->jiffies = jiffies;
2412                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2413                                         bond->dev->name,
2414                                         (bond->params.downdelay - slave->delay) *
2415                                         bond->params.miimon,
2416                                         slave->dev->name);
2417                                 continue;
2418                         }
2419
2420                         if (slave->delay <= 0) {
2421                                 slave->new_link = BOND_LINK_DOWN;
2422                                 commit++;
2423                                 continue;
2424                         }
2425
2426                         slave->delay--;
2427                         break;
2428
2429                 case BOND_LINK_DOWN:
2430                         if (!link_state)
2431                                 continue;
2432
2433                         slave->link = BOND_LINK_BACK;
2434                         slave->delay = bond->params.updelay;
2435
2436                         if (slave->delay) {
2437                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2438                                         bond->dev->name, slave->dev->name,
2439                                         ignore_updelay ? 0 :
2440                                         bond->params.updelay *
2441                                         bond->params.miimon);
2442                         }
2443                         /*FALLTHRU*/
2444                 case BOND_LINK_BACK:
2445                         if (!link_state) {
2446                                 slave->link = BOND_LINK_DOWN;
2447                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2448                                         bond->dev->name,
2449                                         (bond->params.updelay - slave->delay) *
2450                                         bond->params.miimon,
2451                                         slave->dev->name);
2452
2453                                 continue;
2454                         }
2455
2456                         if (ignore_updelay)
2457                                 slave->delay = 0;
2458
2459                         if (slave->delay <= 0) {
2460                                 slave->new_link = BOND_LINK_UP;
2461                                 commit++;
2462                                 ignore_updelay = false;
2463                                 continue;
2464                         }
2465
2466                         slave->delay--;
2467                         break;
2468                 }
2469         }
2470
2471         return commit;
2472 }
2473
2474 static void bond_miimon_commit(struct bonding *bond)
2475 {
2476         struct slave *slave;
2477         int i;
2478
2479         bond_for_each_slave(bond, slave, i) {
2480                 switch (slave->new_link) {
2481                 case BOND_LINK_NOCHANGE:
2482                         continue;
2483
2484                 case BOND_LINK_UP:
2485                         slave->link = BOND_LINK_UP;
2486                         slave->jiffies = jiffies;
2487
2488                         if (bond->params.mode == BOND_MODE_8023AD) {
2489                                 /* prevent it from being the active one */
2490                                 bond_set_backup_slave(slave);
2491                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2492                                 /* make it immediately active */
2493                                 bond_set_active_slave(slave);
2494                         } else if (slave != bond->primary_slave) {
2495                                 /* prevent it from being the active one */
2496                                 bond_set_backup_slave(slave);
2497                         }
2498
2499                         bond_update_speed_duplex(slave);
2500
2501                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2502                                 bond->dev->name, slave->dev->name,
2503                                 slave->speed, slave->duplex ? "full" : "half");
2504
2505                         /* notify ad that the link status has changed */
2506                         if (bond->params.mode == BOND_MODE_8023AD)
2507                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2508
2509                         if (bond_is_lb(bond))
2510                                 bond_alb_handle_link_change(bond, slave,
2511                                                             BOND_LINK_UP);
2512
2513                         if (!bond->curr_active_slave ||
2514                             (slave == bond->primary_slave))
2515                                 goto do_failover;
2516
2517                         continue;
2518
2519                 case BOND_LINK_DOWN:
2520                         if (slave->link_failure_count < UINT_MAX)
2521                                 slave->link_failure_count++;
2522
2523                         slave->link = BOND_LINK_DOWN;
2524
2525                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2526                             bond->params.mode == BOND_MODE_8023AD)
2527                                 bond_set_slave_inactive_flags(slave);
2528
2529                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2530                                 bond->dev->name, slave->dev->name);
2531
2532                         if (bond->params.mode == BOND_MODE_8023AD)
2533                                 bond_3ad_handle_link_change(slave,
2534                                                             BOND_LINK_DOWN);
2535
2536                         if (bond_is_lb(bond))
2537                                 bond_alb_handle_link_change(bond, slave,
2538                                                             BOND_LINK_DOWN);
2539
2540                         if (slave == bond->curr_active_slave)
2541                                 goto do_failover;
2542
2543                         continue;
2544
2545                 default:
2546                         pr_err("%s: invalid new link %d on slave %s\n",
2547                                bond->dev->name, slave->new_link,
2548                                slave->dev->name);
2549                         slave->new_link = BOND_LINK_NOCHANGE;
2550
2551                         continue;
2552                 }
2553
2554 do_failover:
2555                 ASSERT_RTNL();
2556                 block_netpoll_tx();
2557                 write_lock_bh(&bond->curr_slave_lock);
2558                 bond_select_active_slave(bond);
2559                 write_unlock_bh(&bond->curr_slave_lock);
2560                 unblock_netpoll_tx();
2561         }
2562
2563         bond_set_carrier(bond);
2564 }
2565
2566 /*
2567  * bond_mii_monitor
2568  *
2569  * Really a wrapper that splits the mii monitor into two phases: an
2570  * inspection, then (if inspection indicates something needs to be done)
2571  * an acquisition of appropriate locks followed by a commit phase to
2572  * implement whatever link state changes are indicated.
2573  */
2574 void bond_mii_monitor(struct work_struct *work)
2575 {
2576         struct bonding *bond = container_of(work, struct bonding,
2577                                             mii_work.work);
2578         bool should_notify_peers = false;
2579
2580         read_lock(&bond->lock);
2581         if (bond->kill_timers)
2582                 goto out;
2583
2584         if (bond->slave_cnt == 0)
2585                 goto re_arm;
2586
2587         should_notify_peers = bond_should_notify_peers(bond);
2588
2589         if (bond_miimon_inspect(bond)) {
2590                 read_unlock(&bond->lock);
2591                 rtnl_lock();
2592                 read_lock(&bond->lock);
2593
2594                 bond_miimon_commit(bond);
2595
2596                 read_unlock(&bond->lock);
2597                 rtnl_unlock();  /* might sleep, hold no other locks */
2598                 read_lock(&bond->lock);
2599         }
2600
2601 re_arm:
2602         if (bond->params.miimon)
2603                 queue_delayed_work(bond->wq, &bond->mii_work,
2604                                    msecs_to_jiffies(bond->params.miimon));
2605 out:
2606         read_unlock(&bond->lock);
2607
2608         if (should_notify_peers) {
2609                 rtnl_lock();
2610                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2611                 rtnl_unlock();
2612         }
2613 }
2614
2615 static __be32 bond_glean_dev_ip(struct net_device *dev)
2616 {
2617         struct in_device *idev;
2618         struct in_ifaddr *ifa;
2619         __be32 addr = 0;
2620
2621         if (!dev)
2622                 return 0;
2623
2624         rcu_read_lock();
2625         idev = __in_dev_get_rcu(dev);
2626         if (!idev)
2627                 goto out;
2628
2629         ifa = idev->ifa_list;
2630         if (!ifa)
2631                 goto out;
2632
2633         addr = ifa->ifa_local;
2634 out:
2635         rcu_read_unlock();
2636         return addr;
2637 }
2638
2639 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2640 {
2641         struct vlan_entry *vlan;
2642
2643         if (ip == bond->master_ip)
2644                 return 1;
2645
2646         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2647                 if (ip == vlan->vlan_ip)
2648                         return 1;
2649         }
2650
2651         return 0;
2652 }
2653
2654 /*
2655  * We go to the (large) trouble of VLAN tagging ARP frames because
2656  * switches in VLAN mode (especially if ports are configured as
2657  * "native" to a VLAN) might not pass non-tagged frames.
2658  */
2659 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2660 {
2661         struct sk_buff *skb;
2662
2663         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2664                  slave_dev->name, dest_ip, src_ip, vlan_id);
2665
2666         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2667                          NULL, slave_dev->dev_addr, NULL);
2668
2669         if (!skb) {
2670                 pr_err("ARP packet allocation failed\n");
2671                 return;
2672         }
2673         if (vlan_id) {
2674                 skb = vlan_put_tag(skb, vlan_id);
2675                 if (!skb) {
2676                         pr_err("failed to insert VLAN tag\n");
2677                         return;
2678                 }
2679         }
2680         arp_xmit(skb);
2681 }
2682
2683
2684 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2685 {
2686         int i, vlan_id;
2687         __be32 *targets = bond->params.arp_targets;
2688         struct vlan_entry *vlan;
2689         struct net_device *vlan_dev;
2690         struct rtable *rt;
2691
2692         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2693                 if (!targets[i])
2694                         break;
2695                 pr_debug("basa: target %x\n", targets[i]);
2696                 if (!bond->vlgrp) {
2697                         pr_debug("basa: empty vlan: arp_send\n");
2698                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2699                                       bond->master_ip, 0);
2700                         continue;
2701                 }
2702
2703                 /*
2704                  * If VLANs are configured, we do a route lookup to
2705                  * determine which VLAN interface would be used, so we
2706                  * can tag the ARP with the proper VLAN tag.
2707                  */
2708                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2709                                      RTO_ONLINK, 0);
2710                 if (IS_ERR(rt)) {
2711                         if (net_ratelimit()) {
2712                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2713                                            bond->dev->name, &targets[i]);
2714                         }
2715                         continue;
2716                 }
2717
2718                 /*
2719                  * This target is not on a VLAN
2720                  */
2721                 if (rt->dst.dev == bond->dev) {
2722                         ip_rt_put(rt);
2723                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2724                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2725                                       bond->master_ip, 0);
2726                         continue;
2727                 }
2728
2729                 vlan_id = 0;
2730                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2731                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2732                         if (vlan_dev == rt->dst.dev) {
2733                                 vlan_id = vlan->vlan_id;
2734                                 pr_debug("basa: vlan match on %s %d\n",
2735                                        vlan_dev->name, vlan_id);
2736                                 break;
2737                         }
2738                 }
2739
2740                 if (vlan_id) {
2741                         ip_rt_put(rt);
2742                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2743                                       vlan->vlan_ip, vlan_id);
2744                         continue;
2745                 }
2746
2747                 if (net_ratelimit()) {
2748                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2749                                    bond->dev->name, &targets[i],
2750                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2751                 }
2752                 ip_rt_put(rt);
2753         }
2754 }
2755
2756 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2757 {
2758         int i;
2759         __be32 *targets = bond->params.arp_targets;
2760
2761         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2762                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2763                          &sip, &tip, i, &targets[i],
2764                          bond_has_this_ip(bond, tip));
2765                 if (sip == targets[i]) {
2766                         if (bond_has_this_ip(bond, tip))
2767                                 slave->last_arp_rx = jiffies;
2768                         return;
2769                 }
2770         }
2771 }
2772
2773 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2774                          struct slave *slave)
2775 {
2776         struct arphdr *arp;
2777         unsigned char *arp_ptr;
2778         __be32 sip, tip;
2779
2780         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2781                 return;
2782
2783         read_lock(&bond->lock);
2784
2785         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2786                  bond->dev->name, skb->dev->name);
2787
2788         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2789                 goto out_unlock;
2790
2791         arp = arp_hdr(skb);
2792         if (arp->ar_hln != bond->dev->addr_len ||
2793             skb->pkt_type == PACKET_OTHERHOST ||
2794             skb->pkt_type == PACKET_LOOPBACK ||
2795             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2796             arp->ar_pro != htons(ETH_P_IP) ||
2797             arp->ar_pln != 4)
2798                 goto out_unlock;
2799
2800         arp_ptr = (unsigned char *)(arp + 1);
2801         arp_ptr += bond->dev->addr_len;
2802         memcpy(&sip, arp_ptr, 4);
2803         arp_ptr += 4 + bond->dev->addr_len;
2804         memcpy(&tip, arp_ptr, 4);
2805
2806         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2807                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2808                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2809                  &sip, &tip);
2810
2811         /*
2812          * Backup slaves won't see the ARP reply, but do come through
2813          * here for each ARP probe (so we swap the sip/tip to validate
2814          * the probe).  In a "redundant switch, common router" type of
2815          * configuration, the ARP probe will (hopefully) travel from
2816          * the active, through one switch, the router, then the other
2817          * switch before reaching the backup.
2818          */
2819         if (bond_is_active_slave(slave))
2820                 bond_validate_arp(bond, slave, sip, tip);
2821         else
2822                 bond_validate_arp(bond, slave, tip, sip);
2823
2824 out_unlock:
2825         read_unlock(&bond->lock);
2826 }
2827
2828 /*
2829  * this function is called regularly to monitor each slave's link
2830  * ensuring that traffic is being sent and received when arp monitoring
2831  * is used in load-balancing mode. if the adapter has been dormant, then an
2832  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2833  * arp monitoring in active backup mode.
2834  */
2835 void bond_loadbalance_arp_mon(struct work_struct *work)
2836 {
2837         struct bonding *bond = container_of(work, struct bonding,
2838                                             arp_work.work);
2839         struct slave *slave, *oldcurrent;
2840         int do_failover = 0;
2841         int delta_in_ticks;
2842         int i;
2843
2844         read_lock(&bond->lock);
2845
2846         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2847
2848         if (bond->kill_timers)
2849                 goto out;
2850
2851         if (bond->slave_cnt == 0)
2852                 goto re_arm;
2853
2854         read_lock(&bond->curr_slave_lock);
2855         oldcurrent = bond->curr_active_slave;
2856         read_unlock(&bond->curr_slave_lock);
2857
2858         /* see if any of the previous devices are up now (i.e. they have
2859          * xmt and rcv traffic). the curr_active_slave does not come into
2860          * the picture unless it is null. also, slave->jiffies is not needed
2861          * here because we send an arp on each slave and give a slave as
2862          * long as it needs to get the tx/rx within the delta.
2863          * TODO: what about up/down delay in arp mode? it wasn't here before
2864          *       so it can wait
2865          */
2866         bond_for_each_slave(bond, slave, i) {
2867                 unsigned long trans_start = dev_trans_start(slave->dev);
2868
2869                 if (slave->link != BOND_LINK_UP) {
2870                         if (time_in_range(jiffies,
2871                                 trans_start - delta_in_ticks,
2872                                 trans_start + delta_in_ticks) &&
2873                             time_in_range(jiffies,
2874                                 slave->dev->last_rx - delta_in_ticks,
2875                                 slave->dev->last_rx + delta_in_ticks)) {
2876
2877                                 slave->link  = BOND_LINK_UP;
2878                                 bond_set_active_slave(slave);
2879
2880                                 /* primary_slave has no meaning in round-robin
2881                                  * mode. the window of a slave being up and
2882                                  * curr_active_slave being null after enslaving
2883                                  * is closed.
2884                                  */
2885                                 if (!oldcurrent) {
2886                                         pr_info("%s: link status definitely up for interface %s, ",
2887                                                 bond->dev->name,
2888                                                 slave->dev->name);
2889                                         do_failover = 1;
2890                                 } else {
2891                                         pr_info("%s: interface %s is now up\n",
2892                                                 bond->dev->name,
2893                                                 slave->dev->name);
2894                                 }
2895                         }
2896                 } else {
2897                         /* slave->link == BOND_LINK_UP */
2898
2899                         /* not all switches will respond to an arp request
2900                          * when the source ip is 0, so don't take the link down
2901                          * if we don't know our ip yet
2902                          */
2903                         if (!time_in_range(jiffies,
2904                                 trans_start - delta_in_ticks,
2905                                 trans_start + 2 * delta_in_ticks) ||
2906                             !time_in_range(jiffies,
2907                                 slave->dev->last_rx - delta_in_ticks,
2908                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2909
2910                                 slave->link  = BOND_LINK_DOWN;
2911                                 bond_set_backup_slave(slave);
2912
2913                                 if (slave->link_failure_count < UINT_MAX)
2914                                         slave->link_failure_count++;
2915
2916                                 pr_info("%s: interface %s is now down.\n",
2917                                         bond->dev->name,
2918                                         slave->dev->name);
2919
2920                                 if (slave == oldcurrent)
2921                                         do_failover = 1;
2922                         }
2923                 }
2924
2925                 /* note: if switch is in round-robin mode, all links
2926                  * must tx arp to ensure all links rx an arp - otherwise
2927                  * links may oscillate or not come up at all; if switch is
2928                  * in something like xor mode, there is nothing we can
2929                  * do - all replies will be rx'ed on same link causing slaves
2930                  * to be unstable during low/no traffic periods
2931                  */
2932                 if (IS_UP(slave->dev))
2933                         bond_arp_send_all(bond, slave);
2934         }
2935
2936         if (do_failover) {
2937                 block_netpoll_tx();
2938                 write_lock_bh(&bond->curr_slave_lock);
2939
2940                 bond_select_active_slave(bond);
2941
2942                 write_unlock_bh(&bond->curr_slave_lock);
2943                 unblock_netpoll_tx();
2944         }
2945
2946 re_arm:
2947         if (bond->params.arp_interval)
2948                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2949 out:
2950         read_unlock(&bond->lock);
2951 }
2952
2953 /*
2954  * Called to inspect slaves for active-backup mode ARP monitor link state
2955  * changes.  Sets new_link in slaves to specify what action should take
2956  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2957  * to link states must be committed.
2958  *
2959  * Called with bond->lock held for read.
2960  */
2961 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2962 {
2963         struct slave *slave;
2964         int i, commit = 0;
2965         unsigned long trans_start;
2966
2967         bond_for_each_slave(bond, slave, i) {
2968                 slave->new_link = BOND_LINK_NOCHANGE;
2969
2970                 if (slave->link != BOND_LINK_UP) {
2971                         if (time_in_range(jiffies,
2972                                 slave_last_rx(bond, slave) - delta_in_ticks,
2973                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2974
2975                                 slave->new_link = BOND_LINK_UP;
2976                                 commit++;
2977                         }
2978
2979                         continue;
2980                 }
2981
2982                 /*
2983                  * Give slaves 2*delta after being enslaved or made
2984                  * active.  This avoids bouncing, as the last receive
2985                  * times need a full ARP monitor cycle to be updated.
2986                  */
2987                 if (time_in_range(jiffies,
2988                                   slave->jiffies - delta_in_ticks,
2989                                   slave->jiffies + 2 * delta_in_ticks))
2990                         continue;
2991
2992                 /*
2993                  * Backup slave is down if:
2994                  * - No current_arp_slave AND
2995                  * - more than 3*delta since last receive AND
2996                  * - the bond has an IP address
2997                  *
2998                  * Note: a non-null current_arp_slave indicates
2999                  * the curr_active_slave went down and we are
3000                  * searching for a new one; under this condition
3001                  * we only take the curr_active_slave down - this
3002                  * gives each slave a chance to tx/rx traffic
3003                  * before being taken out
3004                  */
3005                 if (!bond_is_active_slave(slave) &&
3006                     !bond->current_arp_slave &&
3007                     !time_in_range(jiffies,
3008                         slave_last_rx(bond, slave) - delta_in_ticks,
3009                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
3010
3011                         slave->new_link = BOND_LINK_DOWN;
3012                         commit++;
3013                 }
3014
3015                 /*
3016                  * Active slave is down if:
3017                  * - more than 2*delta since transmitting OR
3018                  * - (more than 2*delta since receive AND
3019                  *    the bond has an IP address)
3020                  */
3021                 trans_start = dev_trans_start(slave->dev);
3022                 if (bond_is_active_slave(slave) &&
3023                     (!time_in_range(jiffies,
3024                         trans_start - delta_in_ticks,
3025                         trans_start + 2 * delta_in_ticks) ||
3026                      !time_in_range(jiffies,
3027                         slave_last_rx(bond, slave) - delta_in_ticks,
3028                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3029
3030                         slave->new_link = BOND_LINK_DOWN;
3031                         commit++;
3032                 }
3033         }
3034
3035         return commit;
3036 }
3037
3038 /*
3039  * Called to commit link state changes noted by inspection step of
3040  * active-backup mode ARP monitor.
3041  *
3042  * Called with RTNL and bond->lock for read.
3043  */
3044 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3045 {
3046         struct slave *slave;
3047         int i;
3048         unsigned long trans_start;
3049
3050         bond_for_each_slave(bond, slave, i) {
3051                 switch (slave->new_link) {
3052                 case BOND_LINK_NOCHANGE:
3053                         continue;
3054
3055                 case BOND_LINK_UP:
3056                         trans_start = dev_trans_start(slave->dev);
3057                         if ((!bond->curr_active_slave &&
3058                              time_in_range(jiffies,
3059                                            trans_start - delta_in_ticks,
3060                                            trans_start + delta_in_ticks)) ||
3061                             bond->curr_active_slave != slave) {
3062                                 slave->link = BOND_LINK_UP;
3063                                 bond->current_arp_slave = NULL;
3064
3065                                 pr_info("%s: link status definitely up for interface %s.\n",
3066                                         bond->dev->name, slave->dev->name);
3067
3068                                 if (!bond->curr_active_slave ||
3069                                     (slave == bond->primary_slave))
3070                                         goto do_failover;
3071
3072                         }
3073
3074                         continue;
3075
3076                 case BOND_LINK_DOWN:
3077                         if (slave->link_failure_count < UINT_MAX)
3078                                 slave->link_failure_count++;
3079
3080                         slave->link = BOND_LINK_DOWN;
3081                         bond_set_slave_inactive_flags(slave);
3082
3083                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3084                                 bond->dev->name, slave->dev->name);
3085
3086                         if (slave == bond->curr_active_slave) {
3087                                 bond->current_arp_slave = NULL;
3088                                 goto do_failover;
3089                         }
3090
3091                         continue;
3092
3093                 default:
3094                         pr_err("%s: impossible: new_link %d on slave %s\n",
3095                                bond->dev->name, slave->new_link,
3096                                slave->dev->name);
3097                         continue;
3098                 }
3099
3100 do_failover:
3101                 ASSERT_RTNL();
3102                 block_netpoll_tx();
3103                 write_lock_bh(&bond->curr_slave_lock);
3104                 bond_select_active_slave(bond);
3105                 write_unlock_bh(&bond->curr_slave_lock);
3106                 unblock_netpoll_tx();
3107         }
3108
3109         bond_set_carrier(bond);
3110 }
3111
3112 /*
3113  * Send ARP probes for active-backup mode ARP monitor.
3114  *
3115  * Called with bond->lock held for read.
3116  */
3117 static void bond_ab_arp_probe(struct bonding *bond)
3118 {
3119         struct slave *slave;
3120         int i;
3121
3122         read_lock(&bond->curr_slave_lock);
3123
3124         if (bond->current_arp_slave && bond->curr_active_slave)
3125                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3126                         bond->current_arp_slave->dev->name,
3127                         bond->curr_active_slave->dev->name);
3128
3129         if (bond->curr_active_slave) {
3130                 bond_arp_send_all(bond, bond->curr_active_slave);
3131                 read_unlock(&bond->curr_slave_lock);
3132                 return;
3133         }
3134
3135         read_unlock(&bond->curr_slave_lock);
3136
3137         /* if we don't have a curr_active_slave, search for the next available
3138          * backup slave from the current_arp_slave and make it the candidate
3139          * for becoming the curr_active_slave
3140          */
3141
3142         if (!bond->current_arp_slave) {
3143                 bond->current_arp_slave = bond->first_slave;
3144                 if (!bond->current_arp_slave)
3145                         return;
3146         }
3147
3148         bond_set_slave_inactive_flags(bond->current_arp_slave);
3149
3150         /* search for next candidate */
3151         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3152                 if (IS_UP(slave->dev)) {
3153                         slave->link = BOND_LINK_BACK;
3154                         bond_set_slave_active_flags(slave);
3155                         bond_arp_send_all(bond, slave);
3156                         slave->jiffies = jiffies;
3157                         bond->current_arp_slave = slave;
3158                         break;
3159                 }
3160
3161                 /* if the link state is up at this point, we
3162                  * mark it down - this can happen if we have
3163                  * simultaneous link failures and
3164                  * reselect_active_interface doesn't make this
3165                  * one the current slave so it is still marked
3166                  * up when it is actually down
3167                  */
3168                 if (slave->link == BOND_LINK_UP) {
3169                         slave->link = BOND_LINK_DOWN;
3170                         if (slave->link_failure_count < UINT_MAX)
3171                                 slave->link_failure_count++;
3172
3173                         bond_set_slave_inactive_flags(slave);
3174
3175                         pr_info("%s: backup interface %s is now down.\n",
3176                                 bond->dev->name, slave->dev->name);
3177                 }
3178         }
3179 }
3180
3181 void bond_activebackup_arp_mon(struct work_struct *work)
3182 {
3183         struct bonding *bond = container_of(work, struct bonding,
3184                                             arp_work.work);
3185         bool should_notify_peers = false;
3186         int delta_in_ticks;
3187
3188         read_lock(&bond->lock);
3189
3190         if (bond->kill_timers)
3191                 goto out;
3192
3193         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3194
3195         if (bond->slave_cnt == 0)
3196                 goto re_arm;
3197
3198         should_notify_peers = bond_should_notify_peers(bond);
3199
3200         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3201                 read_unlock(&bond->lock);
3202                 rtnl_lock();
3203                 read_lock(&bond->lock);
3204
3205                 bond_ab_arp_commit(bond, delta_in_ticks);
3206
3207                 read_unlock(&bond->lock);
3208                 rtnl_unlock();
3209                 read_lock(&bond->lock);
3210         }
3211
3212         bond_ab_arp_probe(bond);
3213
3214 re_arm:
3215         if (bond->params.arp_interval)
3216                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3217 out:
3218         read_unlock(&bond->lock);
3219
3220         if (should_notify_peers) {
3221                 rtnl_lock();
3222                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3223                 rtnl_unlock();
3224         }
3225 }
3226
3227 /*-------------------------- netdev event handling --------------------------*/
3228
3229 /*
3230  * Change device name
3231  */
3232 static int bond_event_changename(struct bonding *bond)
3233 {
3234         bond_remove_proc_entry(bond);
3235         bond_create_proc_entry(bond);
3236
3237         bond_debug_reregister(bond);
3238
3239         return NOTIFY_DONE;
3240 }
3241
3242 static int bond_master_netdev_event(unsigned long event,
3243                                     struct net_device *bond_dev)
3244 {
3245         struct bonding *event_bond = netdev_priv(bond_dev);
3246
3247         switch (event) {
3248         case NETDEV_CHANGENAME:
3249                 return bond_event_changename(event_bond);
3250         default:
3251                 break;
3252         }
3253
3254         return NOTIFY_DONE;
3255 }
3256
3257 static int bond_slave_netdev_event(unsigned long event,
3258                                    struct net_device *slave_dev)
3259 {
3260         struct net_device *bond_dev = slave_dev->master;
3261         struct bonding *bond = netdev_priv(bond_dev);
3262
3263         switch (event) {
3264         case NETDEV_UNREGISTER:
3265                 if (bond_dev) {
3266                         if (bond->setup_by_slave)
3267                                 bond_release_and_destroy(bond_dev, slave_dev);
3268                         else
3269                                 bond_release(bond_dev, slave_dev);
3270                 }
3271                 break;
3272         case NETDEV_CHANGE:
3273                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3274                         struct slave *slave;
3275
3276                         slave = bond_get_slave_by_dev(bond, slave_dev);
3277                         if (slave) {
3278                                 u32 old_speed = slave->speed;
3279                                 u8  old_duplex = slave->duplex;
3280
3281                                 bond_update_speed_duplex(slave);
3282
3283                                 if (bond_is_lb(bond))
3284                                         break;
3285
3286                                 if (old_speed != slave->speed)
3287                                         bond_3ad_adapter_speed_changed(slave);
3288                                 if (old_duplex != slave->duplex)
3289                                         bond_3ad_adapter_duplex_changed(slave);
3290                         }
3291                 }
3292
3293                 break;
3294         case NETDEV_DOWN:
3295                 /*
3296                  * ... Or is it this?
3297                  */
3298                 break;
3299         case NETDEV_CHANGEMTU:
3300                 /*
3301                  * TODO: Should slaves be allowed to
3302                  * independently alter their MTU?  For
3303                  * an active-backup bond, slaves need
3304                  * not be the same type of device, so
3305                  * MTUs may vary.  For other modes,
3306                  * slaves arguably should have the
3307                  * same MTUs. To do this, we'd need to
3308                  * take over the slave's change_mtu
3309                  * function for the duration of their
3310                  * servitude.
3311                  */
3312                 break;
3313         case NETDEV_CHANGENAME:
3314                 /*
3315                  * TODO: handle changing the primary's name
3316                  */
3317                 break;
3318         case NETDEV_FEAT_CHANGE:
3319                 bond_compute_features(bond);
3320                 break;
3321         default:
3322                 break;
3323         }
3324
3325         return NOTIFY_DONE;
3326 }
3327
3328 /*
3329  * bond_netdev_event: handle netdev notifier chain events.
3330  *
3331  * This function receives events for the netdev chain.  The caller (an
3332  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3333  * locks for us to safely manipulate the slave devices (RTNL lock,
3334  * dev_probe_lock).
3335  */
3336 static int bond_netdev_event(struct notifier_block *this,
3337                              unsigned long event, void *ptr)
3338 {
3339         struct net_device *event_dev = (struct net_device *)ptr;
3340
3341         pr_debug("event_dev: %s, event: %lx\n",
3342                  event_dev ? event_dev->name : "None",
3343                  event);
3344
3345         if (!(event_dev->priv_flags & IFF_BONDING))
3346                 return NOTIFY_DONE;
3347
3348         if (event_dev->flags & IFF_MASTER) {
3349                 pr_debug("IFF_MASTER\n");
3350                 return bond_master_netdev_event(event, event_dev);
3351         }
3352
3353         if (event_dev->flags & IFF_SLAVE) {
3354                 pr_debug("IFF_SLAVE\n");
3355                 return bond_slave_netdev_event(event, event_dev);
3356         }
3357
3358         return NOTIFY_DONE;
3359 }
3360
3361 /*
3362  * bond_inetaddr_event: handle inetaddr notifier chain events.
3363  *
3364  * We keep track of device IPs primarily to use as source addresses in
3365  * ARP monitor probes (rather than spewing out broadcasts all the time).
3366  *
3367  * We track one IP for the main device (if it has one), plus one per VLAN.
3368  */
3369 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3370 {
3371         struct in_ifaddr *ifa = ptr;
3372         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3373         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3374         struct bonding *bond;
3375         struct vlan_entry *vlan;
3376
3377         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3378                 if (bond->dev == event_dev) {
3379                         switch (event) {
3380                         case NETDEV_UP:
3381                                 bond->master_ip = ifa->ifa_local;
3382                                 return NOTIFY_OK;
3383                         case NETDEV_DOWN:
3384                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3385                                 return NOTIFY_OK;
3386                         default:
3387                                 return NOTIFY_DONE;
3388                         }
3389                 }
3390
3391                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3392                         if (!bond->vlgrp)
3393                                 continue;
3394                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3395                         if (vlan_dev == event_dev) {
3396                                 switch (event) {
3397                                 case NETDEV_UP:
3398                                         vlan->vlan_ip = ifa->ifa_local;
3399                                         return NOTIFY_OK;
3400                                 case NETDEV_DOWN:
3401                                         vlan->vlan_ip =
3402                                                 bond_glean_dev_ip(vlan_dev);
3403                                         return NOTIFY_OK;
3404                                 default:
3405                                         return NOTIFY_DONE;
3406                                 }
3407                         }
3408                 }
3409         }
3410         return NOTIFY_DONE;
3411 }
3412
3413 static struct notifier_block bond_netdev_notifier = {
3414         .notifier_call = bond_netdev_event,
3415 };
3416
3417 static struct notifier_block bond_inetaddr_notifier = {
3418         .notifier_call = bond_inetaddr_event,
3419 };
3420
3421 /*---------------------------- Hashing Policies -----------------------------*/
3422
3423 /*
3424  * Hash for the output device based upon layer 2 and layer 3 data. If
3425  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3426  */
3427 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3428 {
3429         struct ethhdr *data = (struct ethhdr *)skb->data;
3430         struct iphdr *iph = ip_hdr(skb);
3431
3432         if (skb->protocol == htons(ETH_P_IP)) {
3433                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3434                         (data->h_dest[5] ^ data->h_source[5])) % count;
3435         }
3436
3437         return (data->h_dest[5] ^ data->h_source[5]) % count;
3438 }
3439
3440 /*
3441  * Hash for the output device based upon layer 3 and layer 4 data. If
3442  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3443  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3444  */
3445 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3446 {
3447         struct ethhdr *data = (struct ethhdr *)skb->data;
3448         struct iphdr *iph = ip_hdr(skb);
3449         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3450         int layer4_xor = 0;
3451
3452         if (skb->protocol == htons(ETH_P_IP)) {
3453                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3454                     (iph->protocol == IPPROTO_TCP ||
3455                      iph->protocol == IPPROTO_UDP)) {
3456                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3457                 }
3458                 return (layer4_xor ^
3459                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3460
3461         }
3462
3463         return (data->h_dest[5] ^ data->h_source[5]) % count;
3464 }
3465
3466 /*
3467  * Hash for the output device based upon layer 2 data
3468  */
3469 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3470 {
3471         struct ethhdr *data = (struct ethhdr *)skb->data;
3472
3473         return (data->h_dest[5] ^ data->h_source[5]) % count;
3474 }
3475
3476 /*-------------------------- Device entry points ----------------------------*/
3477
3478 static int bond_open(struct net_device *bond_dev)
3479 {
3480         struct bonding *bond = netdev_priv(bond_dev);
3481
3482         bond->kill_timers = 0;
3483
3484         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3485
3486         if (bond_is_lb(bond)) {
3487                 /* bond_alb_initialize must be called before the timer
3488                  * is started.
3489                  */
3490                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3491                         /* something went wrong - fail the open operation */
3492                         return -ENOMEM;
3493                 }
3494
3495                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3496                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3497         }
3498
3499         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3500                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3501                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3502         }
3503
3504         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3505                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3506                         INIT_DELAYED_WORK(&bond->arp_work,
3507                                           bond_activebackup_arp_mon);
3508                 else
3509                         INIT_DELAYED_WORK(&bond->arp_work,
3510                                           bond_loadbalance_arp_mon);
3511
3512                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3513                 if (bond->params.arp_validate)
3514                         bond->recv_probe = bond_arp_rcv;
3515         }
3516
3517         if (bond->params.mode == BOND_MODE_8023AD) {
3518                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3519                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3520                 /* register to receive LACPDUs */
3521                 bond->recv_probe = bond_3ad_lacpdu_recv;
3522                 bond_3ad_initiate_agg_selection(bond, 1);
3523         }
3524
3525         return 0;
3526 }
3527
3528 static int bond_close(struct net_device *bond_dev)
3529 {
3530         struct bonding *bond = netdev_priv(bond_dev);
3531
3532         write_lock_bh(&bond->lock);
3533
3534         bond->send_peer_notif = 0;
3535
3536         /* signal timers not to re-arm */
3537         bond->kill_timers = 1;
3538
3539         write_unlock_bh(&bond->lock);
3540
3541         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3542                 cancel_delayed_work(&bond->mii_work);
3543         }
3544
3545         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3546                 cancel_delayed_work(&bond->arp_work);
3547         }
3548
3549         switch (bond->params.mode) {
3550         case BOND_MODE_8023AD:
3551                 cancel_delayed_work(&bond->ad_work);
3552                 break;
3553         case BOND_MODE_TLB:
3554         case BOND_MODE_ALB:
3555                 cancel_delayed_work(&bond->alb_work);
3556                 break;
3557         default:
3558                 break;
3559         }
3560
3561         if (delayed_work_pending(&bond->mcast_work))
3562                 cancel_delayed_work(&bond->mcast_work);
3563
3564         if (bond_is_lb(bond)) {
3565                 /* Must be called only after all
3566                  * slaves have been released
3567                  */
3568                 bond_alb_deinitialize(bond);
3569         }
3570         bond->recv_probe = NULL;
3571
3572         return 0;
3573 }
3574
3575 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3576                                                 struct rtnl_link_stats64 *stats)
3577 {
3578         struct bonding *bond = netdev_priv(bond_dev);
3579         struct rtnl_link_stats64 temp;
3580         struct slave *slave;
3581         int i;
3582
3583         memset(stats, 0, sizeof(*stats));
3584
3585         read_lock_bh(&bond->lock);
3586
3587         bond_for_each_slave(bond, slave, i) {
3588                 const struct rtnl_link_stats64 *sstats =
3589                         dev_get_stats(slave->dev, &temp);
3590
3591                 stats->rx_packets += sstats->rx_packets;
3592                 stats->rx_bytes += sstats->rx_bytes;
3593                 stats->rx_errors += sstats->rx_errors;
3594                 stats->rx_dropped += sstats->rx_dropped;
3595
3596                 stats->tx_packets += sstats->tx_packets;
3597                 stats->tx_bytes += sstats->tx_bytes;
3598                 stats->tx_errors += sstats->tx_errors;
3599                 stats->tx_dropped += sstats->tx_dropped;
3600
3601                 stats->multicast += sstats->multicast;
3602                 stats->collisions += sstats->collisions;
3603
3604                 stats->rx_length_errors += sstats->rx_length_errors;
3605                 stats->rx_over_errors += sstats->rx_over_errors;
3606                 stats->rx_crc_errors += sstats->rx_crc_errors;
3607                 stats->rx_frame_errors += sstats->rx_frame_errors;
3608                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3609                 stats->rx_missed_errors += sstats->rx_missed_errors;
3610
3611                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3612                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3613                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3614                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3615                 stats->tx_window_errors += sstats->tx_window_errors;
3616         }
3617
3618         read_unlock_bh(&bond->lock);
3619
3620         return stats;
3621 }
3622
3623 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3624 {
3625         struct net_device *slave_dev = NULL;
3626         struct ifbond k_binfo;
3627         struct ifbond __user *u_binfo = NULL;
3628         struct ifslave k_sinfo;
3629         struct ifslave __user *u_sinfo = NULL;
3630         struct mii_ioctl_data *mii = NULL;
3631         int res = 0;
3632
3633         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3634
3635         switch (cmd) {
3636         case SIOCGMIIPHY:
3637                 mii = if_mii(ifr);
3638                 if (!mii)
3639                         return -EINVAL;
3640
3641                 mii->phy_id = 0;
3642                 /* Fall Through */
3643         case SIOCGMIIREG:
3644                 /*
3645                  * We do this again just in case we were called by SIOCGMIIREG
3646                  * instead of SIOCGMIIPHY.
3647                  */
3648                 mii = if_mii(ifr);
3649                 if (!mii)
3650                         return -EINVAL;
3651
3652
3653                 if (mii->reg_num == 1) {
3654                         struct bonding *bond = netdev_priv(bond_dev);
3655                         mii->val_out = 0;
3656                         read_lock(&bond->lock);
3657                         read_lock(&bond->curr_slave_lock);
3658                         if (netif_carrier_ok(bond->dev))
3659                                 mii->val_out = BMSR_LSTATUS;
3660
3661                         read_unlock(&bond->curr_slave_lock);
3662                         read_unlock(&bond->lock);
3663                 }
3664
3665                 return 0;
3666         case BOND_INFO_QUERY_OLD:
3667         case SIOCBONDINFOQUERY:
3668                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3669
3670                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3671                         return -EFAULT;
3672
3673                 res = bond_info_query(bond_dev, &k_binfo);
3674                 if (res == 0 &&
3675                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3676                         return -EFAULT;
3677
3678                 return res;
3679         case BOND_SLAVE_INFO_QUERY_OLD:
3680         case SIOCBONDSLAVEINFOQUERY:
3681                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3682
3683                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3684                         return -EFAULT;
3685
3686                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3687                 if (res == 0 &&
3688                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3689                         return -EFAULT;
3690
3691                 return res;
3692         default:
3693                 /* Go on */
3694                 break;
3695         }
3696
3697         if (!capable(CAP_NET_ADMIN))
3698                 return -EPERM;
3699
3700         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3701
3702         pr_debug("slave_dev=%p:\n", slave_dev);
3703
3704         if (!slave_dev)
3705                 res = -ENODEV;
3706         else {
3707                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3708                 switch (cmd) {
3709                 case BOND_ENSLAVE_OLD:
3710                 case SIOCBONDENSLAVE:
3711                         res = bond_enslave(bond_dev, slave_dev);
3712                         break;
3713                 case BOND_RELEASE_OLD:
3714                 case SIOCBONDRELEASE:
3715                         res = bond_release(bond_dev, slave_dev);
3716                         break;
3717                 case BOND_SETHWADDR_OLD:
3718                 case SIOCBONDSETHWADDR:
3719                         res = bond_sethwaddr(bond_dev, slave_dev);
3720                         break;
3721                 case BOND_CHANGE_ACTIVE_OLD:
3722                 case SIOCBONDCHANGEACTIVE:
3723                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3724                         break;
3725                 default:
3726                         res = -EOPNOTSUPP;
3727                 }
3728
3729                 dev_put(slave_dev);
3730         }
3731
3732         return res;
3733 }
3734
3735 static bool bond_addr_in_mc_list(unsigned char *addr,
3736                                  struct netdev_hw_addr_list *list,
3737                                  int addrlen)
3738 {
3739         struct netdev_hw_addr *ha;
3740
3741         netdev_hw_addr_list_for_each(ha, list)
3742                 if (!memcmp(ha->addr, addr, addrlen))
3743                         return true;
3744
3745         return false;
3746 }
3747
3748 static void bond_set_multicast_list(struct net_device *bond_dev)
3749 {
3750         struct bonding *bond = netdev_priv(bond_dev);
3751         struct netdev_hw_addr *ha;
3752         bool found;
3753
3754         /*
3755          * Do promisc before checking multicast_mode
3756          */
3757         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3758                 /*
3759                  * FIXME: Need to handle the error when one of the multi-slaves
3760                  * encounters error.
3761                  */
3762                 bond_set_promiscuity(bond, 1);
3763
3764
3765         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3766                 bond_set_promiscuity(bond, -1);
3767
3768
3769         /* set allmulti flag to slaves */
3770         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3771                 /*
3772                  * FIXME: Need to handle the error when one of the multi-slaves
3773                  * encounters error.
3774                  */
3775                 bond_set_allmulti(bond, 1);
3776
3777
3778         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3779                 bond_set_allmulti(bond, -1);
3780
3781
3782         read_lock(&bond->lock);
3783
3784         bond->flags = bond_dev->flags;
3785
3786         /* looking for addresses to add to slaves' mc list */
3787         netdev_for_each_mc_addr(ha, bond_dev) {
3788                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3789                                              bond_dev->addr_len);
3790                 if (!found)
3791                         bond_mc_add(bond, ha->addr);
3792         }
3793
3794         /* looking for addresses to delete from slaves' list */
3795         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3796                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3797                                              bond_dev->addr_len);
3798                 if (!found)
3799                         bond_mc_del(bond, ha->addr);
3800         }
3801
3802         /* save master's multicast list */
3803         __hw_addr_flush(&bond->mc_list);
3804         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3805                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3806
3807         read_unlock(&bond->lock);
3808 }
3809
3810 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3811 {
3812         struct bonding *bond = netdev_priv(dev);
3813         struct slave *slave = bond->first_slave;
3814
3815         if (slave) {
3816                 const struct net_device_ops *slave_ops
3817                         = slave->dev->netdev_ops;
3818                 if (slave_ops->ndo_neigh_setup)
3819                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3820         }
3821         return 0;
3822 }
3823
3824 /*
3825  * Change the MTU of all of a master's slaves to match the master
3826  */
3827 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3828 {
3829         struct bonding *bond = netdev_priv(bond_dev);
3830         struct slave *slave, *stop_at;
3831         int res = 0;
3832         int i;
3833
3834         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3835                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3836
3837         /* Can't hold bond->lock with bh disabled here since
3838          * some base drivers panic. On the other hand we can't
3839          * hold bond->lock without bh disabled because we'll
3840          * deadlock. The only solution is to rely on the fact
3841          * that we're under rtnl_lock here, and the slaves
3842          * list won't change. This doesn't solve the problem
3843          * of setting the slave's MTU while it is
3844          * transmitting, but the assumption is that the base
3845          * driver can handle that.
3846          *
3847          * TODO: figure out a way to safely iterate the slaves
3848          * list, but without holding a lock around the actual
3849          * call to the base driver.
3850          */
3851
3852         bond_for_each_slave(bond, slave, i) {
3853                 pr_debug("s %p s->p %p c_m %p\n",
3854                          slave,
3855                          slave->prev,
3856                          slave->dev->netdev_ops->ndo_change_mtu);
3857
3858                 res = dev_set_mtu(slave->dev, new_mtu);
3859
3860                 if (res) {
3861                         /* If we failed to set the slave's mtu to the new value
3862                          * we must abort the operation even in ACTIVE_BACKUP
3863                          * mode, because if we allow the backup slaves to have
3864                          * different mtu values than the active slave we'll
3865                          * need to change their mtu when doing a failover. That
3866                          * means changing their mtu from timer context, which
3867                          * is probably not a good idea.
3868                          */
3869                         pr_debug("err %d %s\n", res, slave->dev->name);
3870                         goto unwind;
3871                 }
3872         }
3873
3874         bond_dev->mtu = new_mtu;
3875
3876         return 0;
3877
3878 unwind:
3879         /* unwind from head to the slave that failed */
3880         stop_at = slave;
3881         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3882                 int tmp_res;
3883
3884                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3885                 if (tmp_res) {
3886                         pr_debug("unwind err %d dev %s\n",
3887                                  tmp_res, slave->dev->name);
3888                 }
3889         }
3890
3891         return res;
3892 }
3893
3894 /*
3895  * Change HW address
3896  *
3897  * Note that many devices must be down to change the HW address, and
3898  * downing the master releases all slaves.  We can make bonds full of
3899  * bonding devices to test this, however.
3900  */
3901 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3902 {
3903         struct bonding *bond = netdev_priv(bond_dev);
3904         struct sockaddr *sa = addr, tmp_sa;
3905         struct slave *slave, *stop_at;
3906         int res = 0;
3907         int i;
3908
3909         if (bond->params.mode == BOND_MODE_ALB)
3910                 return bond_alb_set_mac_address(bond_dev, addr);
3911
3912
3913         pr_debug("bond=%p, name=%s\n",
3914                  bond, bond_dev ? bond_dev->name : "None");
3915
3916         /*
3917          * If fail_over_mac is set to active, do nothing and return
3918          * success.  Returning an error causes ifenslave to fail.
3919          */
3920         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3921                 return 0;
3922
3923         if (!is_valid_ether_addr(sa->sa_data))
3924                 return -EADDRNOTAVAIL;
3925
3926         /* Can't hold bond->lock with bh disabled here since
3927          * some base drivers panic. On the other hand we can't
3928          * hold bond->lock without bh disabled because we'll
3929          * deadlock. The only solution is to rely on the fact
3930          * that we're under rtnl_lock here, and the slaves
3931          * list won't change. This doesn't solve the problem
3932          * of setting the slave's hw address while it is
3933          * transmitting, but the assumption is that the base
3934          * driver can handle that.
3935          *
3936          * TODO: figure out a way to safely iterate the slaves
3937          * list, but without holding a lock around the actual
3938          * call to the base driver.
3939          */
3940
3941         bond_for_each_slave(bond, slave, i) {
3942                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3943                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3944
3945                 if (slave_ops->ndo_set_mac_address == NULL) {
3946                         res = -EOPNOTSUPP;
3947                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3948                         goto unwind;
3949                 }
3950
3951                 res = dev_set_mac_address(slave->dev, addr);
3952                 if (res) {
3953                         /* TODO: consider downing the slave
3954                          * and retry ?
3955                          * User should expect communications
3956                          * breakage anyway until ARP finish
3957                          * updating, so...
3958                          */
3959                         pr_debug("err %d %s\n", res, slave->dev->name);
3960                         goto unwind;
3961                 }
3962         }
3963
3964         /* success */
3965         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3966         return 0;
3967
3968 unwind:
3969         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3970         tmp_sa.sa_family = bond_dev->type;
3971
3972         /* unwind from head to the slave that failed */
3973         stop_at = slave;
3974         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3975                 int tmp_res;
3976
3977                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3978                 if (tmp_res) {
3979                         pr_debug("unwind err %d dev %s\n",
3980                                  tmp_res, slave->dev->name);
3981                 }
3982         }
3983
3984         return res;
3985 }
3986
3987 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3988 {
3989         struct bonding *bond = netdev_priv(bond_dev);
3990         struct slave *slave, *start_at;
3991         int i, slave_no, res = 1;
3992         struct iphdr *iph = ip_hdr(skb);
3993
3994         /*
3995          * Start with the curr_active_slave that joined the bond as the
3996          * default for sending IGMP traffic.  For failover purposes one
3997          * needs to maintain some consistency for the interface that will
3998          * send the join/membership reports.  The curr_active_slave found
3999          * will send all of this type of traffic.
4000          */
4001         if ((iph->protocol == IPPROTO_IGMP) &&
4002             (skb->protocol == htons(ETH_P_IP))) {
4003
4004                 read_lock(&bond->curr_slave_lock);
4005                 slave = bond->curr_active_slave;
4006                 read_unlock(&bond->curr_slave_lock);
4007
4008                 if (!slave)
4009                         goto out;
4010         } else {
4011                 /*
4012                  * Concurrent TX may collide on rr_tx_counter; we accept
4013                  * that as being rare enough not to justify using an
4014                  * atomic op here.
4015                  */
4016                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4017
4018                 bond_for_each_slave(bond, slave, i) {
4019                         slave_no--;
4020                         if (slave_no < 0)
4021                                 break;
4022                 }
4023         }
4024
4025         start_at = slave;
4026         bond_for_each_slave_from(bond, slave, i, start_at) {
4027                 if (IS_UP(slave->dev) &&
4028                     (slave->link == BOND_LINK_UP) &&
4029                     bond_is_active_slave(slave)) {
4030                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4031                         break;
4032                 }
4033         }
4034
4035 out:
4036         if (res) {
4037                 /* no suitable interface, frame not sent */
4038                 dev_kfree_skb(skb);
4039         }
4040
4041         return NETDEV_TX_OK;
4042 }
4043
4044
4045 /*
4046  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4047  * the bond has a usable interface.
4048  */
4049 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4050 {
4051         struct bonding *bond = netdev_priv(bond_dev);
4052         int res = 1;
4053
4054         read_lock(&bond->curr_slave_lock);
4055
4056         if (bond->curr_active_slave)
4057                 res = bond_dev_queue_xmit(bond, skb,
4058                         bond->curr_active_slave->dev);
4059
4060         if (res)
4061                 /* no suitable interface, frame not sent */
4062                 dev_kfree_skb(skb);
4063
4064         read_unlock(&bond->curr_slave_lock);
4065
4066         return NETDEV_TX_OK;
4067 }
4068
4069 /*
4070  * In bond_xmit_xor() , we determine the output device by using a pre-
4071  * determined xmit_hash_policy(), If the selected device is not enabled,
4072  * find the next active slave.
4073  */
4074 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4075 {
4076         struct bonding *bond = netdev_priv(bond_dev);
4077         struct slave *slave, *start_at;
4078         int slave_no;
4079         int i;
4080         int res = 1;
4081
4082         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4083
4084         bond_for_each_slave(bond, slave, i) {
4085                 slave_no--;
4086                 if (slave_no < 0)
4087                         break;
4088         }
4089
4090         start_at = slave;
4091
4092         bond_for_each_slave_from(bond, slave, i, start_at) {
4093                 if (IS_UP(slave->dev) &&
4094                     (slave->link == BOND_LINK_UP) &&
4095                     bond_is_active_slave(slave)) {
4096                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4097                         break;
4098                 }
4099         }
4100
4101         if (res) {
4102                 /* no suitable interface, frame not sent */
4103                 dev_kfree_skb(skb);
4104         }
4105
4106         return NETDEV_TX_OK;
4107 }
4108
4109 /*
4110  * in broadcast mode, we send everything to all usable interfaces.
4111  */
4112 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4113 {
4114         struct bonding *bond = netdev_priv(bond_dev);
4115         struct slave *slave, *start_at;
4116         struct net_device *tx_dev = NULL;
4117         int i;
4118         int res = 1;
4119
4120         read_lock(&bond->curr_slave_lock);
4121         start_at = bond->curr_active_slave;
4122         read_unlock(&bond->curr_slave_lock);
4123
4124         if (!start_at)
4125                 goto out;
4126
4127         bond_for_each_slave_from(bond, slave, i, start_at) {
4128                 if (IS_UP(slave->dev) &&
4129                     (slave->link == BOND_LINK_UP) &&
4130                     bond_is_active_slave(slave)) {
4131                         if (tx_dev) {
4132                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4133                                 if (!skb2) {
4134                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4135                                                bond_dev->name);
4136                                         continue;
4137                                 }
4138
4139                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4140                                 if (res) {
4141                                         dev_kfree_skb(skb2);
4142                                         continue;
4143                                 }
4144                         }
4145                         tx_dev = slave->dev;
4146                 }
4147         }
4148
4149         if (tx_dev)
4150                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4151
4152 out:
4153         if (res)
4154                 /* no suitable interface, frame not sent */
4155                 dev_kfree_skb(skb);
4156
4157         /* frame sent to all suitable interfaces */
4158         return NETDEV_TX_OK;
4159 }
4160
4161 /*------------------------- Device initialization ---------------------------*/
4162
4163 static void bond_set_xmit_hash_policy(struct bonding *bond)
4164 {
4165         switch (bond->params.xmit_policy) {
4166         case BOND_XMIT_POLICY_LAYER23:
4167                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4168                 break;
4169         case BOND_XMIT_POLICY_LAYER34:
4170                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4171                 break;
4172         case BOND_XMIT_POLICY_LAYER2:
4173         default:
4174                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4175                 break;
4176         }
4177 }
4178
4179 /*
4180  * Lookup the slave that corresponds to a qid
4181  */
4182 static inline int bond_slave_override(struct bonding *bond,
4183                                       struct sk_buff *skb)
4184 {
4185         int i, res = 1;
4186         struct slave *slave = NULL;
4187         struct slave *check_slave;
4188
4189         if (!skb->queue_mapping)
4190                 return 1;
4191
4192         /* Find out if any slaves have the same mapping as this skb. */
4193         bond_for_each_slave(bond, check_slave, i) {
4194                 if (check_slave->queue_id == skb->queue_mapping) {
4195                         slave = check_slave;
4196                         break;
4197                 }
4198         }
4199
4200         /* If the slave isn't UP, use default transmit policy. */
4201         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4202             (slave->link == BOND_LINK_UP)) {
4203                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4204         }
4205
4206         return res;
4207 }
4208
4209 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4210 {
4211         /*
4212          * This helper function exists to help dev_pick_tx get the correct
4213          * destination queue.  Using a helper function skips a call to
4214          * skb_tx_hash and will put the skbs in the queue we expect on their
4215          * way down to the bonding driver.
4216          */
4217         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4218
4219         if (unlikely(txq >= dev->real_num_tx_queues)) {
4220                 do {
4221                         txq -= dev->real_num_tx_queues;
4222                 } while (txq >= dev->real_num_tx_queues);
4223         }
4224         return txq;
4225 }
4226
4227 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4228 {
4229         struct bonding *bond = netdev_priv(dev);
4230
4231         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4232                 if (!bond_slave_override(bond, skb))
4233                         return NETDEV_TX_OK;
4234         }
4235
4236         switch (bond->params.mode) {
4237         case BOND_MODE_ROUNDROBIN:
4238                 return bond_xmit_roundrobin(skb, dev);
4239         case BOND_MODE_ACTIVEBACKUP:
4240                 return bond_xmit_activebackup(skb, dev);
4241         case BOND_MODE_XOR:
4242                 return bond_xmit_xor(skb, dev);
4243         case BOND_MODE_BROADCAST:
4244                 return bond_xmit_broadcast(skb, dev);
4245         case BOND_MODE_8023AD:
4246                 return bond_3ad_xmit_xor(skb, dev);
4247         case BOND_MODE_ALB:
4248         case BOND_MODE_TLB:
4249                 return bond_alb_xmit(skb, dev);
4250         default:
4251                 /* Should never happen, mode already checked */
4252                 pr_err("%s: Error: Unknown bonding mode %d\n",
4253                        dev->name, bond->params.mode);
4254                 WARN_ON_ONCE(1);
4255                 dev_kfree_skb(skb);
4256                 return NETDEV_TX_OK;
4257         }
4258 }
4259
4260 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4261 {
4262         struct bonding *bond = netdev_priv(dev);
4263         netdev_tx_t ret = NETDEV_TX_OK;
4264
4265         /*
4266          * If we risk deadlock from transmitting this in the
4267          * netpoll path, tell netpoll to queue the frame for later tx
4268          */
4269         if (is_netpoll_tx_blocked(dev))
4270                 return NETDEV_TX_BUSY;
4271
4272         read_lock(&bond->lock);
4273
4274         if (bond->slave_cnt)
4275                 ret = __bond_start_xmit(skb, dev);
4276         else
4277                 dev_kfree_skb(skb);
4278
4279         read_unlock(&bond->lock);
4280
4281         return ret;
4282 }
4283
4284 /*
4285  * set bond mode specific net device operations
4286  */
4287 void bond_set_mode_ops(struct bonding *bond, int mode)
4288 {
4289         struct net_device *bond_dev = bond->dev;
4290
4291         switch (mode) {
4292         case BOND_MODE_ROUNDROBIN:
4293                 break;
4294         case BOND_MODE_ACTIVEBACKUP:
4295                 break;
4296         case BOND_MODE_XOR:
4297                 bond_set_xmit_hash_policy(bond);
4298                 break;
4299         case BOND_MODE_BROADCAST:
4300                 break;
4301         case BOND_MODE_8023AD:
4302                 bond_set_xmit_hash_policy(bond);
4303                 break;
4304         case BOND_MODE_ALB:
4305                 /* FALLTHRU */
4306         case BOND_MODE_TLB:
4307                 break;
4308         default:
4309                 /* Should never happen, mode already checked */
4310                 pr_err("%s: Error: Unknown bonding mode %d\n",
4311                        bond_dev->name, mode);
4312                 break;
4313         }
4314 }
4315
4316 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4317                                     struct ethtool_drvinfo *drvinfo)
4318 {
4319         strncpy(drvinfo->driver, DRV_NAME, 32);
4320         strncpy(drvinfo->version, DRV_VERSION, 32);
4321         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4322 }
4323
4324 static const struct ethtool_ops bond_ethtool_ops = {
4325         .get_drvinfo            = bond_ethtool_get_drvinfo,
4326         .get_link               = ethtool_op_get_link,
4327 };
4328
4329 static const struct net_device_ops bond_netdev_ops = {
4330         .ndo_init               = bond_init,
4331         .ndo_uninit             = bond_uninit,
4332         .ndo_open               = bond_open,
4333         .ndo_stop               = bond_close,
4334         .ndo_start_xmit         = bond_start_xmit,
4335         .ndo_select_queue       = bond_select_queue,
4336         .ndo_get_stats64        = bond_get_stats,
4337         .ndo_do_ioctl           = bond_do_ioctl,
4338         .ndo_set_multicast_list = bond_set_multicast_list,
4339         .ndo_change_mtu         = bond_change_mtu,
4340         .ndo_set_mac_address    = bond_set_mac_address,
4341         .ndo_neigh_setup        = bond_neigh_setup,
4342         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4343         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4344         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4345 #ifdef CONFIG_NET_POLL_CONTROLLER
4346         .ndo_netpoll_setup      = bond_netpoll_setup,
4347         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4348         .ndo_poll_controller    = bond_poll_controller,
4349 #endif
4350         .ndo_add_slave          = bond_enslave,
4351         .ndo_del_slave          = bond_release,
4352         .ndo_fix_features       = bond_fix_features,
4353 };
4354
4355 static void bond_destructor(struct net_device *bond_dev)
4356 {
4357         struct bonding *bond = netdev_priv(bond_dev);
4358         if (bond->wq)
4359                 destroy_workqueue(bond->wq);
4360         free_netdev(bond_dev);
4361 }
4362
4363 static void bond_setup(struct net_device *bond_dev)
4364 {
4365         struct bonding *bond = netdev_priv(bond_dev);
4366
4367         /* initialize rwlocks */
4368         rwlock_init(&bond->lock);
4369         rwlock_init(&bond->curr_slave_lock);
4370
4371         bond->params = bonding_defaults;
4372
4373         /* Initialize pointers */
4374         bond->dev = bond_dev;
4375         INIT_LIST_HEAD(&bond->vlan_list);
4376
4377         /* Initialize the device entry points */
4378         ether_setup(bond_dev);
4379         bond_dev->netdev_ops = &bond_netdev_ops;
4380         bond_dev->ethtool_ops = &bond_ethtool_ops;
4381         bond_set_mode_ops(bond, bond->params.mode);
4382
4383         bond_dev->destructor = bond_destructor;
4384
4385         /* Initialize the device options */
4386         bond_dev->tx_queue_len = 0;
4387         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4388         bond_dev->priv_flags |= IFF_BONDING;
4389         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4390
4391         /* At first, we block adding VLANs. That's the only way to
4392          * prevent problems that occur when adding VLANs over an
4393          * empty bond. The block will be removed once non-challenged
4394          * slaves are enslaved.
4395          */
4396         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4397
4398         /* don't acquire bond device's netif_tx_lock when
4399          * transmitting */
4400         bond_dev->features |= NETIF_F_LLTX;
4401
4402         /* By default, we declare the bond to be fully
4403          * VLAN hardware accelerated capable. Special
4404          * care is taken in the various xmit functions
4405          * when there are slaves that are not hw accel
4406          * capable
4407          */
4408
4409         bond_dev->hw_features = BOND_VLAN_FEATURES |
4410                                 NETIF_F_HW_VLAN_TX |
4411                                 NETIF_F_HW_VLAN_RX |
4412                                 NETIF_F_HW_VLAN_FILTER;
4413
4414         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4415         bond_dev->features |= bond_dev->hw_features;
4416 }
4417
4418 static void bond_work_cancel_all(struct bonding *bond)
4419 {
4420         write_lock_bh(&bond->lock);
4421         bond->kill_timers = 1;
4422         write_unlock_bh(&bond->lock);
4423
4424         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4425                 cancel_delayed_work(&bond->mii_work);
4426
4427         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4428                 cancel_delayed_work(&bond->arp_work);
4429
4430         if (bond->params.mode == BOND_MODE_ALB &&
4431             delayed_work_pending(&bond->alb_work))
4432                 cancel_delayed_work(&bond->alb_work);
4433
4434         if (bond->params.mode == BOND_MODE_8023AD &&
4435             delayed_work_pending(&bond->ad_work))
4436                 cancel_delayed_work(&bond->ad_work);
4437
4438         if (delayed_work_pending(&bond->mcast_work))
4439                 cancel_delayed_work(&bond->mcast_work);
4440 }
4441
4442 /*
4443 * Destroy a bonding device.
4444 * Must be under rtnl_lock when this function is called.
4445 */
4446 static void bond_uninit(struct net_device *bond_dev)
4447 {
4448         struct bonding *bond = netdev_priv(bond_dev);
4449         struct vlan_entry *vlan, *tmp;
4450
4451         bond_netpoll_cleanup(bond_dev);
4452
4453         /* Release the bonded slaves */
4454         bond_release_all(bond_dev);
4455
4456         list_del(&bond->bond_list);
4457
4458         bond_work_cancel_all(bond);
4459
4460         bond_remove_proc_entry(bond);
4461
4462         bond_debug_unregister(bond);
4463
4464         __hw_addr_flush(&bond->mc_list);
4465
4466         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4467                 list_del(&vlan->vlan_list);
4468                 kfree(vlan);
4469         }
4470 }
4471
4472 /*------------------------- Module initialization ---------------------------*/
4473
4474 /*
4475  * Convert string input module parms.  Accept either the
4476  * number of the mode or its string name.  A bit complicated because
4477  * some mode names are substrings of other names, and calls from sysfs
4478  * may have whitespace in the name (trailing newlines, for example).
4479  */
4480 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4481 {
4482         int modeint = -1, i, rv;
4483         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4484
4485         for (p = (char *)buf; *p; p++)
4486                 if (!(isdigit(*p) || isspace(*p)))
4487                         break;
4488
4489         if (*p)
4490                 rv = sscanf(buf, "%20s", modestr);
4491         else
4492                 rv = sscanf(buf, "%d", &modeint);
4493
4494         if (!rv)
4495                 return -1;
4496
4497         for (i = 0; tbl[i].modename; i++) {
4498                 if (modeint == tbl[i].mode)
4499                         return tbl[i].mode;
4500                 if (strcmp(modestr, tbl[i].modename) == 0)
4501                         return tbl[i].mode;
4502         }
4503
4504         return -1;
4505 }
4506
4507 static int bond_check_params(struct bond_params *params)
4508 {
4509         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4510
4511         /*
4512          * Convert string parameters.
4513          */
4514         if (mode) {
4515                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4516                 if (bond_mode == -1) {
4517                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4518                                mode == NULL ? "NULL" : mode);
4519                         return -EINVAL;
4520                 }
4521         }
4522
4523         if (xmit_hash_policy) {
4524                 if ((bond_mode != BOND_MODE_XOR) &&
4525                     (bond_mode != BOND_MODE_8023AD)) {
4526                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4527                                bond_mode_name(bond_mode));
4528                 } else {
4529                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4530                                                         xmit_hashtype_tbl);
4531                         if (xmit_hashtype == -1) {
4532                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4533                                        xmit_hash_policy == NULL ? "NULL" :
4534                                        xmit_hash_policy);
4535                                 return -EINVAL;
4536                         }
4537                 }
4538         }
4539
4540         if (lacp_rate) {
4541                 if (bond_mode != BOND_MODE_8023AD) {
4542                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4543                                 bond_mode_name(bond_mode));
4544                 } else {
4545                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4546                         if (lacp_fast == -1) {
4547                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4548                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4549                                 return -EINVAL;
4550                         }
4551                 }
4552         }
4553
4554         if (ad_select) {
4555                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4556                 if (params->ad_select == -1) {
4557                         pr_err("Error: Invalid ad_select \"%s\"\n",
4558                                ad_select == NULL ? "NULL" : ad_select);
4559                         return -EINVAL;
4560                 }
4561
4562                 if (bond_mode != BOND_MODE_8023AD) {
4563                         pr_warning("ad_select param only affects 802.3ad mode\n");
4564                 }
4565         } else {
4566                 params->ad_select = BOND_AD_STABLE;
4567         }
4568
4569         if (max_bonds < 0) {
4570                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4571                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4572                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4573         }
4574
4575         if (miimon < 0) {
4576                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4577                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4578                 miimon = BOND_LINK_MON_INTERV;
4579         }
4580
4581         if (updelay < 0) {
4582                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4583                            updelay, INT_MAX);
4584                 updelay = 0;
4585         }
4586
4587         if (downdelay < 0) {
4588                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4589                            downdelay, INT_MAX);
4590                 downdelay = 0;
4591         }
4592
4593         if ((use_carrier != 0) && (use_carrier != 1)) {
4594                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4595                            use_carrier);
4596                 use_carrier = 1;
4597         }
4598
4599         if (num_peer_notif < 0 || num_peer_notif > 255) {
4600                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4601                            num_peer_notif);
4602                 num_peer_notif = 1;
4603         }
4604
4605         /* reset values for 802.3ad */
4606         if (bond_mode == BOND_MODE_8023AD) {
4607                 if (!miimon) {
4608                         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");
4609                         pr_warning("Forcing miimon to 100msec\n");
4610                         miimon = 100;
4611                 }
4612         }
4613
4614         if (tx_queues < 1 || tx_queues > 255) {
4615                 pr_warning("Warning: tx_queues (%d) should be between "
4616                            "1 and 255, resetting to %d\n",
4617                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4618                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4619         }
4620
4621         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4622                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4623                            "not of valid value (0/1), so it was set to "
4624                            "0\n", all_slaves_active);
4625                 all_slaves_active = 0;
4626         }
4627
4628         if (resend_igmp < 0 || resend_igmp > 255) {
4629                 pr_warning("Warning: resend_igmp (%d) should be between "
4630                            "0 and 255, resetting to %d\n",
4631                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4632                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4633         }
4634
4635         /* reset values for TLB/ALB */
4636         if ((bond_mode == BOND_MODE_TLB) ||
4637             (bond_mode == BOND_MODE_ALB)) {
4638                 if (!miimon) {
4639                         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");
4640                         pr_warning("Forcing miimon to 100msec\n");
4641                         miimon = 100;
4642                 }
4643         }
4644
4645         if (bond_mode == BOND_MODE_ALB) {
4646                 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",
4647                           updelay);
4648         }
4649
4650         if (!miimon) {
4651                 if (updelay || downdelay) {
4652                         /* just warn the user the up/down delay will have
4653                          * no effect since miimon is zero...
4654                          */
4655                         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",
4656                                    updelay, downdelay);
4657                 }
4658         } else {
4659                 /* don't allow arp monitoring */
4660                 if (arp_interval) {
4661                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4662                                    miimon, arp_interval);
4663                         arp_interval = 0;
4664                 }
4665
4666                 if ((updelay % miimon) != 0) {
4667                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4668                                    updelay, miimon,
4669                                    (updelay / miimon) * miimon);
4670                 }
4671
4672                 updelay /= miimon;
4673
4674                 if ((downdelay % miimon) != 0) {
4675                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4676                                    downdelay, miimon,
4677                                    (downdelay / miimon) * miimon);
4678                 }
4679
4680                 downdelay /= miimon;
4681         }
4682
4683         if (arp_interval < 0) {
4684                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4685                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4686                 arp_interval = BOND_LINK_ARP_INTERV;
4687         }
4688
4689         for (arp_ip_count = 0;
4690              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4691              arp_ip_count++) {
4692                 /* not complete check, but should be good enough to
4693                    catch mistakes */
4694                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4695                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4696                                    arp_ip_target[arp_ip_count]);
4697                         arp_interval = 0;
4698                 } else {
4699                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4700                         arp_target[arp_ip_count] = ip;
4701                 }
4702         }
4703
4704         if (arp_interval && !arp_ip_count) {
4705                 /* don't allow arping if no arp_ip_target given... */
4706                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4707                            arp_interval);
4708                 arp_interval = 0;
4709         }
4710
4711         if (arp_validate) {
4712                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4713                         pr_err("arp_validate only supported in active-backup mode\n");
4714                         return -EINVAL;
4715                 }
4716                 if (!arp_interval) {
4717                         pr_err("arp_validate requires arp_interval\n");
4718                         return -EINVAL;
4719                 }
4720
4721                 arp_validate_value = bond_parse_parm(arp_validate,
4722                                                      arp_validate_tbl);
4723                 if (arp_validate_value == -1) {
4724                         pr_err("Error: invalid arp_validate \"%s\"\n",
4725                                arp_validate == NULL ? "NULL" : arp_validate);
4726                         return -EINVAL;
4727                 }
4728         } else
4729                 arp_validate_value = 0;
4730
4731         if (miimon) {
4732                 pr_info("MII link monitoring set to %d ms\n", miimon);
4733         } else if (arp_interval) {
4734                 int i;
4735
4736                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4737                         arp_interval,
4738                         arp_validate_tbl[arp_validate_value].modename,
4739                         arp_ip_count);
4740
4741                 for (i = 0; i < arp_ip_count; i++)
4742                         pr_info(" %s", arp_ip_target[i]);
4743
4744                 pr_info("\n");
4745
4746         } else if (max_bonds) {
4747                 /* miimon and arp_interval not set, we need one so things
4748                  * work as expected, see bonding.txt for details
4749                  */
4750                 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");
4751         }
4752
4753         if (primary && !USES_PRIMARY(bond_mode)) {
4754                 /* currently, using a primary only makes sense
4755                  * in active backup, TLB or ALB modes
4756                  */
4757                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4758                            primary, bond_mode_name(bond_mode));
4759                 primary = NULL;
4760         }
4761
4762         if (primary && primary_reselect) {
4763                 primary_reselect_value = bond_parse_parm(primary_reselect,
4764                                                          pri_reselect_tbl);
4765                 if (primary_reselect_value == -1) {
4766                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4767                                primary_reselect ==
4768                                         NULL ? "NULL" : primary_reselect);
4769                         return -EINVAL;
4770                 }
4771         } else {
4772                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4773         }
4774
4775         if (fail_over_mac) {
4776                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4777                                                       fail_over_mac_tbl);
4778                 if (fail_over_mac_value == -1) {
4779                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4780                                arp_validate == NULL ? "NULL" : arp_validate);
4781                         return -EINVAL;
4782                 }
4783
4784                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4785                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4786         } else {
4787                 fail_over_mac_value = BOND_FOM_NONE;
4788         }
4789
4790         /* fill params struct with the proper values */
4791         params->mode = bond_mode;
4792         params->xmit_policy = xmit_hashtype;
4793         params->miimon = miimon;
4794         params->num_peer_notif = num_peer_notif;
4795         params->arp_interval = arp_interval;
4796         params->arp_validate = arp_validate_value;
4797         params->updelay = updelay;
4798         params->downdelay = downdelay;
4799         params->use_carrier = use_carrier;
4800         params->lacp_fast = lacp_fast;
4801         params->primary[0] = 0;
4802         params->primary_reselect = primary_reselect_value;
4803         params->fail_over_mac = fail_over_mac_value;
4804         params->tx_queues = tx_queues;
4805         params->all_slaves_active = all_slaves_active;
4806         params->resend_igmp = resend_igmp;
4807
4808         if (primary) {
4809                 strncpy(params->primary, primary, IFNAMSIZ);
4810                 params->primary[IFNAMSIZ - 1] = 0;
4811         }
4812
4813         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4814
4815         return 0;
4816 }
4817
4818 static struct lock_class_key bonding_netdev_xmit_lock_key;
4819 static struct lock_class_key bonding_netdev_addr_lock_key;
4820
4821 static void bond_set_lockdep_class_one(struct net_device *dev,
4822                                        struct netdev_queue *txq,
4823                                        void *_unused)
4824 {
4825         lockdep_set_class(&txq->_xmit_lock,
4826                           &bonding_netdev_xmit_lock_key);
4827 }
4828
4829 static void bond_set_lockdep_class(struct net_device *dev)
4830 {
4831         lockdep_set_class(&dev->addr_list_lock,
4832                           &bonding_netdev_addr_lock_key);
4833         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4834 }
4835
4836 /*
4837  * Called from registration process
4838  */
4839 static int bond_init(struct net_device *bond_dev)
4840 {
4841         struct bonding *bond = netdev_priv(bond_dev);
4842         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4843         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4844
4845         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4846
4847         /*
4848          * Initialize locks that may be required during
4849          * en/deslave operations.  All of the bond_open work
4850          * (of which this is part) should really be moved to
4851          * a phase prior to dev_open
4852          */
4853         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4854         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4855
4856         bond->wq = create_singlethread_workqueue(bond_dev->name);
4857         if (!bond->wq)
4858                 return -ENOMEM;
4859
4860         bond_set_lockdep_class(bond_dev);
4861
4862         bond_create_proc_entry(bond);
4863         list_add_tail(&bond->bond_list, &bn->dev_list);
4864
4865         bond_prepare_sysfs_group(bond);
4866
4867         bond_debug_register(bond);
4868
4869         __hw_addr_init(&bond->mc_list);
4870         return 0;
4871 }
4872
4873 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4874 {
4875         if (tb[IFLA_ADDRESS]) {
4876                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4877                         return -EINVAL;
4878                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4879                         return -EADDRNOTAVAIL;
4880         }
4881         return 0;
4882 }
4883
4884 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4885         .kind           = "bond",
4886         .priv_size      = sizeof(struct bonding),
4887         .setup          = bond_setup,
4888         .validate       = bond_validate,
4889 };
4890
4891 /* Create a new bond based on the specified name and bonding parameters.
4892  * If name is NULL, obtain a suitable "bond%d" name for us.
4893  * Caller must NOT hold rtnl_lock; we need to release it here before we
4894  * set up our sysfs entries.
4895  */
4896 int bond_create(struct net *net, const char *name)
4897 {
4898         struct net_device *bond_dev;
4899         int res;
4900
4901         rtnl_lock();
4902
4903         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4904                                    name ? name : "bond%d",
4905                                    bond_setup, tx_queues);
4906         if (!bond_dev) {
4907                 pr_err("%s: eek! can't alloc netdev!\n", name);
4908                 rtnl_unlock();
4909                 return -ENOMEM;
4910         }
4911
4912         dev_net_set(bond_dev, net);
4913         bond_dev->rtnl_link_ops = &bond_link_ops;
4914
4915         res = register_netdevice(bond_dev);
4916
4917         netif_carrier_off(bond_dev);
4918
4919         rtnl_unlock();
4920         if (res < 0)
4921                 bond_destructor(bond_dev);
4922         return res;
4923 }
4924
4925 static int __net_init bond_net_init(struct net *net)
4926 {
4927         struct bond_net *bn = net_generic(net, bond_net_id);
4928
4929         bn->net = net;
4930         INIT_LIST_HEAD(&bn->dev_list);
4931
4932         bond_create_proc_dir(bn);
4933
4934         return 0;
4935 }
4936
4937 static void __net_exit bond_net_exit(struct net *net)
4938 {
4939         struct bond_net *bn = net_generic(net, bond_net_id);
4940
4941         bond_destroy_proc_dir(bn);
4942 }
4943
4944 static struct pernet_operations bond_net_ops = {
4945         .init = bond_net_init,
4946         .exit = bond_net_exit,
4947         .id   = &bond_net_id,
4948         .size = sizeof(struct bond_net),
4949 };
4950
4951 static int __init bonding_init(void)
4952 {
4953         int i;
4954         int res;
4955
4956         pr_info("%s", bond_version);
4957
4958         res = bond_check_params(&bonding_defaults);
4959         if (res)
4960                 goto out;
4961
4962         res = register_pernet_subsys(&bond_net_ops);
4963         if (res)
4964                 goto out;
4965
4966         res = rtnl_link_register(&bond_link_ops);
4967         if (res)
4968                 goto err_link;
4969
4970         bond_create_debugfs();
4971
4972         for (i = 0; i < max_bonds; i++) {
4973                 res = bond_create(&init_net, NULL);
4974                 if (res)
4975                         goto err;
4976         }
4977
4978         res = bond_create_sysfs();
4979         if (res)
4980                 goto err;
4981
4982         register_netdevice_notifier(&bond_netdev_notifier);
4983         register_inetaddr_notifier(&bond_inetaddr_notifier);
4984 out:
4985         return res;
4986 err:
4987         rtnl_link_unregister(&bond_link_ops);
4988 err_link:
4989         unregister_pernet_subsys(&bond_net_ops);
4990         goto out;
4991
4992 }
4993
4994 static void __exit bonding_exit(void)
4995 {
4996         unregister_netdevice_notifier(&bond_netdev_notifier);
4997         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4998
4999         bond_destroy_sysfs();
5000         bond_destroy_debugfs();
5001
5002         rtnl_link_unregister(&bond_link_ops);
5003         unregister_pernet_subsys(&bond_net_ops);
5004
5005 #ifdef CONFIG_NET_POLL_CONTROLLER
5006         /*
5007          * Make sure we don't have an imbalance on our netpoll blocking
5008          */
5009         WARN_ON(atomic_read(&netpoll_block_tx));
5010 #endif
5011 }
5012
5013 module_init(bonding_init);
5014 module_exit(bonding_exit);
5015 MODULE_LICENSE("GPL");
5016 MODULE_VERSION(DRV_VERSION);
5017 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5018 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5019 MODULE_ALIAS_RTNL_LINK("bond");