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/netpoll.h>
  63 #include <linux/inetdevice.h>
  64 #include <linux/igmp.h>
  65 #include <linux/etherdevice.h>
  66 #include <linux/skbuff.h>
  67 #include <net/sock.h>
  68 #include <linux/rtnetlink.h>
  69 #include <linux/proc_fs.h>
  70 #include <linux/seq_file.h>
  71 #include <linux/smp.h>
  72 #include <linux/if_ether.h>
  73 #include <net/arp.h>
  74 #include <linux/mii.h>
  75 #include <linux/ethtool.h>
  76 #include <linux/if_vlan.h>
  77 #include <linux/if_bonding.h>
  78 #include <linux/jiffies.h>
  79 #include <net/route.h>
  80 #include <net/net_namespace.h>
  81 #include <net/netns/generic.h>
  82 #include "bonding.h"
  83 #include "bond_3ad.h"
  84 #include "bond_alb.h"
  85
  86 /*---------------------------- Module parameters ----------------------------*/
  87
  88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  89 #define BOND_LINK_MON_INTERV    0
  90 #define BOND_LINK_ARP_INTERV    0
  91
  92 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  93 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  94 static int num_grat_arp = 1;
  95 static int num_unsol_na = 1;
  96 static int miimon       = BOND_LINK_MON_INTERV;
  97 static int updelay;
  98 static int downdelay;
  99 static int use_carrier  = 1;
 100 static char *mode;
 101 static char *primary;
 102 static char *primary_reselect;
 103 static char *lacp_rate;
 104 static char *ad_select;
 105 static char *xmit_hash_policy;
 106 static int arp_interval = BOND_LINK_ARP_INTERV;
 107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 108 static char *arp_validate;
 109 static char *fail_over_mac;
 110 static int all_slaves_active = 0;
 111 static struct bond_params bonding_defaults;
 112
 113 module_param(max_bonds, int, 0);
 114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 115 module_param(tx_queues, int, 0);
 116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 117 module_param(num_grat_arp, int, 0644);
 118 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 119 module_param(num_unsol_na, int, 0644);
 120 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
 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                             "(slow/fast)");
 149 module_param(ad_select, charp, 0);
 150 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 151 module_param(xmit_hash_policy, charp, 0);
 152 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 153                                    ", 1 for layer 3+4");
 154 module_param(arp_interval, int, 0);
 155 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 156 module_param_array(arp_ip_target, charp, NULL, 0);
 157 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 158 module_param(arp_validate, charp, 0);
 159 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 160 module_param(fail_over_mac, charp, 0);
 161 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 162 module_param(all_slaves_active, int, 0);
 163 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 164                                      "by setting active flag for all slaves.  "
 165                                      "0 for never (default), 1 for always.");
 166
 167 /*----------------------------- Global variables ----------------------------*/
 168
 169 static const char * const version =
 170         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 171
 172 int bond_net_id __read_mostly;
 173
 174 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 175 static int arp_ip_count;
 176 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 177 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 178 static int lacp_fast;
 179 #ifdef CONFIG_NET_POLL_CONTROLLER
 180 static int disable_netpoll = 1;
 181 #endif
 182
 183 const struct bond_parm_tbl bond_lacp_tbl[] = {
 184 {       "slow",         AD_LACP_SLOW},
 185 {       "fast",         AD_LACP_FAST},
 186 {       NULL,           -1},
 187 };
 188
 189 const struct bond_parm_tbl bond_mode_tbl[] = {
 190 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 191 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 192 {       "balance-xor",          BOND_MODE_XOR},
 193 {       "broadcast",            BOND_MODE_BROADCAST},
 194 {       "802.3ad",              BOND_MODE_8023AD},
 195 {       "balance-tlb",          BOND_MODE_TLB},
 196 {       "balance-alb",          BOND_MODE_ALB},
 197 {       NULL,                   -1},
 198 };
 199
 200 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 201 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 202 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 203 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 204 {       NULL,                   -1},
 205 };
 206
 207 const struct bond_parm_tbl arp_validate_tbl[] = {
 208 {       "none",                 BOND_ARP_VALIDATE_NONE},
 209 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 210 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 211 {       "all",                  BOND_ARP_VALIDATE_ALL},
 212 {       NULL,                   -1},
 213 };
 214
 215 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 216 {       "none",                 BOND_FOM_NONE},
 217 {       "active",               BOND_FOM_ACTIVE},
 218 {       "follow",               BOND_FOM_FOLLOW},
 219 {       NULL,                   -1},
 220 };
 221
 222 const struct bond_parm_tbl pri_reselect_tbl[] = {
 223 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 224 {       "better",               BOND_PRI_RESELECT_BETTER},
 225 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 226 {       NULL,                   -1},
 227 };
 228
 229 struct bond_parm_tbl ad_select_tbl[] = {
 230 {       "stable",       BOND_AD_STABLE},
 231 {       "bandwidth",    BOND_AD_BANDWIDTH},
 232 {       "count",        BOND_AD_COUNT},
 233 {       NULL,           -1},
 234 };
 235
 236 /*-------------------------- Forward declarations ---------------------------*/
 237
 238 static void bond_send_gratuitous_arp(struct bonding *bond);
 239 static int bond_init(struct net_device *bond_dev);
 240 static void bond_uninit(struct net_device *bond_dev);
 241
 242 /*---------------------------- General routines -----------------------------*/
 243
 244 static const char *bond_mode_name(int mode)
 245 {
 246         static const char *names[] = {
 247                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 248                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 249                 [BOND_MODE_XOR] = "load balancing (xor)",
 250                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 251                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 252                 [BOND_MODE_TLB] = "transmit load balancing",
 253                 [BOND_MODE_ALB] = "adaptive load balancing",
 254         };
 255
 256         if (mode < 0 || mode > BOND_MODE_ALB)
 257                 return "unknown";
 258
 259         return names[mode];
 260 }
 261
 262 /*---------------------------------- VLAN -----------------------------------*/
 263
 264 /**
 265  * bond_add_vlan - add a new vlan id on bond
 266  * @bond: bond that got the notification
 267  * @vlan_id: the vlan id to add
 268  *
 269  * Returns -ENOMEM if allocation failed.
 270  */
 271 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 272 {
 273         struct vlan_entry *vlan;
 274
 275         pr_debug("bond: %s, vlan id %d\n",
 276                  (bond ? bond->dev->name : "None"), vlan_id);
 277
 278         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 279         if (!vlan)
 280                 return -ENOMEM;
 281
 282         INIT_LIST_HEAD(&vlan->vlan_list);
 283         vlan->vlan_id = vlan_id;
 284
 285         write_lock_bh(&bond->lock);
 286
 287         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 288
 289         write_unlock_bh(&bond->lock);
 290
 291         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 292
 293         return 0;
 294 }
 295
 296 /**
 297  * bond_del_vlan - delete a vlan id from bond
 298  * @bond: bond that got the notification
 299  * @vlan_id: the vlan id to delete
 300  *
 301  * returns -ENODEV if @vlan_id was not found in @bond.
 302  */
 303 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 304 {
 305         struct vlan_entry *vlan;
 306         int res = -ENODEV;
 307
 308         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 309
 310         write_lock_bh(&bond->lock);
 311
 312         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 313                 if (vlan->vlan_id == vlan_id) {
 314                         list_del(&vlan->vlan_list);
 315
 316                         if (bond_is_lb(bond))
 317                                 bond_alb_clear_vlan(bond, vlan_id);
 318
 319                         pr_debug("removed VLAN ID %d from bond %s\n",
 320                                  vlan_id, bond->dev->name);
 321
 322                         kfree(vlan);
 323
 324                         if (list_empty(&bond->vlan_list) &&
 325                             (bond->slave_cnt == 0)) {
 326                                 /* Last VLAN removed and no slaves, so
 327                                  * restore block on adding VLANs. This will
 328                                  * be removed once new slaves that are not
 329                                  * VLAN challenged will be added.
 330                                  */
 331                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 332                         }
 333
 334                         res = 0;
 335                         goto out;
 336                 }
 337         }
 338
 339         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 340                  vlan_id, bond->dev->name);
 341
 342 out:
 343         write_unlock_bh(&bond->lock);
 344         return res;
 345 }
 346
 347 /**
 348  * bond_has_challenged_slaves
 349  * @bond: the bond we're working on
 350  *
 351  * Searches the slave list. Returns 1 if a vlan challenged slave
 352  * was found, 0 otherwise.
 353  *
 354  * Assumes bond->lock is held.
 355  */
 356 static int bond_has_challenged_slaves(struct bonding *bond)
 357 {
 358         struct slave *slave;
 359         int i;
 360
 361         bond_for_each_slave(bond, slave, i) {
 362                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 363                         pr_debug("found VLAN challenged slave - %s\n",
 364                                  slave->dev->name);
 365                         return 1;
 366                 }
 367         }
 368
 369         pr_debug("no VLAN challenged slaves found\n");
 370         return 0;
 371 }
 372
 373 /**
 374  * bond_next_vlan - safely skip to the next item in the vlans list.
 375  * @bond: the bond we're working on
 376  * @curr: item we're advancing from
 377  *
 378  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 379  * or @curr->next otherwise (even if it is @curr itself again).
 380  *
 381  * Caller must hold bond->lock
 382  */
 383 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 384 {
 385         struct vlan_entry *next, *last;
 386
 387         if (list_empty(&bond->vlan_list))
 388                 return NULL;
 389
 390         if (!curr) {
 391                 next = list_entry(bond->vlan_list.next,
 392                                   struct vlan_entry, vlan_list);
 393         } else {
 394                 last = list_entry(bond->vlan_list.prev,
 395                                   struct vlan_entry, vlan_list);
 396                 if (last == curr) {
 397                         next = list_entry(bond->vlan_list.next,
 398                                           struct vlan_entry, vlan_list);
 399                 } else {
 400                         next = list_entry(curr->vlan_list.next,
 401                                           struct vlan_entry, vlan_list);
 402                 }
 403         }
 404
 405         return next;
 406 }
 407
 408 /**
 409  * bond_dev_queue_xmit - Prepare skb for xmit.
 410  *
 411  * @bond: bond device that got this skb for tx.
 412  * @skb: hw accel VLAN tagged skb to transmit
 413  * @slave_dev: slave that is supposed to xmit this skbuff
 414  *
 415  * When the bond gets an skb to transmit that is
 416  * already hardware accelerated VLAN tagged, and it
 417  * needs to relay this skb to a slave that is not
 418  * hw accel capable, the skb needs to be "unaccelerated",
 419  * i.e. strip the hwaccel tag and re-insert it as part
 420  * of the payload.
 421  */
 422 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 423                         struct net_device *slave_dev)
 424 {
 425         unsigned short uninitialized_var(vlan_id);
 426
 427         /* Test vlan_list not vlgrp to catch and handle 802.1p tags */
 428         if (!list_empty(&bond->vlan_list) &&
 429             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 430             vlan_get_tag(skb, &vlan_id) == 0) {
 431                 skb->dev = slave_dev;
 432                 skb = vlan_put_tag(skb, vlan_id);
 433                 if (!skb) {
 434                         /* vlan_put_tag() frees the skb in case of error,
 435                          * so return success here so the calling functions
 436                          * won't attempt to free is again.
 437                          */
 438                         return 0;
 439                 }
 440         } else {
 441                 skb->dev = slave_dev;
 442         }
 443
 444         skb->priority = 1;
 445 #ifdef CONFIG_NET_POLL_CONTROLLER
 446         if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
 447                 struct netpoll *np = bond->dev->npinfo->netpoll;
 448                 slave_dev->npinfo = bond->dev->npinfo;
 449                 np->real_dev = np->dev = skb->dev;
 450                 slave_dev->priv_flags |= IFF_IN_NETPOLL;
 451                 netpoll_send_skb(np, skb);
 452                 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
 453                 np->dev = bond->dev;
 454         } else
 455 #endif
 456                 dev_queue_xmit(skb);
 457
 458         return 0;
 459 }
 460
 461 /*
 462  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 463  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 464  * lock because:
 465  * a. This operation is performed in IOCTL context,
 466  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 467  * c. Holding a lock with BH disabled while directly calling a base driver
 468  *    entry point is generally a BAD idea.
 469  *
 470  * The design of synchronization/protection for this operation in the 8021q
 471  * module is good for one or more VLAN devices over a single physical device
 472  * and cannot be extended for a teaming solution like bonding, so there is a
 473  * potential race condition here where a net device from the vlan group might
 474  * be referenced (either by a base driver or the 8021q code) while it is being
 475  * removed from the system. However, it turns out we're not making matters
 476  * worse, and if it works for regular VLAN usage it will work here too.
 477 */
 478
 479 /**
 480  * bond_vlan_rx_register - Propagates registration to slaves
 481  * @bond_dev: bonding net device that got called
 482  * @grp: vlan group being registered
 483  */
 484 static void bond_vlan_rx_register(struct net_device *bond_dev,
 485                                   struct vlan_group *grp)
 486 {
 487         struct bonding *bond = netdev_priv(bond_dev);
 488         struct slave *slave;
 489         int i;
 490
 491         write_lock(&bond->lock);
 492         bond->vlgrp = grp;
 493         write_unlock(&bond->lock);
 494
 495         bond_for_each_slave(bond, slave, i) {
 496                 struct net_device *slave_dev = slave->dev;
 497                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 498
 499                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 500                     slave_ops->ndo_vlan_rx_register) {
 501                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 502                 }
 503         }
 504 }
 505
 506 /**
 507  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 508  * @bond_dev: bonding net device that got called
 509  * @vid: vlan id being added
 510  */
 511 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 512 {
 513         struct bonding *bond = netdev_priv(bond_dev);
 514         struct slave *slave;
 515         int i, res;
 516
 517         bond_for_each_slave(bond, slave, i) {
 518                 struct net_device *slave_dev = slave->dev;
 519                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 520
 521                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 522                     slave_ops->ndo_vlan_rx_add_vid) {
 523                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 524                 }
 525         }
 526
 527         res = bond_add_vlan(bond, vid);
 528         if (res) {
 529                 pr_err("%s: Error: Failed to add vlan id %d\n",
 530                        bond_dev->name, vid);
 531         }
 532 }
 533
 534 /**
 535  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 536  * @bond_dev: bonding net device that got called
 537  * @vid: vlan id being removed
 538  */
 539 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 540 {
 541         struct bonding *bond = netdev_priv(bond_dev);
 542         struct slave *slave;
 543         struct net_device *vlan_dev;
 544         int i, res;
 545
 546         bond_for_each_slave(bond, slave, i) {
 547                 struct net_device *slave_dev = slave->dev;
 548                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 549
 550                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 551                     slave_ops->ndo_vlan_rx_kill_vid) {
 552                         /* Save and then restore vlan_dev in the grp array,
 553                          * since the slave's driver might clear it.
 554                          */
 555                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 556                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 557                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 558                 }
 559         }
 560
 561         res = bond_del_vlan(bond, vid);
 562         if (res) {
 563                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 564                        bond_dev->name, vid);
 565         }
 566 }
 567
 568 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 569 {
 570         struct vlan_entry *vlan;
 571         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 572
 573         if (!bond->vlgrp)
 574                 return;
 575
 576         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 577             slave_ops->ndo_vlan_rx_register)
 578                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 579
 580         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 581             !(slave_ops->ndo_vlan_rx_add_vid))
 582                 return;
 583
 584         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 585                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 586 }
 587
 588 static void bond_del_vlans_from_slave(struct bonding *bond,
 589                                       struct net_device *slave_dev)
 590 {
 591         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 592         struct vlan_entry *vlan;
 593         struct net_device *vlan_dev;
 594
 595         if (!bond->vlgrp)
 596                 return;
 597
 598         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 599             !(slave_ops->ndo_vlan_rx_kill_vid))
 600                 goto unreg;
 601
 602         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 603                 if (!vlan->vlan_id)
 604                         continue;
 605                 /* Save and then restore vlan_dev in the grp array,
 606                  * since the slave's driver might clear it.
 607                  */
 608                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 609                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 610                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 611         }
 612
 613 unreg:
 614         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 615             slave_ops->ndo_vlan_rx_register)
 616                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 617 }
 618
 619 /*------------------------------- Link status -------------------------------*/
 620
 621 /*
 622  * Set the carrier state for the master according to the state of its
 623  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 624  * do special 802.3ad magic.
 625  *
 626  * Returns zero if carrier state does not change, nonzero if it does.
 627  */
 628 static int bond_set_carrier(struct bonding *bond)
 629 {
 630         struct slave *slave;
 631         int i;
 632
 633         if (bond->slave_cnt == 0)
 634                 goto down;
 635
 636         if (bond->params.mode == BOND_MODE_8023AD)
 637                 return bond_3ad_set_carrier(bond);
 638
 639         bond_for_each_slave(bond, slave, i) {
 640                 if (slave->link == BOND_LINK_UP) {
 641                         if (!netif_carrier_ok(bond->dev)) {
 642                                 netif_carrier_on(bond->dev);
 643                                 return 1;
 644                         }
 645                         return 0;
 646                 }
 647         }
 648
 649 down:
 650         if (netif_carrier_ok(bond->dev)) {
 651                 netif_carrier_off(bond->dev);
 652                 return 1;
 653         }
 654         return 0;
 655 }
 656
 657 /*
 658  * Get link speed and duplex from the slave's base driver
 659  * using ethtool. If for some reason the call fails or the
 660  * values are invalid, fake speed and duplex to 100/Full
 661  * and return error.
 662  */
 663 static int bond_update_speed_duplex(struct slave *slave)
 664 {
 665         struct net_device *slave_dev = slave->dev;
 666         struct ethtool_cmd etool;
 667         int res;
 668
 669         /* Fake speed and duplex */
 670         slave->speed = SPEED_100;
 671         slave->duplex = DUPLEX_FULL;
 672
 673         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 674                 return -1;
 675
 676         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 677         if (res < 0)
 678                 return -1;
 679
 680         switch (etool.speed) {
 681         case SPEED_10:
 682         case SPEED_100:
 683         case SPEED_1000:
 684         case SPEED_10000:
 685                 break;
 686         default:
 687                 return -1;
 688         }
 689
 690         switch (etool.duplex) {
 691         case DUPLEX_FULL:
 692         case DUPLEX_HALF:
 693                 break;
 694         default:
 695                 return -1;
 696         }
 697
 698         slave->speed = etool.speed;
 699         slave->duplex = etool.duplex;
 700
 701         return 0;
 702 }
 703
 704 /*
 705  * if <dev> supports MII link status reporting, check its link status.
 706  *
 707  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 708  * depending upon the setting of the use_carrier parameter.
 709  *
 710  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 711  * can't tell and just pretend it is), or 0, meaning that the link is
 712  * down.
 713  *
 714  * If reporting is non-zero, instead of faking link up, return -1 if
 715  * both ETHTOOL and MII ioctls fail (meaning the device does not
 716  * support them).  If use_carrier is set, return whatever it says.
 717  * It'd be nice if there was a good way to tell if a driver supports
 718  * netif_carrier, but there really isn't.
 719  */
 720 static int bond_check_dev_link(struct bonding *bond,
 721                                struct net_device *slave_dev, int reporting)
 722 {
 723         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 724         int (*ioctl)(struct net_device *, struct ifreq *, int);
 725         struct ifreq ifr;
 726         struct mii_ioctl_data *mii;
 727
 728         if (!reporting && !netif_running(slave_dev))
 729                 return 0;
 730
 731         if (bond->params.use_carrier)
 732                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 733
 734         /* Try to get link status using Ethtool first. */
 735         if (slave_dev->ethtool_ops) {
 736                 if (slave_dev->ethtool_ops->get_link) {
 737                         u32 link;
 738
 739                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 740
 741                         return link ? BMSR_LSTATUS : 0;
 742                 }
 743         }
 744
 745         /* Ethtool can't be used, fallback to MII ioctls. */
 746         ioctl = slave_ops->ndo_do_ioctl;
 747         if (ioctl) {
 748                 /* TODO: set pointer to correct ioctl on a per team member */
 749                 /*       bases to make this more efficient. that is, once  */
 750                 /*       we determine the correct ioctl, we will always    */
 751                 /*       call it and not the others for that team          */
 752                 /*       member.                                           */
 753
 754                 /*
 755                  * We cannot assume that SIOCGMIIPHY will also read a
 756                  * register; not all network drivers (e.g., e100)
 757                  * support that.
 758                  */
 759
 760                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 761                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 762                 mii = if_mii(&ifr);
 763                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 764                         mii->reg_num = MII_BMSR;
 765                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 766                                 return mii->val_out & BMSR_LSTATUS;
 767                 }
 768         }
 769
 770         /*
 771          * If reporting, report that either there's no dev->do_ioctl,
 772          * or both SIOCGMIIREG and get_link failed (meaning that we
 773          * cannot report link status).  If not reporting, pretend
 774          * we're ok.
 775          */
 776         return reporting ? -1 : BMSR_LSTATUS;
 777 }
 778
 779 /*----------------------------- Multicast list ------------------------------*/
 780
 781 /*
 782  * Push the promiscuity flag down to appropriate slaves
 783  */
 784 static int bond_set_promiscuity(struct bonding *bond, int inc)
 785 {
 786         int err = 0;
 787         if (USES_PRIMARY(bond->params.mode)) {
 788                 /* write lock already acquired */
 789                 if (bond->curr_active_slave) {
 790                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 791                                                   inc);
 792                 }
 793         } else {
 794                 struct slave *slave;
 795                 int i;
 796                 bond_for_each_slave(bond, slave, i) {
 797                         err = dev_set_promiscuity(slave->dev, inc);
 798                         if (err)
 799                                 return err;
 800                 }
 801         }
 802         return err;
 803 }
 804
 805 /*
 806  * Push the allmulti flag down to all slaves
 807  */
 808 static int bond_set_allmulti(struct bonding *bond, int inc)
 809 {
 810         int err = 0;
 811         if (USES_PRIMARY(bond->params.mode)) {
 812                 /* write lock already acquired */
 813                 if (bond->curr_active_slave) {
 814                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 815                                                inc);
 816                 }
 817         } else {
 818                 struct slave *slave;
 819                 int i;
 820                 bond_for_each_slave(bond, slave, i) {
 821                         err = dev_set_allmulti(slave->dev, inc);
 822                         if (err)
 823                                 return err;
 824                 }
 825         }
 826         return err;
 827 }
 828
 829 /*
 830  * Add a Multicast address to slaves
 831  * according to mode
 832  */
 833 static void bond_mc_add(struct bonding *bond, void *addr)
 834 {
 835         if (USES_PRIMARY(bond->params.mode)) {
 836                 /* write lock already acquired */
 837                 if (bond->curr_active_slave)
 838                         dev_mc_add(bond->curr_active_slave->dev, addr);
 839         } else {
 840                 struct slave *slave;
 841                 int i;
 842
 843                 bond_for_each_slave(bond, slave, i)
 844                         dev_mc_add(slave->dev, addr);
 845         }
 846 }
 847
 848 /*
 849  * Remove a multicast address from slave
 850  * according to mode
 851  */
 852 static void bond_mc_del(struct bonding *bond, void *addr)
 853 {
 854         if (USES_PRIMARY(bond->params.mode)) {
 855                 /* write lock already acquired */
 856                 if (bond->curr_active_slave)
 857                         dev_mc_del(bond->curr_active_slave->dev, addr);
 858         } else {
 859                 struct slave *slave;
 860                 int i;
 861                 bond_for_each_slave(bond, slave, i) {
 862                         dev_mc_del(slave->dev, addr);
 863                 }
 864         }
 865 }
 866
 867
 868 /*
 869  * Retrieve the list of registered multicast addresses for the bonding
 870  * device and retransmit an IGMP JOIN request to the current active
 871  * slave.
 872  */
 873 static void bond_resend_igmp_join_requests(struct bonding *bond)
 874 {
 875         struct in_device *in_dev;
 876         struct ip_mc_list *im;
 877
 878         rcu_read_lock();
 879         in_dev = __in_dev_get_rcu(bond->dev);
 880         if (in_dev) {
 881                 for (im = in_dev->mc_list; im; im = im->next)
 882                         ip_mc_rejoin_group(im);
 883         }
 884
 885         rcu_read_unlock();
 886 }
 887
 888 /*
 889  * flush all members of flush->mc_list from device dev->mc_list
 890  */
 891 static void bond_mc_list_flush(struct net_device *bond_dev,
 892                                struct net_device *slave_dev)
 893 {
 894         struct bonding *bond = netdev_priv(bond_dev);
 895         struct netdev_hw_addr *ha;
 896
 897         netdev_for_each_mc_addr(ha, bond_dev)
 898                 dev_mc_del(slave_dev, ha->addr);
 899
 900         if (bond->params.mode == BOND_MODE_8023AD) {
 901                 /* del lacpdu mc addr from mc list */
 902                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 903
 904                 dev_mc_del(slave_dev, lacpdu_multicast);
 905         }
 906 }
 907
 908 /*--------------------------- Active slave change ---------------------------*/
 909
 910 /*
 911  * Update the mc list and multicast-related flags for the new and
 912  * old active slaves (if any) according to the multicast mode, and
 913  * promiscuous flags unconditionally.
 914  */
 915 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 916                          struct slave *old_active)
 917 {
 918         struct netdev_hw_addr *ha;
 919
 920         if (!USES_PRIMARY(bond->params.mode))
 921                 /* nothing to do -  mc list is already up-to-date on
 922                  * all slaves
 923                  */
 924                 return;
 925
 926         if (old_active) {
 927                 if (bond->dev->flags & IFF_PROMISC)
 928                         dev_set_promiscuity(old_active->dev, -1);
 929
 930                 if (bond->dev->flags & IFF_ALLMULTI)
 931                         dev_set_allmulti(old_active->dev, -1);
 932
 933                 netdev_for_each_mc_addr(ha, bond->dev)
 934                         dev_mc_del(old_active->dev, ha->addr);
 935         }
 936
 937         if (new_active) {
 938                 /* FIXME: Signal errors upstream. */
 939                 if (bond->dev->flags & IFF_PROMISC)
 940                         dev_set_promiscuity(new_active->dev, 1);
 941
 942                 if (bond->dev->flags & IFF_ALLMULTI)
 943                         dev_set_allmulti(new_active->dev, 1);
 944
 945                 netdev_for_each_mc_addr(ha, bond->dev)
 946                         dev_mc_add(new_active->dev, ha->addr);
 947                 bond_resend_igmp_join_requests(bond);
 948         }
 949 }
 950
 951 /*
 952  * bond_do_fail_over_mac
 953  *
 954  * Perform special MAC address swapping for fail_over_mac settings
 955  *
 956  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 957  */
 958 static void bond_do_fail_over_mac(struct bonding *bond,
 959                                   struct slave *new_active,
 960                                   struct slave *old_active)
 961         __releases(&bond->curr_slave_lock)
 962         __releases(&bond->lock)
 963         __acquires(&bond->lock)
 964         __acquires(&bond->curr_slave_lock)
 965 {
 966         u8 tmp_mac[ETH_ALEN];
 967         struct sockaddr saddr;
 968         int rv;
 969
 970         switch (bond->params.fail_over_mac) {
 971         case BOND_FOM_ACTIVE:
 972                 if (new_active)
 973                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 974                                new_active->dev->addr_len);
 975                 break;
 976         case BOND_FOM_FOLLOW:
 977                 /*
 978                  * if new_active && old_active, swap them
 979                  * if just old_active, do nothing (going to no active slave)
 980                  * if just new_active, set new_active to bond's MAC
 981                  */
 982                 if (!new_active)
 983                         return;
 984
 985                 write_unlock_bh(&bond->curr_slave_lock);
 986                 read_unlock(&bond->lock);
 987
 988                 if (old_active) {
 989                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 990                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 991                                ETH_ALEN);
 992                         saddr.sa_family = new_active->dev->type;
 993                 } else {
 994                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 995                         saddr.sa_family = bond->dev->type;
 996                 }
 997
 998                 rv = dev_set_mac_address(new_active->dev, &saddr);
 999                 if (rv) {
1000                         pr_err("%s: Error %d setting MAC of slave %s\n",
1001                                bond->dev->name, -rv, new_active->dev->name);
1002                         goto out;
1003                 }
1004
1005                 if (!old_active)
1006                         goto out;
1007
1008                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1009                 saddr.sa_family = old_active->dev->type;
1010
1011                 rv = dev_set_mac_address(old_active->dev, &saddr);
1012                 if (rv)
1013                         pr_err("%s: Error %d setting MAC of slave %s\n",
1014                                bond->dev->name, -rv, new_active->dev->name);
1015 out:
1016                 read_lock(&bond->lock);
1017                 write_lock_bh(&bond->curr_slave_lock);
1018                 break;
1019         default:
1020                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1021                        bond->dev->name, bond->params.fail_over_mac);
1022                 break;
1023         }
1024
1025 }
1026
1027 static bool bond_should_change_active(struct bonding *bond)
1028 {
1029         struct slave *prim = bond->primary_slave;
1030         struct slave *curr = bond->curr_active_slave;
1031
1032         if (!prim || !curr || curr->link != BOND_LINK_UP)
1033                 return true;
1034         if (bond->force_primary) {
1035                 bond->force_primary = false;
1036                 return true;
1037         }
1038         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1039             (prim->speed < curr->speed ||
1040              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1041                 return false;
1042         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1043                 return false;
1044         return true;
1045 }
1046
1047 /**
1048  * find_best_interface - select the best available slave to be the active one
1049  * @bond: our bonding struct
1050  *
1051  * Warning: Caller must hold curr_slave_lock for writing.
1052  */
1053 static struct slave *bond_find_best_slave(struct bonding *bond)
1054 {
1055         struct slave *new_active, *old_active;
1056         struct slave *bestslave = NULL;
1057         int mintime = bond->params.updelay;
1058         int i;
1059
1060         new_active = bond->curr_active_slave;
1061
1062         if (!new_active) { /* there were no active slaves left */
1063                 if (bond->slave_cnt > 0)   /* found one slave */
1064                         new_active = bond->first_slave;
1065                 else
1066                         return NULL; /* still no slave, return NULL */
1067         }
1068
1069         if ((bond->primary_slave) &&
1070             bond->primary_slave->link == BOND_LINK_UP &&
1071             bond_should_change_active(bond)) {
1072                 new_active = bond->primary_slave;
1073         }
1074
1075         /* remember where to stop iterating over the slaves */
1076         old_active = new_active;
1077
1078         bond_for_each_slave_from(bond, new_active, i, old_active) {
1079                 if (new_active->link == BOND_LINK_UP) {
1080                         return new_active;
1081                 } else if (new_active->link == BOND_LINK_BACK &&
1082                            IS_UP(new_active->dev)) {
1083                         /* link up, but waiting for stabilization */
1084                         if (new_active->delay < mintime) {
1085                                 mintime = new_active->delay;
1086                                 bestslave = new_active;
1087                         }
1088                 }
1089         }
1090
1091         return bestslave;
1092 }
1093
1094 /**
1095  * change_active_interface - change the active slave into the specified one
1096  * @bond: our bonding struct
1097  * @new: the new slave to make the active one
1098  *
1099  * Set the new slave to the bond's settings and unset them on the old
1100  * curr_active_slave.
1101  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1102  *
1103  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1104  * because it is apparently the best available slave we have, even though its
1105  * updelay hasn't timed out yet.
1106  *
1107  * If new_active is not NULL, caller must hold bond->lock for read and
1108  * curr_slave_lock for write_bh.
1109  */
1110 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1111 {
1112         struct slave *old_active = bond->curr_active_slave;
1113
1114         if (old_active == new_active)
1115                 return;
1116
1117         if (new_active) {
1118                 new_active->jiffies = jiffies;
1119
1120                 if (new_active->link == BOND_LINK_BACK) {
1121                         if (USES_PRIMARY(bond->params.mode)) {
1122                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1123                                         bond->dev->name, new_active->dev->name,
1124                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1125                         }
1126
1127                         new_active->delay = 0;
1128                         new_active->link = BOND_LINK_UP;
1129
1130                         if (bond->params.mode == BOND_MODE_8023AD)
1131                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1132
1133                         if (bond_is_lb(bond))
1134                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1135                 } else {
1136                         if (USES_PRIMARY(bond->params.mode)) {
1137                                 pr_info("%s: making interface %s the new active one.\n",
1138                                         bond->dev->name, new_active->dev->name);
1139                         }
1140                 }
1141         }
1142
1143         if (USES_PRIMARY(bond->params.mode))
1144                 bond_mc_swap(bond, new_active, old_active);
1145
1146         if (bond_is_lb(bond)) {
1147                 bond_alb_handle_active_change(bond, new_active);
1148                 if (old_active)
1149                         bond_set_slave_inactive_flags(old_active);
1150                 if (new_active)
1151                         bond_set_slave_active_flags(new_active);
1152         } else {
1153                 bond->curr_active_slave = new_active;
1154         }
1155
1156         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1157                 if (old_active)
1158                         bond_set_slave_inactive_flags(old_active);
1159
1160                 if (new_active) {
1161                         bond_set_slave_active_flags(new_active);
1162
1163                         if (bond->params.fail_over_mac)
1164                                 bond_do_fail_over_mac(bond, new_active,
1165                                                       old_active);
1166
1167                         bond->send_grat_arp = bond->params.num_grat_arp;
1168                         bond_send_gratuitous_arp(bond);
1169
1170                         bond->send_unsol_na = bond->params.num_unsol_na;
1171                         bond_send_unsolicited_na(bond);
1172
1173                         write_unlock_bh(&bond->curr_slave_lock);
1174                         read_unlock(&bond->lock);
1175
1176                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1177
1178                         read_lock(&bond->lock);
1179                         write_lock_bh(&bond->curr_slave_lock);
1180                 }
1181         }
1182
1183         /* resend IGMP joins since all were sent on curr_active_slave */
1184         if (bond->params.mode == BOND_MODE_ROUNDROBIN) {
1185                 bond_resend_igmp_join_requests(bond);
1186         }
1187 }
1188
1189 /**
1190  * bond_select_active_slave - select a new active slave, if needed
1191  * @bond: our bonding struct
1192  *
1193  * This functions should be called when one of the following occurs:
1194  * - The old curr_active_slave has been released or lost its link.
1195  * - The primary_slave has got its link back.
1196  * - A slave has got its link back and there's no old curr_active_slave.
1197  *
1198  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1199  */
1200 void bond_select_active_slave(struct bonding *bond)
1201 {
1202         struct slave *best_slave;
1203         int rv;
1204
1205         best_slave = bond_find_best_slave(bond);
1206         if (best_slave != bond->curr_active_slave) {
1207                 bond_change_active_slave(bond, best_slave);
1208                 rv = bond_set_carrier(bond);
1209                 if (!rv)
1210                         return;
1211
1212                 if (netif_carrier_ok(bond->dev)) {
1213                         pr_info("%s: first active interface up!\n",
1214                                 bond->dev->name);
1215                 } else {
1216                         pr_info("%s: now running without any active interface !\n",
1217                                 bond->dev->name);
1218                 }
1219         }
1220 }
1221
1222 /*--------------------------- slave list handling ---------------------------*/
1223
1224 /*
1225  * This function attaches the slave to the end of list.
1226  *
1227  * bond->lock held for writing by caller.
1228  */
1229 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1230 {
1231         if (bond->first_slave == NULL) { /* attaching the first slave */
1232                 new_slave->next = new_slave;
1233                 new_slave->prev = new_slave;
1234                 bond->first_slave = new_slave;
1235         } else {
1236                 new_slave->next = bond->first_slave;
1237                 new_slave->prev = bond->first_slave->prev;
1238                 new_slave->next->prev = new_slave;
1239                 new_slave->prev->next = new_slave;
1240         }
1241
1242         bond->slave_cnt++;
1243 }
1244
1245 /*
1246  * This function detaches the slave from the list.
1247  * WARNING: no check is made to verify if the slave effectively
1248  * belongs to <bond>.
1249  * Nothing is freed on return, structures are just unchained.
1250  * If any slave pointer in bond was pointing to <slave>,
1251  * it should be changed by the calling function.
1252  *
1253  * bond->lock held for writing by caller.
1254  */
1255 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1256 {
1257         if (slave->next)
1258                 slave->next->prev = slave->prev;
1259
1260         if (slave->prev)
1261                 slave->prev->next = slave->next;
1262
1263         if (bond->first_slave == slave) { /* slave is the first slave */
1264                 if (bond->slave_cnt > 1) { /* there are more slave */
1265                         bond->first_slave = slave->next;
1266                 } else {
1267                         bond->first_slave = NULL; /* slave was the last one */
1268                 }
1269         }
1270
1271         slave->next = NULL;
1272         slave->prev = NULL;
1273         bond->slave_cnt--;
1274 }
1275
1276 #ifdef CONFIG_NET_POLL_CONTROLLER
1277 /*
1278  * You must hold read lock on bond->lock before calling this.
1279  */
1280 static bool slaves_support_netpoll(struct net_device *bond_dev)
1281 {
1282         struct bonding *bond = netdev_priv(bond_dev);
1283         struct slave *slave;
1284         int i = 0;
1285         bool ret = true;
1286
1287         bond_for_each_slave(bond, slave, i) {
1288                 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1289                     !slave->dev->netdev_ops->ndo_poll_controller)
1290                         ret = false;
1291         }
1292         return i != 0 && ret;
1293 }
1294
1295 static void bond_poll_controller(struct net_device *bond_dev)
1296 {
1297         struct net_device *dev = bond_dev->npinfo->netpoll->real_dev;
1298         if (dev != bond_dev)
1299                 netpoll_poll_dev(dev);
1300 }
1301
1302 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1303 {
1304         struct bonding *bond = netdev_priv(bond_dev);
1305         struct slave *slave;
1306         const struct net_device_ops *ops;
1307         int i;
1308
1309         read_lock(&bond->lock);
1310         bond_dev->npinfo = NULL;
1311         bond_for_each_slave(bond, slave, i) {
1312                 if (slave->dev) {
1313                         ops = slave->dev->netdev_ops;
1314                         if (ops->ndo_netpoll_cleanup)
1315                                 ops->ndo_netpoll_cleanup(slave->dev);
1316                         else
1317                                 slave->dev->npinfo = NULL;
1318                 }
1319         }
1320         read_unlock(&bond->lock);
1321 }
1322
1323 #else
1324
1325 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1326 {
1327 }
1328
1329 #endif
1330
1331 /*---------------------------------- IOCTL ----------------------------------*/
1332
1333 static int bond_sethwaddr(struct net_device *bond_dev,
1334                           struct net_device *slave_dev)
1335 {
1336         pr_debug("bond_dev=%p\n", bond_dev);
1337         pr_debug("slave_dev=%p\n", slave_dev);
1338         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1339         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1340         return 0;
1341 }
1342
1343 #define BOND_VLAN_FEATURES \
1344         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1345          NETIF_F_HW_VLAN_FILTER)
1346
1347 /*
1348  * Compute the common dev->feature set available to all slaves.  Some
1349  * feature bits are managed elsewhere, so preserve those feature bits
1350  * on the master device.
1351  */
1352 static int bond_compute_features(struct bonding *bond)
1353 {
1354         struct slave *slave;
1355         struct net_device *bond_dev = bond->dev;
1356         unsigned long features = bond_dev->features;
1357         unsigned long vlan_features = 0;
1358         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1359                                                 bond_dev->hard_header_len);
1360         int i;
1361
1362         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1363         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1364
1365         if (!bond->first_slave)
1366                 goto done;
1367
1368         features &= ~NETIF_F_ONE_FOR_ALL;
1369
1370         vlan_features = bond->first_slave->dev->vlan_features;
1371         bond_for_each_slave(bond, slave, i) {
1372                 features = netdev_increment_features(features,
1373                                                      slave->dev->features,
1374                                                      NETIF_F_ONE_FOR_ALL);
1375                 vlan_features = netdev_increment_features(vlan_features,
1376                                                         slave->dev->vlan_features,
1377                                                         NETIF_F_ONE_FOR_ALL);
1378                 if (slave->dev->hard_header_len > max_hard_header_len)
1379                         max_hard_header_len = slave->dev->hard_header_len;
1380         }
1381
1382 done:
1383         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1384         bond_dev->features = netdev_fix_features(features, NULL);
1385         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1386         bond_dev->hard_header_len = max_hard_header_len;
1387
1388         return 0;
1389 }
1390
1391 static void bond_setup_by_slave(struct net_device *bond_dev,
1392                                 struct net_device *slave_dev)
1393 {
1394         struct bonding *bond = netdev_priv(bond_dev);
1395
1396         bond_dev->header_ops        = slave_dev->header_ops;
1397
1398         bond_dev->type              = slave_dev->type;
1399         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1400         bond_dev->addr_len          = slave_dev->addr_len;
1401
1402         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1403                 slave_dev->addr_len);
1404         bond->setup_by_slave = 1;
1405 }
1406
1407 /* enslave device <slave> to bond device <master> */
1408 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1409 {
1410         struct bonding *bond = netdev_priv(bond_dev);
1411         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1412         struct slave *new_slave = NULL;
1413         struct netdev_hw_addr *ha;
1414         struct sockaddr addr;
1415         int link_reporting;
1416         int old_features = bond_dev->features;
1417         int res = 0;
1418
1419         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1420                 slave_ops->ndo_do_ioctl == NULL) {
1421                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1422                            bond_dev->name, slave_dev->name);
1423         }
1424
1425         /* bond must be initialized by bond_open() before enslaving */
1426         if (!(bond_dev->flags & IFF_UP)) {
1427                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1428                            bond_dev->name);
1429         }
1430
1431         /* already enslaved */
1432         if (slave_dev->flags & IFF_SLAVE) {
1433                 pr_debug("Error, Device was already enslaved\n");
1434                 return -EBUSY;
1435         }
1436
1437         /* vlan challenged mutual exclusion */
1438         /* no need to lock since we're protected by rtnl_lock */
1439         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1440                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1441                 if (bond->vlgrp) {
1442                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1443                                bond_dev->name, slave_dev->name, bond_dev->name);
1444                         return -EPERM;
1445                 } else {
1446                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1447                                    bond_dev->name, slave_dev->name,
1448                                    slave_dev->name, bond_dev->name);
1449                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1450                 }
1451         } else {
1452                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1453                 if (bond->slave_cnt == 0) {
1454                         /* First slave, and it is not VLAN challenged,
1455                          * so remove the block of adding VLANs over the bond.
1456                          */
1457                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1458                 }
1459         }
1460
1461         /*
1462          * Old ifenslave binaries are no longer supported.  These can
1463          * be identified with moderate accuracy by the state of the slave:
1464          * the current ifenslave will set the interface down prior to
1465          * enslaving it; the old ifenslave will not.
1466          */
1467         if ((slave_dev->flags & IFF_UP)) {
1468                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1469                        slave_dev->name);
1470                 res = -EPERM;
1471                 goto err_undo_flags;
1472         }
1473
1474         /* set bonding device ether type by slave - bonding netdevices are
1475          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1476          * there is a need to override some of the type dependent attribs/funcs.
1477          *
1478          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1479          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1480          */
1481         if (bond->slave_cnt == 0) {
1482                 if (bond_dev->type != slave_dev->type) {
1483                         pr_debug("%s: change device type from %d to %d\n",
1484                                  bond_dev->name,
1485                                  bond_dev->type, slave_dev->type);
1486
1487                         res = netdev_bonding_change(bond_dev,
1488                                                     NETDEV_PRE_TYPE_CHANGE);
1489                         res = notifier_to_errno(res);
1490                         if (res) {
1491                                 pr_err("%s: refused to change device type\n",
1492                                        bond_dev->name);
1493                                 res = -EBUSY;
1494                                 goto err_undo_flags;
1495                         }
1496
1497                         /* Flush unicast and multicast addresses */
1498                         dev_uc_flush(bond_dev);
1499                         dev_mc_flush(bond_dev);
1500
1501                         if (slave_dev->type != ARPHRD_ETHER)
1502                                 bond_setup_by_slave(bond_dev, slave_dev);
1503                         else
1504                                 ether_setup(bond_dev);
1505
1506                         netdev_bonding_change(bond_dev,
1507                                               NETDEV_POST_TYPE_CHANGE);
1508                 }
1509         } else if (bond_dev->type != slave_dev->type) {
1510                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1511                        slave_dev->name,
1512                        slave_dev->type, bond_dev->type);
1513                 res = -EINVAL;
1514                 goto err_undo_flags;
1515         }
1516
1517         if (slave_ops->ndo_set_mac_address == NULL) {
1518                 if (bond->slave_cnt == 0) {
1519                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1520                                    bond_dev->name);
1521                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1522                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1523                         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",
1524                                bond_dev->name);
1525                         res = -EOPNOTSUPP;
1526                         goto err_undo_flags;
1527                 }
1528         }
1529
1530         /* If this is the first slave, then we need to set the master's hardware
1531          * address to be the same as the slave's. */
1532         if (bond->slave_cnt == 0)
1533                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1534                        slave_dev->addr_len);
1535
1536
1537         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1538         if (!new_slave) {
1539                 res = -ENOMEM;
1540                 goto err_undo_flags;
1541         }
1542
1543         /*
1544          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1545          * is set via sysfs or module option if desired.
1546          */
1547         new_slave->queue_id = 0;
1548
1549         /* Save slave's original mtu and then set it to match the bond */
1550         new_slave->original_mtu = slave_dev->mtu;
1551         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1552         if (res) {
1553                 pr_debug("Error %d calling dev_set_mtu\n", res);
1554                 goto err_free;
1555         }
1556
1557         /*
1558          * Save slave's original ("permanent") mac address for modes
1559          * that need it, and for restoring it upon release, and then
1560          * set it to the master's address
1561          */
1562         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1563
1564         if (!bond->params.fail_over_mac) {
1565                 /*
1566                  * Set slave to master's mac address.  The application already
1567                  * set the master's mac address to that of the first slave
1568                  */
1569                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1570                 addr.sa_family = slave_dev->type;
1571                 res = dev_set_mac_address(slave_dev, &addr);
1572                 if (res) {
1573                         pr_debug("Error %d calling set_mac_address\n", res);
1574                         goto err_restore_mtu;
1575                 }
1576         }
1577
1578         res = netdev_set_master(slave_dev, bond_dev);
1579         if (res) {
1580                 pr_debug("Error %d calling netdev_set_master\n", res);
1581                 goto err_restore_mac;
1582         }
1583         /* open the slave since the application closed it */
1584         res = dev_open(slave_dev);
1585         if (res) {
1586                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1587                 goto err_unset_master;
1588         }
1589
1590         new_slave->dev = slave_dev;
1591         slave_dev->priv_flags |= IFF_BONDING;
1592
1593         if (bond_is_lb(bond)) {
1594                 /* bond_alb_init_slave() must be called before all other stages since
1595                  * it might fail and we do not want to have to undo everything
1596                  */
1597                 res = bond_alb_init_slave(bond, new_slave);
1598                 if (res)
1599                         goto err_close;
1600         }
1601
1602         /* If the mode USES_PRIMARY, then the new slave gets the
1603          * master's promisc (and mc) settings only if it becomes the
1604          * curr_active_slave, and that is taken care of later when calling
1605          * bond_change_active()
1606          */
1607         if (!USES_PRIMARY(bond->params.mode)) {
1608                 /* set promiscuity level to new slave */
1609                 if (bond_dev->flags & IFF_PROMISC) {
1610                         res = dev_set_promiscuity(slave_dev, 1);
1611                         if (res)
1612                                 goto err_close;
1613                 }
1614
1615                 /* set allmulti level to new slave */
1616                 if (bond_dev->flags & IFF_ALLMULTI) {
1617                         res = dev_set_allmulti(slave_dev, 1);
1618                         if (res)
1619                                 goto err_close;
1620                 }
1621
1622                 netif_addr_lock_bh(bond_dev);
1623                 /* upload master's mc_list to new slave */
1624                 netdev_for_each_mc_addr(ha, bond_dev)
1625                         dev_mc_add(slave_dev, ha->addr);
1626                 netif_addr_unlock_bh(bond_dev);
1627         }
1628
1629         if (bond->params.mode == BOND_MODE_8023AD) {
1630                 /* add lacpdu mc addr to mc list */
1631                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1632
1633                 dev_mc_add(slave_dev, lacpdu_multicast);
1634         }
1635
1636         bond_add_vlans_on_slave(bond, slave_dev);
1637
1638         write_lock_bh(&bond->lock);
1639
1640         bond_attach_slave(bond, new_slave);
1641
1642         new_slave->delay = 0;
1643         new_slave->link_failure_count = 0;
1644
1645         bond_compute_features(bond);
1646
1647         write_unlock_bh(&bond->lock);
1648
1649         read_lock(&bond->lock);
1650
1651         new_slave->last_arp_rx = jiffies;
1652
1653         if (bond->params.miimon && !bond->params.use_carrier) {
1654                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1655
1656                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1657                         /*
1658                          * miimon is set but a bonded network driver
1659                          * does not support ETHTOOL/MII and
1660                          * arp_interval is not set.  Note: if
1661                          * use_carrier is enabled, we will never go
1662                          * here (because netif_carrier is always
1663                          * supported); thus, we don't need to change
1664                          * the messages for netif_carrier.
1665                          */
1666                         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",
1667                                bond_dev->name, slave_dev->name);
1668                 } else if (link_reporting == -1) {
1669                         /* unable get link status using mii/ethtool */
1670                         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",
1671                                    bond_dev->name, slave_dev->name);
1672                 }
1673         }
1674
1675         /* check for initial state */
1676         if (!bond->params.miimon ||
1677             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1678                 if (bond->params.updelay) {
1679                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1680                         new_slave->link  = BOND_LINK_BACK;
1681                         new_slave->delay = bond->params.updelay;
1682                 } else {
1683                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1684                         new_slave->link  = BOND_LINK_UP;
1685                 }
1686                 new_slave->jiffies = jiffies;
1687         } else {
1688                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1689                 new_slave->link  = BOND_LINK_DOWN;
1690         }
1691
1692         if (bond_update_speed_duplex(new_slave) &&
1693             (new_slave->link != BOND_LINK_DOWN)) {
1694                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1695                            bond_dev->name, new_slave->dev->name);
1696
1697                 if (bond->params.mode == BOND_MODE_8023AD) {
1698                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1699                                    bond_dev->name);
1700                 }
1701         }
1702
1703         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1704                 /* if there is a primary slave, remember it */
1705                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1706                         bond->primary_slave = new_slave;
1707                         bond->force_primary = true;
1708                 }
1709         }
1710
1711         write_lock_bh(&bond->curr_slave_lock);
1712
1713         switch (bond->params.mode) {
1714         case BOND_MODE_ACTIVEBACKUP:
1715                 bond_set_slave_inactive_flags(new_slave);
1716                 bond_select_active_slave(bond);
1717                 break;
1718         case BOND_MODE_8023AD:
1719                 /* in 802.3ad mode, the internal mechanism
1720                  * will activate the slaves in the selected
1721                  * aggregator
1722                  */
1723                 bond_set_slave_inactive_flags(new_slave);
1724                 /* if this is the first slave */
1725                 if (bond->slave_cnt == 1) {
1726                         SLAVE_AD_INFO(new_slave).id = 1;
1727                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1728                          * can be called only after the mac address of the bond is set
1729                          */
1730                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1731                                             bond->params.lacp_fast);
1732                 } else {
1733                         SLAVE_AD_INFO(new_slave).id =
1734                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1735                 }
1736
1737                 bond_3ad_bind_slave(new_slave);
1738                 break;
1739         case BOND_MODE_TLB:
1740         case BOND_MODE_ALB:
1741                 new_slave->state = BOND_STATE_ACTIVE;
1742                 bond_set_slave_inactive_flags(new_slave);
1743                 bond_select_active_slave(bond);
1744                 break;
1745         default:
1746                 pr_debug("This slave is always active in trunk mode\n");
1747
1748                 /* always active in trunk mode */
1749                 new_slave->state = BOND_STATE_ACTIVE;
1750
1751                 /* In trunking mode there is little meaning to curr_active_slave
1752                  * anyway (it holds no special properties of the bond device),
1753                  * so we can change it without calling change_active_interface()
1754                  */
1755                 if (!bond->curr_active_slave)
1756                         bond->curr_active_slave = new_slave;
1757
1758                 break;
1759         } /* switch(bond_mode) */
1760
1761         write_unlock_bh(&bond->curr_slave_lock);
1762
1763         bond_set_carrier(bond);
1764
1765 #ifdef CONFIG_NET_POLL_CONTROLLER
1766         /*
1767          * Netpoll and bonding is broken, make sure it is not initialized
1768          * until it is fixed.
1769          */
1770         if (disable_netpoll) {
1771                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1772         } else {
1773                 if (slaves_support_netpoll(bond_dev)) {
1774                         bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1775                         if (bond_dev->npinfo)
1776                                 slave_dev->npinfo = bond_dev->npinfo;
1777                 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1778                         bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1779                         pr_info("New slave device %s does not support netpoll\n",
1780                                 slave_dev->name);
1781                         pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1782                 }
1783         }
1784 #endif
1785         read_unlock(&bond->lock);
1786
1787         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1788         if (res)
1789                 goto err_close;
1790
1791         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1792                 bond_dev->name, slave_dev->name,
1793                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1794                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1795
1796         /* enslave is successful */
1797         return 0;
1798
1799 /* Undo stages on error */
1800 err_close:
1801         dev_close(slave_dev);
1802
1803 err_unset_master:
1804         netdev_set_master(slave_dev, NULL);
1805
1806 err_restore_mac:
1807         if (!bond->params.fail_over_mac) {
1808                 /* XXX TODO - fom follow mode needs to change master's
1809                  * MAC if this slave's MAC is in use by the bond, or at
1810                  * least print a warning.
1811                  */
1812                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1813                 addr.sa_family = slave_dev->type;
1814                 dev_set_mac_address(slave_dev, &addr);
1815         }
1816
1817 err_restore_mtu:
1818         dev_set_mtu(slave_dev, new_slave->original_mtu);
1819
1820 err_free:
1821         kfree(new_slave);
1822
1823 err_undo_flags:
1824         bond_dev->features = old_features;
1825
1826         return res;
1827 }
1828
1829 /*
1830  * Try to release the slave device <slave> from the bond device <master>
1831  * It is legal to access curr_active_slave without a lock because all the function
1832  * is write-locked.
1833  *
1834  * The rules for slave state should be:
1835  *   for Active/Backup:
1836  *     Active stays on all backups go down
1837  *   for Bonded connections:
1838  *     The first up interface should be left on and all others downed.
1839  */
1840 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1841 {
1842         struct bonding *bond = netdev_priv(bond_dev);
1843         struct slave *slave, *oldcurrent;
1844         struct sockaddr addr;
1845
1846         /* slave is not a slave or master is not master of this slave */
1847         if (!(slave_dev->flags & IFF_SLAVE) ||
1848             (slave_dev->master != bond_dev)) {
1849                 pr_err("%s: Error: cannot release %s.\n",
1850                        bond_dev->name, slave_dev->name);
1851                 return -EINVAL;
1852         }
1853
1854         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1855         write_lock_bh(&bond->lock);
1856
1857         slave = bond_get_slave_by_dev(bond, slave_dev);
1858         if (!slave) {
1859                 /* not a slave of this bond */
1860                 pr_info("%s: %s not enslaved\n",
1861                         bond_dev->name, slave_dev->name);
1862                 write_unlock_bh(&bond->lock);
1863                 return -EINVAL;
1864         }
1865
1866         if (!bond->params.fail_over_mac) {
1867                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1868                     bond->slave_cnt > 1)
1869                         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",
1870                                    bond_dev->name, slave_dev->name,
1871                                    slave->perm_hwaddr,
1872                                    bond_dev->name, slave_dev->name);
1873         }
1874
1875         /* Inform AD package of unbinding of slave. */
1876         if (bond->params.mode == BOND_MODE_8023AD) {
1877                 /* must be called before the slave is
1878                  * detached from the list
1879                  */
1880                 bond_3ad_unbind_slave(slave);
1881         }
1882
1883         pr_info("%s: releasing %s interface %s\n",
1884                 bond_dev->name,
1885                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1886                 slave_dev->name);
1887
1888         oldcurrent = bond->curr_active_slave;
1889
1890         bond->current_arp_slave = NULL;
1891
1892         /* release the slave from its bond */
1893         bond_detach_slave(bond, slave);
1894
1895         bond_compute_features(bond);
1896
1897         if (bond->primary_slave == slave)
1898                 bond->primary_slave = NULL;
1899
1900         if (oldcurrent == slave)
1901                 bond_change_active_slave(bond, NULL);
1902
1903         if (bond_is_lb(bond)) {
1904                 /* Must be called only after the slave has been
1905                  * detached from the list and the curr_active_slave
1906                  * has been cleared (if our_slave == old_current),
1907                  * but before a new active slave is selected.
1908                  */
1909                 write_unlock_bh(&bond->lock);
1910                 bond_alb_deinit_slave(bond, slave);
1911                 write_lock_bh(&bond->lock);
1912         }
1913
1914         if (oldcurrent == slave) {
1915                 /*
1916                  * Note that we hold RTNL over this sequence, so there
1917                  * is no concern that another slave add/remove event
1918                  * will interfere.
1919                  */
1920                 write_unlock_bh(&bond->lock);
1921                 read_lock(&bond->lock);
1922                 write_lock_bh(&bond->curr_slave_lock);
1923
1924                 bond_select_active_slave(bond);
1925
1926                 write_unlock_bh(&bond->curr_slave_lock);
1927                 read_unlock(&bond->lock);
1928                 write_lock_bh(&bond->lock);
1929         }
1930
1931         if (bond->slave_cnt == 0) {
1932                 bond_set_carrier(bond);
1933
1934                 /* if the last slave was removed, zero the mac address
1935                  * of the master so it will be set by the application
1936                  * to the mac address of the first slave
1937                  */
1938                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1939
1940                 if (!bond->vlgrp) {
1941                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1942                 } else {
1943                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1944                                    bond_dev->name, bond_dev->name);
1945                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1946                                    bond_dev->name);
1947                 }
1948         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1949                    !bond_has_challenged_slaves(bond)) {
1950                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1951                         bond_dev->name, slave_dev->name, bond_dev->name);
1952                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1953         }
1954
1955         write_unlock_bh(&bond->lock);
1956
1957         /* must do this from outside any spinlocks */
1958         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1959
1960         bond_del_vlans_from_slave(bond, slave_dev);
1961
1962         /* If the mode USES_PRIMARY, then we should only remove its
1963          * promisc and mc settings if it was the curr_active_slave, but that was
1964          * already taken care of above when we detached the slave
1965          */
1966         if (!USES_PRIMARY(bond->params.mode)) {
1967                 /* unset promiscuity level from slave */
1968                 if (bond_dev->flags & IFF_PROMISC)
1969                         dev_set_promiscuity(slave_dev, -1);
1970
1971                 /* unset allmulti level from slave */
1972                 if (bond_dev->flags & IFF_ALLMULTI)
1973                         dev_set_allmulti(slave_dev, -1);
1974
1975                 /* flush master's mc_list from slave */
1976                 netif_addr_lock_bh(bond_dev);
1977                 bond_mc_list_flush(bond_dev, slave_dev);
1978                 netif_addr_unlock_bh(bond_dev);
1979         }
1980
1981         netdev_set_master(slave_dev, NULL);
1982
1983 #ifdef CONFIG_NET_POLL_CONTROLLER
1984         read_lock_bh(&bond->lock);
1985
1986          /* Make sure netpoll over stays disabled until fixed. */
1987         if (!disable_netpoll)
1988                 if (slaves_support_netpoll(bond_dev))
1989                                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1990         read_unlock_bh(&bond->lock);
1991         if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
1992                 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
1993         else
1994                 slave_dev->npinfo = NULL;
1995 #endif
1996
1997         /* close slave before restoring its mac address */
1998         dev_close(slave_dev);
1999
2000         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2001                 /* restore original ("permanent") mac address */
2002                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2003                 addr.sa_family = slave_dev->type;
2004                 dev_set_mac_address(slave_dev, &addr);
2005         }
2006
2007         dev_set_mtu(slave_dev, slave->original_mtu);
2008
2009         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2010                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
2011                                    IFF_SLAVE_NEEDARP);
2012
2013         kfree(slave);
2014
2015         return 0;  /* deletion OK */
2016 }
2017
2018 /*
2019 * First release a slave and than destroy the bond if no more slaves are left.
2020 * Must be under rtnl_lock when this function is called.
2021 */
2022 int  bond_release_and_destroy(struct net_device *bond_dev,
2023                               struct net_device *slave_dev)
2024 {
2025         struct bonding *bond = netdev_priv(bond_dev);
2026         int ret;
2027
2028         ret = bond_release(bond_dev, slave_dev);
2029         if ((ret == 0) && (bond->slave_cnt == 0)) {
2030                 pr_info("%s: destroying bond %s.\n",
2031                         bond_dev->name, bond_dev->name);
2032                 unregister_netdevice(bond_dev);
2033         }
2034         return ret;
2035 }
2036
2037 /*
2038  * This function releases all slaves.
2039  */
2040 static int bond_release_all(struct net_device *bond_dev)
2041 {
2042         struct bonding *bond = netdev_priv(bond_dev);
2043         struct slave *slave;
2044         struct net_device *slave_dev;
2045         struct sockaddr addr;
2046
2047         write_lock_bh(&bond->lock);
2048
2049         netif_carrier_off(bond_dev);
2050
2051         if (bond->slave_cnt == 0)
2052                 goto out;
2053
2054         bond->current_arp_slave = NULL;
2055         bond->primary_slave = NULL;
2056         bond_change_active_slave(bond, NULL);
2057
2058         while ((slave = bond->first_slave) != NULL) {
2059                 /* Inform AD package of unbinding of slave
2060                  * before slave is detached from the list.
2061                  */
2062                 if (bond->params.mode == BOND_MODE_8023AD)
2063                         bond_3ad_unbind_slave(slave);
2064
2065                 slave_dev = slave->dev;
2066                 bond_detach_slave(bond, slave);
2067
2068                 /* now that the slave is detached, unlock and perform
2069                  * all the undo steps that should not be called from
2070                  * within a lock.
2071                  */
2072                 write_unlock_bh(&bond->lock);
2073
2074                 if (bond_is_lb(bond)) {
2075                         /* must be called only after the slave
2076                          * has been detached from the list
2077                          */
2078                         bond_alb_deinit_slave(bond, slave);
2079                 }
2080
2081                 bond_compute_features(bond);
2082
2083                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2084                 bond_del_vlans_from_slave(bond, slave_dev);
2085
2086                 /* If the mode USES_PRIMARY, then we should only remove its
2087                  * promisc and mc settings if it was the curr_active_slave, but that was
2088                  * already taken care of above when we detached the slave
2089                  */
2090                 if (!USES_PRIMARY(bond->params.mode)) {
2091                         /* unset promiscuity level from slave */
2092                         if (bond_dev->flags & IFF_PROMISC)
2093                                 dev_set_promiscuity(slave_dev, -1);
2094
2095                         /* unset allmulti level from slave */
2096                         if (bond_dev->flags & IFF_ALLMULTI)
2097                                 dev_set_allmulti(slave_dev, -1);
2098
2099                         /* flush master's mc_list from slave */
2100                         netif_addr_lock_bh(bond_dev);
2101                         bond_mc_list_flush(bond_dev, slave_dev);
2102                         netif_addr_unlock_bh(bond_dev);
2103                 }
2104
2105                 netdev_set_master(slave_dev, NULL);
2106
2107                 /* close slave before restoring its mac address */
2108                 dev_close(slave_dev);
2109
2110                 if (!bond->params.fail_over_mac) {
2111                         /* restore original ("permanent") mac address*/
2112                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2113                         addr.sa_family = slave_dev->type;
2114                         dev_set_mac_address(slave_dev, &addr);
2115                 }
2116
2117                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2118                                            IFF_SLAVE_INACTIVE);
2119
2120                 kfree(slave);
2121
2122                 /* re-acquire the lock before getting the next slave */
2123                 write_lock_bh(&bond->lock);
2124         }
2125
2126         /* zero the mac address of the master so it will be
2127          * set by the application to the mac address of the
2128          * first slave
2129          */
2130         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2131
2132         if (!bond->vlgrp) {
2133                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2134         } else {
2135                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2136                            bond_dev->name, bond_dev->name);
2137                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2138                            bond_dev->name);
2139         }
2140
2141         pr_info("%s: released all slaves\n", bond_dev->name);
2142
2143 out:
2144         write_unlock_bh(&bond->lock);
2145
2146         return 0;
2147 }
2148
2149 /*
2150  * This function changes the active slave to slave <slave_dev>.
2151  * It returns -EINVAL in the following cases.
2152  *  - <slave_dev> is not found in the list.
2153  *  - There is not active slave now.
2154  *  - <slave_dev> is already active.
2155  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2156  *  - <slave_dev> is not running.
2157  * In these cases, this function does nothing.
2158  * In the other cases, current_slave pointer is changed and 0 is returned.
2159  */
2160 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2161 {
2162         struct bonding *bond = netdev_priv(bond_dev);
2163         struct slave *old_active = NULL;
2164         struct slave *new_active = NULL;
2165         int res = 0;
2166
2167         if (!USES_PRIMARY(bond->params.mode))
2168                 return -EINVAL;
2169
2170         /* Verify that master_dev is indeed the master of slave_dev */
2171         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2172                 return -EINVAL;
2173
2174         read_lock(&bond->lock);
2175
2176         read_lock(&bond->curr_slave_lock);
2177         old_active = bond->curr_active_slave;
2178         read_unlock(&bond->curr_slave_lock);
2179
2180         new_active = bond_get_slave_by_dev(bond, slave_dev);
2181
2182         /*
2183          * Changing to the current active: do nothing; return success.
2184          */
2185         if (new_active && (new_active == old_active)) {
2186                 read_unlock(&bond->lock);
2187                 return 0;
2188         }
2189
2190         if ((new_active) &&
2191             (old_active) &&
2192             (new_active->link == BOND_LINK_UP) &&
2193             IS_UP(new_active->dev)) {
2194                 write_lock_bh(&bond->curr_slave_lock);
2195                 bond_change_active_slave(bond, new_active);
2196                 write_unlock_bh(&bond->curr_slave_lock);
2197         } else
2198                 res = -EINVAL;
2199
2200         read_unlock(&bond->lock);
2201
2202         return res;
2203 }
2204
2205 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2206 {
2207         struct bonding *bond = netdev_priv(bond_dev);
2208
2209         info->bond_mode = bond->params.mode;
2210         info->miimon = bond->params.miimon;
2211
2212         read_lock(&bond->lock);
2213         info->num_slaves = bond->slave_cnt;
2214         read_unlock(&bond->lock);
2215
2216         return 0;
2217 }
2218
2219 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2220 {
2221         struct bonding *bond = netdev_priv(bond_dev);
2222         struct slave *slave;
2223         int i, res = -ENODEV;
2224
2225         read_lock(&bond->lock);
2226
2227         bond_for_each_slave(bond, slave, i) {
2228                 if (i == (int)info->slave_id) {
2229                         res = 0;
2230                         strcpy(info->slave_name, slave->dev->name);
2231                         info->link = slave->link;
2232                         info->state = slave->state;
2233                         info->link_failure_count = slave->link_failure_count;
2234                         break;
2235                 }
2236         }
2237
2238         read_unlock(&bond->lock);
2239
2240         return res;
2241 }
2242
2243 /*-------------------------------- Monitoring -------------------------------*/
2244
2245
2246 static int bond_miimon_inspect(struct bonding *bond)
2247 {
2248         struct slave *slave;
2249         int i, link_state, commit = 0;
2250         bool ignore_updelay;
2251
2252         ignore_updelay = !bond->curr_active_slave ? true : false;
2253
2254         bond_for_each_slave(bond, slave, i) {
2255                 slave->new_link = BOND_LINK_NOCHANGE;
2256
2257                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2258
2259                 switch (slave->link) {
2260                 case BOND_LINK_UP:
2261                         if (link_state)
2262                                 continue;
2263
2264                         slave->link = BOND_LINK_FAIL;
2265                         slave->delay = bond->params.downdelay;
2266                         if (slave->delay) {
2267                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2268                                         bond->dev->name,
2269                                         (bond->params.mode ==
2270                                          BOND_MODE_ACTIVEBACKUP) ?
2271                                         ((slave->state == BOND_STATE_ACTIVE) ?
2272                                          "active " : "backup ") : "",
2273                                         slave->dev->name,
2274                                         bond->params.downdelay * bond->params.miimon);
2275                         }
2276                         /*FALLTHRU*/
2277                 case BOND_LINK_FAIL:
2278                         if (link_state) {
2279                                 /*
2280                                  * recovered before downdelay expired
2281                                  */
2282                                 slave->link = BOND_LINK_UP;
2283                                 slave->jiffies = jiffies;
2284                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2285                                         bond->dev->name,
2286                                         (bond->params.downdelay - slave->delay) *
2287                                         bond->params.miimon,
2288                                         slave->dev->name);
2289                                 continue;
2290                         }
2291
2292                         if (slave->delay <= 0) {
2293                                 slave->new_link = BOND_LINK_DOWN;
2294                                 commit++;
2295                                 continue;
2296                         }
2297
2298                         slave->delay--;
2299                         break;
2300
2301                 case BOND_LINK_DOWN:
2302                         if (!link_state)
2303                                 continue;
2304
2305                         slave->link = BOND_LINK_BACK;
2306                         slave->delay = bond->params.updelay;
2307
2308                         if (slave->delay) {
2309                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2310                                         bond->dev->name, slave->dev->name,
2311                                         ignore_updelay ? 0 :
2312                                         bond->params.updelay *
2313                                         bond->params.miimon);
2314                         }
2315                         /*FALLTHRU*/
2316                 case BOND_LINK_BACK:
2317                         if (!link_state) {
2318                                 slave->link = BOND_LINK_DOWN;
2319                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2320                                         bond->dev->name,
2321                                         (bond->params.updelay - slave->delay) *
2322                                         bond->params.miimon,
2323                                         slave->dev->name);
2324
2325                                 continue;
2326                         }
2327
2328                         if (ignore_updelay)
2329                                 slave->delay = 0;
2330
2331                         if (slave->delay <= 0) {
2332                                 slave->new_link = BOND_LINK_UP;
2333                                 commit++;
2334                                 ignore_updelay = false;
2335                                 continue;
2336                         }
2337
2338                         slave->delay--;
2339                         break;
2340                 }
2341         }
2342
2343         return commit;
2344 }
2345
2346 static void bond_miimon_commit(struct bonding *bond)
2347 {
2348         struct slave *slave;
2349         int i;
2350
2351         bond_for_each_slave(bond, slave, i) {
2352                 switch (slave->new_link) {
2353                 case BOND_LINK_NOCHANGE:
2354                         continue;
2355
2356                 case BOND_LINK_UP:
2357                         slave->link = BOND_LINK_UP;
2358                         slave->jiffies = jiffies;
2359
2360                         if (bond->params.mode == BOND_MODE_8023AD) {
2361                                 /* prevent it from being the active one */
2362                                 slave->state = BOND_STATE_BACKUP;
2363                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2364                                 /* make it immediately active */
2365                                 slave->state = BOND_STATE_ACTIVE;
2366                         } else if (slave != bond->primary_slave) {
2367                                 /* prevent it from being the active one */
2368                                 slave->state = BOND_STATE_BACKUP;
2369                         }
2370
2371                         pr_info("%s: link status definitely up for interface %s.\n",
2372                                 bond->dev->name, slave->dev->name);
2373
2374                         /* notify ad that the link status has changed */
2375                         if (bond->params.mode == BOND_MODE_8023AD)
2376                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2377
2378                         if (bond_is_lb(bond))
2379                                 bond_alb_handle_link_change(bond, slave,
2380                                                             BOND_LINK_UP);
2381
2382                         if (!bond->curr_active_slave ||
2383                             (slave == bond->primary_slave))
2384                                 goto do_failover;
2385
2386                         continue;
2387
2388                 case BOND_LINK_DOWN:
2389                         if (slave->link_failure_count < UINT_MAX)
2390                                 slave->link_failure_count++;
2391
2392                         slave->link = BOND_LINK_DOWN;
2393
2394                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2395                             bond->params.mode == BOND_MODE_8023AD)
2396                                 bond_set_slave_inactive_flags(slave);
2397
2398                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2399                                 bond->dev->name, slave->dev->name);
2400
2401                         if (bond->params.mode == BOND_MODE_8023AD)
2402                                 bond_3ad_handle_link_change(slave,
2403                                                             BOND_LINK_DOWN);
2404
2405                         if (bond_is_lb(bond))
2406                                 bond_alb_handle_link_change(bond, slave,
2407                                                             BOND_LINK_DOWN);
2408
2409                         if (slave == bond->curr_active_slave)
2410                                 goto do_failover;
2411
2412                         continue;
2413
2414                 default:
2415                         pr_err("%s: invalid new link %d on slave %s\n",
2416                                bond->dev->name, slave->new_link,
2417                                slave->dev->name);
2418                         slave->new_link = BOND_LINK_NOCHANGE;
2419
2420                         continue;
2421                 }
2422
2423 do_failover:
2424                 ASSERT_RTNL();
2425                 write_lock_bh(&bond->curr_slave_lock);
2426                 bond_select_active_slave(bond);
2427                 write_unlock_bh(&bond->curr_slave_lock);
2428         }
2429
2430         bond_set_carrier(bond);
2431 }
2432
2433 /*
2434  * bond_mii_monitor
2435  *
2436  * Really a wrapper that splits the mii monitor into two phases: an
2437  * inspection, then (if inspection indicates something needs to be done)
2438  * an acquisition of appropriate locks followed by a commit phase to
2439  * implement whatever link state changes are indicated.
2440  */
2441 void bond_mii_monitor(struct work_struct *work)
2442 {
2443         struct bonding *bond = container_of(work, struct bonding,
2444                                             mii_work.work);
2445
2446         read_lock(&bond->lock);
2447         if (bond->kill_timers)
2448                 goto out;
2449
2450         if (bond->slave_cnt == 0)
2451                 goto re_arm;
2452
2453         if (bond->send_grat_arp) {
2454                 read_lock(&bond->curr_slave_lock);
2455                 bond_send_gratuitous_arp(bond);
2456                 read_unlock(&bond->curr_slave_lock);
2457         }
2458
2459         if (bond->send_unsol_na) {
2460                 read_lock(&bond->curr_slave_lock);
2461                 bond_send_unsolicited_na(bond);
2462                 read_unlock(&bond->curr_slave_lock);
2463         }
2464
2465         if (bond_miimon_inspect(bond)) {
2466                 read_unlock(&bond->lock);
2467                 rtnl_lock();
2468                 read_lock(&bond->lock);
2469
2470                 bond_miimon_commit(bond);
2471
2472                 read_unlock(&bond->lock);
2473                 rtnl_unlock();  /* might sleep, hold no other locks */
2474                 read_lock(&bond->lock);
2475         }
2476
2477 re_arm:
2478         if (bond->params.miimon)
2479                 queue_delayed_work(bond->wq, &bond->mii_work,
2480                                    msecs_to_jiffies(bond->params.miimon));
2481 out:
2482         read_unlock(&bond->lock);
2483 }
2484
2485 static __be32 bond_glean_dev_ip(struct net_device *dev)
2486 {
2487         struct in_device *idev;
2488         struct in_ifaddr *ifa;
2489         __be32 addr = 0;
2490
2491         if (!dev)
2492                 return 0;
2493
2494         rcu_read_lock();
2495         idev = __in_dev_get_rcu(dev);
2496         if (!idev)
2497                 goto out;
2498
2499         ifa = idev->ifa_list;
2500         if (!ifa)
2501                 goto out;
2502
2503         addr = ifa->ifa_local;
2504 out:
2505         rcu_read_unlock();
2506         return addr;
2507 }
2508
2509 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2510 {
2511         struct vlan_entry *vlan;
2512
2513         if (ip == bond->master_ip)
2514                 return 1;
2515
2516         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2517                 if (ip == vlan->vlan_ip)
2518                         return 1;
2519         }
2520
2521         return 0;
2522 }
2523
2524 /*
2525  * We go to the (large) trouble of VLAN tagging ARP frames because
2526  * switches in VLAN mode (especially if ports are configured as
2527  * "native" to a VLAN) might not pass non-tagged frames.
2528  */
2529 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2530 {
2531         struct sk_buff *skb;
2532
2533         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2534                  slave_dev->name, dest_ip, src_ip, vlan_id);
2535
2536         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2537                          NULL, slave_dev->dev_addr, NULL);
2538
2539         if (!skb) {
2540                 pr_err("ARP packet allocation failed\n");
2541                 return;
2542         }
2543         if (vlan_id) {
2544                 skb = vlan_put_tag(skb, vlan_id);
2545                 if (!skb) {
2546                         pr_err("failed to insert VLAN tag\n");
2547                         return;
2548                 }
2549         }
2550         arp_xmit(skb);
2551 }
2552
2553
2554 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2555 {
2556         int i, vlan_id, rv;
2557         __be32 *targets = bond->params.arp_targets;
2558         struct vlan_entry *vlan;
2559         struct net_device *vlan_dev;
2560         struct flowi fl;
2561         struct rtable *rt;
2562
2563         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2564                 if (!targets[i])
2565                         break;
2566                 pr_debug("basa: target %x\n", targets[i]);
2567                 if (!bond->vlgrp) {
2568                         pr_debug("basa: empty vlan: arp_send\n");
2569                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2570                                       bond->master_ip, 0);
2571                         continue;
2572                 }
2573
2574                 /*
2575                  * If VLANs are configured, we do a route lookup to
2576                  * determine which VLAN interface would be used, so we
2577                  * can tag the ARP with the proper VLAN tag.
2578                  */
2579                 memset(&fl, 0, sizeof(fl));
2580                 fl.fl4_dst = targets[i];
2581                 fl.fl4_tos = RTO_ONLINK;
2582
2583                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2584                 if (rv) {
2585                         if (net_ratelimit()) {
2586                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2587                                            bond->dev->name, &fl.fl4_dst);
2588                         }
2589                         continue;
2590                 }
2591
2592                 /*
2593                  * This target is not on a VLAN
2594                  */
2595                 if (rt->dst.dev == bond->dev) {
2596                         ip_rt_put(rt);
2597                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2598                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2599                                       bond->master_ip, 0);
2600                         continue;
2601                 }
2602
2603                 vlan_id = 0;
2604                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2605                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2606                         if (vlan_dev == rt->dst.dev) {
2607                                 vlan_id = vlan->vlan_id;
2608                                 pr_debug("basa: vlan match on %s %d\n",
2609                                        vlan_dev->name, vlan_id);
2610                                 break;
2611                         }
2612                 }
2613
2614                 if (vlan_id) {
2615                         ip_rt_put(rt);
2616                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2617                                       vlan->vlan_ip, vlan_id);
2618                         continue;
2619                 }
2620
2621                 if (net_ratelimit()) {
2622                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2623                                    bond->dev->name, &fl.fl4_dst,
2624                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2625                 }
2626                 ip_rt_put(rt);
2627         }
2628 }
2629
2630 /*
2631  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2632  * for each VLAN above us.
2633  *
2634  * Caller must hold curr_slave_lock for read or better
2635  */
2636 static void bond_send_gratuitous_arp(struct bonding *bond)
2637 {
2638         struct slave *slave = bond->curr_active_slave;
2639         struct vlan_entry *vlan;
2640         struct net_device *vlan_dev;
2641
2642         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2643                  bond->dev->name, slave ? slave->dev->name : "NULL");
2644
2645         if (!slave || !bond->send_grat_arp ||
2646             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2647                 return;
2648
2649         bond->send_grat_arp--;
2650
2651         if (bond->master_ip) {
2652                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2653                                 bond->master_ip, 0);
2654         }
2655
2656         if (!bond->vlgrp)
2657                 return;
2658
2659         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2660                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2661                 if (vlan->vlan_ip) {
2662                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2663                                       vlan->vlan_ip, vlan->vlan_id);
2664                 }
2665         }
2666 }
2667
2668 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2669 {
2670         int i;
2671         __be32 *targets = bond->params.arp_targets;
2672
2673         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2674                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2675                          &sip, &tip, i, &targets[i],
2676                          bond_has_this_ip(bond, tip));
2677                 if (sip == targets[i]) {
2678                         if (bond_has_this_ip(bond, tip))
2679                                 slave->last_arp_rx = jiffies;
2680                         return;
2681                 }
2682         }
2683 }
2684
2685 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2686 {
2687         struct arphdr *arp;
2688         struct slave *slave;
2689         struct bonding *bond;
2690         unsigned char *arp_ptr;
2691         __be32 sip, tip;
2692
2693         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2694                 /*
2695                  * When using VLANS and bonding, dev and oriv_dev may be
2696                  * incorrect if the physical interface supports VLAN
2697                  * acceleration.  With this change ARP validation now
2698                  * works for hosts only reachable on the VLAN interface.
2699                  */
2700                 dev = vlan_dev_real_dev(dev);
2701                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2702         }
2703
2704         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2705                 goto out;
2706
2707         bond = netdev_priv(dev);
2708         read_lock(&bond->lock);
2709
2710         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2711                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2712                  orig_dev ? orig_dev->name : "NULL");
2713
2714         slave = bond_get_slave_by_dev(bond, orig_dev);
2715         if (!slave || !slave_do_arp_validate(bond, slave))
2716                 goto out_unlock;
2717
2718         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2719                 goto out_unlock;
2720
2721         arp = arp_hdr(skb);
2722         if (arp->ar_hln != dev->addr_len ||
2723             skb->pkt_type == PACKET_OTHERHOST ||
2724             skb->pkt_type == PACKET_LOOPBACK ||
2725             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2726             arp->ar_pro != htons(ETH_P_IP) ||
2727             arp->ar_pln != 4)
2728                 goto out_unlock;
2729
2730         arp_ptr = (unsigned char *)(arp + 1);
2731         arp_ptr += dev->addr_len;
2732         memcpy(&sip, arp_ptr, 4);
2733         arp_ptr += 4 + dev->addr_len;
2734         memcpy(&tip, arp_ptr, 4);
2735
2736         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2737                  bond->dev->name, slave->dev->name, slave->state,
2738                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2739                  &sip, &tip);
2740
2741         /*
2742          * Backup slaves won't see the ARP reply, but do come through
2743          * here for each ARP probe (so we swap the sip/tip to validate
2744          * the probe).  In a "redundant switch, common router" type of
2745          * configuration, the ARP probe will (hopefully) travel from
2746          * the active, through one switch, the router, then the other
2747          * switch before reaching the backup.
2748          */
2749         if (slave->state == BOND_STATE_ACTIVE)
2750                 bond_validate_arp(bond, slave, sip, tip);
2751         else
2752                 bond_validate_arp(bond, slave, tip, sip);
2753
2754 out_unlock:
2755         read_unlock(&bond->lock);
2756 out:
2757         dev_kfree_skb(skb);
2758         return NET_RX_SUCCESS;
2759 }
2760
2761 /*
2762  * this function is called regularly to monitor each slave's link
2763  * ensuring that traffic is being sent and received when arp monitoring
2764  * is used in load-balancing mode. if the adapter has been dormant, then an
2765  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2766  * arp monitoring in active backup mode.
2767  */
2768 void bond_loadbalance_arp_mon(struct work_struct *work)
2769 {
2770         struct bonding *bond = container_of(work, struct bonding,
2771                                             arp_work.work);
2772         struct slave *slave, *oldcurrent;
2773         int do_failover = 0;
2774         int delta_in_ticks;
2775         int i;
2776
2777         read_lock(&bond->lock);
2778
2779         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2780
2781         if (bond->kill_timers)
2782                 goto out;
2783
2784         if (bond->slave_cnt == 0)
2785                 goto re_arm;
2786
2787         read_lock(&bond->curr_slave_lock);
2788         oldcurrent = bond->curr_active_slave;
2789         read_unlock(&bond->curr_slave_lock);
2790
2791         /* see if any of the previous devices are up now (i.e. they have
2792          * xmt and rcv traffic). the curr_active_slave does not come into
2793          * the picture unless it is null. also, slave->jiffies is not needed
2794          * here because we send an arp on each slave and give a slave as
2795          * long as it needs to get the tx/rx within the delta.
2796          * TODO: what about up/down delay in arp mode? it wasn't here before
2797          *       so it can wait
2798          */
2799         bond_for_each_slave(bond, slave, i) {
2800                 if (slave->link != BOND_LINK_UP) {
2801                         if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2802                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2803
2804                                 slave->link  = BOND_LINK_UP;
2805                                 slave->state = BOND_STATE_ACTIVE;
2806
2807                                 /* primary_slave has no meaning in round-robin
2808                                  * mode. the window of a slave being up and
2809                                  * curr_active_slave being null after enslaving
2810                                  * is closed.
2811                                  */
2812                                 if (!oldcurrent) {
2813                                         pr_info("%s: link status definitely up for interface %s, ",
2814                                                 bond->dev->name,
2815                                                 slave->dev->name);
2816                                         do_failover = 1;
2817                                 } else {
2818                                         pr_info("%s: interface %s is now up\n",
2819                                                 bond->dev->name,
2820                                                 slave->dev->name);
2821                                 }
2822                         }
2823                 } else {
2824                         /* slave->link == BOND_LINK_UP */
2825
2826                         /* not all switches will respond to an arp request
2827                          * when the source ip is 0, so don't take the link down
2828                          * if we don't know our ip yet
2829                          */
2830                         if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2831                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2832
2833                                 slave->link  = BOND_LINK_DOWN;
2834                                 slave->state = BOND_STATE_BACKUP;
2835
2836                                 if (slave->link_failure_count < UINT_MAX)
2837                                         slave->link_failure_count++;
2838
2839                                 pr_info("%s: interface %s is now down.\n",
2840                                         bond->dev->name,
2841                                         slave->dev->name);
2842
2843                                 if (slave == oldcurrent)
2844                                         do_failover = 1;
2845                         }
2846                 }
2847
2848                 /* note: if switch is in round-robin mode, all links
2849                  * must tx arp to ensure all links rx an arp - otherwise
2850                  * links may oscillate or not come up at all; if switch is
2851                  * in something like xor mode, there is nothing we can
2852                  * do - all replies will be rx'ed on same link causing slaves
2853                  * to be unstable during low/no traffic periods
2854                  */
2855                 if (IS_UP(slave->dev))
2856                         bond_arp_send_all(bond, slave);
2857         }
2858
2859         if (do_failover) {
2860                 write_lock_bh(&bond->curr_slave_lock);
2861
2862                 bond_select_active_slave(bond);
2863
2864                 write_unlock_bh(&bond->curr_slave_lock);
2865         }
2866
2867 re_arm:
2868         if (bond->params.arp_interval)
2869                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2870 out:
2871         read_unlock(&bond->lock);
2872 }
2873
2874 /*
2875  * Called to inspect slaves for active-backup mode ARP monitor link state
2876  * changes.  Sets new_link in slaves to specify what action should take
2877  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2878  * to link states must be committed.
2879  *
2880  * Called with bond->lock held for read.
2881  */
2882 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2883 {
2884         struct slave *slave;
2885         int i, commit = 0;
2886
2887         bond_for_each_slave(bond, slave, i) {
2888                 slave->new_link = BOND_LINK_NOCHANGE;
2889
2890                 if (slave->link != BOND_LINK_UP) {
2891                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2892                                            delta_in_ticks)) {
2893                                 slave->new_link = BOND_LINK_UP;
2894                                 commit++;
2895                         }
2896
2897                         continue;
2898                 }
2899
2900                 /*
2901                  * Give slaves 2*delta after being enslaved or made
2902                  * active.  This avoids bouncing, as the last receive
2903                  * times need a full ARP monitor cycle to be updated.
2904                  */
2905                 if (!time_after_eq(jiffies, slave->jiffies +
2906                                    2 * delta_in_ticks))
2907                         continue;
2908
2909                 /*
2910                  * Backup slave is down if:
2911                  * - No current_arp_slave AND
2912                  * - more than 3*delta since last receive AND
2913                  * - the bond has an IP address
2914                  *
2915                  * Note: a non-null current_arp_slave indicates
2916                  * the curr_active_slave went down and we are
2917                  * searching for a new one; under this condition
2918                  * we only take the curr_active_slave down - this
2919                  * gives each slave a chance to tx/rx traffic
2920                  * before being taken out
2921                  */
2922                 if (slave->state == BOND_STATE_BACKUP &&
2923                     !bond->current_arp_slave &&
2924                     time_after(jiffies, slave_last_rx(bond, slave) +
2925                                3 * delta_in_ticks)) {
2926                         slave->new_link = BOND_LINK_DOWN;
2927                         commit++;
2928                 }
2929
2930                 /*
2931                  * Active slave is down if:
2932                  * - more than 2*delta since transmitting OR
2933                  * - (more than 2*delta since receive AND
2934                  *    the bond has an IP address)
2935                  */
2936                 if ((slave->state == BOND_STATE_ACTIVE) &&
2937                     (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2938                                     2 * delta_in_ticks) ||
2939                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2940                                      + 2 * delta_in_ticks)))) {
2941                         slave->new_link = BOND_LINK_DOWN;
2942                         commit++;
2943                 }
2944         }
2945
2946         return commit;
2947 }
2948
2949 /*
2950  * Called to commit link state changes noted by inspection step of
2951  * active-backup mode ARP monitor.
2952  *
2953  * Called with RTNL and bond->lock for read.
2954  */
2955 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2956 {
2957         struct slave *slave;
2958         int i;
2959
2960         bond_for_each_slave(bond, slave, i) {
2961                 switch (slave->new_link) {
2962                 case BOND_LINK_NOCHANGE:
2963                         continue;
2964
2965                 case BOND_LINK_UP:
2966                         if ((!bond->curr_active_slave &&
2967                              time_before_eq(jiffies,
2968                                             dev_trans_start(slave->dev) +
2969                                             delta_in_ticks)) ||
2970                             bond->curr_active_slave != slave) {
2971                                 slave->link = BOND_LINK_UP;
2972                                 bond->current_arp_slave = NULL;
2973
2974                                 pr_info("%s: link status definitely up for interface %s.\n",
2975                                         bond->dev->name, slave->dev->name);
2976
2977                                 if (!bond->curr_active_slave ||
2978                                     (slave == bond->primary_slave))
2979                                         goto do_failover;
2980
2981                         }
2982
2983                         continue;
2984
2985                 case BOND_LINK_DOWN:
2986                         if (slave->link_failure_count < UINT_MAX)
2987                                 slave->link_failure_count++;
2988
2989                         slave->link = BOND_LINK_DOWN;
2990                         bond_set_slave_inactive_flags(slave);
2991
2992                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2993                                 bond->dev->name, slave->dev->name);
2994
2995                         if (slave == bond->curr_active_slave) {
2996                                 bond->current_arp_slave = NULL;
2997                                 goto do_failover;
2998                         }
2999
3000                         continue;
3001
3002                 default:
3003                         pr_err("%s: impossible: new_link %d on slave %s\n",
3004                                bond->dev->name, slave->new_link,
3005                                slave->dev->name);
3006                         continue;
3007                 }
3008
3009 do_failover:
3010                 ASSERT_RTNL();
3011                 write_lock_bh(&bond->curr_slave_lock);
3012                 bond_select_active_slave(bond);
3013                 write_unlock_bh(&bond->curr_slave_lock);
3014         }
3015
3016         bond_set_carrier(bond);
3017 }
3018
3019 /*
3020  * Send ARP probes for active-backup mode ARP monitor.
3021  *
3022  * Called with bond->lock held for read.
3023  */
3024 static void bond_ab_arp_probe(struct bonding *bond)
3025 {
3026         struct slave *slave;
3027         int i;
3028
3029         read_lock(&bond->curr_slave_lock);
3030
3031         if (bond->current_arp_slave && bond->curr_active_slave)
3032                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3033                         bond->current_arp_slave->dev->name,
3034                         bond->curr_active_slave->dev->name);
3035
3036         if (bond->curr_active_slave) {
3037                 bond_arp_send_all(bond, bond->curr_active_slave);
3038                 read_unlock(&bond->curr_slave_lock);
3039                 return;
3040         }
3041
3042         read_unlock(&bond->curr_slave_lock);
3043
3044         /* if we don't have a curr_active_slave, search for the next available
3045          * backup slave from the current_arp_slave and make it the candidate
3046          * for becoming the curr_active_slave
3047          */
3048
3049         if (!bond->current_arp_slave) {
3050                 bond->current_arp_slave = bond->first_slave;
3051                 if (!bond->current_arp_slave)
3052                         return;
3053         }
3054
3055         bond_set_slave_inactive_flags(bond->current_arp_slave);
3056
3057         /* search for next candidate */
3058         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3059                 if (IS_UP(slave->dev)) {
3060                         slave->link = BOND_LINK_BACK;
3061                         bond_set_slave_active_flags(slave);
3062                         bond_arp_send_all(bond, slave);
3063                         slave->jiffies = jiffies;
3064                         bond->current_arp_slave = slave;
3065                         break;
3066                 }
3067
3068                 /* if the link state is up at this point, we
3069                  * mark it down - this can happen if we have
3070                  * simultaneous link failures and
3071                  * reselect_active_interface doesn't make this
3072                  * one the current slave so it is still marked
3073                  * up when it is actually down
3074                  */
3075                 if (slave->link == BOND_LINK_UP) {
3076                         slave->link = BOND_LINK_DOWN;
3077                         if (slave->link_failure_count < UINT_MAX)
3078                                 slave->link_failure_count++;
3079
3080                         bond_set_slave_inactive_flags(slave);
3081
3082                         pr_info("%s: backup interface %s is now down.\n",
3083                                 bond->dev->name, slave->dev->name);
3084                 }
3085         }
3086 }
3087
3088 void bond_activebackup_arp_mon(struct work_struct *work)
3089 {
3090         struct bonding *bond = container_of(work, struct bonding,
3091                                             arp_work.work);
3092         int delta_in_ticks;
3093
3094         read_lock(&bond->lock);
3095
3096         if (bond->kill_timers)
3097                 goto out;
3098
3099         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3100
3101         if (bond->slave_cnt == 0)
3102                 goto re_arm;
3103
3104         if (bond->send_grat_arp) {
3105                 read_lock(&bond->curr_slave_lock);
3106                 bond_send_gratuitous_arp(bond);
3107                 read_unlock(&bond->curr_slave_lock);
3108         }
3109
3110         if (bond->send_unsol_na) {
3111                 read_lock(&bond->curr_slave_lock);
3112                 bond_send_unsolicited_na(bond);
3113                 read_unlock(&bond->curr_slave_lock);
3114         }
3115
3116         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3117                 read_unlock(&bond->lock);
3118                 rtnl_lock();
3119                 read_lock(&bond->lock);
3120
3121                 bond_ab_arp_commit(bond, delta_in_ticks);
3122
3123                 read_unlock(&bond->lock);
3124                 rtnl_unlock();
3125                 read_lock(&bond->lock);
3126         }
3127
3128         bond_ab_arp_probe(bond);
3129
3130 re_arm:
3131         if (bond->params.arp_interval)
3132                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3133 out:
3134         read_unlock(&bond->lock);
3135 }
3136
3137 /*------------------------------ proc/seq_file-------------------------------*/
3138
3139 #ifdef CONFIG_PROC_FS
3140
3141 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3142         __acquires(&dev_base_lock)
3143         __acquires(&bond->lock)
3144 {
3145         struct bonding *bond = seq->private;
3146         loff_t off = 0;
3147         struct slave *slave;
3148         int i;
3149
3150         /* make sure the bond won't be taken away */
3151         read_lock(&dev_base_lock);
3152         read_lock(&bond->lock);
3153
3154         if (*pos == 0)
3155                 return SEQ_START_TOKEN;
3156
3157         bond_for_each_slave(bond, slave, i) {
3158                 if (++off == *pos)
3159                         return slave;
3160         }
3161
3162         return NULL;
3163 }
3164
3165 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3166 {
3167         struct bonding *bond = seq->private;
3168         struct slave *slave = v;
3169
3170         ++*pos;
3171         if (v == SEQ_START_TOKEN)
3172                 return bond->first_slave;
3173
3174         slave = slave->next;
3175
3176         return (slave == bond->first_slave) ? NULL : slave;
3177 }
3178
3179 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3180         __releases(&bond->lock)
3181         __releases(&dev_base_lock)
3182 {
3183         struct bonding *bond = seq->private;
3184
3185         read_unlock(&bond->lock);
3186         read_unlock(&dev_base_lock);
3187 }
3188
3189 static void bond_info_show_master(struct seq_file *seq)
3190 {
3191         struct bonding *bond = seq->private;
3192         struct slave *curr;
3193         int i;
3194
3195         read_lock(&bond->curr_slave_lock);
3196         curr = bond->curr_active_slave;
3197         read_unlock(&bond->curr_slave_lock);
3198
3199         seq_printf(seq, "Bonding Mode: %s",
3200                    bond_mode_name(bond->params.mode));
3201
3202         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3203             bond->params.fail_over_mac)
3204                 seq_printf(seq, " (fail_over_mac %s)",
3205                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3206
3207         seq_printf(seq, "\n");
3208
3209         if (bond->params.mode == BOND_MODE_XOR ||
3210                 bond->params.mode == BOND_MODE_8023AD) {
3211                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3212                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3213                         bond->params.xmit_policy);
3214         }
3215
3216         if (USES_PRIMARY(bond->params.mode)) {
3217                 seq_printf(seq, "Primary Slave: %s",
3218                            (bond->primary_slave) ?
3219                            bond->primary_slave->dev->name : "None");
3220                 if (bond->primary_slave)
3221                         seq_printf(seq, " (primary_reselect %s)",
3222                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3223
3224                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3225                            (curr) ? curr->dev->name : "None");
3226         }
3227
3228         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3229                    "up" : "down");
3230         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3231         seq_printf(seq, "Up Delay (ms): %d\n",
3232                    bond->params.updelay * bond->params.miimon);
3233         seq_printf(seq, "Down Delay (ms): %d\n",
3234                    bond->params.downdelay * bond->params.miimon);
3235
3236
3237         /* ARP information */
3238         if (bond->params.arp_interval > 0) {
3239                 int printed = 0;
3240                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3241                                 bond->params.arp_interval);
3242
3243                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3244
3245                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3246                         if (!bond->params.arp_targets[i])
3247                                 break;
3248                         if (printed)
3249                                 seq_printf(seq, ",");
3250                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3251                         printed = 1;
3252                 }
3253                 seq_printf(seq, "\n");
3254         }
3255
3256         if (bond->params.mode == BOND_MODE_8023AD) {
3257                 struct ad_info ad_info;
3258
3259                 seq_puts(seq, "\n802.3ad info\n");
3260                 seq_printf(seq, "LACP rate: %s\n",
3261                            (bond->params.lacp_fast) ? "fast" : "slow");
3262                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3263                            ad_select_tbl[bond->params.ad_select].modename);
3264
3265                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3266                         seq_printf(seq, "bond %s has no active aggregator\n",
3267                                    bond->dev->name);
3268                 } else {
3269                         seq_printf(seq, "Active Aggregator Info:\n");
3270
3271                         seq_printf(seq, "\tAggregator ID: %d\n",
3272                                    ad_info.aggregator_id);
3273                         seq_printf(seq, "\tNumber of ports: %d\n",
3274                                    ad_info.ports);
3275                         seq_printf(seq, "\tActor Key: %d\n",
3276                                    ad_info.actor_key);
3277                         seq_printf(seq, "\tPartner Key: %d\n",
3278                                    ad_info.partner_key);
3279                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3280                                    ad_info.partner_system);
3281                 }
3282         }
3283 }
3284
3285 static void bond_info_show_slave(struct seq_file *seq,
3286                                  const struct slave *slave)
3287 {
3288         struct bonding *bond = seq->private;
3289
3290         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3291         seq_printf(seq, "MII Status: %s\n",
3292                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3293         seq_printf(seq, "Link Failure Count: %u\n",
3294                    slave->link_failure_count);
3295
3296         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3297
3298         if (bond->params.mode == BOND_MODE_8023AD) {
3299                 const struct aggregator *agg
3300                         = SLAVE_AD_INFO(slave).port.aggregator;
3301
3302                 if (agg)
3303                         seq_printf(seq, "Aggregator ID: %d\n",
3304                                    agg->aggregator_identifier);
3305                 else
3306                         seq_puts(seq, "Aggregator ID: N/A\n");
3307         }
3308         seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3309 }
3310
3311 static int bond_info_seq_show(struct seq_file *seq, void *v)
3312 {
3313         if (v == SEQ_START_TOKEN) {
3314                 seq_printf(seq, "%s\n", version);
3315                 bond_info_show_master(seq);
3316         } else
3317                 bond_info_show_slave(seq, v);
3318
3319         return 0;
3320 }
3321
3322 static const struct seq_operations bond_info_seq_ops = {
3323         .start = bond_info_seq_start,
3324         .next  = bond_info_seq_next,
3325         .stop  = bond_info_seq_stop,
3326         .show  = bond_info_seq_show,
3327 };
3328
3329 static int bond_info_open(struct inode *inode, struct file *file)
3330 {
3331         struct seq_file *seq;
3332         struct proc_dir_entry *proc;
3333         int res;
3334
3335         res = seq_open(file, &bond_info_seq_ops);
3336         if (!res) {
3337                 /* recover the pointer buried in proc_dir_entry data */
3338                 seq = file->private_data;
3339                 proc = PDE(inode);
3340                 seq->private = proc->data;
3341         }
3342
3343         return res;
3344 }
3345
3346 static const struct file_operations bond_info_fops = {
3347         .owner   = THIS_MODULE,
3348         .open    = bond_info_open,
3349         .read    = seq_read,
3350         .llseek  = seq_lseek,
3351         .release = seq_release,
3352 };
3353
3354 static void bond_create_proc_entry(struct bonding *bond)
3355 {
3356         struct net_device *bond_dev = bond->dev;
3357         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3358
3359         if (bn->proc_dir) {
3360                 bond->proc_entry = proc_create_data(bond_dev->name,
3361                                                     S_IRUGO, bn->proc_dir,
3362                                                     &bond_info_fops, bond);
3363                 if (bond->proc_entry == NULL)
3364                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3365                                    DRV_NAME, bond_dev->name);
3366                 else
3367                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3368         }
3369 }
3370
3371 static void bond_remove_proc_entry(struct bonding *bond)
3372 {
3373         struct net_device *bond_dev = bond->dev;
3374         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3375
3376         if (bn->proc_dir && bond->proc_entry) {
3377                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3378                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3379                 bond->proc_entry = NULL;
3380         }
3381 }
3382
3383 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3384  * Caller must hold rtnl_lock.
3385  */
3386 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3387 {
3388         if (!bn->proc_dir) {
3389                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3390                 if (!bn->proc_dir)
3391                         pr_warning("Warning: cannot create /proc/net/%s\n",
3392                                    DRV_NAME);
3393         }
3394 }
3395
3396 /* Destroy the bonding directory under /proc/net, if empty.
3397  * Caller must hold rtnl_lock.
3398  */
3399 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3400 {
3401         if (bn->proc_dir) {
3402                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3403                 bn->proc_dir = NULL;
3404         }
3405 }
3406
3407 #else /* !CONFIG_PROC_FS */
3408
3409 static void bond_create_proc_entry(struct bonding *bond)
3410 {
3411 }
3412
3413 static void bond_remove_proc_entry(struct bonding *bond)
3414 {
3415 }
3416
3417 static inline void bond_create_proc_dir(struct bond_net *bn)
3418 {
3419 }
3420
3421 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3422 {
3423 }
3424
3425 #endif /* CONFIG_PROC_FS */
3426
3427
3428 /*-------------------------- netdev event handling --------------------------*/
3429
3430 /*
3431  * Change device name
3432  */
3433 static int bond_event_changename(struct bonding *bond)
3434 {
3435         bond_remove_proc_entry(bond);
3436         bond_create_proc_entry(bond);
3437
3438         return NOTIFY_DONE;
3439 }
3440
3441 static int bond_master_netdev_event(unsigned long event,
3442                                     struct net_device *bond_dev)
3443 {
3444         struct bonding *event_bond = netdev_priv(bond_dev);
3445
3446         switch (event) {
3447         case NETDEV_CHANGENAME:
3448                 return bond_event_changename(event_bond);
3449         default:
3450                 break;
3451         }
3452
3453         return NOTIFY_DONE;
3454 }
3455
3456 static int bond_slave_netdev_event(unsigned long event,
3457                                    struct net_device *slave_dev)
3458 {
3459         struct net_device *bond_dev = slave_dev->master;
3460         struct bonding *bond = netdev_priv(bond_dev);
3461
3462         switch (event) {
3463         case NETDEV_UNREGISTER:
3464                 if (bond_dev) {
3465                         if (bond->setup_by_slave)
3466                                 bond_release_and_destroy(bond_dev, slave_dev);
3467                         else
3468                                 bond_release(bond_dev, slave_dev);
3469                 }
3470                 break;
3471         case NETDEV_CHANGE:
3472                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3473                         struct slave *slave;
3474
3475                         slave = bond_get_slave_by_dev(bond, slave_dev);
3476                         if (slave) {
3477                                 u16 old_speed = slave->speed;
3478                                 u16 old_duplex = slave->duplex;
3479
3480                                 bond_update_speed_duplex(slave);
3481
3482                                 if (bond_is_lb(bond))
3483                                         break;
3484
3485                                 if (old_speed != slave->speed)
3486                                         bond_3ad_adapter_speed_changed(slave);
3487                                 if (old_duplex != slave->duplex)
3488                                         bond_3ad_adapter_duplex_changed(slave);
3489                         }
3490                 }
3491
3492                 break;
3493         case NETDEV_DOWN:
3494                 /*
3495                  * ... Or is it this?
3496                  */
3497                 break;
3498         case NETDEV_CHANGEMTU:
3499                 /*
3500                  * TODO: Should slaves be allowed to
3501                  * independently alter their MTU?  For
3502                  * an active-backup bond, slaves need
3503                  * not be the same type of device, so
3504                  * MTUs may vary.  For other modes,
3505                  * slaves arguably should have the
3506                  * same MTUs. To do this, we'd need to
3507                  * take over the slave's change_mtu
3508                  * function for the duration of their
3509                  * servitude.
3510                  */
3511                 break;
3512         case NETDEV_CHANGENAME:
3513                 /*
3514                  * TODO: handle changing the primary's name
3515                  */
3516                 break;
3517         case NETDEV_FEAT_CHANGE:
3518                 bond_compute_features(bond);
3519                 break;
3520         default:
3521                 break;
3522         }
3523
3524         return NOTIFY_DONE;
3525 }
3526
3527 /*
3528  * bond_netdev_event: handle netdev notifier chain events.
3529  *
3530  * This function receives events for the netdev chain.  The caller (an
3531  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3532  * locks for us to safely manipulate the slave devices (RTNL lock,
3533  * dev_probe_lock).
3534  */
3535 static int bond_netdev_event(struct notifier_block *this,
3536                              unsigned long event, void *ptr)
3537 {
3538         struct net_device *event_dev = (struct net_device *)ptr;
3539
3540         pr_debug("event_dev: %s, event: %lx\n",
3541                  event_dev ? event_dev->name : "None",
3542                  event);
3543
3544         if (!(event_dev->priv_flags & IFF_BONDING))
3545                 return NOTIFY_DONE;
3546
3547         if (event_dev->flags & IFF_MASTER) {
3548                 pr_debug("IFF_MASTER\n");
3549                 return bond_master_netdev_event(event, event_dev);
3550         }
3551
3552         if (event_dev->flags & IFF_SLAVE) {
3553                 pr_debug("IFF_SLAVE\n");
3554                 return bond_slave_netdev_event(event, event_dev);
3555         }
3556
3557         return NOTIFY_DONE;
3558 }
3559
3560 /*
3561  * bond_inetaddr_event: handle inetaddr notifier chain events.
3562  *
3563  * We keep track of device IPs primarily to use as source addresses in
3564  * ARP monitor probes (rather than spewing out broadcasts all the time).
3565  *
3566  * We track one IP for the main device (if it has one), plus one per VLAN.
3567  */
3568 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3569 {
3570         struct in_ifaddr *ifa = ptr;
3571         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3572         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3573         struct bonding *bond;
3574         struct vlan_entry *vlan;
3575
3576         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3577                 if (bond->dev == event_dev) {
3578                         switch (event) {
3579                         case NETDEV_UP:
3580                                 bond->master_ip = ifa->ifa_local;
3581                                 return NOTIFY_OK;
3582                         case NETDEV_DOWN:
3583                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3584                                 return NOTIFY_OK;
3585                         default:
3586                                 return NOTIFY_DONE;
3587                         }
3588                 }
3589
3590                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3591                         if (!bond->vlgrp)
3592                                 continue;
3593                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3594                         if (vlan_dev == event_dev) {
3595                                 switch (event) {
3596                                 case NETDEV_UP:
3597                                         vlan->vlan_ip = ifa->ifa_local;
3598                                         return NOTIFY_OK;
3599                                 case NETDEV_DOWN:
3600                                         vlan->vlan_ip =
3601                                                 bond_glean_dev_ip(vlan_dev);
3602                                         return NOTIFY_OK;
3603                                 default:
3604                                         return NOTIFY_DONE;
3605                                 }
3606                         }
3607                 }
3608         }
3609         return NOTIFY_DONE;
3610 }
3611
3612 static struct notifier_block bond_netdev_notifier = {
3613         .notifier_call = bond_netdev_event,
3614 };
3615
3616 static struct notifier_block bond_inetaddr_notifier = {
3617         .notifier_call = bond_inetaddr_event,
3618 };
3619
3620 /*-------------------------- Packet type handling ---------------------------*/
3621
3622 /* register to receive lacpdus on a bond */
3623 static void bond_register_lacpdu(struct bonding *bond)
3624 {
3625         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3626
3627         /* initialize packet type */
3628         pk_type->type = PKT_TYPE_LACPDU;
3629         pk_type->dev = bond->dev;
3630         pk_type->func = bond_3ad_lacpdu_recv;
3631
3632         dev_add_pack(pk_type);
3633 }
3634
3635 /* unregister to receive lacpdus on a bond */
3636 static void bond_unregister_lacpdu(struct bonding *bond)
3637 {
3638         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3639 }
3640
3641 void bond_register_arp(struct bonding *bond)
3642 {
3643         struct packet_type *pt = &bond->arp_mon_pt;
3644
3645         if (pt->type)
3646                 return;
3647
3648         pt->type = htons(ETH_P_ARP);
3649         pt->dev = bond->dev;
3650         pt->func = bond_arp_rcv;
3651         dev_add_pack(pt);
3652 }
3653
3654 void bond_unregister_arp(struct bonding *bond)
3655 {
3656         struct packet_type *pt = &bond->arp_mon_pt;
3657
3658         dev_remove_pack(pt);
3659         pt->type = 0;
3660 }
3661
3662 /*---------------------------- Hashing Policies -----------------------------*/
3663
3664 /*
3665  * Hash for the output device based upon layer 2 and layer 3 data. If
3666  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3667  */
3668 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3669 {
3670         struct ethhdr *data = (struct ethhdr *)skb->data;
3671         struct iphdr *iph = ip_hdr(skb);
3672
3673         if (skb->protocol == htons(ETH_P_IP)) {
3674                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3675                         (data->h_dest[5] ^ data->h_source[5])) % count;
3676         }
3677
3678         return (data->h_dest[5] ^ data->h_source[5]) % count;
3679 }
3680
3681 /*
3682  * Hash for the output device based upon layer 3 and layer 4 data. If
3683  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3684  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3685  */
3686 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3687 {
3688         struct ethhdr *data = (struct ethhdr *)skb->data;
3689         struct iphdr *iph = ip_hdr(skb);
3690         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3691         int layer4_xor = 0;
3692
3693         if (skb->protocol == htons(ETH_P_IP)) {
3694                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3695                     (iph->protocol == IPPROTO_TCP ||
3696                      iph->protocol == IPPROTO_UDP)) {
3697                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3698                 }
3699                 return (layer4_xor ^
3700                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3701
3702         }
3703
3704         return (data->h_dest[5] ^ data->h_source[5]) % count;
3705 }
3706
3707 /*
3708  * Hash for the output device based upon layer 2 data
3709  */
3710 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3711 {
3712         struct ethhdr *data = (struct ethhdr *)skb->data;
3713
3714         return (data->h_dest[5] ^ data->h_source[5]) % count;
3715 }
3716
3717 /*-------------------------- Device entry points ----------------------------*/
3718
3719 static int bond_open(struct net_device *bond_dev)
3720 {
3721         struct bonding *bond = netdev_priv(bond_dev);
3722
3723         bond->kill_timers = 0;
3724
3725         if (bond_is_lb(bond)) {
3726                 /* bond_alb_initialize must be called before the timer
3727                  * is started.
3728                  */
3729                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3730                         /* something went wrong - fail the open operation */
3731                         return -ENOMEM;
3732                 }
3733
3734                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3735                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3736         }
3737
3738         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3739                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3740                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3741         }
3742
3743         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3744                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3745                         INIT_DELAYED_WORK(&bond->arp_work,
3746                                           bond_activebackup_arp_mon);
3747                 else
3748                         INIT_DELAYED_WORK(&bond->arp_work,
3749                                           bond_loadbalance_arp_mon);
3750
3751                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3752                 if (bond->params.arp_validate)
3753                         bond_register_arp(bond);
3754         }
3755
3756         if (bond->params.mode == BOND_MODE_8023AD) {
3757                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3758                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3759                 /* register to receive LACPDUs */
3760                 bond_register_lacpdu(bond);
3761                 bond_3ad_initiate_agg_selection(bond, 1);
3762         }
3763
3764         return 0;
3765 }
3766
3767 static int bond_close(struct net_device *bond_dev)
3768 {
3769         struct bonding *bond = netdev_priv(bond_dev);
3770
3771         if (bond->params.mode == BOND_MODE_8023AD) {
3772                 /* Unregister the receive of LACPDUs */
3773                 bond_unregister_lacpdu(bond);
3774         }
3775
3776         if (bond->params.arp_validate)
3777                 bond_unregister_arp(bond);
3778
3779         write_lock_bh(&bond->lock);
3780
3781         bond->send_grat_arp = 0;
3782         bond->send_unsol_na = 0;
3783
3784         /* signal timers not to re-arm */
3785         bond->kill_timers = 1;
3786
3787         write_unlock_bh(&bond->lock);
3788
3789         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3790                 cancel_delayed_work(&bond->mii_work);
3791         }
3792
3793         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3794                 cancel_delayed_work(&bond->arp_work);
3795         }
3796
3797         switch (bond->params.mode) {
3798         case BOND_MODE_8023AD:
3799                 cancel_delayed_work(&bond->ad_work);
3800                 break;
3801         case BOND_MODE_TLB:
3802         case BOND_MODE_ALB:
3803                 cancel_delayed_work(&bond->alb_work);
3804                 break;
3805         default:
3806                 break;
3807         }
3808
3809
3810         if (bond_is_lb(bond)) {
3811                 /* Must be called only after all
3812                  * slaves have been released
3813                  */
3814                 bond_alb_deinitialize(bond);
3815         }
3816
3817         return 0;
3818 }
3819
3820 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3821                                                 struct rtnl_link_stats64 *stats)
3822 {
3823         struct bonding *bond = netdev_priv(bond_dev);
3824         struct rtnl_link_stats64 temp;
3825         struct slave *slave;
3826         int i;
3827
3828         memset(stats, 0, sizeof(*stats));
3829
3830         read_lock_bh(&bond->lock);
3831
3832         bond_for_each_slave(bond, slave, i) {
3833                 const struct rtnl_link_stats64 *sstats =
3834                         dev_get_stats(slave->dev, &temp);
3835
3836                 stats->rx_packets += sstats->rx_packets;
3837                 stats->rx_bytes += sstats->rx_bytes;
3838                 stats->rx_errors += sstats->rx_errors;
3839                 stats->rx_dropped += sstats->rx_dropped;
3840
3841                 stats->tx_packets += sstats->tx_packets;
3842                 stats->tx_bytes += sstats->tx_bytes;
3843                 stats->tx_errors += sstats->tx_errors;
3844                 stats->tx_dropped += sstats->tx_dropped;
3845
3846                 stats->multicast += sstats->multicast;
3847                 stats->collisions += sstats->collisions;
3848
3849                 stats->rx_length_errors += sstats->rx_length_errors;
3850                 stats->rx_over_errors += sstats->rx_over_errors;
3851                 stats->rx_crc_errors += sstats->rx_crc_errors;
3852                 stats->rx_frame_errors += sstats->rx_frame_errors;
3853                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3854                 stats->rx_missed_errors += sstats->rx_missed_errors;
3855
3856                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3857                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3858                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3859                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3860                 stats->tx_window_errors += sstats->tx_window_errors;
3861         }
3862
3863         read_unlock_bh(&bond->lock);
3864
3865         return stats;
3866 }
3867
3868 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3869 {
3870         struct net_device *slave_dev = NULL;
3871         struct ifbond k_binfo;
3872         struct ifbond __user *u_binfo = NULL;
3873         struct ifslave k_sinfo;
3874         struct ifslave __user *u_sinfo = NULL;
3875         struct mii_ioctl_data *mii = NULL;
3876         int res = 0;
3877
3878         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3879
3880         switch (cmd) {
3881         case SIOCGMIIPHY:
3882                 mii = if_mii(ifr);
3883                 if (!mii)
3884                         return -EINVAL;
3885
3886                 mii->phy_id = 0;
3887                 /* Fall Through */
3888         case SIOCGMIIREG:
3889                 /*
3890                  * We do this again just in case we were called by SIOCGMIIREG
3891                  * instead of SIOCGMIIPHY.
3892                  */
3893                 mii = if_mii(ifr);
3894                 if (!mii)
3895                         return -EINVAL;
3896
3897
3898                 if (mii->reg_num == 1) {
3899                         struct bonding *bond = netdev_priv(bond_dev);
3900                         mii->val_out = 0;
3901                         read_lock(&bond->lock);
3902                         read_lock(&bond->curr_slave_lock);
3903                         if (netif_carrier_ok(bond->dev))
3904                                 mii->val_out = BMSR_LSTATUS;
3905
3906                         read_unlock(&bond->curr_slave_lock);
3907                         read_unlock(&bond->lock);
3908                 }
3909
3910                 return 0;
3911         case BOND_INFO_QUERY_OLD:
3912         case SIOCBONDINFOQUERY:
3913                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3914
3915                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3916                         return -EFAULT;
3917
3918                 res = bond_info_query(bond_dev, &k_binfo);
3919                 if (res == 0 &&
3920                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3921                         return -EFAULT;
3922
3923                 return res;
3924         case BOND_SLAVE_INFO_QUERY_OLD:
3925         case SIOCBONDSLAVEINFOQUERY:
3926                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3927
3928                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3929                         return -EFAULT;
3930
3931                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3932                 if (res == 0 &&
3933                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3934                         return -EFAULT;
3935
3936                 return res;
3937         default:
3938                 /* Go on */
3939                 break;
3940         }
3941
3942         if (!capable(CAP_NET_ADMIN))
3943                 return -EPERM;
3944
3945         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3946
3947         pr_debug("slave_dev=%p:\n", slave_dev);
3948
3949         if (!slave_dev)
3950                 res = -ENODEV;
3951         else {
3952                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3953                 switch (cmd) {
3954                 case BOND_ENSLAVE_OLD:
3955                 case SIOCBONDENSLAVE:
3956                         res = bond_enslave(bond_dev, slave_dev);
3957                         break;
3958                 case BOND_RELEASE_OLD:
3959                 case SIOCBONDRELEASE:
3960                         res = bond_release(bond_dev, slave_dev);
3961                         break;
3962                 case BOND_SETHWADDR_OLD:
3963                 case SIOCBONDSETHWADDR:
3964                         res = bond_sethwaddr(bond_dev, slave_dev);
3965                         break;
3966                 case BOND_CHANGE_ACTIVE_OLD:
3967                 case SIOCBONDCHANGEACTIVE:
3968                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3969                         break;
3970                 default:
3971                         res = -EOPNOTSUPP;
3972                 }
3973
3974                 dev_put(slave_dev);
3975         }
3976
3977         return res;
3978 }
3979
3980 static bool bond_addr_in_mc_list(unsigned char *addr,
3981                                  struct netdev_hw_addr_list *list,
3982                                  int addrlen)
3983 {
3984         struct netdev_hw_addr *ha;
3985
3986         netdev_hw_addr_list_for_each(ha, list)
3987                 if (!memcmp(ha->addr, addr, addrlen))
3988                         return true;
3989
3990         return false;
3991 }
3992
3993 static void bond_set_multicast_list(struct net_device *bond_dev)
3994 {
3995         struct bonding *bond = netdev_priv(bond_dev);
3996         struct netdev_hw_addr *ha;
3997         bool found;
3998
3999         /*
4000          * Do promisc before checking multicast_mode
4001          */
4002         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4003                 /*
4004                  * FIXME: Need to handle the error when one of the multi-slaves
4005                  * encounters error.
4006                  */
4007                 bond_set_promiscuity(bond, 1);
4008
4009
4010         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4011                 bond_set_promiscuity(bond, -1);
4012
4013
4014         /* set allmulti flag to slaves */
4015         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4016                 /*
4017                  * FIXME: Need to handle the error when one of the multi-slaves
4018                  * encounters error.
4019                  */
4020                 bond_set_allmulti(bond, 1);
4021
4022
4023         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4024                 bond_set_allmulti(bond, -1);
4025
4026
4027         read_lock(&bond->lock);
4028
4029         bond->flags = bond_dev->flags;
4030
4031         /* looking for addresses to add to slaves' mc list */
4032         netdev_for_each_mc_addr(ha, bond_dev) {
4033                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4034                                              bond_dev->addr_len);
4035                 if (!found)
4036                         bond_mc_add(bond, ha->addr);
4037         }
4038
4039         /* looking for addresses to delete from slaves' list */
4040         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4041                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4042                                              bond_dev->addr_len);
4043                 if (!found)
4044                         bond_mc_del(bond, ha->addr);
4045         }
4046
4047         /* save master's multicast list */
4048         __hw_addr_flush(&bond->mc_list);
4049         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4050                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4051
4052         read_unlock(&bond->lock);
4053 }
4054
4055 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4056 {
4057         struct bonding *bond = netdev_priv(dev);
4058         struct slave *slave = bond->first_slave;
4059
4060         if (slave) {
4061                 const struct net_device_ops *slave_ops
4062                         = slave->dev->netdev_ops;
4063                 if (slave_ops->ndo_neigh_setup)
4064                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4065         }
4066         return 0;
4067 }
4068
4069 /*
4070  * Change the MTU of all of a master's slaves to match the master
4071  */
4072 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4073 {
4074         struct bonding *bond = netdev_priv(bond_dev);
4075         struct slave *slave, *stop_at;
4076         int res = 0;
4077         int i;
4078
4079         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4080                  (bond_dev ? bond_dev->name : "None"), new_mtu);
4081
4082         /* Can't hold bond->lock with bh disabled here since
4083          * some base drivers panic. On the other hand we can't
4084          * hold bond->lock without bh disabled because we'll
4085          * deadlock. The only solution is to rely on the fact
4086          * that we're under rtnl_lock here, and the slaves
4087          * list won't change. This doesn't solve the problem
4088          * of setting the slave's MTU while it is
4089          * transmitting, but the assumption is that the base
4090          * driver can handle that.
4091          *
4092          * TODO: figure out a way to safely iterate the slaves
4093          * list, but without holding a lock around the actual
4094          * call to the base driver.
4095          */
4096
4097         bond_for_each_slave(bond, slave, i) {
4098                 pr_debug("s %p s->p %p c_m %p\n",
4099                          slave,
4100                          slave->prev,
4101                          slave->dev->netdev_ops->ndo_change_mtu);
4102
4103                 res = dev_set_mtu(slave->dev, new_mtu);
4104
4105                 if (res) {
4106                         /* If we failed to set the slave's mtu to the new value
4107                          * we must abort the operation even in ACTIVE_BACKUP
4108                          * mode, because if we allow the backup slaves to have
4109                          * different mtu values than the active slave we'll
4110                          * need to change their mtu when doing a failover. That
4111                          * means changing their mtu from timer context, which
4112                          * is probably not a good idea.
4113                          */
4114                         pr_debug("err %d %s\n", res, slave->dev->name);
4115                         goto unwind;
4116                 }
4117         }
4118
4119         bond_dev->mtu = new_mtu;
4120
4121         return 0;
4122
4123 unwind:
4124         /* unwind from head to the slave that failed */
4125         stop_at = slave;
4126         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4127                 int tmp_res;
4128
4129                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4130                 if (tmp_res) {
4131                         pr_debug("unwind err %d dev %s\n",
4132                                  tmp_res, slave->dev->name);
4133                 }
4134         }
4135
4136         return res;
4137 }
4138
4139 /*
4140  * Change HW address
4141  *
4142  * Note that many devices must be down to change the HW address, and
4143  * downing the master releases all slaves.  We can make bonds full of
4144  * bonding devices to test this, however.
4145  */
4146 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4147 {
4148         struct bonding *bond = netdev_priv(bond_dev);
4149         struct sockaddr *sa = addr, tmp_sa;
4150         struct slave *slave, *stop_at;
4151         int res = 0;
4152         int i;
4153
4154         if (bond->params.mode == BOND_MODE_ALB)
4155                 return bond_alb_set_mac_address(bond_dev, addr);
4156
4157
4158         pr_debug("bond=%p, name=%s\n",
4159                  bond, bond_dev ? bond_dev->name : "None");
4160
4161         /*
4162          * If fail_over_mac is set to active, do nothing and return
4163          * success.  Returning an error causes ifenslave to fail.
4164          */
4165         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4166                 return 0;
4167
4168         if (!is_valid_ether_addr(sa->sa_data))
4169                 return -EADDRNOTAVAIL;
4170
4171         /* Can't hold bond->lock with bh disabled here since
4172          * some base drivers panic. On the other hand we can't
4173          * hold bond->lock without bh disabled because we'll
4174          * deadlock. The only solution is to rely on the fact
4175          * that we're under rtnl_lock here, and the slaves
4176          * list won't change. This doesn't solve the problem
4177          * of setting the slave's hw address while it is
4178          * transmitting, but the assumption is that the base
4179          * driver can handle that.
4180          *
4181          * TODO: figure out a way to safely iterate the slaves
4182          * list, but without holding a lock around the actual
4183          * call to the base driver.
4184          */
4185
4186         bond_for_each_slave(bond, slave, i) {
4187                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4188                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4189
4190                 if (slave_ops->ndo_set_mac_address == NULL) {
4191                         res = -EOPNOTSUPP;
4192                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4193                         goto unwind;
4194                 }
4195
4196                 res = dev_set_mac_address(slave->dev, addr);
4197                 if (res) {
4198                         /* TODO: consider downing the slave
4199                          * and retry ?
4200                          * User should expect communications
4201                          * breakage anyway until ARP finish
4202                          * updating, so...
4203                          */
4204                         pr_debug("err %d %s\n", res, slave->dev->name);
4205                         goto unwind;
4206                 }
4207         }
4208
4209         /* success */
4210         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4211         return 0;
4212
4213 unwind:
4214         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4215         tmp_sa.sa_family = bond_dev->type;
4216
4217         /* unwind from head to the slave that failed */
4218         stop_at = slave;
4219         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4220                 int tmp_res;
4221
4222                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4223                 if (tmp_res) {
4224                         pr_debug("unwind err %d dev %s\n",
4225                                  tmp_res, slave->dev->name);
4226                 }
4227         }
4228
4229         return res;
4230 }
4231
4232 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4233 {
4234         struct bonding *bond = netdev_priv(bond_dev);
4235         struct slave *slave, *start_at;
4236         int i, slave_no, res = 1;
4237         struct iphdr *iph = ip_hdr(skb);
4238
4239         read_lock(&bond->lock);
4240
4241         if (!BOND_IS_OK(bond))
4242                 goto out;
4243         /*
4244          * Start with the curr_active_slave that joined the bond as the
4245          * default for sending IGMP traffic.  For failover purposes one
4246          * needs to maintain some consistency for the interface that will
4247          * send the join/membership reports.  The curr_active_slave found
4248          * will send all of this type of traffic.
4249          */
4250         if ((iph->protocol == IPPROTO_IGMP) &&
4251             (skb->protocol == htons(ETH_P_IP))) {
4252
4253                 read_lock(&bond->curr_slave_lock);
4254                 slave = bond->curr_active_slave;
4255                 read_unlock(&bond->curr_slave_lock);
4256
4257                 if (!slave)
4258                         goto out;
4259         } else {
4260                 /*
4261                  * Concurrent TX may collide on rr_tx_counter; we accept
4262                  * that as being rare enough not to justify using an
4263                  * atomic op here.
4264                  */
4265                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4266
4267                 bond_for_each_slave(bond, slave, i) {
4268                         slave_no--;
4269                         if (slave_no < 0)
4270                                 break;
4271                 }
4272         }
4273
4274         start_at = slave;
4275         bond_for_each_slave_from(bond, slave, i, start_at) {
4276                 if (IS_UP(slave->dev) &&
4277                     (slave->link == BOND_LINK_UP) &&
4278                     (slave->state == BOND_STATE_ACTIVE)) {
4279                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4280                         break;
4281                 }
4282         }
4283
4284 out:
4285         if (res) {
4286                 /* no suitable interface, frame not sent */
4287                 dev_kfree_skb(skb);
4288         }
4289         read_unlock(&bond->lock);
4290         return NETDEV_TX_OK;
4291 }
4292
4293
4294 /*
4295  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4296  * the bond has a usable interface.
4297  */
4298 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4299 {
4300         struct bonding *bond = netdev_priv(bond_dev);
4301         int res = 1;
4302
4303         read_lock(&bond->lock);
4304         read_lock(&bond->curr_slave_lock);
4305
4306         if (!BOND_IS_OK(bond))
4307                 goto out;
4308
4309         if (!bond->curr_active_slave)
4310                 goto out;
4311
4312         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4313
4314 out:
4315         if (res)
4316                 /* no suitable interface, frame not sent */
4317                 dev_kfree_skb(skb);
4318
4319         read_unlock(&bond->curr_slave_lock);
4320         read_unlock(&bond->lock);
4321         return NETDEV_TX_OK;
4322 }
4323
4324 /*
4325  * In bond_xmit_xor() , we determine the output device by using a pre-
4326  * determined xmit_hash_policy(), If the selected device is not enabled,
4327  * find the next active slave.
4328  */
4329 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4330 {
4331         struct bonding *bond = netdev_priv(bond_dev);
4332         struct slave *slave, *start_at;
4333         int slave_no;
4334         int i;
4335         int res = 1;
4336
4337         read_lock(&bond->lock);
4338
4339         if (!BOND_IS_OK(bond))
4340                 goto out;
4341
4342         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4343
4344         bond_for_each_slave(bond, slave, i) {
4345                 slave_no--;
4346                 if (slave_no < 0)
4347                         break;
4348         }
4349
4350         start_at = slave;
4351
4352         bond_for_each_slave_from(bond, slave, i, start_at) {
4353                 if (IS_UP(slave->dev) &&
4354                     (slave->link == BOND_LINK_UP) &&
4355                     (slave->state == BOND_STATE_ACTIVE)) {
4356                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4357                         break;
4358                 }
4359         }
4360
4361 out:
4362         if (res) {
4363                 /* no suitable interface, frame not sent */
4364                 dev_kfree_skb(skb);
4365         }
4366         read_unlock(&bond->lock);
4367         return NETDEV_TX_OK;
4368 }
4369
4370 /*
4371  * in broadcast mode, we send everything to all usable interfaces.
4372  */
4373 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4374 {
4375         struct bonding *bond = netdev_priv(bond_dev);
4376         struct slave *slave, *start_at;
4377         struct net_device *tx_dev = NULL;
4378         int i;
4379         int res = 1;
4380
4381         read_lock(&bond->lock);
4382
4383         if (!BOND_IS_OK(bond))
4384                 goto out;
4385
4386         read_lock(&bond->curr_slave_lock);
4387         start_at = bond->curr_active_slave;
4388         read_unlock(&bond->curr_slave_lock);
4389
4390         if (!start_at)
4391                 goto out;
4392
4393         bond_for_each_slave_from(bond, slave, i, start_at) {
4394                 if (IS_UP(slave->dev) &&
4395                     (slave->link == BOND_LINK_UP) &&
4396                     (slave->state == BOND_STATE_ACTIVE)) {
4397                         if (tx_dev) {
4398                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4399                                 if (!skb2) {
4400                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4401                                                bond_dev->name);
4402                                         continue;
4403                                 }
4404
4405                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4406                                 if (res) {
4407                                         dev_kfree_skb(skb2);
4408                                         continue;
4409                                 }
4410                         }
4411                         tx_dev = slave->dev;
4412                 }
4413         }
4414
4415         if (tx_dev)
4416                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4417
4418 out:
4419         if (res)
4420                 /* no suitable interface, frame not sent */
4421                 dev_kfree_skb(skb);
4422
4423         /* frame sent to all suitable interfaces */
4424         read_unlock(&bond->lock);
4425         return NETDEV_TX_OK;
4426 }
4427
4428 /*------------------------- Device initialization ---------------------------*/
4429
4430 static void bond_set_xmit_hash_policy(struct bonding *bond)
4431 {
4432         switch (bond->params.xmit_policy) {
4433         case BOND_XMIT_POLICY_LAYER23:
4434                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4435                 break;
4436         case BOND_XMIT_POLICY_LAYER34:
4437                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4438                 break;
4439         case BOND_XMIT_POLICY_LAYER2:
4440         default:
4441                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4442                 break;
4443         }
4444 }
4445
4446 /*
4447  * Lookup the slave that corresponds to a qid
4448  */
4449 static inline int bond_slave_override(struct bonding *bond,
4450                                       struct sk_buff *skb)
4451 {
4452         int i, res = 1;
4453         struct slave *slave = NULL;
4454         struct slave *check_slave;
4455
4456         read_lock(&bond->lock);
4457
4458         if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4459                 goto out;
4460
4461         /* Find out if any slaves have the same mapping as this skb. */
4462         bond_for_each_slave(bond, check_slave, i) {
4463                 if (check_slave->queue_id == skb->queue_mapping) {
4464                         slave = check_slave;
4465                         break;
4466                 }
4467         }
4468
4469         /* If the slave isn't UP, use default transmit policy. */
4470         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4471             (slave->link == BOND_LINK_UP)) {
4472                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4473         }
4474
4475 out:
4476         read_unlock(&bond->lock);
4477         return res;
4478 }
4479
4480 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4481 {
4482         /*
4483          * This helper function exists to help dev_pick_tx get the correct
4484          * destination queue.  Using a helper function skips the a call to
4485          * skb_tx_hash and will put the skbs in the queue we expect on their
4486          * way down to the bonding driver.
4487          */
4488         return skb->queue_mapping;
4489 }
4490
4491 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4492 {
4493         struct bonding *bond = netdev_priv(dev);
4494
4495         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4496                 if (!bond_slave_override(bond, skb))
4497                         return NETDEV_TX_OK;
4498         }
4499
4500         switch (bond->params.mode) {
4501         case BOND_MODE_ROUNDROBIN:
4502                 return bond_xmit_roundrobin(skb, dev);
4503         case BOND_MODE_ACTIVEBACKUP:
4504                 return bond_xmit_activebackup(skb, dev);
4505         case BOND_MODE_XOR:
4506                 return bond_xmit_xor(skb, dev);
4507         case BOND_MODE_BROADCAST:
4508                 return bond_xmit_broadcast(skb, dev);
4509         case BOND_MODE_8023AD:
4510                 return bond_3ad_xmit_xor(skb, dev);
4511         case BOND_MODE_ALB:
4512         case BOND_MODE_TLB:
4513                 return bond_alb_xmit(skb, dev);
4514         default:
4515                 /* Should never happen, mode already checked */
4516                 pr_err("%s: Error: Unknown bonding mode %d\n",
4517                        dev->name, bond->params.mode);
4518                 WARN_ON_ONCE(1);
4519                 dev_kfree_skb(skb);
4520                 return NETDEV_TX_OK;
4521         }
4522 }
4523
4524
4525 /*
4526  * set bond mode specific net device operations
4527  */
4528 void bond_set_mode_ops(struct bonding *bond, int mode)
4529 {
4530         struct net_device *bond_dev = bond->dev;
4531
4532         switch (mode) {
4533         case BOND_MODE_ROUNDROBIN:
4534                 break;
4535         case BOND_MODE_ACTIVEBACKUP:
4536                 break;
4537         case BOND_MODE_XOR:
4538                 bond_set_xmit_hash_policy(bond);
4539                 break;
4540         case BOND_MODE_BROADCAST:
4541                 break;
4542         case BOND_MODE_8023AD:
4543                 bond_set_master_3ad_flags(bond);
4544                 bond_set_xmit_hash_policy(bond);
4545                 break;
4546         case BOND_MODE_ALB:
4547                 bond_set_master_alb_flags(bond);
4548                 /* FALLTHRU */
4549         case BOND_MODE_TLB:
4550                 break;
4551         default:
4552                 /* Should never happen, mode already checked */
4553                 pr_err("%s: Error: Unknown bonding mode %d\n",
4554                        bond_dev->name, mode);
4555                 break;
4556         }
4557 }
4558
4559 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4560                                     struct ethtool_drvinfo *drvinfo)
4561 {
4562         strncpy(drvinfo->driver, DRV_NAME, 32);
4563         strncpy(drvinfo->version, DRV_VERSION, 32);
4564         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4565 }
4566
4567 static const struct ethtool_ops bond_ethtool_ops = {
4568         .get_drvinfo            = bond_ethtool_get_drvinfo,
4569         .get_link               = ethtool_op_get_link,
4570         .get_tx_csum            = ethtool_op_get_tx_csum,
4571         .get_sg                 = ethtool_op_get_sg,
4572         .get_tso                = ethtool_op_get_tso,
4573         .get_ufo                = ethtool_op_get_ufo,
4574         .get_flags              = ethtool_op_get_flags,
4575 };
4576
4577 static const struct net_device_ops bond_netdev_ops = {
4578         .ndo_init               = bond_init,
4579         .ndo_uninit             = bond_uninit,
4580         .ndo_open               = bond_open,
4581         .ndo_stop               = bond_close,
4582         .ndo_start_xmit         = bond_start_xmit,
4583         .ndo_select_queue       = bond_select_queue,
4584         .ndo_get_stats64        = bond_get_stats,
4585         .ndo_do_ioctl           = bond_do_ioctl,
4586         .ndo_set_multicast_list = bond_set_multicast_list,
4587         .ndo_change_mtu         = bond_change_mtu,
4588         .ndo_set_mac_address    = bond_set_mac_address,
4589         .ndo_neigh_setup        = bond_neigh_setup,
4590         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4591         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4592         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4593 #ifdef CONFIG_NET_POLL_CONTROLLER
4594         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4595         .ndo_poll_controller    = bond_poll_controller,
4596 #endif
4597 };
4598
4599 static void bond_destructor(struct net_device *bond_dev)
4600 {
4601         struct bonding *bond = netdev_priv(bond_dev);
4602         if (bond->wq)
4603                 destroy_workqueue(bond->wq);
4604         free_netdev(bond_dev);
4605 }
4606
4607 static void bond_setup(struct net_device *bond_dev)
4608 {
4609         struct bonding *bond = netdev_priv(bond_dev);
4610
4611         /* initialize rwlocks */
4612         rwlock_init(&bond->lock);
4613         rwlock_init(&bond->curr_slave_lock);
4614
4615         bond->params = bonding_defaults;
4616
4617         /* Initialize pointers */
4618         bond->dev = bond_dev;
4619         INIT_LIST_HEAD(&bond->vlan_list);
4620
4621         /* Initialize the device entry points */
4622         ether_setup(bond_dev);
4623         bond_dev->netdev_ops = &bond_netdev_ops;
4624         bond_dev->ethtool_ops = &bond_ethtool_ops;
4625         bond_set_mode_ops(bond, bond->params.mode);
4626
4627         bond_dev->destructor = bond_destructor;
4628
4629         /* Initialize the device options */
4630         bond_dev->tx_queue_len = 0;
4631         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4632         bond_dev->priv_flags |= IFF_BONDING;
4633         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4634
4635         if (bond->params.arp_interval)
4636                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4637
4638         /* At first, we block adding VLANs. That's the only way to
4639          * prevent problems that occur when adding VLANs over an
4640          * empty bond. The block will be removed once non-challenged
4641          * slaves are enslaved.
4642          */
4643         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4644
4645         /* don't acquire bond device's netif_tx_lock when
4646          * transmitting */
4647         bond_dev->features |= NETIF_F_LLTX;
4648
4649         /* By default, we declare the bond to be fully
4650          * VLAN hardware accelerated capable. Special
4651          * care is taken in the various xmit functions
4652          * when there are slaves that are not hw accel
4653          * capable
4654          */
4655         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4656                                NETIF_F_HW_VLAN_RX |
4657                                NETIF_F_HW_VLAN_FILTER);
4658
4659 }
4660
4661 static void bond_work_cancel_all(struct bonding *bond)
4662 {
4663         write_lock_bh(&bond->lock);
4664         bond->kill_timers = 1;
4665         write_unlock_bh(&bond->lock);
4666
4667         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4668                 cancel_delayed_work(&bond->mii_work);
4669
4670         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4671                 cancel_delayed_work(&bond->arp_work);
4672
4673         if (bond->params.mode == BOND_MODE_ALB &&
4674             delayed_work_pending(&bond->alb_work))
4675                 cancel_delayed_work(&bond->alb_work);
4676
4677         if (bond->params.mode == BOND_MODE_8023AD &&
4678             delayed_work_pending(&bond->ad_work))
4679                 cancel_delayed_work(&bond->ad_work);
4680 }
4681
4682 /*
4683 * Destroy a bonding device.
4684 * Must be under rtnl_lock when this function is called.
4685 */
4686 static void bond_uninit(struct net_device *bond_dev)
4687 {
4688         struct bonding *bond = netdev_priv(bond_dev);
4689         struct vlan_entry *vlan, *tmp;
4690
4691         bond_netpoll_cleanup(bond_dev);
4692
4693         /* Release the bonded slaves */
4694         bond_release_all(bond_dev);
4695
4696         list_del(&bond->bond_list);
4697
4698         bond_work_cancel_all(bond);
4699
4700         bond_remove_proc_entry(bond);
4701
4702         __hw_addr_flush(&bond->mc_list);
4703
4704         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4705                 list_del(&vlan->vlan_list);
4706                 kfree(vlan);
4707         }
4708 }
4709
4710 /*------------------------- Module initialization ---------------------------*/
4711
4712 /*
4713  * Convert string input module parms.  Accept either the
4714  * number of the mode or its string name.  A bit complicated because
4715  * some mode names are substrings of other names, and calls from sysfs
4716  * may have whitespace in the name (trailing newlines, for example).
4717  */
4718 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4719 {
4720         int modeint = -1, i, rv;
4721         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4722
4723         for (p = (char *)buf; *p; p++)
4724                 if (!(isdigit(*p) || isspace(*p)))
4725                         break;
4726
4727         if (*p)
4728                 rv = sscanf(buf, "%20s", modestr);
4729         else
4730                 rv = sscanf(buf, "%d", &modeint);
4731
4732         if (!rv)
4733                 return -1;
4734
4735         for (i = 0; tbl[i].modename; i++) {
4736                 if (modeint == tbl[i].mode)
4737                         return tbl[i].mode;
4738                 if (strcmp(modestr, tbl[i].modename) == 0)
4739                         return tbl[i].mode;
4740         }
4741
4742         return -1;
4743 }
4744
4745 static int bond_check_params(struct bond_params *params)
4746 {
4747         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4748
4749         /*
4750          * Convert string parameters.
4751          */
4752         if (mode) {
4753                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4754                 if (bond_mode == -1) {
4755                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4756                                mode == NULL ? "NULL" : mode);
4757                         return -EINVAL;
4758                 }
4759         }
4760
4761         if (xmit_hash_policy) {
4762                 if ((bond_mode != BOND_MODE_XOR) &&
4763                     (bond_mode != BOND_MODE_8023AD)) {
4764                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4765                                bond_mode_name(bond_mode));
4766                 } else {
4767                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4768                                                         xmit_hashtype_tbl);
4769                         if (xmit_hashtype == -1) {
4770                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4771                                        xmit_hash_policy == NULL ? "NULL" :
4772                                        xmit_hash_policy);
4773                                 return -EINVAL;
4774                         }
4775                 }
4776         }
4777
4778         if (lacp_rate) {
4779                 if (bond_mode != BOND_MODE_8023AD) {
4780                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4781                                 bond_mode_name(bond_mode));
4782                 } else {
4783                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4784                         if (lacp_fast == -1) {
4785                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4786                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4787                                 return -EINVAL;
4788                         }
4789                 }
4790         }
4791
4792         if (ad_select) {
4793                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4794                 if (params->ad_select == -1) {
4795                         pr_err("Error: Invalid ad_select \"%s\"\n",
4796                                ad_select == NULL ? "NULL" : ad_select);
4797                         return -EINVAL;
4798                 }
4799
4800                 if (bond_mode != BOND_MODE_8023AD) {
4801                         pr_warning("ad_select param only affects 802.3ad mode\n");
4802                 }
4803         } else {
4804                 params->ad_select = BOND_AD_STABLE;
4805         }
4806
4807         if (max_bonds < 0) {
4808                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4809                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4810                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4811         }
4812
4813         if (miimon < 0) {
4814                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4815                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4816                 miimon = BOND_LINK_MON_INTERV;
4817         }
4818
4819         if (updelay < 0) {
4820                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4821                            updelay, INT_MAX);
4822                 updelay = 0;
4823         }
4824
4825         if (downdelay < 0) {
4826                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4827                            downdelay, INT_MAX);
4828                 downdelay = 0;
4829         }
4830
4831         if ((use_carrier != 0) && (use_carrier != 1)) {
4832                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4833                            use_carrier);
4834                 use_carrier = 1;
4835         }
4836
4837         if (num_grat_arp < 0 || num_grat_arp > 255) {
4838                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4839                            num_grat_arp);
4840                 num_grat_arp = 1;
4841         }
4842
4843         if (num_unsol_na < 0 || num_unsol_na > 255) {
4844                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4845                            num_unsol_na);
4846                 num_unsol_na = 1;
4847         }
4848
4849         /* reset values for 802.3ad */
4850         if (bond_mode == BOND_MODE_8023AD) {
4851                 if (!miimon) {
4852                         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");
4853                         pr_warning("Forcing miimon to 100msec\n");
4854                         miimon = 100;
4855                 }
4856         }
4857
4858         if (tx_queues < 1 || tx_queues > 255) {
4859                 pr_warning("Warning: tx_queues (%d) should be between "
4860                            "1 and 255, resetting to %d\n",
4861                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4862                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4863         }
4864
4865         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4866                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4867                            "not of valid value (0/1), so it was set to "
4868                            "0\n", all_slaves_active);
4869                 all_slaves_active = 0;
4870         }
4871
4872         /* reset values for TLB/ALB */
4873         if ((bond_mode == BOND_MODE_TLB) ||
4874             (bond_mode == BOND_MODE_ALB)) {
4875                 if (!miimon) {
4876                         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");
4877                         pr_warning("Forcing miimon to 100msec\n");
4878                         miimon = 100;
4879                 }
4880         }
4881
4882         if (bond_mode == BOND_MODE_ALB) {
4883                 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",
4884                           updelay);
4885         }
4886
4887         if (!miimon) {
4888                 if (updelay || downdelay) {
4889                         /* just warn the user the up/down delay will have
4890                          * no effect since miimon is zero...
4891                          */
4892                         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",
4893                                    updelay, downdelay);
4894                 }
4895         } else {
4896                 /* don't allow arp monitoring */
4897                 if (arp_interval) {
4898                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4899                                    miimon, arp_interval);
4900                         arp_interval = 0;
4901                 }
4902
4903                 if ((updelay % miimon) != 0) {
4904                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4905                                    updelay, miimon,
4906                                    (updelay / miimon) * miimon);
4907                 }
4908
4909                 updelay /= miimon;
4910
4911                 if ((downdelay % miimon) != 0) {
4912                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4913                                    downdelay, miimon,
4914                                    (downdelay / miimon) * miimon);
4915                 }
4916
4917                 downdelay /= miimon;
4918         }
4919
4920         if (arp_interval < 0) {
4921                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4922                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4923                 arp_interval = BOND_LINK_ARP_INTERV;
4924         }
4925
4926         for (arp_ip_count = 0;
4927              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4928              arp_ip_count++) {
4929                 /* not complete check, but should be good enough to
4930                    catch mistakes */
4931                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4932                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4933                                    arp_ip_target[arp_ip_count]);
4934                         arp_interval = 0;
4935                 } else {
4936                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4937                         arp_target[arp_ip_count] = ip;
4938                 }
4939         }
4940
4941         if (arp_interval && !arp_ip_count) {
4942                 /* don't allow arping if no arp_ip_target given... */
4943                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4944                            arp_interval);
4945                 arp_interval = 0;
4946         }
4947
4948         if (arp_validate) {
4949                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4950                         pr_err("arp_validate only supported in active-backup mode\n");
4951                         return -EINVAL;
4952                 }
4953                 if (!arp_interval) {
4954                         pr_err("arp_validate requires arp_interval\n");
4955                         return -EINVAL;
4956                 }
4957
4958                 arp_validate_value = bond_parse_parm(arp_validate,
4959                                                      arp_validate_tbl);
4960                 if (arp_validate_value == -1) {
4961                         pr_err("Error: invalid arp_validate \"%s\"\n",
4962                                arp_validate == NULL ? "NULL" : arp_validate);
4963                         return -EINVAL;
4964                 }
4965         } else
4966                 arp_validate_value = 0;
4967
4968         if (miimon) {
4969                 pr_info("MII link monitoring set to %d ms\n", miimon);
4970         } else if (arp_interval) {
4971                 int i;
4972
4973                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4974                         arp_interval,
4975                         arp_validate_tbl[arp_validate_value].modename,
4976                         arp_ip_count);
4977
4978                 for (i = 0; i < arp_ip_count; i++)
4979                         pr_info(" %s", arp_ip_target[i]);
4980
4981                 pr_info("\n");
4982
4983         } else if (max_bonds) {
4984                 /* miimon and arp_interval not set, we need one so things
4985                  * work as expected, see bonding.txt for details
4986                  */
4987                 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");
4988         }
4989
4990         if (primary && !USES_PRIMARY(bond_mode)) {
4991                 /* currently, using a primary only makes sense
4992                  * in active backup, TLB or ALB modes
4993                  */
4994                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4995                            primary, bond_mode_name(bond_mode));
4996                 primary = NULL;
4997         }
4998
4999         if (primary && primary_reselect) {
5000                 primary_reselect_value = bond_parse_parm(primary_reselect,
5001                                                          pri_reselect_tbl);
5002                 if (primary_reselect_value == -1) {
5003                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5004                                primary_reselect ==
5005                                         NULL ? "NULL" : primary_reselect);
5006                         return -EINVAL;
5007                 }
5008         } else {
5009                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5010         }
5011
5012         if (fail_over_mac) {
5013                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5014                                                       fail_over_mac_tbl);
5015                 if (fail_over_mac_value == -1) {
5016                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5017                                arp_validate == NULL ? "NULL" : arp_validate);
5018                         return -EINVAL;
5019                 }
5020
5021                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5022                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5023         } else {
5024                 fail_over_mac_value = BOND_FOM_NONE;
5025         }
5026
5027         /* fill params struct with the proper values */
5028         params->mode = bond_mode;
5029         params->xmit_policy = xmit_hashtype;
5030         params->miimon = miimon;
5031         params->num_grat_arp = num_grat_arp;
5032         params->num_unsol_na = num_unsol_na;
5033         params->arp_interval = arp_interval;
5034         params->arp_validate = arp_validate_value;
5035         params->updelay = updelay;
5036         params->downdelay = downdelay;
5037         params->use_carrier = use_carrier;
5038         params->lacp_fast = lacp_fast;
5039         params->primary[0] = 0;
5040         params->primary_reselect = primary_reselect_value;
5041         params->fail_over_mac = fail_over_mac_value;
5042         params->tx_queues = tx_queues;
5043         params->all_slaves_active = all_slaves_active;
5044
5045         if (primary) {
5046                 strncpy(params->primary, primary, IFNAMSIZ);
5047                 params->primary[IFNAMSIZ - 1] = 0;
5048         }
5049
5050         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5051
5052         return 0;
5053 }
5054
5055 static struct lock_class_key bonding_netdev_xmit_lock_key;
5056 static struct lock_class_key bonding_netdev_addr_lock_key;
5057
5058 static void bond_set_lockdep_class_one(struct net_device *dev,
5059                                        struct netdev_queue *txq,
5060                                        void *_unused)
5061 {
5062         lockdep_set_class(&txq->_xmit_lock,
5063                           &bonding_netdev_xmit_lock_key);
5064 }
5065
5066 static void bond_set_lockdep_class(struct net_device *dev)
5067 {
5068         lockdep_set_class(&dev->addr_list_lock,
5069                           &bonding_netdev_addr_lock_key);
5070         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5071 }
5072
5073 /*
5074  * Called from registration process
5075  */
5076 static int bond_init(struct net_device *bond_dev)
5077 {
5078         struct bonding *bond = netdev_priv(bond_dev);
5079         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5080
5081         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5082
5083         bond->wq = create_singlethread_workqueue(bond_dev->name);
5084         if (!bond->wq)
5085                 return -ENOMEM;
5086
5087         bond_set_lockdep_class(bond_dev);
5088
5089         netif_carrier_off(bond_dev);
5090
5091         bond_create_proc_entry(bond);
5092         list_add_tail(&bond->bond_list, &bn->dev_list);
5093
5094         bond_prepare_sysfs_group(bond);
5095
5096         __hw_addr_init(&bond->mc_list);
5097         return 0;
5098 }
5099
5100 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5101 {
5102         if (tb[IFLA_ADDRESS]) {
5103                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5104                         return -EINVAL;
5105                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5106                         return -EADDRNOTAVAIL;
5107         }
5108         return 0;
5109 }
5110
5111 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5112         .kind           = "bond",
5113         .priv_size      = sizeof(struct bonding),
5114         .setup          = bond_setup,
5115         .validate       = bond_validate,
5116 };
5117
5118 /* Create a new bond based on the specified name and bonding parameters.
5119  * If name is NULL, obtain a suitable "bond%d" name for us.
5120  * Caller must NOT hold rtnl_lock; we need to release it here before we
5121  * set up our sysfs entries.
5122  */
5123 int bond_create(struct net *net, const char *name)
5124 {
5125         struct net_device *bond_dev;
5126         int res;
5127
5128         rtnl_lock();
5129
5130         bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5131                                 bond_setup, tx_queues);
5132         if (!bond_dev) {
5133                 pr_err("%s: eek! can't alloc netdev!\n", name);
5134                 rtnl_unlock();
5135                 return -ENOMEM;
5136         }
5137
5138         dev_net_set(bond_dev, net);
5139         bond_dev->rtnl_link_ops = &bond_link_ops;
5140
5141         if (!name) {
5142                 res = dev_alloc_name(bond_dev, "bond%d");
5143                 if (res < 0)
5144                         goto out;
5145         }
5146
5147         res = register_netdevice(bond_dev);
5148
5149 out:
5150         rtnl_unlock();
5151         if (res < 0)
5152                 bond_destructor(bond_dev);
5153         return res;
5154 }
5155
5156 static int __net_init bond_net_init(struct net *net)
5157 {
5158         struct bond_net *bn = net_generic(net, bond_net_id);
5159
5160         bn->net = net;
5161         INIT_LIST_HEAD(&bn->dev_list);
5162
5163         bond_create_proc_dir(bn);
5164
5165         return 0;
5166 }
5167
5168 static void __net_exit bond_net_exit(struct net *net)
5169 {
5170         struct bond_net *bn = net_generic(net, bond_net_id);
5171
5172         bond_destroy_proc_dir(bn);
5173 }
5174
5175 static struct pernet_operations bond_net_ops = {
5176         .init = bond_net_init,
5177         .exit = bond_net_exit,
5178         .id   = &bond_net_id,
5179         .size = sizeof(struct bond_net),
5180 };
5181
5182 static int __init bonding_init(void)
5183 {
5184         int i;
5185         int res;
5186
5187         pr_info("%s", version);
5188
5189         res = bond_check_params(&bonding_defaults);
5190         if (res)
5191                 goto out;
5192
5193         res = register_pernet_subsys(&bond_net_ops);
5194         if (res)
5195                 goto out;
5196
5197         res = rtnl_link_register(&bond_link_ops);
5198         if (res)
5199                 goto err_link;
5200
5201         for (i = 0; i < max_bonds; i++) {
5202                 res = bond_create(&init_net, NULL);
5203                 if (res)
5204                         goto err;
5205         }
5206
5207         res = bond_create_sysfs();
5208         if (res)
5209                 goto err;
5210
5211         register_netdevice_notifier(&bond_netdev_notifier);
5212         register_inetaddr_notifier(&bond_inetaddr_notifier);
5213         bond_register_ipv6_notifier();
5214 out:
5215         return res;
5216 err:
5217         rtnl_link_unregister(&bond_link_ops);
5218 err_link:
5219         unregister_pernet_subsys(&bond_net_ops);
5220         goto out;
5221
5222 }
5223
5224 static void __exit bonding_exit(void)
5225 {
5226         unregister_netdevice_notifier(&bond_netdev_notifier);
5227         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5228         bond_unregister_ipv6_notifier();
5229
5230         bond_destroy_sysfs();
5231
5232         rtnl_link_unregister(&bond_link_ops);
5233         unregister_pernet_subsys(&bond_net_ops);
5234 }
5235
5236 module_init(bonding_init);
5237 module_exit(bonding_exit);
5238 MODULE_LICENSE("GPL");
5239 MODULE_VERSION(DRV_VERSION);
5240 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5241 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5242 MODULE_ALIAS_RTNL_LINK("bond");