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 num_grat_arp = 1;
  94 static int num_unsol_na = 1;
  95 static int miimon       = BOND_LINK_MON_INTERV;
  96 static int updelay;
  97 static int downdelay;
  98 static int use_carrier  = 1;
  99 static char *mode;
 100 static char *primary;
 101 static char *primary_reselect;
 102 static char *lacp_rate;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval = BOND_LINK_ARP_INTERV;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *fail_over_mac;
 109 static struct bond_params bonding_defaults;
 110
 111 module_param(max_bonds, int, 0);
 112 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 113 module_param(num_grat_arp, int, 0644);
 114 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
 115 module_param(num_unsol_na, int, 0644);
 116 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
 117 module_param(miimon, int, 0);
 118 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 119 module_param(updelay, int, 0);
 120 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 121 module_param(downdelay, int, 0);
 122 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 123                             "in milliseconds");
 124 module_param(use_carrier, int, 0);
 125 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 126                               "0 for off, 1 for on (default)");
 127 module_param(mode, charp, 0);
 128 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
 129                        "1 for active-backup, 2 for balance-xor, "
 130                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 131                        "6 for balance-alb");
 132 module_param(primary, charp, 0);
 133 MODULE_PARM_DESC(primary, "Primary network device to use");
 134 module_param(primary_reselect, charp, 0);
 135 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 136                                    "once it comes up; "
 137                                    "0 for always (default), "
 138                                    "1 for only if speed of primary is "
 139                                    "better, "
 140                                    "2 for only on active slave "
 141                                    "failure");
 142 module_param(lacp_rate, charp, 0);
 143 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
 144                             "(slow/fast)");
 145 module_param(ad_select, charp, 0);
 146 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
 147 module_param(xmit_hash_policy, charp, 0);
 148 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
 149                                    ", 1 for layer 3+4");
 150 module_param(arp_interval, int, 0);
 151 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 152 module_param_array(arp_ip_target, charp, NULL, 0);
 153 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 154 module_param(arp_validate, charp, 0);
 155 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
 156 module_param(fail_over_mac, charp, 0);
 157 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
 158
 159 /*----------------------------- Global variables ----------------------------*/
 160
 161 static const char * const version =
 162         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
 163
 164 int bond_net_id __read_mostly;
 165
 166 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 167 static int arp_ip_count;
 168 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 169 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 170 static int lacp_fast;
 171
 172
 173 const struct bond_parm_tbl bond_lacp_tbl[] = {
 174 {       "slow",         AD_LACP_SLOW},
 175 {       "fast",         AD_LACP_FAST},
 176 {       NULL,           -1},
 177 };
 178
 179 const struct bond_parm_tbl bond_mode_tbl[] = {
 180 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 181 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 182 {       "balance-xor",          BOND_MODE_XOR},
 183 {       "broadcast",            BOND_MODE_BROADCAST},
 184 {       "802.3ad",              BOND_MODE_8023AD},
 185 {       "balance-tlb",          BOND_MODE_TLB},
 186 {       "balance-alb",          BOND_MODE_ALB},
 187 {       NULL,                   -1},
 188 };
 189
 190 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 191 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 192 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 193 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 194 {       NULL,                   -1},
 195 };
 196
 197 const struct bond_parm_tbl arp_validate_tbl[] = {
 198 {       "none",                 BOND_ARP_VALIDATE_NONE},
 199 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 200 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 201 {       "all",                  BOND_ARP_VALIDATE_ALL},
 202 {       NULL,                   -1},
 203 };
 204
 205 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 206 {       "none",                 BOND_FOM_NONE},
 207 {       "active",               BOND_FOM_ACTIVE},
 208 {       "follow",               BOND_FOM_FOLLOW},
 209 {       NULL,                   -1},
 210 };
 211
 212 const struct bond_parm_tbl pri_reselect_tbl[] = {
 213 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 214 {       "better",               BOND_PRI_RESELECT_BETTER},
 215 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 216 {       NULL,                   -1},
 217 };
 218
 219 struct bond_parm_tbl ad_select_tbl[] = {
 220 {       "stable",       BOND_AD_STABLE},
 221 {       "bandwidth",    BOND_AD_BANDWIDTH},
 222 {       "count",        BOND_AD_COUNT},
 223 {       NULL,           -1},
 224 };
 225
 226 /*-------------------------- Forward declarations ---------------------------*/
 227
 228 static void bond_send_gratuitous_arp(struct bonding *bond);
 229 static int bond_init(struct net_device *bond_dev);
 230 static void bond_uninit(struct net_device *bond_dev);
 231
 232 /*---------------------------- General routines -----------------------------*/
 233
 234 static const char *bond_mode_name(int mode)
 235 {
 236         static const char *names[] = {
 237                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 238                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 239                 [BOND_MODE_XOR] = "load balancing (xor)",
 240                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 241                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 242                 [BOND_MODE_TLB] = "transmit load balancing",
 243                 [BOND_MODE_ALB] = "adaptive load balancing",
 244         };
 245
 246         if (mode < 0 || mode > BOND_MODE_ALB)
 247                 return "unknown";
 248
 249         return names[mode];
 250 }
 251
 252 /*---------------------------------- VLAN -----------------------------------*/
 253
 254 /**
 255  * bond_add_vlan - add a new vlan id on bond
 256  * @bond: bond that got the notification
 257  * @vlan_id: the vlan id to add
 258  *
 259  * Returns -ENOMEM if allocation failed.
 260  */
 261 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 262 {
 263         struct vlan_entry *vlan;
 264
 265         pr_debug("bond: %s, vlan id %d\n",
 266                  (bond ? bond->dev->name : "None"), vlan_id);
 267
 268         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 269         if (!vlan)
 270                 return -ENOMEM;
 271
 272         INIT_LIST_HEAD(&vlan->vlan_list);
 273         vlan->vlan_id = vlan_id;
 274
 275         write_lock_bh(&bond->lock);
 276
 277         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 278
 279         write_unlock_bh(&bond->lock);
 280
 281         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 282
 283         return 0;
 284 }
 285
 286 /**
 287  * bond_del_vlan - delete a vlan id from bond
 288  * @bond: bond that got the notification
 289  * @vlan_id: the vlan id to delete
 290  *
 291  * returns -ENODEV if @vlan_id was not found in @bond.
 292  */
 293 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 294 {
 295         struct vlan_entry *vlan;
 296         int res = -ENODEV;
 297
 298         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 299
 300         write_lock_bh(&bond->lock);
 301
 302         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 303                 if (vlan->vlan_id == vlan_id) {
 304                         list_del(&vlan->vlan_list);
 305
 306                         if (bond_is_lb(bond))
 307                                 bond_alb_clear_vlan(bond, vlan_id);
 308
 309                         pr_debug("removed VLAN ID %d from bond %s\n",
 310                                  vlan_id, bond->dev->name);
 311
 312                         kfree(vlan);
 313
 314                         if (list_empty(&bond->vlan_list) &&
 315                             (bond->slave_cnt == 0)) {
 316                                 /* Last VLAN removed and no slaves, so
 317                                  * restore block on adding VLANs. This will
 318                                  * be removed once new slaves that are not
 319                                  * VLAN challenged will be added.
 320                                  */
 321                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
 322                         }
 323
 324                         res = 0;
 325                         goto out;
 326                 }
 327         }
 328
 329         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 330                  vlan_id, bond->dev->name);
 331
 332 out:
 333         write_unlock_bh(&bond->lock);
 334         return res;
 335 }
 336
 337 /**
 338  * bond_has_challenged_slaves
 339  * @bond: the bond we're working on
 340  *
 341  * Searches the slave list. Returns 1 if a vlan challenged slave
 342  * was found, 0 otherwise.
 343  *
 344  * Assumes bond->lock is held.
 345  */
 346 static int bond_has_challenged_slaves(struct bonding *bond)
 347 {
 348         struct slave *slave;
 349         int i;
 350
 351         bond_for_each_slave(bond, slave, i) {
 352                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
 353                         pr_debug("found VLAN challenged slave - %s\n",
 354                                  slave->dev->name);
 355                         return 1;
 356                 }
 357         }
 358
 359         pr_debug("no VLAN challenged slaves found\n");
 360         return 0;
 361 }
 362
 363 /**
 364  * bond_next_vlan - safely skip to the next item in the vlans list.
 365  * @bond: the bond we're working on
 366  * @curr: item we're advancing from
 367  *
 368  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 369  * or @curr->next otherwise (even if it is @curr itself again).
 370  *
 371  * Caller must hold bond->lock
 372  */
 373 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 374 {
 375         struct vlan_entry *next, *last;
 376
 377         if (list_empty(&bond->vlan_list))
 378                 return NULL;
 379
 380         if (!curr) {
 381                 next = list_entry(bond->vlan_list.next,
 382                                   struct vlan_entry, vlan_list);
 383         } else {
 384                 last = list_entry(bond->vlan_list.prev,
 385                                   struct vlan_entry, vlan_list);
 386                 if (last == curr) {
 387                         next = list_entry(bond->vlan_list.next,
 388                                           struct vlan_entry, vlan_list);
 389                 } else {
 390                         next = list_entry(curr->vlan_list.next,
 391                                           struct vlan_entry, vlan_list);
 392                 }
 393         }
 394
 395         return next;
 396 }
 397
 398 /**
 399  * bond_dev_queue_xmit - Prepare skb for xmit.
 400  *
 401  * @bond: bond device that got this skb for tx.
 402  * @skb: hw accel VLAN tagged skb to transmit
 403  * @slave_dev: slave that is supposed to xmit this skbuff
 404  *
 405  * When the bond gets an skb to transmit that is
 406  * already hardware accelerated VLAN tagged, and it
 407  * needs to relay this skb to a slave that is not
 408  * hw accel capable, the skb needs to be "unaccelerated",
 409  * i.e. strip the hwaccel tag and re-insert it as part
 410  * of the payload.
 411  */
 412 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 413                         struct net_device *slave_dev)
 414 {
 415         unsigned short uninitialized_var(vlan_id);
 416
 417         if (!list_empty(&bond->vlan_list) &&
 418             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
 419             vlan_get_tag(skb, &vlan_id) == 0) {
 420                 skb->dev = slave_dev;
 421                 skb = vlan_put_tag(skb, vlan_id);
 422                 if (!skb) {
 423                         /* vlan_put_tag() frees the skb in case of error,
 424                          * so return success here so the calling functions
 425                          * won't attempt to free is again.
 426                          */
 427                         return 0;
 428                 }
 429         } else {
 430                 skb->dev = slave_dev;
 431         }
 432
 433         skb->priority = 1;
 434 #ifdef CONFIG_NET_POLL_CONTROLLER
 435         if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
 436                 struct netpoll *np = bond->dev->npinfo->netpoll;
 437                 slave_dev->npinfo = bond->dev->npinfo;
 438                 np->real_dev = np->dev = skb->dev;
 439                 slave_dev->priv_flags |= IFF_IN_NETPOLL;
 440                 netpoll_send_skb(np, skb);
 441                 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
 442                 np->dev = bond->dev;
 443         } else
 444 #endif
 445                 dev_queue_xmit(skb);
 446
 447         return 0;
 448 }
 449
 450 /*
 451  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
 452  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
 453  * lock because:
 454  * a. This operation is performed in IOCTL context,
 455  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 456  * c. Holding a lock with BH disabled while directly calling a base driver
 457  *    entry point is generally a BAD idea.
 458  *
 459  * The design of synchronization/protection for this operation in the 8021q
 460  * module is good for one or more VLAN devices over a single physical device
 461  * and cannot be extended for a teaming solution like bonding, so there is a
 462  * potential race condition here where a net device from the vlan group might
 463  * be referenced (either by a base driver or the 8021q code) while it is being
 464  * removed from the system. However, it turns out we're not making matters
 465  * worse, and if it works for regular VLAN usage it will work here too.
 466 */
 467
 468 /**
 469  * bond_vlan_rx_register - Propagates registration to slaves
 470  * @bond_dev: bonding net device that got called
 471  * @grp: vlan group being registered
 472  */
 473 static void bond_vlan_rx_register(struct net_device *bond_dev,
 474                                   struct vlan_group *grp)
 475 {
 476         struct bonding *bond = netdev_priv(bond_dev);
 477         struct slave *slave;
 478         int i;
 479
 480         bond->vlgrp = grp;
 481
 482         bond_for_each_slave(bond, slave, i) {
 483                 struct net_device *slave_dev = slave->dev;
 484                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 485
 486                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 487                     slave_ops->ndo_vlan_rx_register) {
 488                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
 489                 }
 490         }
 491 }
 492
 493 /**
 494  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 495  * @bond_dev: bonding net device that got called
 496  * @vid: vlan id being added
 497  */
 498 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 499 {
 500         struct bonding *bond = netdev_priv(bond_dev);
 501         struct slave *slave;
 502         int i, res;
 503
 504         bond_for_each_slave(bond, slave, i) {
 505                 struct net_device *slave_dev = slave->dev;
 506                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 507
 508                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 509                     slave_ops->ndo_vlan_rx_add_vid) {
 510                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 511                 }
 512         }
 513
 514         res = bond_add_vlan(bond, vid);
 515         if (res) {
 516                 pr_err("%s: Error: Failed to add vlan id %d\n",
 517                        bond_dev->name, vid);
 518         }
 519 }
 520
 521 /**
 522  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 523  * @bond_dev: bonding net device that got called
 524  * @vid: vlan id being removed
 525  */
 526 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 527 {
 528         struct bonding *bond = netdev_priv(bond_dev);
 529         struct slave *slave;
 530         struct net_device *vlan_dev;
 531         int i, res;
 532
 533         bond_for_each_slave(bond, slave, i) {
 534                 struct net_device *slave_dev = slave->dev;
 535                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 536
 537                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 538                     slave_ops->ndo_vlan_rx_kill_vid) {
 539                         /* Save and then restore vlan_dev in the grp array,
 540                          * since the slave's driver might clear it.
 541                          */
 542                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
 543                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 544                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
 545                 }
 546         }
 547
 548         res = bond_del_vlan(bond, vid);
 549         if (res) {
 550                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 551                        bond_dev->name, vid);
 552         }
 553 }
 554
 555 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 556 {
 557         struct vlan_entry *vlan;
 558         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 559
 560         write_lock_bh(&bond->lock);
 561
 562         if (list_empty(&bond->vlan_list))
 563                 goto out;
 564
 565         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 566             slave_ops->ndo_vlan_rx_register)
 567                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
 568
 569         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 570             !(slave_ops->ndo_vlan_rx_add_vid))
 571                 goto out;
 572
 573         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 574                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 575
 576 out:
 577         write_unlock_bh(&bond->lock);
 578 }
 579
 580 static void bond_del_vlans_from_slave(struct bonding *bond,
 581                                       struct net_device *slave_dev)
 582 {
 583         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 584         struct vlan_entry *vlan;
 585         struct net_device *vlan_dev;
 586
 587         write_lock_bh(&bond->lock);
 588
 589         if (list_empty(&bond->vlan_list))
 590                 goto out;
 591
 592         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 593             !(slave_ops->ndo_vlan_rx_kill_vid))
 594                 goto unreg;
 595
 596         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 597                 /* Save and then restore vlan_dev in the grp array,
 598                  * since the slave's driver might clear it.
 599                  */
 600                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
 601                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 602                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
 603         }
 604
 605 unreg:
 606         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
 607             slave_ops->ndo_vlan_rx_register)
 608                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
 609
 610 out:
 611         write_unlock_bh(&bond->lock);
 612 }
 613
 614 /*------------------------------- Link status -------------------------------*/
 615
 616 /*
 617  * Set the carrier state for the master according to the state of its
 618  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 619  * do special 802.3ad magic.
 620  *
 621  * Returns zero if carrier state does not change, nonzero if it does.
 622  */
 623 static int bond_set_carrier(struct bonding *bond)
 624 {
 625         struct slave *slave;
 626         int i;
 627
 628         if (bond->slave_cnt == 0)
 629                 goto down;
 630
 631         if (bond->params.mode == BOND_MODE_8023AD)
 632                 return bond_3ad_set_carrier(bond);
 633
 634         bond_for_each_slave(bond, slave, i) {
 635                 if (slave->link == BOND_LINK_UP) {
 636                         if (!netif_carrier_ok(bond->dev)) {
 637                                 netif_carrier_on(bond->dev);
 638                                 return 1;
 639                         }
 640                         return 0;
 641                 }
 642         }
 643
 644 down:
 645         if (netif_carrier_ok(bond->dev)) {
 646                 netif_carrier_off(bond->dev);
 647                 return 1;
 648         }
 649         return 0;
 650 }
 651
 652 /*
 653  * Get link speed and duplex from the slave's base driver
 654  * using ethtool. If for some reason the call fails or the
 655  * values are invalid, fake speed and duplex to 100/Full
 656  * and return error.
 657  */
 658 static int bond_update_speed_duplex(struct slave *slave)
 659 {
 660         struct net_device *slave_dev = slave->dev;
 661         struct ethtool_cmd etool;
 662         int res;
 663
 664         /* Fake speed and duplex */
 665         slave->speed = SPEED_100;
 666         slave->duplex = DUPLEX_FULL;
 667
 668         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 669                 return -1;
 670
 671         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 672         if (res < 0)
 673                 return -1;
 674
 675         switch (etool.speed) {
 676         case SPEED_10:
 677         case SPEED_100:
 678         case SPEED_1000:
 679         case SPEED_10000:
 680                 break;
 681         default:
 682                 return -1;
 683         }
 684
 685         switch (etool.duplex) {
 686         case DUPLEX_FULL:
 687         case DUPLEX_HALF:
 688                 break;
 689         default:
 690                 return -1;
 691         }
 692
 693         slave->speed = etool.speed;
 694         slave->duplex = etool.duplex;
 695
 696         return 0;
 697 }
 698
 699 /*
 700  * if <dev> supports MII link status reporting, check its link status.
 701  *
 702  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 703  * depending upon the setting of the use_carrier parameter.
 704  *
 705  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 706  * can't tell and just pretend it is), or 0, meaning that the link is
 707  * down.
 708  *
 709  * If reporting is non-zero, instead of faking link up, return -1 if
 710  * both ETHTOOL and MII ioctls fail (meaning the device does not
 711  * support them).  If use_carrier is set, return whatever it says.
 712  * It'd be nice if there was a good way to tell if a driver supports
 713  * netif_carrier, but there really isn't.
 714  */
 715 static int bond_check_dev_link(struct bonding *bond,
 716                                struct net_device *slave_dev, int reporting)
 717 {
 718         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 719         int (*ioctl)(struct net_device *, struct ifreq *, int);
 720         struct ifreq ifr;
 721         struct mii_ioctl_data *mii;
 722
 723         if (!reporting && !netif_running(slave_dev))
 724                 return 0;
 725
 726         if (bond->params.use_carrier)
 727                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 728
 729         /* Try to get link status using Ethtool first. */
 730         if (slave_dev->ethtool_ops) {
 731                 if (slave_dev->ethtool_ops->get_link) {
 732                         u32 link;
 733
 734                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 735
 736                         return link ? BMSR_LSTATUS : 0;
 737                 }
 738         }
 739
 740         /* Ethtool can't be used, fallback to MII ioctls. */
 741         ioctl = slave_ops->ndo_do_ioctl;
 742         if (ioctl) {
 743                 /* TODO: set pointer to correct ioctl on a per team member */
 744                 /*       bases to make this more efficient. that is, once  */
 745                 /*       we determine the correct ioctl, we will always    */
 746                 /*       call it and not the others for that team          */
 747                 /*       member.                                           */
 748
 749                 /*
 750                  * We cannot assume that SIOCGMIIPHY will also read a
 751                  * register; not all network drivers (e.g., e100)
 752                  * support that.
 753                  */
 754
 755                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 756                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 757                 mii = if_mii(&ifr);
 758                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 759                         mii->reg_num = MII_BMSR;
 760                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 761                                 return mii->val_out & BMSR_LSTATUS;
 762                 }
 763         }
 764
 765         /*
 766          * If reporting, report that either there's no dev->do_ioctl,
 767          * or both SIOCGMIIREG and get_link failed (meaning that we
 768          * cannot report link status).  If not reporting, pretend
 769          * we're ok.
 770          */
 771         return reporting ? -1 : BMSR_LSTATUS;
 772 }
 773
 774 /*----------------------------- Multicast list ------------------------------*/
 775
 776 /*
 777  * Push the promiscuity flag down to appropriate slaves
 778  */
 779 static int bond_set_promiscuity(struct bonding *bond, int inc)
 780 {
 781         int err = 0;
 782         if (USES_PRIMARY(bond->params.mode)) {
 783                 /* write lock already acquired */
 784                 if (bond->curr_active_slave) {
 785                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 786                                                   inc);
 787                 }
 788         } else {
 789                 struct slave *slave;
 790                 int i;
 791                 bond_for_each_slave(bond, slave, i) {
 792                         err = dev_set_promiscuity(slave->dev, inc);
 793                         if (err)
 794                                 return err;
 795                 }
 796         }
 797         return err;
 798 }
 799
 800 /*
 801  * Push the allmulti flag down to all slaves
 802  */
 803 static int bond_set_allmulti(struct bonding *bond, int inc)
 804 {
 805         int err = 0;
 806         if (USES_PRIMARY(bond->params.mode)) {
 807                 /* write lock already acquired */
 808                 if (bond->curr_active_slave) {
 809                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 810                                                inc);
 811                 }
 812         } else {
 813                 struct slave *slave;
 814                 int i;
 815                 bond_for_each_slave(bond, slave, i) {
 816                         err = dev_set_allmulti(slave->dev, inc);
 817                         if (err)
 818                                 return err;
 819                 }
 820         }
 821         return err;
 822 }
 823
 824 /*
 825  * Add a Multicast address to slaves
 826  * according to mode
 827  */
 828 static void bond_mc_add(struct bonding *bond, void *addr)
 829 {
 830         if (USES_PRIMARY(bond->params.mode)) {
 831                 /* write lock already acquired */
 832                 if (bond->curr_active_slave)
 833                         dev_mc_add(bond->curr_active_slave->dev, addr);
 834         } else {
 835                 struct slave *slave;
 836                 int i;
 837
 838                 bond_for_each_slave(bond, slave, i)
 839                         dev_mc_add(slave->dev, addr);
 840         }
 841 }
 842
 843 /*
 844  * Remove a multicast address from slave
 845  * according to mode
 846  */
 847 static void bond_mc_del(struct bonding *bond, void *addr)
 848 {
 849         if (USES_PRIMARY(bond->params.mode)) {
 850                 /* write lock already acquired */
 851                 if (bond->curr_active_slave)
 852                         dev_mc_del(bond->curr_active_slave->dev, addr);
 853         } else {
 854                 struct slave *slave;
 855                 int i;
 856                 bond_for_each_slave(bond, slave, i) {
 857                         dev_mc_del(slave->dev, addr);
 858                 }
 859         }
 860 }
 861
 862
 863 /*
 864  * Retrieve the list of registered multicast addresses for the bonding
 865  * device and retransmit an IGMP JOIN request to the current active
 866  * slave.
 867  */
 868 static void bond_resend_igmp_join_requests(struct bonding *bond)
 869 {
 870         struct in_device *in_dev;
 871         struct ip_mc_list *im;
 872
 873         rcu_read_lock();
 874         in_dev = __in_dev_get_rcu(bond->dev);
 875         if (in_dev) {
 876                 for (im = in_dev->mc_list; im; im = im->next)
 877                         ip_mc_rejoin_group(im);
 878         }
 879
 880         rcu_read_unlock();
 881 }
 882
 883 /*
 884  * flush all members of flush->mc_list from device dev->mc_list
 885  */
 886 static void bond_mc_list_flush(struct net_device *bond_dev,
 887                                struct net_device *slave_dev)
 888 {
 889         struct bonding *bond = netdev_priv(bond_dev);
 890         struct netdev_hw_addr *ha;
 891
 892         netdev_for_each_mc_addr(ha, bond_dev)
 893                 dev_mc_del(slave_dev, ha->addr);
 894
 895         if (bond->params.mode == BOND_MODE_8023AD) {
 896                 /* del lacpdu mc addr from mc list */
 897                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 898
 899                 dev_mc_del(slave_dev, lacpdu_multicast);
 900         }
 901 }
 902
 903 /*--------------------------- Active slave change ---------------------------*/
 904
 905 /*
 906  * Update the mc list and multicast-related flags for the new and
 907  * old active slaves (if any) according to the multicast mode, and
 908  * promiscuous flags unconditionally.
 909  */
 910 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 911                          struct slave *old_active)
 912 {
 913         struct netdev_hw_addr *ha;
 914
 915         if (!USES_PRIMARY(bond->params.mode))
 916                 /* nothing to do -  mc list is already up-to-date on
 917                  * all slaves
 918                  */
 919                 return;
 920
 921         if (old_active) {
 922                 if (bond->dev->flags & IFF_PROMISC)
 923                         dev_set_promiscuity(old_active->dev, -1);
 924
 925                 if (bond->dev->flags & IFF_ALLMULTI)
 926                         dev_set_allmulti(old_active->dev, -1);
 927
 928                 netdev_for_each_mc_addr(ha, bond->dev)
 929                         dev_mc_del(old_active->dev, ha->addr);
 930         }
 931
 932         if (new_active) {
 933                 /* FIXME: Signal errors upstream. */
 934                 if (bond->dev->flags & IFF_PROMISC)
 935                         dev_set_promiscuity(new_active->dev, 1);
 936
 937                 if (bond->dev->flags & IFF_ALLMULTI)
 938                         dev_set_allmulti(new_active->dev, 1);
 939
 940                 netdev_for_each_mc_addr(ha, bond->dev)
 941                         dev_mc_add(new_active->dev, ha->addr);
 942                 bond_resend_igmp_join_requests(bond);
 943         }
 944 }
 945
 946 /*
 947  * bond_do_fail_over_mac
 948  *
 949  * Perform special MAC address swapping for fail_over_mac settings
 950  *
 951  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 952  */
 953 static void bond_do_fail_over_mac(struct bonding *bond,
 954                                   struct slave *new_active,
 955                                   struct slave *old_active)
 956         __releases(&bond->curr_slave_lock)
 957         __releases(&bond->lock)
 958         __acquires(&bond->lock)
 959         __acquires(&bond->curr_slave_lock)
 960 {
 961         u8 tmp_mac[ETH_ALEN];
 962         struct sockaddr saddr;
 963         int rv;
 964
 965         switch (bond->params.fail_over_mac) {
 966         case BOND_FOM_ACTIVE:
 967                 if (new_active)
 968                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 969                                new_active->dev->addr_len);
 970                 break;
 971         case BOND_FOM_FOLLOW:
 972                 /*
 973                  * if new_active && old_active, swap them
 974                  * if just old_active, do nothing (going to no active slave)
 975                  * if just new_active, set new_active to bond's MAC
 976                  */
 977                 if (!new_active)
 978                         return;
 979
 980                 write_unlock_bh(&bond->curr_slave_lock);
 981                 read_unlock(&bond->lock);
 982
 983                 if (old_active) {
 984                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 985                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 986                                ETH_ALEN);
 987                         saddr.sa_family = new_active->dev->type;
 988                 } else {
 989                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 990                         saddr.sa_family = bond->dev->type;
 991                 }
 992
 993                 rv = dev_set_mac_address(new_active->dev, &saddr);
 994                 if (rv) {
 995                         pr_err("%s: Error %d setting MAC of slave %s\n",
 996                                bond->dev->name, -rv, new_active->dev->name);
 997                         goto out;
 998                 }
 999
1000                 if (!old_active)
1001                         goto out;
1002
1003                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1004                 saddr.sa_family = old_active->dev->type;
1005
1006                 rv = dev_set_mac_address(old_active->dev, &saddr);
1007                 if (rv)
1008                         pr_err("%s: Error %d setting MAC of slave %s\n",
1009                                bond->dev->name, -rv, new_active->dev->name);
1010 out:
1011                 read_lock(&bond->lock);
1012                 write_lock_bh(&bond->curr_slave_lock);
1013                 break;
1014         default:
1015                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1016                        bond->dev->name, bond->params.fail_over_mac);
1017                 break;
1018         }
1019
1020 }
1021
1022 static bool bond_should_change_active(struct bonding *bond)
1023 {
1024         struct slave *prim = bond->primary_slave;
1025         struct slave *curr = bond->curr_active_slave;
1026
1027         if (!prim || !curr || curr->link != BOND_LINK_UP)
1028                 return true;
1029         if (bond->force_primary) {
1030                 bond->force_primary = false;
1031                 return true;
1032         }
1033         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1034             (prim->speed < curr->speed ||
1035              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1036                 return false;
1037         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1038                 return false;
1039         return true;
1040 }
1041
1042 /**
1043  * find_best_interface - select the best available slave to be the active one
1044  * @bond: our bonding struct
1045  *
1046  * Warning: Caller must hold curr_slave_lock for writing.
1047  */
1048 static struct slave *bond_find_best_slave(struct bonding *bond)
1049 {
1050         struct slave *new_active, *old_active;
1051         struct slave *bestslave = NULL;
1052         int mintime = bond->params.updelay;
1053         int i;
1054
1055         new_active = bond->curr_active_slave;
1056
1057         if (!new_active) { /* there were no active slaves left */
1058                 if (bond->slave_cnt > 0)   /* found one slave */
1059                         new_active = bond->first_slave;
1060                 else
1061                         return NULL; /* still no slave, return NULL */
1062         }
1063
1064         if ((bond->primary_slave) &&
1065             bond->primary_slave->link == BOND_LINK_UP &&
1066             bond_should_change_active(bond)) {
1067                 new_active = bond->primary_slave;
1068         }
1069
1070         /* remember where to stop iterating over the slaves */
1071         old_active = new_active;
1072
1073         bond_for_each_slave_from(bond, new_active, i, old_active) {
1074                 if (new_active->link == BOND_LINK_UP) {
1075                         return new_active;
1076                 } else if (new_active->link == BOND_LINK_BACK &&
1077                            IS_UP(new_active->dev)) {
1078                         /* link up, but waiting for stabilization */
1079                         if (new_active->delay < mintime) {
1080                                 mintime = new_active->delay;
1081                                 bestslave = new_active;
1082                         }
1083                 }
1084         }
1085
1086         return bestslave;
1087 }
1088
1089 /**
1090  * change_active_interface - change the active slave into the specified one
1091  * @bond: our bonding struct
1092  * @new: the new slave to make the active one
1093  *
1094  * Set the new slave to the bond's settings and unset them on the old
1095  * curr_active_slave.
1096  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1097  *
1098  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1099  * because it is apparently the best available slave we have, even though its
1100  * updelay hasn't timed out yet.
1101  *
1102  * If new_active is not NULL, caller must hold bond->lock for read and
1103  * curr_slave_lock for write_bh.
1104  */
1105 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1106 {
1107         struct slave *old_active = bond->curr_active_slave;
1108
1109         if (old_active == new_active)
1110                 return;
1111
1112         if (new_active) {
1113                 new_active->jiffies = jiffies;
1114
1115                 if (new_active->link == BOND_LINK_BACK) {
1116                         if (USES_PRIMARY(bond->params.mode)) {
1117                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1118                                         bond->dev->name, new_active->dev->name,
1119                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1120                         }
1121
1122                         new_active->delay = 0;
1123                         new_active->link = BOND_LINK_UP;
1124
1125                         if (bond->params.mode == BOND_MODE_8023AD)
1126                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1127
1128                         if (bond_is_lb(bond))
1129                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1130                 } else {
1131                         if (USES_PRIMARY(bond->params.mode)) {
1132                                 pr_info("%s: making interface %s the new active one.\n",
1133                                         bond->dev->name, new_active->dev->name);
1134                         }
1135                 }
1136         }
1137
1138         if (USES_PRIMARY(bond->params.mode))
1139                 bond_mc_swap(bond, new_active, old_active);
1140
1141         if (bond_is_lb(bond)) {
1142                 bond_alb_handle_active_change(bond, new_active);
1143                 if (old_active)
1144                         bond_set_slave_inactive_flags(old_active);
1145                 if (new_active)
1146                         bond_set_slave_active_flags(new_active);
1147         } else {
1148                 bond->curr_active_slave = new_active;
1149         }
1150
1151         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1152                 if (old_active)
1153                         bond_set_slave_inactive_flags(old_active);
1154
1155                 if (new_active) {
1156                         bond_set_slave_active_flags(new_active);
1157
1158                         if (bond->params.fail_over_mac)
1159                                 bond_do_fail_over_mac(bond, new_active,
1160                                                       old_active);
1161
1162                         bond->send_grat_arp = bond->params.num_grat_arp;
1163                         bond_send_gratuitous_arp(bond);
1164
1165                         bond->send_unsol_na = bond->params.num_unsol_na;
1166                         bond_send_unsolicited_na(bond);
1167
1168                         write_unlock_bh(&bond->curr_slave_lock);
1169                         read_unlock(&bond->lock);
1170
1171                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1172
1173                         read_lock(&bond->lock);
1174                         write_lock_bh(&bond->curr_slave_lock);
1175                 }
1176         }
1177
1178         /* resend IGMP joins since all were sent on curr_active_slave */
1179         if (bond->params.mode == BOND_MODE_ROUNDROBIN) {
1180                 bond_resend_igmp_join_requests(bond);
1181         }
1182 }
1183
1184 /**
1185  * bond_select_active_slave - select a new active slave, if needed
1186  * @bond: our bonding struct
1187  *
1188  * This functions should be called when one of the following occurs:
1189  * - The old curr_active_slave has been released or lost its link.
1190  * - The primary_slave has got its link back.
1191  * - A slave has got its link back and there's no old curr_active_slave.
1192  *
1193  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1194  */
1195 void bond_select_active_slave(struct bonding *bond)
1196 {
1197         struct slave *best_slave;
1198         int rv;
1199
1200         best_slave = bond_find_best_slave(bond);
1201         if (best_slave != bond->curr_active_slave) {
1202                 bond_change_active_slave(bond, best_slave);
1203                 rv = bond_set_carrier(bond);
1204                 if (!rv)
1205                         return;
1206
1207                 if (netif_carrier_ok(bond->dev)) {
1208                         pr_info("%s: first active interface up!\n",
1209                                 bond->dev->name);
1210                 } else {
1211                         pr_info("%s: now running without any active interface !\n",
1212                                 bond->dev->name);
1213                 }
1214         }
1215 }
1216
1217 /*--------------------------- slave list handling ---------------------------*/
1218
1219 /*
1220  * This function attaches the slave to the end of list.
1221  *
1222  * bond->lock held for writing by caller.
1223  */
1224 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1225 {
1226         if (bond->first_slave == NULL) { /* attaching the first slave */
1227                 new_slave->next = new_slave;
1228                 new_slave->prev = new_slave;
1229                 bond->first_slave = new_slave;
1230         } else {
1231                 new_slave->next = bond->first_slave;
1232                 new_slave->prev = bond->first_slave->prev;
1233                 new_slave->next->prev = new_slave;
1234                 new_slave->prev->next = new_slave;
1235         }
1236
1237         bond->slave_cnt++;
1238 }
1239
1240 /*
1241  * This function detaches the slave from the list.
1242  * WARNING: no check is made to verify if the slave effectively
1243  * belongs to <bond>.
1244  * Nothing is freed on return, structures are just unchained.
1245  * If any slave pointer in bond was pointing to <slave>,
1246  * it should be changed by the calling function.
1247  *
1248  * bond->lock held for writing by caller.
1249  */
1250 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1251 {
1252         if (slave->next)
1253                 slave->next->prev = slave->prev;
1254
1255         if (slave->prev)
1256                 slave->prev->next = slave->next;
1257
1258         if (bond->first_slave == slave) { /* slave is the first slave */
1259                 if (bond->slave_cnt > 1) { /* there are more slave */
1260                         bond->first_slave = slave->next;
1261                 } else {
1262                         bond->first_slave = NULL; /* slave was the last one */
1263                 }
1264         }
1265
1266         slave->next = NULL;
1267         slave->prev = NULL;
1268         bond->slave_cnt--;
1269 }
1270
1271 #ifdef CONFIG_NET_POLL_CONTROLLER
1272 /*
1273  * You must hold read lock on bond->lock before calling this.
1274  */
1275 static bool slaves_support_netpoll(struct net_device *bond_dev)
1276 {
1277         struct bonding *bond = netdev_priv(bond_dev);
1278         struct slave *slave;
1279         int i = 0;
1280         bool ret = true;
1281
1282         bond_for_each_slave(bond, slave, i) {
1283                 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1284                     !slave->dev->netdev_ops->ndo_poll_controller)
1285                         ret = false;
1286         }
1287         return i != 0 && ret;
1288 }
1289
1290 static void bond_poll_controller(struct net_device *bond_dev)
1291 {
1292         struct net_device *dev = bond_dev->npinfo->netpoll->real_dev;
1293         if (dev != bond_dev)
1294                 netpoll_poll_dev(dev);
1295 }
1296
1297 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1298 {
1299         struct bonding *bond = netdev_priv(bond_dev);
1300         struct slave *slave;
1301         const struct net_device_ops *ops;
1302         int i;
1303
1304         read_lock(&bond->lock);
1305         bond_dev->npinfo = NULL;
1306         bond_for_each_slave(bond, slave, i) {
1307                 if (slave->dev) {
1308                         ops = slave->dev->netdev_ops;
1309                         if (ops->ndo_netpoll_cleanup)
1310                                 ops->ndo_netpoll_cleanup(slave->dev);
1311                         else
1312                                 slave->dev->npinfo = NULL;
1313                 }
1314         }
1315         read_unlock(&bond->lock);
1316 }
1317
1318 #else
1319
1320 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1321 {
1322 }
1323
1324 #endif
1325
1326 /*---------------------------------- IOCTL ----------------------------------*/
1327
1328 static int bond_sethwaddr(struct net_device *bond_dev,
1329                           struct net_device *slave_dev)
1330 {
1331         pr_debug("bond_dev=%p\n", bond_dev);
1332         pr_debug("slave_dev=%p\n", slave_dev);
1333         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1334         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1335         return 0;
1336 }
1337
1338 #define BOND_VLAN_FEATURES \
1339         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1340          NETIF_F_HW_VLAN_FILTER)
1341
1342 /*
1343  * Compute the common dev->feature set available to all slaves.  Some
1344  * feature bits are managed elsewhere, so preserve those feature bits
1345  * on the master device.
1346  */
1347 static int bond_compute_features(struct bonding *bond)
1348 {
1349         struct slave *slave;
1350         struct net_device *bond_dev = bond->dev;
1351         unsigned long features = bond_dev->features;
1352         unsigned long vlan_features = 0;
1353         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1354                                                 bond_dev->hard_header_len);
1355         int i;
1356
1357         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1358         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1359
1360         if (!bond->first_slave)
1361                 goto done;
1362
1363         features &= ~NETIF_F_ONE_FOR_ALL;
1364
1365         vlan_features = bond->first_slave->dev->vlan_features;
1366         bond_for_each_slave(bond, slave, i) {
1367                 features = netdev_increment_features(features,
1368                                                      slave->dev->features,
1369                                                      NETIF_F_ONE_FOR_ALL);
1370                 vlan_features = netdev_increment_features(vlan_features,
1371                                                         slave->dev->vlan_features,
1372                                                         NETIF_F_ONE_FOR_ALL);
1373                 if (slave->dev->hard_header_len > max_hard_header_len)
1374                         max_hard_header_len = slave->dev->hard_header_len;
1375         }
1376
1377 done:
1378         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1379         bond_dev->features = netdev_fix_features(features, NULL);
1380         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1381         bond_dev->hard_header_len = max_hard_header_len;
1382
1383         return 0;
1384 }
1385
1386 static void bond_setup_by_slave(struct net_device *bond_dev,
1387                                 struct net_device *slave_dev)
1388 {
1389         struct bonding *bond = netdev_priv(bond_dev);
1390
1391         bond_dev->header_ops        = slave_dev->header_ops;
1392
1393         bond_dev->type              = slave_dev->type;
1394         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1395         bond_dev->addr_len          = slave_dev->addr_len;
1396
1397         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1398                 slave_dev->addr_len);
1399         bond->setup_by_slave = 1;
1400 }
1401
1402 /* enslave device <slave> to bond device <master> */
1403 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1404 {
1405         struct bonding *bond = netdev_priv(bond_dev);
1406         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1407         struct slave *new_slave = NULL;
1408         struct netdev_hw_addr *ha;
1409         struct sockaddr addr;
1410         int link_reporting;
1411         int old_features = bond_dev->features;
1412         int res = 0;
1413
1414         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1415                 slave_ops->ndo_do_ioctl == NULL) {
1416                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1417                            bond_dev->name, slave_dev->name);
1418         }
1419
1420         /* bond must be initialized by bond_open() before enslaving */
1421         if (!(bond_dev->flags & IFF_UP)) {
1422                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1423                            bond_dev->name);
1424         }
1425
1426         /* already enslaved */
1427         if (slave_dev->flags & IFF_SLAVE) {
1428                 pr_debug("Error, Device was already enslaved\n");
1429                 return -EBUSY;
1430         }
1431
1432         /* vlan challenged mutual exclusion */
1433         /* no need to lock since we're protected by rtnl_lock */
1434         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1435                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1436                 if (!list_empty(&bond->vlan_list)) {
1437                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1438                                bond_dev->name, slave_dev->name, bond_dev->name);
1439                         return -EPERM;
1440                 } else {
1441                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1442                                    bond_dev->name, slave_dev->name,
1443                                    slave_dev->name, bond_dev->name);
1444                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1445                 }
1446         } else {
1447                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1448                 if (bond->slave_cnt == 0) {
1449                         /* First slave, and it is not VLAN challenged,
1450                          * so remove the block of adding VLANs over the bond.
1451                          */
1452                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1453                 }
1454         }
1455
1456         /*
1457          * Old ifenslave binaries are no longer supported.  These can
1458          * be identified with moderate accuracy by the state of the slave:
1459          * the current ifenslave will set the interface down prior to
1460          * enslaving it; the old ifenslave will not.
1461          */
1462         if ((slave_dev->flags & IFF_UP)) {
1463                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1464                        slave_dev->name);
1465                 res = -EPERM;
1466                 goto err_undo_flags;
1467         }
1468
1469         /* set bonding device ether type by slave - bonding netdevices are
1470          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1471          * there is a need to override some of the type dependent attribs/funcs.
1472          *
1473          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1474          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1475          */
1476         if (bond->slave_cnt == 0) {
1477                 if (bond_dev->type != slave_dev->type) {
1478                         pr_debug("%s: change device type from %d to %d\n",
1479                                  bond_dev->name,
1480                                  bond_dev->type, slave_dev->type);
1481
1482                         res = netdev_bonding_change(bond_dev,
1483                                                     NETDEV_PRE_TYPE_CHANGE);
1484                         res = notifier_to_errno(res);
1485                         if (res) {
1486                                 pr_err("%s: refused to change device type\n",
1487                                        bond_dev->name);
1488                                 res = -EBUSY;
1489                                 goto err_undo_flags;
1490                         }
1491
1492                         /* Flush unicast and multicast addresses */
1493                         dev_uc_flush(bond_dev);
1494                         dev_mc_flush(bond_dev);
1495
1496                         if (slave_dev->type != ARPHRD_ETHER)
1497                                 bond_setup_by_slave(bond_dev, slave_dev);
1498                         else
1499                                 ether_setup(bond_dev);
1500
1501                         netdev_bonding_change(bond_dev,
1502                                               NETDEV_POST_TYPE_CHANGE);
1503                 }
1504         } else if (bond_dev->type != slave_dev->type) {
1505                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1506                        slave_dev->name,
1507                        slave_dev->type, bond_dev->type);
1508                 res = -EINVAL;
1509                 goto err_undo_flags;
1510         }
1511
1512         if (slave_ops->ndo_set_mac_address == NULL) {
1513                 if (bond->slave_cnt == 0) {
1514                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1515                                    bond_dev->name);
1516                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1517                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1518                         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",
1519                                bond_dev->name);
1520                         res = -EOPNOTSUPP;
1521                         goto err_undo_flags;
1522                 }
1523         }
1524
1525         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1526         if (!new_slave) {
1527                 res = -ENOMEM;
1528                 goto err_undo_flags;
1529         }
1530
1531         /* save slave's original flags before calling
1532          * netdev_set_master and dev_open
1533          */
1534         new_slave->original_flags = slave_dev->flags;
1535
1536         /*
1537          * Save slave's original ("permanent") mac address for modes
1538          * that need it, and for restoring it upon release, and then
1539          * set it to the master's address
1540          */
1541         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1542
1543         if (!bond->params.fail_over_mac) {
1544                 /*
1545                  * Set slave to master's mac address.  The application already
1546                  * set the master's mac address to that of the first slave
1547                  */
1548                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1549                 addr.sa_family = slave_dev->type;
1550                 res = dev_set_mac_address(slave_dev, &addr);
1551                 if (res) {
1552                         pr_debug("Error %d calling set_mac_address\n", res);
1553                         goto err_free;
1554                 }
1555         }
1556
1557         res = netdev_set_master(slave_dev, bond_dev);
1558         if (res) {
1559                 pr_debug("Error %d calling netdev_set_master\n", res);
1560                 goto err_restore_mac;
1561         }
1562         /* open the slave since the application closed it */
1563         res = dev_open(slave_dev);
1564         if (res) {
1565                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1566                 goto err_unset_master;
1567         }
1568
1569         new_slave->dev = slave_dev;
1570         slave_dev->priv_flags |= IFF_BONDING;
1571
1572         if (bond_is_lb(bond)) {
1573                 /* bond_alb_init_slave() must be called before all other stages since
1574                  * it might fail and we do not want to have to undo everything
1575                  */
1576                 res = bond_alb_init_slave(bond, new_slave);
1577                 if (res)
1578                         goto err_close;
1579         }
1580
1581         /* If the mode USES_PRIMARY, then the new slave gets the
1582          * master's promisc (and mc) settings only if it becomes the
1583          * curr_active_slave, and that is taken care of later when calling
1584          * bond_change_active()
1585          */
1586         if (!USES_PRIMARY(bond->params.mode)) {
1587                 /* set promiscuity level to new slave */
1588                 if (bond_dev->flags & IFF_PROMISC) {
1589                         res = dev_set_promiscuity(slave_dev, 1);
1590                         if (res)
1591                                 goto err_close;
1592                 }
1593
1594                 /* set allmulti level to new slave */
1595                 if (bond_dev->flags & IFF_ALLMULTI) {
1596                         res = dev_set_allmulti(slave_dev, 1);
1597                         if (res)
1598                                 goto err_close;
1599                 }
1600
1601                 netif_addr_lock_bh(bond_dev);
1602                 /* upload master's mc_list to new slave */
1603                 netdev_for_each_mc_addr(ha, bond_dev)
1604                         dev_mc_add(slave_dev, ha->addr);
1605                 netif_addr_unlock_bh(bond_dev);
1606         }
1607
1608         if (bond->params.mode == BOND_MODE_8023AD) {
1609                 /* add lacpdu mc addr to mc list */
1610                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1611
1612                 dev_mc_add(slave_dev, lacpdu_multicast);
1613         }
1614
1615         bond_add_vlans_on_slave(bond, slave_dev);
1616
1617         write_lock_bh(&bond->lock);
1618
1619         bond_attach_slave(bond, new_slave);
1620
1621         new_slave->delay = 0;
1622         new_slave->link_failure_count = 0;
1623
1624         bond_compute_features(bond);
1625
1626         write_unlock_bh(&bond->lock);
1627
1628         read_lock(&bond->lock);
1629
1630         new_slave->last_arp_rx = jiffies;
1631
1632         if (bond->params.miimon && !bond->params.use_carrier) {
1633                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1634
1635                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1636                         /*
1637                          * miimon is set but a bonded network driver
1638                          * does not support ETHTOOL/MII and
1639                          * arp_interval is not set.  Note: if
1640                          * use_carrier is enabled, we will never go
1641                          * here (because netif_carrier is always
1642                          * supported); thus, we don't need to change
1643                          * the messages for netif_carrier.
1644                          */
1645                         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",
1646                                bond_dev->name, slave_dev->name);
1647                 } else if (link_reporting == -1) {
1648                         /* unable get link status using mii/ethtool */
1649                         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",
1650                                    bond_dev->name, slave_dev->name);
1651                 }
1652         }
1653
1654         /* check for initial state */
1655         if (!bond->params.miimon ||
1656             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1657                 if (bond->params.updelay) {
1658                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1659                         new_slave->link  = BOND_LINK_BACK;
1660                         new_slave->delay = bond->params.updelay;
1661                 } else {
1662                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1663                         new_slave->link  = BOND_LINK_UP;
1664                 }
1665                 new_slave->jiffies = jiffies;
1666         } else {
1667                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1668                 new_slave->link  = BOND_LINK_DOWN;
1669         }
1670
1671         if (bond_update_speed_duplex(new_slave) &&
1672             (new_slave->link != BOND_LINK_DOWN)) {
1673                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1674                            bond_dev->name, new_slave->dev->name);
1675
1676                 if (bond->params.mode == BOND_MODE_8023AD) {
1677                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1678                                    bond_dev->name);
1679                 }
1680         }
1681
1682         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1683                 /* if there is a primary slave, remember it */
1684                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1685                         bond->primary_slave = new_slave;
1686                         bond->force_primary = true;
1687                 }
1688         }
1689
1690         write_lock_bh(&bond->curr_slave_lock);
1691
1692         switch (bond->params.mode) {
1693         case BOND_MODE_ACTIVEBACKUP:
1694                 bond_set_slave_inactive_flags(new_slave);
1695                 bond_select_active_slave(bond);
1696                 break;
1697         case BOND_MODE_8023AD:
1698                 /* in 802.3ad mode, the internal mechanism
1699                  * will activate the slaves in the selected
1700                  * aggregator
1701                  */
1702                 bond_set_slave_inactive_flags(new_slave);
1703                 /* if this is the first slave */
1704                 if (bond->slave_cnt == 1) {
1705                         SLAVE_AD_INFO(new_slave).id = 1;
1706                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1707                          * can be called only after the mac address of the bond is set
1708                          */
1709                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1710                                             bond->params.lacp_fast);
1711                 } else {
1712                         SLAVE_AD_INFO(new_slave).id =
1713                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1714                 }
1715
1716                 bond_3ad_bind_slave(new_slave);
1717                 break;
1718         case BOND_MODE_TLB:
1719         case BOND_MODE_ALB:
1720                 new_slave->state = BOND_STATE_ACTIVE;
1721                 bond_set_slave_inactive_flags(new_slave);
1722                 bond_select_active_slave(bond);
1723                 break;
1724         default:
1725                 pr_debug("This slave is always active in trunk mode\n");
1726
1727                 /* always active in trunk mode */
1728                 new_slave->state = BOND_STATE_ACTIVE;
1729
1730                 /* In trunking mode there is little meaning to curr_active_slave
1731                  * anyway (it holds no special properties of the bond device),
1732                  * so we can change it without calling change_active_interface()
1733                  */
1734                 if (!bond->curr_active_slave)
1735                         bond->curr_active_slave = new_slave;
1736
1737                 break;
1738         } /* switch(bond_mode) */
1739
1740         write_unlock_bh(&bond->curr_slave_lock);
1741
1742         bond_set_carrier(bond);
1743
1744 #ifdef CONFIG_NET_POLL_CONTROLLER
1745         if (slaves_support_netpoll(bond_dev)) {
1746                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1747                 if (bond_dev->npinfo)
1748                         slave_dev->npinfo = bond_dev->npinfo;
1749         } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1750                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1751                 pr_info("New slave device %s does not support netpoll\n",
1752                         slave_dev->name);
1753                 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1754         }
1755 #endif
1756         read_unlock(&bond->lock);
1757
1758         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1759         if (res)
1760                 goto err_close;
1761
1762         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1763                 bond_dev->name, slave_dev->name,
1764                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1765                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1766
1767         /* enslave is successful */
1768         return 0;
1769
1770 /* Undo stages on error */
1771 err_close:
1772         dev_close(slave_dev);
1773
1774 err_unset_master:
1775         netdev_set_master(slave_dev, NULL);
1776
1777 err_restore_mac:
1778         if (!bond->params.fail_over_mac) {
1779                 /* XXX TODO - fom follow mode needs to change master's
1780                  * MAC if this slave's MAC is in use by the bond, or at
1781                  * least print a warning.
1782                  */
1783                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1784                 addr.sa_family = slave_dev->type;
1785                 dev_set_mac_address(slave_dev, &addr);
1786         }
1787
1788 err_free:
1789         kfree(new_slave);
1790
1791 err_undo_flags:
1792         bond_dev->features = old_features;
1793
1794         return res;
1795 }
1796
1797 /*
1798  * Try to release the slave device <slave> from the bond device <master>
1799  * It is legal to access curr_active_slave without a lock because all the function
1800  * is write-locked.
1801  *
1802  * The rules for slave state should be:
1803  *   for Active/Backup:
1804  *     Active stays on all backups go down
1805  *   for Bonded connections:
1806  *     The first up interface should be left on and all others downed.
1807  */
1808 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1809 {
1810         struct bonding *bond = netdev_priv(bond_dev);
1811         struct slave *slave, *oldcurrent;
1812         struct sockaddr addr;
1813
1814         /* slave is not a slave or master is not master of this slave */
1815         if (!(slave_dev->flags & IFF_SLAVE) ||
1816             (slave_dev->master != bond_dev)) {
1817                 pr_err("%s: Error: cannot release %s.\n",
1818                        bond_dev->name, slave_dev->name);
1819                 return -EINVAL;
1820         }
1821
1822         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1823         write_lock_bh(&bond->lock);
1824
1825         slave = bond_get_slave_by_dev(bond, slave_dev);
1826         if (!slave) {
1827                 /* not a slave of this bond */
1828                 pr_info("%s: %s not enslaved\n",
1829                         bond_dev->name, slave_dev->name);
1830                 write_unlock_bh(&bond->lock);
1831                 return -EINVAL;
1832         }
1833
1834         if (!bond->params.fail_over_mac) {
1835                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1836                     bond->slave_cnt > 1)
1837                         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",
1838                                    bond_dev->name, slave_dev->name,
1839                                    slave->perm_hwaddr,
1840                                    bond_dev->name, slave_dev->name);
1841         }
1842
1843         /* Inform AD package of unbinding of slave. */
1844         if (bond->params.mode == BOND_MODE_8023AD) {
1845                 /* must be called before the slave is
1846                  * detached from the list
1847                  */
1848                 bond_3ad_unbind_slave(slave);
1849         }
1850
1851         pr_info("%s: releasing %s interface %s\n",
1852                 bond_dev->name,
1853                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1854                 slave_dev->name);
1855
1856         oldcurrent = bond->curr_active_slave;
1857
1858         bond->current_arp_slave = NULL;
1859
1860         /* release the slave from its bond */
1861         bond_detach_slave(bond, slave);
1862
1863         bond_compute_features(bond);
1864
1865         if (bond->primary_slave == slave)
1866                 bond->primary_slave = NULL;
1867
1868         if (oldcurrent == slave)
1869                 bond_change_active_slave(bond, NULL);
1870
1871         if (bond_is_lb(bond)) {
1872                 /* Must be called only after the slave has been
1873                  * detached from the list and the curr_active_slave
1874                  * has been cleared (if our_slave == old_current),
1875                  * but before a new active slave is selected.
1876                  */
1877                 write_unlock_bh(&bond->lock);
1878                 bond_alb_deinit_slave(bond, slave);
1879                 write_lock_bh(&bond->lock);
1880         }
1881
1882         if (oldcurrent == slave) {
1883                 /*
1884                  * Note that we hold RTNL over this sequence, so there
1885                  * is no concern that another slave add/remove event
1886                  * will interfere.
1887                  */
1888                 write_unlock_bh(&bond->lock);
1889                 read_lock(&bond->lock);
1890                 write_lock_bh(&bond->curr_slave_lock);
1891
1892                 bond_select_active_slave(bond);
1893
1894                 write_unlock_bh(&bond->curr_slave_lock);
1895                 read_unlock(&bond->lock);
1896                 write_lock_bh(&bond->lock);
1897         }
1898
1899         if (bond->slave_cnt == 0) {
1900                 bond_set_carrier(bond);
1901
1902                 /* if the last slave was removed, zero the mac address
1903                  * of the master so it will be set by the application
1904                  * to the mac address of the first slave
1905                  */
1906                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1907
1908                 if (list_empty(&bond->vlan_list)) {
1909                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1910                 } else {
1911                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1912                                    bond_dev->name, bond_dev->name);
1913                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1914                                    bond_dev->name);
1915                 }
1916         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1917                    !bond_has_challenged_slaves(bond)) {
1918                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1919                         bond_dev->name, slave_dev->name, bond_dev->name);
1920                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1921         }
1922
1923         write_unlock_bh(&bond->lock);
1924
1925         /* must do this from outside any spinlocks */
1926         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1927
1928         bond_del_vlans_from_slave(bond, slave_dev);
1929
1930         /* If the mode USES_PRIMARY, then we should only remove its
1931          * promisc and mc settings if it was the curr_active_slave, but that was
1932          * already taken care of above when we detached the slave
1933          */
1934         if (!USES_PRIMARY(bond->params.mode)) {
1935                 /* unset promiscuity level from slave */
1936                 if (bond_dev->flags & IFF_PROMISC)
1937                         dev_set_promiscuity(slave_dev, -1);
1938
1939                 /* unset allmulti level from slave */
1940                 if (bond_dev->flags & IFF_ALLMULTI)
1941                         dev_set_allmulti(slave_dev, -1);
1942
1943                 /* flush master's mc_list from slave */
1944                 netif_addr_lock_bh(bond_dev);
1945                 bond_mc_list_flush(bond_dev, slave_dev);
1946                 netif_addr_unlock_bh(bond_dev);
1947         }
1948
1949         netdev_set_master(slave_dev, NULL);
1950
1951 #ifdef CONFIG_NET_POLL_CONTROLLER
1952         read_lock_bh(&bond->lock);
1953         if (slaves_support_netpoll(bond_dev))
1954                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1955         read_unlock_bh(&bond->lock);
1956         if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
1957                 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
1958         else
1959                 slave_dev->npinfo = NULL;
1960 #endif
1961
1962         /* close slave before restoring its mac address */
1963         dev_close(slave_dev);
1964
1965         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1966                 /* restore original ("permanent") mac address */
1967                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1968                 addr.sa_family = slave_dev->type;
1969                 dev_set_mac_address(slave_dev, &addr);
1970         }
1971
1972         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1973                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1974                                    IFF_SLAVE_NEEDARP);
1975
1976         kfree(slave);
1977
1978         return 0;  /* deletion OK */
1979 }
1980
1981 /*
1982 * First release a slave and than destroy the bond if no more slaves are left.
1983 * Must be under rtnl_lock when this function is called.
1984 */
1985 int  bond_release_and_destroy(struct net_device *bond_dev,
1986                               struct net_device *slave_dev)
1987 {
1988         struct bonding *bond = netdev_priv(bond_dev);
1989         int ret;
1990
1991         ret = bond_release(bond_dev, slave_dev);
1992         if ((ret == 0) && (bond->slave_cnt == 0)) {
1993                 pr_info("%s: destroying bond %s.\n",
1994                         bond_dev->name, bond_dev->name);
1995                 unregister_netdevice(bond_dev);
1996         }
1997         return ret;
1998 }
1999
2000 /*
2001  * This function releases all slaves.
2002  */
2003 static int bond_release_all(struct net_device *bond_dev)
2004 {
2005         struct bonding *bond = netdev_priv(bond_dev);
2006         struct slave *slave;
2007         struct net_device *slave_dev;
2008         struct sockaddr addr;
2009
2010         write_lock_bh(&bond->lock);
2011
2012         netif_carrier_off(bond_dev);
2013
2014         if (bond->slave_cnt == 0)
2015                 goto out;
2016
2017         bond->current_arp_slave = NULL;
2018         bond->primary_slave = NULL;
2019         bond_change_active_slave(bond, NULL);
2020
2021         while ((slave = bond->first_slave) != NULL) {
2022                 /* Inform AD package of unbinding of slave
2023                  * before slave is detached from the list.
2024                  */
2025                 if (bond->params.mode == BOND_MODE_8023AD)
2026                         bond_3ad_unbind_slave(slave);
2027
2028                 slave_dev = slave->dev;
2029                 bond_detach_slave(bond, slave);
2030
2031                 /* now that the slave is detached, unlock and perform
2032                  * all the undo steps that should not be called from
2033                  * within a lock.
2034                  */
2035                 write_unlock_bh(&bond->lock);
2036
2037                 if (bond_is_lb(bond)) {
2038                         /* must be called only after the slave
2039                          * has been detached from the list
2040                          */
2041                         bond_alb_deinit_slave(bond, slave);
2042                 }
2043
2044                 bond_compute_features(bond);
2045
2046                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2047                 bond_del_vlans_from_slave(bond, slave_dev);
2048
2049                 /* If the mode USES_PRIMARY, then we should only remove its
2050                  * promisc and mc settings if it was the curr_active_slave, but that was
2051                  * already taken care of above when we detached the slave
2052                  */
2053                 if (!USES_PRIMARY(bond->params.mode)) {
2054                         /* unset promiscuity level from slave */
2055                         if (bond_dev->flags & IFF_PROMISC)
2056                                 dev_set_promiscuity(slave_dev, -1);
2057
2058                         /* unset allmulti level from slave */
2059                         if (bond_dev->flags & IFF_ALLMULTI)
2060                                 dev_set_allmulti(slave_dev, -1);
2061
2062                         /* flush master's mc_list from slave */
2063                         netif_addr_lock_bh(bond_dev);
2064                         bond_mc_list_flush(bond_dev, slave_dev);
2065                         netif_addr_unlock_bh(bond_dev);
2066                 }
2067
2068                 netdev_set_master(slave_dev, NULL);
2069
2070                 /* close slave before restoring its mac address */
2071                 dev_close(slave_dev);
2072
2073                 if (!bond->params.fail_over_mac) {
2074                         /* restore original ("permanent") mac address*/
2075                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2076                         addr.sa_family = slave_dev->type;
2077                         dev_set_mac_address(slave_dev, &addr);
2078                 }
2079
2080                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2081                                            IFF_SLAVE_INACTIVE);
2082
2083                 kfree(slave);
2084
2085                 /* re-acquire the lock before getting the next slave */
2086                 write_lock_bh(&bond->lock);
2087         }
2088
2089         /* zero the mac address of the master so it will be
2090          * set by the application to the mac address of the
2091          * first slave
2092          */
2093         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2094
2095         if (list_empty(&bond->vlan_list))
2096                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2097         else {
2098                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2099                            bond_dev->name, bond_dev->name);
2100                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2101                            bond_dev->name);
2102         }
2103
2104         pr_info("%s: released all slaves\n", bond_dev->name);
2105
2106 out:
2107         write_unlock_bh(&bond->lock);
2108
2109         return 0;
2110 }
2111
2112 /*
2113  * This function changes the active slave to slave <slave_dev>.
2114  * It returns -EINVAL in the following cases.
2115  *  - <slave_dev> is not found in the list.
2116  *  - There is not active slave now.
2117  *  - <slave_dev> is already active.
2118  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2119  *  - <slave_dev> is not running.
2120  * In these cases, this function does nothing.
2121  * In the other cases, current_slave pointer is changed and 0 is returned.
2122  */
2123 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2124 {
2125         struct bonding *bond = netdev_priv(bond_dev);
2126         struct slave *old_active = NULL;
2127         struct slave *new_active = NULL;
2128         int res = 0;
2129
2130         if (!USES_PRIMARY(bond->params.mode))
2131                 return -EINVAL;
2132
2133         /* Verify that master_dev is indeed the master of slave_dev */
2134         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2135                 return -EINVAL;
2136
2137         read_lock(&bond->lock);
2138
2139         read_lock(&bond->curr_slave_lock);
2140         old_active = bond->curr_active_slave;
2141         read_unlock(&bond->curr_slave_lock);
2142
2143         new_active = bond_get_slave_by_dev(bond, slave_dev);
2144
2145         /*
2146          * Changing to the current active: do nothing; return success.
2147          */
2148         if (new_active && (new_active == old_active)) {
2149                 read_unlock(&bond->lock);
2150                 return 0;
2151         }
2152
2153         if ((new_active) &&
2154             (old_active) &&
2155             (new_active->link == BOND_LINK_UP) &&
2156             IS_UP(new_active->dev)) {
2157                 write_lock_bh(&bond->curr_slave_lock);
2158                 bond_change_active_slave(bond, new_active);
2159                 write_unlock_bh(&bond->curr_slave_lock);
2160         } else
2161                 res = -EINVAL;
2162
2163         read_unlock(&bond->lock);
2164
2165         return res;
2166 }
2167
2168 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2169 {
2170         struct bonding *bond = netdev_priv(bond_dev);
2171
2172         info->bond_mode = bond->params.mode;
2173         info->miimon = bond->params.miimon;
2174
2175         read_lock(&bond->lock);
2176         info->num_slaves = bond->slave_cnt;
2177         read_unlock(&bond->lock);
2178
2179         return 0;
2180 }
2181
2182 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2183 {
2184         struct bonding *bond = netdev_priv(bond_dev);
2185         struct slave *slave;
2186         int i, res = -ENODEV;
2187
2188         read_lock(&bond->lock);
2189
2190         bond_for_each_slave(bond, slave, i) {
2191                 if (i == (int)info->slave_id) {
2192                         res = 0;
2193                         strcpy(info->slave_name, slave->dev->name);
2194                         info->link = slave->link;
2195                         info->state = slave->state;
2196                         info->link_failure_count = slave->link_failure_count;
2197                         break;
2198                 }
2199         }
2200
2201         read_unlock(&bond->lock);
2202
2203         return res;
2204 }
2205
2206 /*-------------------------------- Monitoring -------------------------------*/
2207
2208
2209 static int bond_miimon_inspect(struct bonding *bond)
2210 {
2211         struct slave *slave;
2212         int i, link_state, commit = 0;
2213         bool ignore_updelay;
2214
2215         ignore_updelay = !bond->curr_active_slave ? true : false;
2216
2217         bond_for_each_slave(bond, slave, i) {
2218                 slave->new_link = BOND_LINK_NOCHANGE;
2219
2220                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2221
2222                 switch (slave->link) {
2223                 case BOND_LINK_UP:
2224                         if (link_state)
2225                                 continue;
2226
2227                         slave->link = BOND_LINK_FAIL;
2228                         slave->delay = bond->params.downdelay;
2229                         if (slave->delay) {
2230                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2231                                         bond->dev->name,
2232                                         (bond->params.mode ==
2233                                          BOND_MODE_ACTIVEBACKUP) ?
2234                                         ((slave->state == BOND_STATE_ACTIVE) ?
2235                                          "active " : "backup ") : "",
2236                                         slave->dev->name,
2237                                         bond->params.downdelay * bond->params.miimon);
2238                         }
2239                         /*FALLTHRU*/
2240                 case BOND_LINK_FAIL:
2241                         if (link_state) {
2242                                 /*
2243                                  * recovered before downdelay expired
2244                                  */
2245                                 slave->link = BOND_LINK_UP;
2246                                 slave->jiffies = jiffies;
2247                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2248                                         bond->dev->name,
2249                                         (bond->params.downdelay - slave->delay) *
2250                                         bond->params.miimon,
2251                                         slave->dev->name);
2252                                 continue;
2253                         }
2254
2255                         if (slave->delay <= 0) {
2256                                 slave->new_link = BOND_LINK_DOWN;
2257                                 commit++;
2258                                 continue;
2259                         }
2260
2261                         slave->delay--;
2262                         break;
2263
2264                 case BOND_LINK_DOWN:
2265                         if (!link_state)
2266                                 continue;
2267
2268                         slave->link = BOND_LINK_BACK;
2269                         slave->delay = bond->params.updelay;
2270
2271                         if (slave->delay) {
2272                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2273                                         bond->dev->name, slave->dev->name,
2274                                         ignore_updelay ? 0 :
2275                                         bond->params.updelay *
2276                                         bond->params.miimon);
2277                         }
2278                         /*FALLTHRU*/
2279                 case BOND_LINK_BACK:
2280                         if (!link_state) {
2281                                 slave->link = BOND_LINK_DOWN;
2282                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2283                                         bond->dev->name,
2284                                         (bond->params.updelay - slave->delay) *
2285                                         bond->params.miimon,
2286                                         slave->dev->name);
2287
2288                                 continue;
2289                         }
2290
2291                         if (ignore_updelay)
2292                                 slave->delay = 0;
2293
2294                         if (slave->delay <= 0) {
2295                                 slave->new_link = BOND_LINK_UP;
2296                                 commit++;
2297                                 ignore_updelay = false;
2298                                 continue;
2299                         }
2300
2301                         slave->delay--;
2302                         break;
2303                 }
2304         }
2305
2306         return commit;
2307 }
2308
2309 static void bond_miimon_commit(struct bonding *bond)
2310 {
2311         struct slave *slave;
2312         int i;
2313
2314         bond_for_each_slave(bond, slave, i) {
2315                 switch (slave->new_link) {
2316                 case BOND_LINK_NOCHANGE:
2317                         continue;
2318
2319                 case BOND_LINK_UP:
2320                         slave->link = BOND_LINK_UP;
2321                         slave->jiffies = jiffies;
2322
2323                         if (bond->params.mode == BOND_MODE_8023AD) {
2324                                 /* prevent it from being the active one */
2325                                 slave->state = BOND_STATE_BACKUP;
2326                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2327                                 /* make it immediately active */
2328                                 slave->state = BOND_STATE_ACTIVE;
2329                         } else if (slave != bond->primary_slave) {
2330                                 /* prevent it from being the active one */
2331                                 slave->state = BOND_STATE_BACKUP;
2332                         }
2333
2334                         pr_info("%s: link status definitely up for interface %s.\n",
2335                                 bond->dev->name, slave->dev->name);
2336
2337                         /* notify ad that the link status has changed */
2338                         if (bond->params.mode == BOND_MODE_8023AD)
2339                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2340
2341                         if (bond_is_lb(bond))
2342                                 bond_alb_handle_link_change(bond, slave,
2343                                                             BOND_LINK_UP);
2344
2345                         if (!bond->curr_active_slave ||
2346                             (slave == bond->primary_slave))
2347                                 goto do_failover;
2348
2349                         continue;
2350
2351                 case BOND_LINK_DOWN:
2352                         if (slave->link_failure_count < UINT_MAX)
2353                                 slave->link_failure_count++;
2354
2355                         slave->link = BOND_LINK_DOWN;
2356
2357                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2358                             bond->params.mode == BOND_MODE_8023AD)
2359                                 bond_set_slave_inactive_flags(slave);
2360
2361                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2362                                 bond->dev->name, slave->dev->name);
2363
2364                         if (bond->params.mode == BOND_MODE_8023AD)
2365                                 bond_3ad_handle_link_change(slave,
2366                                                             BOND_LINK_DOWN);
2367
2368                         if (bond_is_lb(bond))
2369                                 bond_alb_handle_link_change(bond, slave,
2370                                                             BOND_LINK_DOWN);
2371
2372                         if (slave == bond->curr_active_slave)
2373                                 goto do_failover;
2374
2375                         continue;
2376
2377                 default:
2378                         pr_err("%s: invalid new link %d on slave %s\n",
2379                                bond->dev->name, slave->new_link,
2380                                slave->dev->name);
2381                         slave->new_link = BOND_LINK_NOCHANGE;
2382
2383                         continue;
2384                 }
2385
2386 do_failover:
2387                 ASSERT_RTNL();
2388                 write_lock_bh(&bond->curr_slave_lock);
2389                 bond_select_active_slave(bond);
2390                 write_unlock_bh(&bond->curr_slave_lock);
2391         }
2392
2393         bond_set_carrier(bond);
2394 }
2395
2396 /*
2397  * bond_mii_monitor
2398  *
2399  * Really a wrapper that splits the mii monitor into two phases: an
2400  * inspection, then (if inspection indicates something needs to be done)
2401  * an acquisition of appropriate locks followed by a commit phase to
2402  * implement whatever link state changes are indicated.
2403  */
2404 void bond_mii_monitor(struct work_struct *work)
2405 {
2406         struct bonding *bond = container_of(work, struct bonding,
2407                                             mii_work.work);
2408
2409         read_lock(&bond->lock);
2410         if (bond->kill_timers)
2411                 goto out;
2412
2413         if (bond->slave_cnt == 0)
2414                 goto re_arm;
2415
2416         if (bond->send_grat_arp) {
2417                 read_lock(&bond->curr_slave_lock);
2418                 bond_send_gratuitous_arp(bond);
2419                 read_unlock(&bond->curr_slave_lock);
2420         }
2421
2422         if (bond->send_unsol_na) {
2423                 read_lock(&bond->curr_slave_lock);
2424                 bond_send_unsolicited_na(bond);
2425                 read_unlock(&bond->curr_slave_lock);
2426         }
2427
2428         if (bond_miimon_inspect(bond)) {
2429                 read_unlock(&bond->lock);
2430                 rtnl_lock();
2431                 read_lock(&bond->lock);
2432
2433                 bond_miimon_commit(bond);
2434
2435                 read_unlock(&bond->lock);
2436                 rtnl_unlock();  /* might sleep, hold no other locks */
2437                 read_lock(&bond->lock);
2438         }
2439
2440 re_arm:
2441         if (bond->params.miimon)
2442                 queue_delayed_work(bond->wq, &bond->mii_work,
2443                                    msecs_to_jiffies(bond->params.miimon));
2444 out:
2445         read_unlock(&bond->lock);
2446 }
2447
2448 static __be32 bond_glean_dev_ip(struct net_device *dev)
2449 {
2450         struct in_device *idev;
2451         struct in_ifaddr *ifa;
2452         __be32 addr = 0;
2453
2454         if (!dev)
2455                 return 0;
2456
2457         rcu_read_lock();
2458         idev = __in_dev_get_rcu(dev);
2459         if (!idev)
2460                 goto out;
2461
2462         ifa = idev->ifa_list;
2463         if (!ifa)
2464                 goto out;
2465
2466         addr = ifa->ifa_local;
2467 out:
2468         rcu_read_unlock();
2469         return addr;
2470 }
2471
2472 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2473 {
2474         struct vlan_entry *vlan;
2475
2476         if (ip == bond->master_ip)
2477                 return 1;
2478
2479         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2480                 if (ip == vlan->vlan_ip)
2481                         return 1;
2482         }
2483
2484         return 0;
2485 }
2486
2487 /*
2488  * We go to the (large) trouble of VLAN tagging ARP frames because
2489  * switches in VLAN mode (especially if ports are configured as
2490  * "native" to a VLAN) might not pass non-tagged frames.
2491  */
2492 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2493 {
2494         struct sk_buff *skb;
2495
2496         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2497                  slave_dev->name, dest_ip, src_ip, vlan_id);
2498
2499         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2500                          NULL, slave_dev->dev_addr, NULL);
2501
2502         if (!skb) {
2503                 pr_err("ARP packet allocation failed\n");
2504                 return;
2505         }
2506         if (vlan_id) {
2507                 skb = vlan_put_tag(skb, vlan_id);
2508                 if (!skb) {
2509                         pr_err("failed to insert VLAN tag\n");
2510                         return;
2511                 }
2512         }
2513         arp_xmit(skb);
2514 }
2515
2516
2517 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2518 {
2519         int i, vlan_id, rv;
2520         __be32 *targets = bond->params.arp_targets;
2521         struct vlan_entry *vlan;
2522         struct net_device *vlan_dev;
2523         struct flowi fl;
2524         struct rtable *rt;
2525
2526         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2527                 if (!targets[i])
2528                         break;
2529                 pr_debug("basa: target %x\n", targets[i]);
2530                 if (list_empty(&bond->vlan_list)) {
2531                         pr_debug("basa: empty vlan: arp_send\n");
2532                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2533                                       bond->master_ip, 0);
2534                         continue;
2535                 }
2536
2537                 /*
2538                  * If VLANs are configured, we do a route lookup to
2539                  * determine which VLAN interface would be used, so we
2540                  * can tag the ARP with the proper VLAN tag.
2541                  */
2542                 memset(&fl, 0, sizeof(fl));
2543                 fl.fl4_dst = targets[i];
2544                 fl.fl4_tos = RTO_ONLINK;
2545
2546                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2547                 if (rv) {
2548                         if (net_ratelimit()) {
2549                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2550                                            bond->dev->name, &fl.fl4_dst);
2551                         }
2552                         continue;
2553                 }
2554
2555                 /*
2556                  * This target is not on a VLAN
2557                  */
2558                 if (rt->u.dst.dev == bond->dev) {
2559                         ip_rt_put(rt);
2560                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2561                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2562                                       bond->master_ip, 0);
2563                         continue;
2564                 }
2565
2566                 vlan_id = 0;
2567                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2568                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2569                         if (vlan_dev == rt->u.dst.dev) {
2570                                 vlan_id = vlan->vlan_id;
2571                                 pr_debug("basa: vlan match on %s %d\n",
2572                                        vlan_dev->name, vlan_id);
2573                                 break;
2574                         }
2575                 }
2576
2577                 if (vlan_id) {
2578                         ip_rt_put(rt);
2579                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2580                                       vlan->vlan_ip, vlan_id);
2581                         continue;
2582                 }
2583
2584                 if (net_ratelimit()) {
2585                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2586                                    bond->dev->name, &fl.fl4_dst,
2587                                    rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2588                 }
2589                 ip_rt_put(rt);
2590         }
2591 }
2592
2593 /*
2594  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2595  * for each VLAN above us.
2596  *
2597  * Caller must hold curr_slave_lock for read or better
2598  */
2599 static void bond_send_gratuitous_arp(struct bonding *bond)
2600 {
2601         struct slave *slave = bond->curr_active_slave;
2602         struct vlan_entry *vlan;
2603         struct net_device *vlan_dev;
2604
2605         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2606                  bond->dev->name, slave ? slave->dev->name : "NULL");
2607
2608         if (!slave || !bond->send_grat_arp ||
2609             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2610                 return;
2611
2612         bond->send_grat_arp--;
2613
2614         if (bond->master_ip) {
2615                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2616                                 bond->master_ip, 0);
2617         }
2618
2619         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2620                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2621                 if (vlan->vlan_ip) {
2622                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2623                                       vlan->vlan_ip, vlan->vlan_id);
2624                 }
2625         }
2626 }
2627
2628 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2629 {
2630         int i;
2631         __be32 *targets = bond->params.arp_targets;
2632
2633         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2634                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2635                          &sip, &tip, i, &targets[i],
2636                          bond_has_this_ip(bond, tip));
2637                 if (sip == targets[i]) {
2638                         if (bond_has_this_ip(bond, tip))
2639                                 slave->last_arp_rx = jiffies;
2640                         return;
2641                 }
2642         }
2643 }
2644
2645 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2646 {
2647         struct arphdr *arp;
2648         struct slave *slave;
2649         struct bonding *bond;
2650         unsigned char *arp_ptr;
2651         __be32 sip, tip;
2652
2653         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2654                 /*
2655                  * When using VLANS and bonding, dev and oriv_dev may be
2656                  * incorrect if the physical interface supports VLAN
2657                  * acceleration.  With this change ARP validation now
2658                  * works for hosts only reachable on the VLAN interface.
2659                  */
2660                 dev = vlan_dev_real_dev(dev);
2661                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2662         }
2663
2664         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2665                 goto out;
2666
2667         bond = netdev_priv(dev);
2668         read_lock(&bond->lock);
2669
2670         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2671                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2672                  orig_dev ? orig_dev->name : "NULL");
2673
2674         slave = bond_get_slave_by_dev(bond, orig_dev);
2675         if (!slave || !slave_do_arp_validate(bond, slave))
2676                 goto out_unlock;
2677
2678         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2679                 goto out_unlock;
2680
2681         arp = arp_hdr(skb);
2682         if (arp->ar_hln != dev->addr_len ||
2683             skb->pkt_type == PACKET_OTHERHOST ||
2684             skb->pkt_type == PACKET_LOOPBACK ||
2685             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2686             arp->ar_pro != htons(ETH_P_IP) ||
2687             arp->ar_pln != 4)
2688                 goto out_unlock;
2689
2690         arp_ptr = (unsigned char *)(arp + 1);
2691         arp_ptr += dev->addr_len;
2692         memcpy(&sip, arp_ptr, 4);
2693         arp_ptr += 4 + dev->addr_len;
2694         memcpy(&tip, arp_ptr, 4);
2695
2696         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2697                  bond->dev->name, slave->dev->name, slave->state,
2698                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2699                  &sip, &tip);
2700
2701         /*
2702          * Backup slaves won't see the ARP reply, but do come through
2703          * here for each ARP probe (so we swap the sip/tip to validate
2704          * the probe).  In a "redundant switch, common router" type of
2705          * configuration, the ARP probe will (hopefully) travel from
2706          * the active, through one switch, the router, then the other
2707          * switch before reaching the backup.
2708          */
2709         if (slave->state == BOND_STATE_ACTIVE)
2710                 bond_validate_arp(bond, slave, sip, tip);
2711         else
2712                 bond_validate_arp(bond, slave, tip, sip);
2713
2714 out_unlock:
2715         read_unlock(&bond->lock);
2716 out:
2717         dev_kfree_skb(skb);
2718         return NET_RX_SUCCESS;
2719 }
2720
2721 /*
2722  * this function is called regularly to monitor each slave's link
2723  * ensuring that traffic is being sent and received when arp monitoring
2724  * is used in load-balancing mode. if the adapter has been dormant, then an
2725  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2726  * arp monitoring in active backup mode.
2727  */
2728 void bond_loadbalance_arp_mon(struct work_struct *work)
2729 {
2730         struct bonding *bond = container_of(work, struct bonding,
2731                                             arp_work.work);
2732         struct slave *slave, *oldcurrent;
2733         int do_failover = 0;
2734         int delta_in_ticks;
2735         int i;
2736
2737         read_lock(&bond->lock);
2738
2739         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2740
2741         if (bond->kill_timers)
2742                 goto out;
2743
2744         if (bond->slave_cnt == 0)
2745                 goto re_arm;
2746
2747         read_lock(&bond->curr_slave_lock);
2748         oldcurrent = bond->curr_active_slave;
2749         read_unlock(&bond->curr_slave_lock);
2750
2751         /* see if any of the previous devices are up now (i.e. they have
2752          * xmt and rcv traffic). the curr_active_slave does not come into
2753          * the picture unless it is null. also, slave->jiffies is not needed
2754          * here because we send an arp on each slave and give a slave as
2755          * long as it needs to get the tx/rx within the delta.
2756          * TODO: what about up/down delay in arp mode? it wasn't here before
2757          *       so it can wait
2758          */
2759         bond_for_each_slave(bond, slave, i) {
2760                 if (slave->link != BOND_LINK_UP) {
2761                         if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2762                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2763
2764                                 slave->link  = BOND_LINK_UP;
2765                                 slave->state = BOND_STATE_ACTIVE;
2766
2767                                 /* primary_slave has no meaning in round-robin
2768                                  * mode. the window of a slave being up and
2769                                  * curr_active_slave being null after enslaving
2770                                  * is closed.
2771                                  */
2772                                 if (!oldcurrent) {
2773                                         pr_info("%s: link status definitely up for interface %s, ",
2774                                                 bond->dev->name,
2775                                                 slave->dev->name);
2776                                         do_failover = 1;
2777                                 } else {
2778                                         pr_info("%s: interface %s is now up\n",
2779                                                 bond->dev->name,
2780                                                 slave->dev->name);
2781                                 }
2782                         }
2783                 } else {
2784                         /* slave->link == BOND_LINK_UP */
2785
2786                         /* not all switches will respond to an arp request
2787                          * when the source ip is 0, so don't take the link down
2788                          * if we don't know our ip yet
2789                          */
2790                         if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2791                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2792
2793                                 slave->link  = BOND_LINK_DOWN;
2794                                 slave->state = BOND_STATE_BACKUP;
2795
2796                                 if (slave->link_failure_count < UINT_MAX)
2797                                         slave->link_failure_count++;
2798
2799                                 pr_info("%s: interface %s is now down.\n",
2800                                         bond->dev->name,
2801                                         slave->dev->name);
2802
2803                                 if (slave == oldcurrent)
2804                                         do_failover = 1;
2805                         }
2806                 }
2807
2808                 /* note: if switch is in round-robin mode, all links
2809                  * must tx arp to ensure all links rx an arp - otherwise
2810                  * links may oscillate or not come up at all; if switch is
2811                  * in something like xor mode, there is nothing we can
2812                  * do - all replies will be rx'ed on same link causing slaves
2813                  * to be unstable during low/no traffic periods
2814                  */
2815                 if (IS_UP(slave->dev))
2816                         bond_arp_send_all(bond, slave);
2817         }
2818
2819         if (do_failover) {
2820                 write_lock_bh(&bond->curr_slave_lock);
2821
2822                 bond_select_active_slave(bond);
2823
2824                 write_unlock_bh(&bond->curr_slave_lock);
2825         }
2826
2827 re_arm:
2828         if (bond->params.arp_interval)
2829                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2830 out:
2831         read_unlock(&bond->lock);
2832 }
2833
2834 /*
2835  * Called to inspect slaves for active-backup mode ARP monitor link state
2836  * changes.  Sets new_link in slaves to specify what action should take
2837  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2838  * to link states must be committed.
2839  *
2840  * Called with bond->lock held for read.
2841  */
2842 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2843 {
2844         struct slave *slave;
2845         int i, commit = 0;
2846
2847         bond_for_each_slave(bond, slave, i) {
2848                 slave->new_link = BOND_LINK_NOCHANGE;
2849
2850                 if (slave->link != BOND_LINK_UP) {
2851                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2852                                            delta_in_ticks)) {
2853                                 slave->new_link = BOND_LINK_UP;
2854                                 commit++;
2855                         }
2856
2857                         continue;
2858                 }
2859
2860                 /*
2861                  * Give slaves 2*delta after being enslaved or made
2862                  * active.  This avoids bouncing, as the last receive
2863                  * times need a full ARP monitor cycle to be updated.
2864                  */
2865                 if (!time_after_eq(jiffies, slave->jiffies +
2866                                    2 * delta_in_ticks))
2867                         continue;
2868
2869                 /*
2870                  * Backup slave is down if:
2871                  * - No current_arp_slave AND
2872                  * - more than 3*delta since last receive AND
2873                  * - the bond has an IP address
2874                  *
2875                  * Note: a non-null current_arp_slave indicates
2876                  * the curr_active_slave went down and we are
2877                  * searching for a new one; under this condition
2878                  * we only take the curr_active_slave down - this
2879                  * gives each slave a chance to tx/rx traffic
2880                  * before being taken out
2881                  */
2882                 if (slave->state == BOND_STATE_BACKUP &&
2883                     !bond->current_arp_slave &&
2884                     time_after(jiffies, slave_last_rx(bond, slave) +
2885                                3 * delta_in_ticks)) {
2886                         slave->new_link = BOND_LINK_DOWN;
2887                         commit++;
2888                 }
2889
2890                 /*
2891                  * Active slave is down if:
2892                  * - more than 2*delta since transmitting OR
2893                  * - (more than 2*delta since receive AND
2894                  *    the bond has an IP address)
2895                  */
2896                 if ((slave->state == BOND_STATE_ACTIVE) &&
2897                     (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2898                                     2 * delta_in_ticks) ||
2899                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2900                                      + 2 * delta_in_ticks)))) {
2901                         slave->new_link = BOND_LINK_DOWN;
2902                         commit++;
2903                 }
2904         }
2905
2906         return commit;
2907 }
2908
2909 /*
2910  * Called to commit link state changes noted by inspection step of
2911  * active-backup mode ARP monitor.
2912  *
2913  * Called with RTNL and bond->lock for read.
2914  */
2915 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2916 {
2917         struct slave *slave;
2918         int i;
2919
2920         bond_for_each_slave(bond, slave, i) {
2921                 switch (slave->new_link) {
2922                 case BOND_LINK_NOCHANGE:
2923                         continue;
2924
2925                 case BOND_LINK_UP:
2926                         if ((!bond->curr_active_slave &&
2927                              time_before_eq(jiffies,
2928                                             dev_trans_start(slave->dev) +
2929                                             delta_in_ticks)) ||
2930                             bond->curr_active_slave != slave) {
2931                                 slave->link = BOND_LINK_UP;
2932                                 bond->current_arp_slave = NULL;
2933
2934                                 pr_info("%s: link status definitely up for interface %s.\n",
2935                                         bond->dev->name, slave->dev->name);
2936
2937                                 if (!bond->curr_active_slave ||
2938                                     (slave == bond->primary_slave))
2939                                         goto do_failover;
2940
2941                         }
2942
2943                         continue;
2944
2945                 case BOND_LINK_DOWN:
2946                         if (slave->link_failure_count < UINT_MAX)
2947                                 slave->link_failure_count++;
2948
2949                         slave->link = BOND_LINK_DOWN;
2950                         bond_set_slave_inactive_flags(slave);
2951
2952                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2953                                 bond->dev->name, slave->dev->name);
2954
2955                         if (slave == bond->curr_active_slave) {
2956                                 bond->current_arp_slave = NULL;
2957                                 goto do_failover;
2958                         }
2959
2960                         continue;
2961
2962                 default:
2963                         pr_err("%s: impossible: new_link %d on slave %s\n",
2964                                bond->dev->name, slave->new_link,
2965                                slave->dev->name);
2966                         continue;
2967                 }
2968
2969 do_failover:
2970                 ASSERT_RTNL();
2971                 write_lock_bh(&bond->curr_slave_lock);
2972                 bond_select_active_slave(bond);
2973                 write_unlock_bh(&bond->curr_slave_lock);
2974         }
2975
2976         bond_set_carrier(bond);
2977 }
2978
2979 /*
2980  * Send ARP probes for active-backup mode ARP monitor.
2981  *
2982  * Called with bond->lock held for read.
2983  */
2984 static void bond_ab_arp_probe(struct bonding *bond)
2985 {
2986         struct slave *slave;
2987         int i;
2988
2989         read_lock(&bond->curr_slave_lock);
2990
2991         if (bond->current_arp_slave && bond->curr_active_slave)
2992                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2993                         bond->current_arp_slave->dev->name,
2994                         bond->curr_active_slave->dev->name);
2995
2996         if (bond->curr_active_slave) {
2997                 bond_arp_send_all(bond, bond->curr_active_slave);
2998                 read_unlock(&bond->curr_slave_lock);
2999                 return;
3000         }
3001
3002         read_unlock(&bond->curr_slave_lock);
3003
3004         /* if we don't have a curr_active_slave, search for the next available
3005          * backup slave from the current_arp_slave and make it the candidate
3006          * for becoming the curr_active_slave
3007          */
3008
3009         if (!bond->current_arp_slave) {
3010                 bond->current_arp_slave = bond->first_slave;
3011                 if (!bond->current_arp_slave)
3012                         return;
3013         }
3014
3015         bond_set_slave_inactive_flags(bond->current_arp_slave);
3016
3017         /* search for next candidate */
3018         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3019                 if (IS_UP(slave->dev)) {
3020                         slave->link = BOND_LINK_BACK;
3021                         bond_set_slave_active_flags(slave);
3022                         bond_arp_send_all(bond, slave);
3023                         slave->jiffies = jiffies;
3024                         bond->current_arp_slave = slave;
3025                         break;
3026                 }
3027
3028                 /* if the link state is up at this point, we
3029                  * mark it down - this can happen if we have
3030                  * simultaneous link failures and
3031                  * reselect_active_interface doesn't make this
3032                  * one the current slave so it is still marked
3033                  * up when it is actually down
3034                  */
3035                 if (slave->link == BOND_LINK_UP) {
3036                         slave->link = BOND_LINK_DOWN;
3037                         if (slave->link_failure_count < UINT_MAX)
3038                                 slave->link_failure_count++;
3039
3040                         bond_set_slave_inactive_flags(slave);
3041
3042                         pr_info("%s: backup interface %s is now down.\n",
3043                                 bond->dev->name, slave->dev->name);
3044                 }
3045         }
3046 }
3047
3048 void bond_activebackup_arp_mon(struct work_struct *work)
3049 {
3050         struct bonding *bond = container_of(work, struct bonding,
3051                                             arp_work.work);
3052         int delta_in_ticks;
3053
3054         read_lock(&bond->lock);
3055
3056         if (bond->kill_timers)
3057                 goto out;
3058
3059         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3060
3061         if (bond->slave_cnt == 0)
3062                 goto re_arm;
3063
3064         if (bond->send_grat_arp) {
3065                 read_lock(&bond->curr_slave_lock);
3066                 bond_send_gratuitous_arp(bond);
3067                 read_unlock(&bond->curr_slave_lock);
3068         }
3069
3070         if (bond->send_unsol_na) {
3071                 read_lock(&bond->curr_slave_lock);
3072                 bond_send_unsolicited_na(bond);
3073                 read_unlock(&bond->curr_slave_lock);
3074         }
3075
3076         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3077                 read_unlock(&bond->lock);
3078                 rtnl_lock();
3079                 read_lock(&bond->lock);
3080
3081                 bond_ab_arp_commit(bond, delta_in_ticks);
3082
3083                 read_unlock(&bond->lock);
3084                 rtnl_unlock();
3085                 read_lock(&bond->lock);
3086         }
3087
3088         bond_ab_arp_probe(bond);
3089
3090 re_arm:
3091         if (bond->params.arp_interval)
3092                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3093 out:
3094         read_unlock(&bond->lock);
3095 }
3096
3097 /*------------------------------ proc/seq_file-------------------------------*/
3098
3099 #ifdef CONFIG_PROC_FS
3100
3101 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3102         __acquires(&dev_base_lock)
3103         __acquires(&bond->lock)
3104 {
3105         struct bonding *bond = seq->private;
3106         loff_t off = 0;
3107         struct slave *slave;
3108         int i;
3109
3110         /* make sure the bond won't be taken away */
3111         read_lock(&dev_base_lock);
3112         read_lock(&bond->lock);
3113
3114         if (*pos == 0)
3115                 return SEQ_START_TOKEN;
3116
3117         bond_for_each_slave(bond, slave, i) {
3118                 if (++off == *pos)
3119                         return slave;
3120         }
3121
3122         return NULL;
3123 }
3124
3125 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3126 {
3127         struct bonding *bond = seq->private;
3128         struct slave *slave = v;
3129
3130         ++*pos;
3131         if (v == SEQ_START_TOKEN)
3132                 return bond->first_slave;
3133
3134         slave = slave->next;
3135
3136         return (slave == bond->first_slave) ? NULL : slave;
3137 }
3138
3139 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3140         __releases(&bond->lock)
3141         __releases(&dev_base_lock)
3142 {
3143         struct bonding *bond = seq->private;
3144
3145         read_unlock(&bond->lock);
3146         read_unlock(&dev_base_lock);
3147 }
3148
3149 static void bond_info_show_master(struct seq_file *seq)
3150 {
3151         struct bonding *bond = seq->private;
3152         struct slave *curr;
3153         int i;
3154
3155         read_lock(&bond->curr_slave_lock);
3156         curr = bond->curr_active_slave;
3157         read_unlock(&bond->curr_slave_lock);
3158
3159         seq_printf(seq, "Bonding Mode: %s",
3160                    bond_mode_name(bond->params.mode));
3161
3162         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3163             bond->params.fail_over_mac)
3164                 seq_printf(seq, " (fail_over_mac %s)",
3165                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3166
3167         seq_printf(seq, "\n");
3168
3169         if (bond->params.mode == BOND_MODE_XOR ||
3170                 bond->params.mode == BOND_MODE_8023AD) {
3171                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3172                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3173                         bond->params.xmit_policy);
3174         }
3175
3176         if (USES_PRIMARY(bond->params.mode)) {
3177                 seq_printf(seq, "Primary Slave: %s",
3178                            (bond->primary_slave) ?
3179                            bond->primary_slave->dev->name : "None");
3180                 if (bond->primary_slave)
3181                         seq_printf(seq, " (primary_reselect %s)",
3182                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3183
3184                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3185                            (curr) ? curr->dev->name : "None");
3186         }
3187
3188         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3189                    "up" : "down");
3190         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3191         seq_printf(seq, "Up Delay (ms): %d\n",
3192                    bond->params.updelay * bond->params.miimon);
3193         seq_printf(seq, "Down Delay (ms): %d\n",
3194                    bond->params.downdelay * bond->params.miimon);
3195
3196
3197         /* ARP information */
3198         if (bond->params.arp_interval > 0) {
3199                 int printed = 0;
3200                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3201                                 bond->params.arp_interval);
3202
3203                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3204
3205                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3206                         if (!bond->params.arp_targets[i])
3207                                 break;
3208                         if (printed)
3209                                 seq_printf(seq, ",");
3210                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3211                         printed = 1;
3212                 }
3213                 seq_printf(seq, "\n");
3214         }
3215
3216         if (bond->params.mode == BOND_MODE_8023AD) {
3217                 struct ad_info ad_info;
3218
3219                 seq_puts(seq, "\n802.3ad info\n");
3220                 seq_printf(seq, "LACP rate: %s\n",
3221                            (bond->params.lacp_fast) ? "fast" : "slow");
3222                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3223                            ad_select_tbl[bond->params.ad_select].modename);
3224
3225                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3226                         seq_printf(seq, "bond %s has no active aggregator\n",
3227                                    bond->dev->name);
3228                 } else {
3229                         seq_printf(seq, "Active Aggregator Info:\n");
3230
3231                         seq_printf(seq, "\tAggregator ID: %d\n",
3232                                    ad_info.aggregator_id);
3233                         seq_printf(seq, "\tNumber of ports: %d\n",
3234                                    ad_info.ports);
3235                         seq_printf(seq, "\tActor Key: %d\n",
3236                                    ad_info.actor_key);
3237                         seq_printf(seq, "\tPartner Key: %d\n",
3238                                    ad_info.partner_key);
3239                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3240                                    ad_info.partner_system);
3241                 }
3242         }
3243 }
3244
3245 static void bond_info_show_slave(struct seq_file *seq,
3246                                  const struct slave *slave)
3247 {
3248         struct bonding *bond = seq->private;
3249
3250         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3251         seq_printf(seq, "MII Status: %s\n",
3252                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3253         seq_printf(seq, "Link Failure Count: %u\n",
3254                    slave->link_failure_count);
3255
3256         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3257
3258         if (bond->params.mode == BOND_MODE_8023AD) {
3259                 const struct aggregator *agg
3260                         = SLAVE_AD_INFO(slave).port.aggregator;
3261
3262                 if (agg)
3263                         seq_printf(seq, "Aggregator ID: %d\n",
3264                                    agg->aggregator_identifier);
3265                 else
3266                         seq_puts(seq, "Aggregator ID: N/A\n");
3267         }
3268 }
3269
3270 static int bond_info_seq_show(struct seq_file *seq, void *v)
3271 {
3272         if (v == SEQ_START_TOKEN) {
3273                 seq_printf(seq, "%s\n", version);
3274                 bond_info_show_master(seq);
3275         } else
3276                 bond_info_show_slave(seq, v);
3277
3278         return 0;
3279 }
3280
3281 static const struct seq_operations bond_info_seq_ops = {
3282         .start = bond_info_seq_start,
3283         .next  = bond_info_seq_next,
3284         .stop  = bond_info_seq_stop,
3285         .show  = bond_info_seq_show,
3286 };
3287
3288 static int bond_info_open(struct inode *inode, struct file *file)
3289 {
3290         struct seq_file *seq;
3291         struct proc_dir_entry *proc;
3292         int res;
3293
3294         res = seq_open(file, &bond_info_seq_ops);
3295         if (!res) {
3296                 /* recover the pointer buried in proc_dir_entry data */
3297                 seq = file->private_data;
3298                 proc = PDE(inode);
3299                 seq->private = proc->data;
3300         }
3301
3302         return res;
3303 }
3304
3305 static const struct file_operations bond_info_fops = {
3306         .owner   = THIS_MODULE,
3307         .open    = bond_info_open,
3308         .read    = seq_read,
3309         .llseek  = seq_lseek,
3310         .release = seq_release,
3311 };
3312
3313 static void bond_create_proc_entry(struct bonding *bond)
3314 {
3315         struct net_device *bond_dev = bond->dev;
3316         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3317
3318         if (bn->proc_dir) {
3319                 bond->proc_entry = proc_create_data(bond_dev->name,
3320                                                     S_IRUGO, bn->proc_dir,
3321                                                     &bond_info_fops, bond);
3322                 if (bond->proc_entry == NULL)
3323                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3324                                    DRV_NAME, bond_dev->name);
3325                 else
3326                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3327         }
3328 }
3329
3330 static void bond_remove_proc_entry(struct bonding *bond)
3331 {
3332         struct net_device *bond_dev = bond->dev;
3333         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3334
3335         if (bn->proc_dir && bond->proc_entry) {
3336                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3337                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3338                 bond->proc_entry = NULL;
3339         }
3340 }
3341
3342 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3343  * Caller must hold rtnl_lock.
3344  */
3345 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3346 {
3347         if (!bn->proc_dir) {
3348                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3349                 if (!bn->proc_dir)
3350                         pr_warning("Warning: cannot create /proc/net/%s\n",
3351                                    DRV_NAME);
3352         }
3353 }
3354
3355 /* Destroy the bonding directory under /proc/net, if empty.
3356  * Caller must hold rtnl_lock.
3357  */
3358 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3359 {
3360         if (bn->proc_dir) {
3361                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3362                 bn->proc_dir = NULL;
3363         }
3364 }
3365
3366 #else /* !CONFIG_PROC_FS */
3367
3368 static void bond_create_proc_entry(struct bonding *bond)
3369 {
3370 }
3371
3372 static void bond_remove_proc_entry(struct bonding *bond)
3373 {
3374 }
3375
3376 static inline void bond_create_proc_dir(struct bond_net *bn)
3377 {
3378 }
3379
3380 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3381 {
3382 }
3383
3384 #endif /* CONFIG_PROC_FS */
3385
3386
3387 /*-------------------------- netdev event handling --------------------------*/
3388
3389 /*
3390  * Change device name
3391  */
3392 static int bond_event_changename(struct bonding *bond)
3393 {
3394         bond_remove_proc_entry(bond);
3395         bond_create_proc_entry(bond);
3396
3397         return NOTIFY_DONE;
3398 }
3399
3400 static int bond_master_netdev_event(unsigned long event,
3401                                     struct net_device *bond_dev)
3402 {
3403         struct bonding *event_bond = netdev_priv(bond_dev);
3404
3405         switch (event) {
3406         case NETDEV_CHANGENAME:
3407                 return bond_event_changename(event_bond);
3408         default:
3409                 break;
3410         }
3411
3412         return NOTIFY_DONE;
3413 }
3414
3415 static int bond_slave_netdev_event(unsigned long event,
3416                                    struct net_device *slave_dev)
3417 {
3418         struct net_device *bond_dev = slave_dev->master;
3419         struct bonding *bond = netdev_priv(bond_dev);
3420
3421         switch (event) {
3422         case NETDEV_UNREGISTER:
3423                 if (bond_dev) {
3424                         if (bond->setup_by_slave)
3425                                 bond_release_and_destroy(bond_dev, slave_dev);
3426                         else
3427                                 bond_release(bond_dev, slave_dev);
3428                 }
3429                 break;
3430         case NETDEV_CHANGE:
3431                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3432                         struct slave *slave;
3433
3434                         slave = bond_get_slave_by_dev(bond, slave_dev);
3435                         if (slave) {
3436                                 u16 old_speed = slave->speed;
3437                                 u16 old_duplex = slave->duplex;
3438
3439                                 bond_update_speed_duplex(slave);
3440
3441                                 if (bond_is_lb(bond))
3442                                         break;
3443
3444                                 if (old_speed != slave->speed)
3445                                         bond_3ad_adapter_speed_changed(slave);
3446                                 if (old_duplex != slave->duplex)
3447                                         bond_3ad_adapter_duplex_changed(slave);
3448                         }
3449                 }
3450
3451                 break;
3452         case NETDEV_DOWN:
3453                 /*
3454                  * ... Or is it this?
3455                  */
3456                 break;
3457         case NETDEV_CHANGEMTU:
3458                 /*
3459                  * TODO: Should slaves be allowed to
3460                  * independently alter their MTU?  For
3461                  * an active-backup bond, slaves need
3462                  * not be the same type of device, so
3463                  * MTUs may vary.  For other modes,
3464                  * slaves arguably should have the
3465                  * same MTUs. To do this, we'd need to
3466                  * take over the slave's change_mtu
3467                  * function for the duration of their
3468                  * servitude.
3469                  */
3470                 break;
3471         case NETDEV_CHANGENAME:
3472                 /*
3473                  * TODO: handle changing the primary's name
3474                  */
3475                 break;
3476         case NETDEV_FEAT_CHANGE:
3477                 bond_compute_features(bond);
3478                 break;
3479         default:
3480                 break;
3481         }
3482
3483         return NOTIFY_DONE;
3484 }
3485
3486 /*
3487  * bond_netdev_event: handle netdev notifier chain events.
3488  *
3489  * This function receives events for the netdev chain.  The caller (an
3490  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3491  * locks for us to safely manipulate the slave devices (RTNL lock,
3492  * dev_probe_lock).
3493  */
3494 static int bond_netdev_event(struct notifier_block *this,
3495                              unsigned long event, void *ptr)
3496 {
3497         struct net_device *event_dev = (struct net_device *)ptr;
3498
3499         pr_debug("event_dev: %s, event: %lx\n",
3500                  event_dev ? event_dev->name : "None",
3501                  event);
3502
3503         if (!(event_dev->priv_flags & IFF_BONDING))
3504                 return NOTIFY_DONE;
3505
3506         if (event_dev->flags & IFF_MASTER) {
3507                 pr_debug("IFF_MASTER\n");
3508                 return bond_master_netdev_event(event, event_dev);
3509         }
3510
3511         if (event_dev->flags & IFF_SLAVE) {
3512                 pr_debug("IFF_SLAVE\n");
3513                 return bond_slave_netdev_event(event, event_dev);
3514         }
3515
3516         return NOTIFY_DONE;
3517 }
3518
3519 /*
3520  * bond_inetaddr_event: handle inetaddr notifier chain events.
3521  *
3522  * We keep track of device IPs primarily to use as source addresses in
3523  * ARP monitor probes (rather than spewing out broadcasts all the time).
3524  *
3525  * We track one IP for the main device (if it has one), plus one per VLAN.
3526  */
3527 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3528 {
3529         struct in_ifaddr *ifa = ptr;
3530         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3531         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3532         struct bonding *bond;
3533         struct vlan_entry *vlan;
3534
3535         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3536                 if (bond->dev == event_dev) {
3537                         switch (event) {
3538                         case NETDEV_UP:
3539                                 bond->master_ip = ifa->ifa_local;
3540                                 return NOTIFY_OK;
3541                         case NETDEV_DOWN:
3542                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3543                                 return NOTIFY_OK;
3544                         default:
3545                                 return NOTIFY_DONE;
3546                         }
3547                 }
3548
3549                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3550                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3551                         if (vlan_dev == event_dev) {
3552                                 switch (event) {
3553                                 case NETDEV_UP:
3554                                         vlan->vlan_ip = ifa->ifa_local;
3555                                         return NOTIFY_OK;
3556                                 case NETDEV_DOWN:
3557                                         vlan->vlan_ip =
3558                                                 bond_glean_dev_ip(vlan_dev);
3559                                         return NOTIFY_OK;
3560                                 default:
3561                                         return NOTIFY_DONE;
3562                                 }
3563                         }
3564                 }
3565         }
3566         return NOTIFY_DONE;
3567 }
3568
3569 static struct notifier_block bond_netdev_notifier = {
3570         .notifier_call = bond_netdev_event,
3571 };
3572
3573 static struct notifier_block bond_inetaddr_notifier = {
3574         .notifier_call = bond_inetaddr_event,
3575 };
3576
3577 /*-------------------------- Packet type handling ---------------------------*/
3578
3579 /* register to receive lacpdus on a bond */
3580 static void bond_register_lacpdu(struct bonding *bond)
3581 {
3582         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3583
3584         /* initialize packet type */
3585         pk_type->type = PKT_TYPE_LACPDU;
3586         pk_type->dev = bond->dev;
3587         pk_type->func = bond_3ad_lacpdu_recv;
3588
3589         dev_add_pack(pk_type);
3590 }
3591
3592 /* unregister to receive lacpdus on a bond */
3593 static void bond_unregister_lacpdu(struct bonding *bond)
3594 {
3595         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3596 }
3597
3598 void bond_register_arp(struct bonding *bond)
3599 {
3600         struct packet_type *pt = &bond->arp_mon_pt;
3601
3602         if (pt->type)
3603                 return;
3604
3605         pt->type = htons(ETH_P_ARP);
3606         pt->dev = bond->dev;
3607         pt->func = bond_arp_rcv;
3608         dev_add_pack(pt);
3609 }
3610
3611 void bond_unregister_arp(struct bonding *bond)
3612 {
3613         struct packet_type *pt = &bond->arp_mon_pt;
3614
3615         dev_remove_pack(pt);
3616         pt->type = 0;
3617 }
3618
3619 /*---------------------------- Hashing Policies -----------------------------*/
3620
3621 /*
3622  * Hash for the output device based upon layer 2 and layer 3 data. If
3623  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3624  */
3625 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3626 {
3627         struct ethhdr *data = (struct ethhdr *)skb->data;
3628         struct iphdr *iph = ip_hdr(skb);
3629
3630         if (skb->protocol == htons(ETH_P_IP)) {
3631                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3632                         (data->h_dest[5] ^ data->h_source[5])) % count;
3633         }
3634
3635         return (data->h_dest[5] ^ data->h_source[5]) % count;
3636 }
3637
3638 /*
3639  * Hash for the output device based upon layer 3 and layer 4 data. If
3640  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3641  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3642  */
3643 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3644 {
3645         struct ethhdr *data = (struct ethhdr *)skb->data;
3646         struct iphdr *iph = ip_hdr(skb);
3647         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3648         int layer4_xor = 0;
3649
3650         if (skb->protocol == htons(ETH_P_IP)) {
3651                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3652                     (iph->protocol == IPPROTO_TCP ||
3653                      iph->protocol == IPPROTO_UDP)) {
3654                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3655                 }
3656                 return (layer4_xor ^
3657                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3658
3659         }
3660
3661         return (data->h_dest[5] ^ data->h_source[5]) % count;
3662 }
3663
3664 /*
3665  * Hash for the output device based upon layer 2 data
3666  */
3667 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3668 {
3669         struct ethhdr *data = (struct ethhdr *)skb->data;
3670
3671         return (data->h_dest[5] ^ data->h_source[5]) % count;
3672 }
3673
3674 /*-------------------------- Device entry points ----------------------------*/
3675
3676 static int bond_open(struct net_device *bond_dev)
3677 {
3678         struct bonding *bond = netdev_priv(bond_dev);
3679
3680         bond->kill_timers = 0;
3681
3682         if (bond_is_lb(bond)) {
3683                 /* bond_alb_initialize must be called before the timer
3684                  * is started.
3685                  */
3686                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3687                         /* something went wrong - fail the open operation */
3688                         return -ENOMEM;
3689                 }
3690
3691                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3692                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3693         }
3694
3695         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3696                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3697                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3698         }
3699
3700         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3701                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3702                         INIT_DELAYED_WORK(&bond->arp_work,
3703                                           bond_activebackup_arp_mon);
3704                 else
3705                         INIT_DELAYED_WORK(&bond->arp_work,
3706                                           bond_loadbalance_arp_mon);
3707
3708                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3709                 if (bond->params.arp_validate)
3710                         bond_register_arp(bond);
3711         }
3712
3713         if (bond->params.mode == BOND_MODE_8023AD) {
3714                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3715                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3716                 /* register to receive LACPDUs */
3717                 bond_register_lacpdu(bond);
3718                 bond_3ad_initiate_agg_selection(bond, 1);
3719         }
3720
3721         return 0;
3722 }
3723
3724 static int bond_close(struct net_device *bond_dev)
3725 {
3726         struct bonding *bond = netdev_priv(bond_dev);
3727
3728         if (bond->params.mode == BOND_MODE_8023AD) {
3729                 /* Unregister the receive of LACPDUs */
3730                 bond_unregister_lacpdu(bond);
3731         }
3732
3733         if (bond->params.arp_validate)
3734                 bond_unregister_arp(bond);
3735
3736         write_lock_bh(&bond->lock);
3737
3738         bond->send_grat_arp = 0;
3739         bond->send_unsol_na = 0;
3740
3741         /* signal timers not to re-arm */
3742         bond->kill_timers = 1;
3743
3744         write_unlock_bh(&bond->lock);
3745
3746         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3747                 cancel_delayed_work(&bond->mii_work);
3748         }
3749
3750         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3751                 cancel_delayed_work(&bond->arp_work);
3752         }
3753
3754         switch (bond->params.mode) {
3755         case BOND_MODE_8023AD:
3756                 cancel_delayed_work(&bond->ad_work);
3757                 break;
3758         case BOND_MODE_TLB:
3759         case BOND_MODE_ALB:
3760                 cancel_delayed_work(&bond->alb_work);
3761                 break;
3762         default:
3763                 break;
3764         }
3765
3766
3767         if (bond_is_lb(bond)) {
3768                 /* Must be called only after all
3769                  * slaves have been released
3770                  */
3771                 bond_alb_deinitialize(bond);
3772         }
3773
3774         return 0;
3775 }
3776
3777 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3778 {
3779         struct bonding *bond = netdev_priv(bond_dev);
3780         struct net_device_stats *stats = &bond_dev->stats;
3781         struct net_device_stats local_stats;
3782         struct slave *slave;
3783         int i;
3784
3785         memset(&local_stats, 0, sizeof(struct net_device_stats));
3786
3787         read_lock_bh(&bond->lock);
3788
3789         bond_for_each_slave(bond, slave, i) {
3790                 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3791
3792                 local_stats.rx_packets += sstats->rx_packets;
3793                 local_stats.rx_bytes += sstats->rx_bytes;
3794                 local_stats.rx_errors += sstats->rx_errors;
3795                 local_stats.rx_dropped += sstats->rx_dropped;
3796
3797                 local_stats.tx_packets += sstats->tx_packets;
3798                 local_stats.tx_bytes += sstats->tx_bytes;
3799                 local_stats.tx_errors += sstats->tx_errors;
3800                 local_stats.tx_dropped += sstats->tx_dropped;
3801
3802                 local_stats.multicast += sstats->multicast;
3803                 local_stats.collisions += sstats->collisions;
3804
3805                 local_stats.rx_length_errors += sstats->rx_length_errors;
3806                 local_stats.rx_over_errors += sstats->rx_over_errors;
3807                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3808                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3809                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3810                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3811
3812                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3813                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3814                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3815                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3816                 local_stats.tx_window_errors += sstats->tx_window_errors;
3817         }
3818
3819         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3820
3821         read_unlock_bh(&bond->lock);
3822
3823         return stats;
3824 }
3825
3826 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3827 {
3828         struct net_device *slave_dev = NULL;
3829         struct ifbond k_binfo;
3830         struct ifbond __user *u_binfo = NULL;
3831         struct ifslave k_sinfo;
3832         struct ifslave __user *u_sinfo = NULL;
3833         struct mii_ioctl_data *mii = NULL;
3834         int res = 0;
3835
3836         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3837
3838         switch (cmd) {
3839         case SIOCGMIIPHY:
3840                 mii = if_mii(ifr);
3841                 if (!mii)
3842                         return -EINVAL;
3843
3844                 mii->phy_id = 0;
3845                 /* Fall Through */
3846         case SIOCGMIIREG:
3847                 /*
3848                  * We do this again just in case we were called by SIOCGMIIREG
3849                  * instead of SIOCGMIIPHY.
3850                  */
3851                 mii = if_mii(ifr);
3852                 if (!mii)
3853                         return -EINVAL;
3854
3855
3856                 if (mii->reg_num == 1) {
3857                         struct bonding *bond = netdev_priv(bond_dev);
3858                         mii->val_out = 0;
3859                         read_lock(&bond->lock);
3860                         read_lock(&bond->curr_slave_lock);
3861                         if (netif_carrier_ok(bond->dev))
3862                                 mii->val_out = BMSR_LSTATUS;
3863
3864                         read_unlock(&bond->curr_slave_lock);
3865                         read_unlock(&bond->lock);
3866                 }
3867
3868                 return 0;
3869         case BOND_INFO_QUERY_OLD:
3870         case SIOCBONDINFOQUERY:
3871                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3872
3873                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3874                         return -EFAULT;
3875
3876                 res = bond_info_query(bond_dev, &k_binfo);
3877                 if (res == 0 &&
3878                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3879                         return -EFAULT;
3880
3881                 return res;
3882         case BOND_SLAVE_INFO_QUERY_OLD:
3883         case SIOCBONDSLAVEINFOQUERY:
3884                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3885
3886                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3887                         return -EFAULT;
3888
3889                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3890                 if (res == 0 &&
3891                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3892                         return -EFAULT;
3893
3894                 return res;
3895         default:
3896                 /* Go on */
3897                 break;
3898         }
3899
3900         if (!capable(CAP_NET_ADMIN))
3901                 return -EPERM;
3902
3903         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3904
3905         pr_debug("slave_dev=%p:\n", slave_dev);
3906
3907         if (!slave_dev)
3908                 res = -ENODEV;
3909         else {
3910                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3911                 switch (cmd) {
3912                 case BOND_ENSLAVE_OLD:
3913                 case SIOCBONDENSLAVE:
3914                         res = bond_enslave(bond_dev, slave_dev);
3915                         break;
3916                 case BOND_RELEASE_OLD:
3917                 case SIOCBONDRELEASE:
3918                         res = bond_release(bond_dev, slave_dev);
3919                         break;
3920                 case BOND_SETHWADDR_OLD:
3921                 case SIOCBONDSETHWADDR:
3922                         res = bond_sethwaddr(bond_dev, slave_dev);
3923                         break;
3924                 case BOND_CHANGE_ACTIVE_OLD:
3925                 case SIOCBONDCHANGEACTIVE:
3926                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3927                         break;
3928                 default:
3929                         res = -EOPNOTSUPP;
3930                 }
3931
3932                 dev_put(slave_dev);
3933         }
3934
3935         return res;
3936 }
3937
3938 static bool bond_addr_in_mc_list(unsigned char *addr,
3939                                  struct netdev_hw_addr_list *list,
3940                                  int addrlen)
3941 {
3942         struct netdev_hw_addr *ha;
3943
3944         netdev_hw_addr_list_for_each(ha, list)
3945                 if (!memcmp(ha->addr, addr, addrlen))
3946                         return true;
3947
3948         return false;
3949 }
3950
3951 static void bond_set_multicast_list(struct net_device *bond_dev)
3952 {
3953         struct bonding *bond = netdev_priv(bond_dev);
3954         struct netdev_hw_addr *ha;
3955         bool found;
3956
3957         /*
3958          * Do promisc before checking multicast_mode
3959          */
3960         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3961                 /*
3962                  * FIXME: Need to handle the error when one of the multi-slaves
3963                  * encounters error.
3964                  */
3965                 bond_set_promiscuity(bond, 1);
3966
3967
3968         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3969                 bond_set_promiscuity(bond, -1);
3970
3971
3972         /* set allmulti flag to slaves */
3973         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3974                 /*
3975                  * FIXME: Need to handle the error when one of the multi-slaves
3976                  * encounters error.
3977                  */
3978                 bond_set_allmulti(bond, 1);
3979
3980
3981         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3982                 bond_set_allmulti(bond, -1);
3983
3984
3985         read_lock(&bond->lock);
3986
3987         bond->flags = bond_dev->flags;
3988
3989         /* looking for addresses to add to slaves' mc list */
3990         netdev_for_each_mc_addr(ha, bond_dev) {
3991                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3992                                              bond_dev->addr_len);
3993                 if (!found)
3994                         bond_mc_add(bond, ha->addr);
3995         }
3996
3997         /* looking for addresses to delete from slaves' list */
3998         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3999                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4000                                              bond_dev->addr_len);
4001                 if (!found)
4002                         bond_mc_del(bond, ha->addr);
4003         }
4004
4005         /* save master's multicast list */
4006         __hw_addr_flush(&bond->mc_list);
4007         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4008                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4009
4010         read_unlock(&bond->lock);
4011 }
4012
4013 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4014 {
4015         struct bonding *bond = netdev_priv(dev);
4016         struct slave *slave = bond->first_slave;
4017
4018         if (slave) {
4019                 const struct net_device_ops *slave_ops
4020                         = slave->dev->netdev_ops;
4021                 if (slave_ops->ndo_neigh_setup)
4022                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4023         }
4024         return 0;
4025 }
4026
4027 /*
4028  * Change the MTU of all of a master's slaves to match the master
4029  */
4030 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4031 {
4032         struct bonding *bond = netdev_priv(bond_dev);
4033         struct slave *slave, *stop_at;
4034         int res = 0;
4035         int i;
4036
4037         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4038                  (bond_dev ? bond_dev->name : "None"), new_mtu);
4039
4040         /* Can't hold bond->lock with bh disabled here since
4041          * some base drivers panic. On the other hand we can't
4042          * hold bond->lock without bh disabled because we'll
4043          * deadlock. The only solution is to rely on the fact
4044          * that we're under rtnl_lock here, and the slaves
4045          * list won't change. This doesn't solve the problem
4046          * of setting the slave's MTU while it is
4047          * transmitting, but the assumption is that the base
4048          * driver can handle that.
4049          *
4050          * TODO: figure out a way to safely iterate the slaves
4051          * list, but without holding a lock around the actual
4052          * call to the base driver.
4053          */
4054
4055         bond_for_each_slave(bond, slave, i) {
4056                 pr_debug("s %p s->p %p c_m %p\n",
4057                          slave,
4058                          slave->prev,
4059                          slave->dev->netdev_ops->ndo_change_mtu);
4060
4061                 res = dev_set_mtu(slave->dev, new_mtu);
4062
4063                 if (res) {
4064                         /* If we failed to set the slave's mtu to the new value
4065                          * we must abort the operation even in ACTIVE_BACKUP
4066                          * mode, because if we allow the backup slaves to have
4067                          * different mtu values than the active slave we'll
4068                          * need to change their mtu when doing a failover. That
4069                          * means changing their mtu from timer context, which
4070                          * is probably not a good idea.
4071                          */
4072                         pr_debug("err %d %s\n", res, slave->dev->name);
4073                         goto unwind;
4074                 }
4075         }
4076
4077         bond_dev->mtu = new_mtu;
4078
4079         return 0;
4080
4081 unwind:
4082         /* unwind from head to the slave that failed */
4083         stop_at = slave;
4084         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4085                 int tmp_res;
4086
4087                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4088                 if (tmp_res) {
4089                         pr_debug("unwind err %d dev %s\n",
4090                                  tmp_res, slave->dev->name);
4091                 }
4092         }
4093
4094         return res;
4095 }
4096
4097 /*
4098  * Change HW address
4099  *
4100  * Note that many devices must be down to change the HW address, and
4101  * downing the master releases all slaves.  We can make bonds full of
4102  * bonding devices to test this, however.
4103  */
4104 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4105 {
4106         struct bonding *bond = netdev_priv(bond_dev);
4107         struct sockaddr *sa = addr, tmp_sa;
4108         struct slave *slave, *stop_at;
4109         int res = 0;
4110         int i;
4111
4112         if (bond->params.mode == BOND_MODE_ALB)
4113                 return bond_alb_set_mac_address(bond_dev, addr);
4114
4115
4116         pr_debug("bond=%p, name=%s\n",
4117                  bond, bond_dev ? bond_dev->name : "None");
4118
4119         /*
4120          * If fail_over_mac is set to active, do nothing and return
4121          * success.  Returning an error causes ifenslave to fail.
4122          */
4123         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4124                 return 0;
4125
4126         if (!is_valid_ether_addr(sa->sa_data))
4127                 return -EADDRNOTAVAIL;
4128
4129         /* Can't hold bond->lock with bh disabled here since
4130          * some base drivers panic. On the other hand we can't
4131          * hold bond->lock without bh disabled because we'll
4132          * deadlock. The only solution is to rely on the fact
4133          * that we're under rtnl_lock here, and the slaves
4134          * list won't change. This doesn't solve the problem
4135          * of setting the slave's hw address while it is
4136          * transmitting, but the assumption is that the base
4137          * driver can handle that.
4138          *
4139          * TODO: figure out a way to safely iterate the slaves
4140          * list, but without holding a lock around the actual
4141          * call to the base driver.
4142          */
4143
4144         bond_for_each_slave(bond, slave, i) {
4145                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4146                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4147
4148                 if (slave_ops->ndo_set_mac_address == NULL) {
4149                         res = -EOPNOTSUPP;
4150                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4151                         goto unwind;
4152                 }
4153
4154                 res = dev_set_mac_address(slave->dev, addr);
4155                 if (res) {
4156                         /* TODO: consider downing the slave
4157                          * and retry ?
4158                          * User should expect communications
4159                          * breakage anyway until ARP finish
4160                          * updating, so...
4161                          */
4162                         pr_debug("err %d %s\n", res, slave->dev->name);
4163                         goto unwind;
4164                 }
4165         }
4166
4167         /* success */
4168         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4169         return 0;
4170
4171 unwind:
4172         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4173         tmp_sa.sa_family = bond_dev->type;
4174
4175         /* unwind from head to the slave that failed */
4176         stop_at = slave;
4177         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4178                 int tmp_res;
4179
4180                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4181                 if (tmp_res) {
4182                         pr_debug("unwind err %d dev %s\n",
4183                                  tmp_res, slave->dev->name);
4184                 }
4185         }
4186
4187         return res;
4188 }
4189
4190 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4191 {
4192         struct bonding *bond = netdev_priv(bond_dev);
4193         struct slave *slave, *start_at;
4194         int i, slave_no, res = 1;
4195         struct iphdr *iph = ip_hdr(skb);
4196
4197         read_lock(&bond->lock);
4198
4199         if (!BOND_IS_OK(bond))
4200                 goto out;
4201         /*
4202          * Start with the curr_active_slave that joined the bond as the
4203          * default for sending IGMP traffic.  For failover purposes one
4204          * needs to maintain some consistency for the interface that will
4205          * send the join/membership reports.  The curr_active_slave found
4206          * will send all of this type of traffic.
4207          */
4208         if ((iph->protocol == IPPROTO_IGMP) &&
4209             (skb->protocol == htons(ETH_P_IP))) {
4210
4211                 read_lock(&bond->curr_slave_lock);
4212                 slave = bond->curr_active_slave;
4213                 read_unlock(&bond->curr_slave_lock);
4214
4215                 if (!slave)
4216                         goto out;
4217         } else {
4218                 /*
4219                  * Concurrent TX may collide on rr_tx_counter; we accept
4220                  * that as being rare enough not to justify using an
4221                  * atomic op here.
4222                  */
4223                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4224
4225                 bond_for_each_slave(bond, slave, i) {
4226                         slave_no--;
4227                         if (slave_no < 0)
4228                                 break;
4229                 }
4230         }
4231
4232         start_at = slave;
4233         bond_for_each_slave_from(bond, slave, i, start_at) {
4234                 if (IS_UP(slave->dev) &&
4235                     (slave->link == BOND_LINK_UP) &&
4236                     (slave->state == BOND_STATE_ACTIVE)) {
4237                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4238                         break;
4239                 }
4240         }
4241
4242 out:
4243         if (res) {
4244                 /* no suitable interface, frame not sent */
4245                 dev_kfree_skb(skb);
4246         }
4247         read_unlock(&bond->lock);
4248         return NETDEV_TX_OK;
4249 }
4250
4251
4252 /*
4253  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4254  * the bond has a usable interface.
4255  */
4256 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4257 {
4258         struct bonding *bond = netdev_priv(bond_dev);
4259         int res = 1;
4260
4261         read_lock(&bond->lock);
4262         read_lock(&bond->curr_slave_lock);
4263
4264         if (!BOND_IS_OK(bond))
4265                 goto out;
4266
4267         if (!bond->curr_active_slave)
4268                 goto out;
4269
4270         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4271
4272 out:
4273         if (res)
4274                 /* no suitable interface, frame not sent */
4275                 dev_kfree_skb(skb);
4276
4277         read_unlock(&bond->curr_slave_lock);
4278         read_unlock(&bond->lock);
4279         return NETDEV_TX_OK;
4280 }
4281
4282 /*
4283  * In bond_xmit_xor() , we determine the output device by using a pre-
4284  * determined xmit_hash_policy(), If the selected device is not enabled,
4285  * find the next active slave.
4286  */
4287 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4288 {
4289         struct bonding *bond = netdev_priv(bond_dev);
4290         struct slave *slave, *start_at;
4291         int slave_no;
4292         int i;
4293         int res = 1;
4294
4295         read_lock(&bond->lock);
4296
4297         if (!BOND_IS_OK(bond))
4298                 goto out;
4299
4300         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4301
4302         bond_for_each_slave(bond, slave, i) {
4303                 slave_no--;
4304                 if (slave_no < 0)
4305                         break;
4306         }
4307
4308         start_at = slave;
4309
4310         bond_for_each_slave_from(bond, slave, i, start_at) {
4311                 if (IS_UP(slave->dev) &&
4312                     (slave->link == BOND_LINK_UP) &&
4313                     (slave->state == BOND_STATE_ACTIVE)) {
4314                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4315                         break;
4316                 }
4317         }
4318
4319 out:
4320         if (res) {
4321                 /* no suitable interface, frame not sent */
4322                 dev_kfree_skb(skb);
4323         }
4324         read_unlock(&bond->lock);
4325         return NETDEV_TX_OK;
4326 }
4327
4328 /*
4329  * in broadcast mode, we send everything to all usable interfaces.
4330  */
4331 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4332 {
4333         struct bonding *bond = netdev_priv(bond_dev);
4334         struct slave *slave, *start_at;
4335         struct net_device *tx_dev = NULL;
4336         int i;
4337         int res = 1;
4338
4339         read_lock(&bond->lock);
4340
4341         if (!BOND_IS_OK(bond))
4342                 goto out;
4343
4344         read_lock(&bond->curr_slave_lock);
4345         start_at = bond->curr_active_slave;
4346         read_unlock(&bond->curr_slave_lock);
4347
4348         if (!start_at)
4349                 goto out;
4350
4351         bond_for_each_slave_from(bond, slave, i, start_at) {
4352                 if (IS_UP(slave->dev) &&
4353                     (slave->link == BOND_LINK_UP) &&
4354                     (slave->state == BOND_STATE_ACTIVE)) {
4355                         if (tx_dev) {
4356                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4357                                 if (!skb2) {
4358                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4359                                                bond_dev->name);
4360                                         continue;
4361                                 }
4362
4363                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4364                                 if (res) {
4365                                         dev_kfree_skb(skb2);
4366                                         continue;
4367                                 }
4368                         }
4369                         tx_dev = slave->dev;
4370                 }
4371         }
4372
4373         if (tx_dev)
4374                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4375
4376 out:
4377         if (res)
4378                 /* no suitable interface, frame not sent */
4379                 dev_kfree_skb(skb);
4380
4381         /* frame sent to all suitable interfaces */
4382         read_unlock(&bond->lock);
4383         return NETDEV_TX_OK;
4384 }
4385
4386 /*------------------------- Device initialization ---------------------------*/
4387
4388 static void bond_set_xmit_hash_policy(struct bonding *bond)
4389 {
4390         switch (bond->params.xmit_policy) {
4391         case BOND_XMIT_POLICY_LAYER23:
4392                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4393                 break;
4394         case BOND_XMIT_POLICY_LAYER34:
4395                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4396                 break;
4397         case BOND_XMIT_POLICY_LAYER2:
4398         default:
4399                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4400                 break;
4401         }
4402 }
4403
4404 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4405 {
4406         const struct bonding *bond = netdev_priv(dev);
4407
4408         switch (bond->params.mode) {
4409         case BOND_MODE_ROUNDROBIN:
4410                 return bond_xmit_roundrobin(skb, dev);
4411         case BOND_MODE_ACTIVEBACKUP:
4412                 return bond_xmit_activebackup(skb, dev);
4413         case BOND_MODE_XOR:
4414                 return bond_xmit_xor(skb, dev);
4415         case BOND_MODE_BROADCAST:
4416                 return bond_xmit_broadcast(skb, dev);
4417         case BOND_MODE_8023AD:
4418                 return bond_3ad_xmit_xor(skb, dev);
4419         case BOND_MODE_ALB:
4420         case BOND_MODE_TLB:
4421                 return bond_alb_xmit(skb, dev);
4422         default:
4423                 /* Should never happen, mode already checked */
4424                 pr_err("%s: Error: Unknown bonding mode %d\n",
4425                        dev->name, bond->params.mode);
4426                 WARN_ON_ONCE(1);
4427                 dev_kfree_skb(skb);
4428                 return NETDEV_TX_OK;
4429         }
4430 }
4431
4432
4433 /*
4434  * set bond mode specific net device operations
4435  */
4436 void bond_set_mode_ops(struct bonding *bond, int mode)
4437 {
4438         struct net_device *bond_dev = bond->dev;
4439
4440         switch (mode) {
4441         case BOND_MODE_ROUNDROBIN:
4442                 break;
4443         case BOND_MODE_ACTIVEBACKUP:
4444                 break;
4445         case BOND_MODE_XOR:
4446                 bond_set_xmit_hash_policy(bond);
4447                 break;
4448         case BOND_MODE_BROADCAST:
4449                 break;
4450         case BOND_MODE_8023AD:
4451                 bond_set_master_3ad_flags(bond);
4452                 bond_set_xmit_hash_policy(bond);
4453                 break;
4454         case BOND_MODE_ALB:
4455                 bond_set_master_alb_flags(bond);
4456                 /* FALLTHRU */
4457         case BOND_MODE_TLB:
4458                 break;
4459         default:
4460                 /* Should never happen, mode already checked */
4461                 pr_err("%s: Error: Unknown bonding mode %d\n",
4462                        bond_dev->name, mode);
4463                 break;
4464         }
4465 }
4466
4467 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4468                                     struct ethtool_drvinfo *drvinfo)
4469 {
4470         strncpy(drvinfo->driver, DRV_NAME, 32);
4471         strncpy(drvinfo->version, DRV_VERSION, 32);
4472         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4473 }
4474
4475 static const struct ethtool_ops bond_ethtool_ops = {
4476         .get_drvinfo            = bond_ethtool_get_drvinfo,
4477         .get_link               = ethtool_op_get_link,
4478         .get_tx_csum            = ethtool_op_get_tx_csum,
4479         .get_sg                 = ethtool_op_get_sg,
4480         .get_tso                = ethtool_op_get_tso,
4481         .get_ufo                = ethtool_op_get_ufo,
4482         .get_flags              = ethtool_op_get_flags,
4483 };
4484
4485 static const struct net_device_ops bond_netdev_ops = {
4486         .ndo_init               = bond_init,
4487         .ndo_uninit             = bond_uninit,
4488         .ndo_open               = bond_open,
4489         .ndo_stop               = bond_close,
4490         .ndo_start_xmit         = bond_start_xmit,
4491         .ndo_get_stats          = bond_get_stats,
4492         .ndo_do_ioctl           = bond_do_ioctl,
4493         .ndo_set_multicast_list = bond_set_multicast_list,
4494         .ndo_change_mtu         = bond_change_mtu,
4495         .ndo_set_mac_address    = bond_set_mac_address,
4496         .ndo_neigh_setup        = bond_neigh_setup,
4497         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4498         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4499         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4500 #ifdef CONFIG_NET_POLL_CONTROLLER
4501         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4502         .ndo_poll_controller    = bond_poll_controller,
4503 #endif
4504 };
4505
4506 static void bond_destructor(struct net_device *bond_dev)
4507 {
4508         struct bonding *bond = netdev_priv(bond_dev);
4509         if (bond->wq)
4510                 destroy_workqueue(bond->wq);
4511         free_netdev(bond_dev);
4512 }
4513
4514 static void bond_setup(struct net_device *bond_dev)
4515 {
4516         struct bonding *bond = netdev_priv(bond_dev);
4517
4518         /* initialize rwlocks */
4519         rwlock_init(&bond->lock);
4520         rwlock_init(&bond->curr_slave_lock);
4521
4522         bond->params = bonding_defaults;
4523
4524         /* Initialize pointers */
4525         bond->dev = bond_dev;
4526         INIT_LIST_HEAD(&bond->vlan_list);
4527
4528         /* Initialize the device entry points */
4529         ether_setup(bond_dev);
4530         bond_dev->netdev_ops = &bond_netdev_ops;
4531         bond_dev->ethtool_ops = &bond_ethtool_ops;
4532         bond_set_mode_ops(bond, bond->params.mode);
4533
4534         bond_dev->destructor = bond_destructor;
4535
4536         /* Initialize the device options */
4537         bond_dev->tx_queue_len = 0;
4538         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4539         bond_dev->priv_flags |= IFF_BONDING;
4540         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4541
4542         if (bond->params.arp_interval)
4543                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4544
4545         /* At first, we block adding VLANs. That's the only way to
4546          * prevent problems that occur when adding VLANs over an
4547          * empty bond. The block will be removed once non-challenged
4548          * slaves are enslaved.
4549          */
4550         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4551
4552         /* don't acquire bond device's netif_tx_lock when
4553          * transmitting */
4554         bond_dev->features |= NETIF_F_LLTX;
4555
4556         /* By default, we declare the bond to be fully
4557          * VLAN hardware accelerated capable. Special
4558          * care is taken in the various xmit functions
4559          * when there are slaves that are not hw accel
4560          * capable
4561          */
4562         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4563                                NETIF_F_HW_VLAN_RX |
4564                                NETIF_F_HW_VLAN_FILTER);
4565
4566 }
4567
4568 static void bond_work_cancel_all(struct bonding *bond)
4569 {
4570         write_lock_bh(&bond->lock);
4571         bond->kill_timers = 1;
4572         write_unlock_bh(&bond->lock);
4573
4574         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4575                 cancel_delayed_work(&bond->mii_work);
4576
4577         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4578                 cancel_delayed_work(&bond->arp_work);
4579
4580         if (bond->params.mode == BOND_MODE_ALB &&
4581             delayed_work_pending(&bond->alb_work))
4582                 cancel_delayed_work(&bond->alb_work);
4583
4584         if (bond->params.mode == BOND_MODE_8023AD &&
4585             delayed_work_pending(&bond->ad_work))
4586                 cancel_delayed_work(&bond->ad_work);
4587 }
4588
4589 /*
4590 * Destroy a bonding device.
4591 * Must be under rtnl_lock when this function is called.
4592 */
4593 static void bond_uninit(struct net_device *bond_dev)
4594 {
4595         struct bonding *bond = netdev_priv(bond_dev);
4596
4597         bond_netpoll_cleanup(bond_dev);
4598
4599         /* Release the bonded slaves */
4600         bond_release_all(bond_dev);
4601
4602         list_del(&bond->bond_list);
4603
4604         bond_work_cancel_all(bond);
4605
4606         bond_remove_proc_entry(bond);
4607
4608         __hw_addr_flush(&bond->mc_list);
4609 }
4610
4611 /*------------------------- Module initialization ---------------------------*/
4612
4613 /*
4614  * Convert string input module parms.  Accept either the
4615  * number of the mode or its string name.  A bit complicated because
4616  * some mode names are substrings of other names, and calls from sysfs
4617  * may have whitespace in the name (trailing newlines, for example).
4618  */
4619 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4620 {
4621         int modeint = -1, i, rv;
4622         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4623
4624         for (p = (char *)buf; *p; p++)
4625                 if (!(isdigit(*p) || isspace(*p)))
4626                         break;
4627
4628         if (*p)
4629                 rv = sscanf(buf, "%20s", modestr);
4630         else
4631                 rv = sscanf(buf, "%d", &modeint);
4632
4633         if (!rv)
4634                 return -1;
4635
4636         for (i = 0; tbl[i].modename; i++) {
4637                 if (modeint == tbl[i].mode)
4638                         return tbl[i].mode;
4639                 if (strcmp(modestr, tbl[i].modename) == 0)
4640                         return tbl[i].mode;
4641         }
4642
4643         return -1;
4644 }
4645
4646 static int bond_check_params(struct bond_params *params)
4647 {
4648         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4649
4650         /*
4651          * Convert string parameters.
4652          */
4653         if (mode) {
4654                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4655                 if (bond_mode == -1) {
4656                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4657                                mode == NULL ? "NULL" : mode);
4658                         return -EINVAL;
4659                 }
4660         }
4661
4662         if (xmit_hash_policy) {
4663                 if ((bond_mode != BOND_MODE_XOR) &&
4664                     (bond_mode != BOND_MODE_8023AD)) {
4665                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4666                                bond_mode_name(bond_mode));
4667                 } else {
4668                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4669                                                         xmit_hashtype_tbl);
4670                         if (xmit_hashtype == -1) {
4671                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4672                                        xmit_hash_policy == NULL ? "NULL" :
4673                                        xmit_hash_policy);
4674                                 return -EINVAL;
4675                         }
4676                 }
4677         }
4678
4679         if (lacp_rate) {
4680                 if (bond_mode != BOND_MODE_8023AD) {
4681                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4682                                 bond_mode_name(bond_mode));
4683                 } else {
4684                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4685                         if (lacp_fast == -1) {
4686                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4687                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4688                                 return -EINVAL;
4689                         }
4690                 }
4691         }
4692
4693         if (ad_select) {
4694                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4695                 if (params->ad_select == -1) {
4696                         pr_err("Error: Invalid ad_select \"%s\"\n",
4697                                ad_select == NULL ? "NULL" : ad_select);
4698                         return -EINVAL;
4699                 }
4700
4701                 if (bond_mode != BOND_MODE_8023AD) {
4702                         pr_warning("ad_select param only affects 802.3ad mode\n");
4703                 }
4704         } else {
4705                 params->ad_select = BOND_AD_STABLE;
4706         }
4707
4708         if (max_bonds < 0) {
4709                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4710                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4711                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4712         }
4713
4714         if (miimon < 0) {
4715                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4716                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4717                 miimon = BOND_LINK_MON_INTERV;
4718         }
4719
4720         if (updelay < 0) {
4721                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4722                            updelay, INT_MAX);
4723                 updelay = 0;
4724         }
4725
4726         if (downdelay < 0) {
4727                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4728                            downdelay, INT_MAX);
4729                 downdelay = 0;
4730         }
4731
4732         if ((use_carrier != 0) && (use_carrier != 1)) {
4733                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4734                            use_carrier);
4735                 use_carrier = 1;
4736         }
4737
4738         if (num_grat_arp < 0 || num_grat_arp > 255) {
4739                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4740                            num_grat_arp);
4741                 num_grat_arp = 1;
4742         }
4743
4744         if (num_unsol_na < 0 || num_unsol_na > 255) {
4745                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4746                            num_unsol_na);
4747                 num_unsol_na = 1;
4748         }
4749
4750         /* reset values for 802.3ad */
4751         if (bond_mode == BOND_MODE_8023AD) {
4752                 if (!miimon) {
4753                         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");
4754                         pr_warning("Forcing miimon to 100msec\n");
4755                         miimon = 100;
4756                 }
4757         }
4758
4759         /* reset values for TLB/ALB */
4760         if ((bond_mode == BOND_MODE_TLB) ||
4761             (bond_mode == BOND_MODE_ALB)) {
4762                 if (!miimon) {
4763                         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");
4764                         pr_warning("Forcing miimon to 100msec\n");
4765                         miimon = 100;
4766                 }
4767         }
4768
4769         if (bond_mode == BOND_MODE_ALB) {
4770                 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",
4771                           updelay);
4772         }
4773
4774         if (!miimon) {
4775                 if (updelay || downdelay) {
4776                         /* just warn the user the up/down delay will have
4777                          * no effect since miimon is zero...
4778                          */
4779                         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",
4780                                    updelay, downdelay);
4781                 }
4782         } else {
4783                 /* don't allow arp monitoring */
4784                 if (arp_interval) {
4785                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4786                                    miimon, arp_interval);
4787                         arp_interval = 0;
4788                 }
4789
4790                 if ((updelay % miimon) != 0) {
4791                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4792                                    updelay, miimon,
4793                                    (updelay / miimon) * miimon);
4794                 }
4795
4796                 updelay /= miimon;
4797
4798                 if ((downdelay % miimon) != 0) {
4799                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4800                                    downdelay, miimon,
4801                                    (downdelay / miimon) * miimon);
4802                 }
4803
4804                 downdelay /= miimon;
4805         }
4806
4807         if (arp_interval < 0) {
4808                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4809                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4810                 arp_interval = BOND_LINK_ARP_INTERV;
4811         }
4812
4813         for (arp_ip_count = 0;
4814              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4815              arp_ip_count++) {
4816                 /* not complete check, but should be good enough to
4817                    catch mistakes */
4818                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4819                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4820                                    arp_ip_target[arp_ip_count]);
4821                         arp_interval = 0;
4822                 } else {
4823                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4824                         arp_target[arp_ip_count] = ip;
4825                 }
4826         }
4827
4828         if (arp_interval && !arp_ip_count) {
4829                 /* don't allow arping if no arp_ip_target given... */
4830                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4831                            arp_interval);
4832                 arp_interval = 0;
4833         }
4834
4835         if (arp_validate) {
4836                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4837                         pr_err("arp_validate only supported in active-backup mode\n");
4838                         return -EINVAL;
4839                 }
4840                 if (!arp_interval) {
4841                         pr_err("arp_validate requires arp_interval\n");
4842                         return -EINVAL;
4843                 }
4844
4845                 arp_validate_value = bond_parse_parm(arp_validate,
4846                                                      arp_validate_tbl);
4847                 if (arp_validate_value == -1) {
4848                         pr_err("Error: invalid arp_validate \"%s\"\n",
4849                                arp_validate == NULL ? "NULL" : arp_validate);
4850                         return -EINVAL;
4851                 }
4852         } else
4853                 arp_validate_value = 0;
4854
4855         if (miimon) {
4856                 pr_info("MII link monitoring set to %d ms\n", miimon);
4857         } else if (arp_interval) {
4858                 int i;
4859
4860                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4861                         arp_interval,
4862                         arp_validate_tbl[arp_validate_value].modename,
4863                         arp_ip_count);
4864
4865                 for (i = 0; i < arp_ip_count; i++)
4866                         pr_info(" %s", arp_ip_target[i]);
4867
4868                 pr_info("\n");
4869
4870         } else if (max_bonds) {
4871                 /* miimon and arp_interval not set, we need one so things
4872                  * work as expected, see bonding.txt for details
4873                  */
4874                 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");
4875         }
4876
4877         if (primary && !USES_PRIMARY(bond_mode)) {
4878                 /* currently, using a primary only makes sense
4879                  * in active backup, TLB or ALB modes
4880                  */
4881                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4882                            primary, bond_mode_name(bond_mode));
4883                 primary = NULL;
4884         }
4885
4886         if (primary && primary_reselect) {
4887                 primary_reselect_value = bond_parse_parm(primary_reselect,
4888                                                          pri_reselect_tbl);
4889                 if (primary_reselect_value == -1) {
4890                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4891                                primary_reselect ==
4892                                         NULL ? "NULL" : primary_reselect);
4893                         return -EINVAL;
4894                 }
4895         } else {
4896                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4897         }
4898
4899         if (fail_over_mac) {
4900                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4901                                                       fail_over_mac_tbl);
4902                 if (fail_over_mac_value == -1) {
4903                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4904                                arp_validate == NULL ? "NULL" : arp_validate);
4905                         return -EINVAL;
4906                 }
4907
4908                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4909                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4910         } else {
4911                 fail_over_mac_value = BOND_FOM_NONE;
4912         }
4913
4914         /* fill params struct with the proper values */
4915         params->mode = bond_mode;
4916         params->xmit_policy = xmit_hashtype;
4917         params->miimon = miimon;
4918         params->num_grat_arp = num_grat_arp;
4919         params->num_unsol_na = num_unsol_na;
4920         params->arp_interval = arp_interval;
4921         params->arp_validate = arp_validate_value;
4922         params->updelay = updelay;
4923         params->downdelay = downdelay;
4924         params->use_carrier = use_carrier;
4925         params->lacp_fast = lacp_fast;
4926         params->primary[0] = 0;
4927         params->primary_reselect = primary_reselect_value;
4928         params->fail_over_mac = fail_over_mac_value;
4929
4930         if (primary) {
4931                 strncpy(params->primary, primary, IFNAMSIZ);
4932                 params->primary[IFNAMSIZ - 1] = 0;
4933         }
4934
4935         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4936
4937         return 0;
4938 }
4939
4940 static struct lock_class_key bonding_netdev_xmit_lock_key;
4941 static struct lock_class_key bonding_netdev_addr_lock_key;
4942
4943 static void bond_set_lockdep_class_one(struct net_device *dev,
4944                                        struct netdev_queue *txq,
4945                                        void *_unused)
4946 {
4947         lockdep_set_class(&txq->_xmit_lock,
4948                           &bonding_netdev_xmit_lock_key);
4949 }
4950
4951 static void bond_set_lockdep_class(struct net_device *dev)
4952 {
4953         lockdep_set_class(&dev->addr_list_lock,
4954                           &bonding_netdev_addr_lock_key);
4955         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4956 }
4957
4958 /*
4959  * Called from registration process
4960  */
4961 static int bond_init(struct net_device *bond_dev)
4962 {
4963         struct bonding *bond = netdev_priv(bond_dev);
4964         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4965
4966         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4967
4968         bond->wq = create_singlethread_workqueue(bond_dev->name);
4969         if (!bond->wq)
4970                 return -ENOMEM;
4971
4972         bond_set_lockdep_class(bond_dev);
4973
4974         netif_carrier_off(bond_dev);
4975
4976         bond_create_proc_entry(bond);
4977         list_add_tail(&bond->bond_list, &bn->dev_list);
4978
4979         bond_prepare_sysfs_group(bond);
4980
4981         __hw_addr_init(&bond->mc_list);
4982         return 0;
4983 }
4984
4985 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4986 {
4987         if (tb[IFLA_ADDRESS]) {
4988                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4989                         return -EINVAL;
4990                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4991                         return -EADDRNOTAVAIL;
4992         }
4993         return 0;
4994 }
4995
4996 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4997         .kind           = "bond",
4998         .priv_size      = sizeof(struct bonding),
4999         .setup          = bond_setup,
5000         .validate       = bond_validate,
5001 };
5002
5003 /* Create a new bond based on the specified name and bonding parameters.
5004  * If name is NULL, obtain a suitable "bond%d" name for us.
5005  * Caller must NOT hold rtnl_lock; we need to release it here before we
5006  * set up our sysfs entries.
5007  */
5008 int bond_create(struct net *net, const char *name)
5009 {
5010         struct net_device *bond_dev;
5011         int res;
5012
5013         rtnl_lock();
5014
5015         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5016                                 bond_setup);
5017         if (!bond_dev) {
5018                 pr_err("%s: eek! can't alloc netdev!\n", name);
5019                 rtnl_unlock();
5020                 return -ENOMEM;
5021         }
5022
5023         dev_net_set(bond_dev, net);
5024         bond_dev->rtnl_link_ops = &bond_link_ops;
5025
5026         if (!name) {
5027                 res = dev_alloc_name(bond_dev, "bond%d");
5028                 if (res < 0)
5029                         goto out;
5030         }
5031
5032         res = register_netdevice(bond_dev);
5033
5034 out:
5035         rtnl_unlock();
5036         if (res < 0)
5037                 bond_destructor(bond_dev);
5038         return res;
5039 }
5040
5041 static int __net_init bond_net_init(struct net *net)
5042 {
5043         struct bond_net *bn = net_generic(net, bond_net_id);
5044
5045         bn->net = net;
5046         INIT_LIST_HEAD(&bn->dev_list);
5047
5048         bond_create_proc_dir(bn);
5049
5050         return 0;
5051 }
5052
5053 static void __net_exit bond_net_exit(struct net *net)
5054 {
5055         struct bond_net *bn = net_generic(net, bond_net_id);
5056
5057         bond_destroy_proc_dir(bn);
5058 }
5059
5060 static struct pernet_operations bond_net_ops = {
5061         .init = bond_net_init,
5062         .exit = bond_net_exit,
5063         .id   = &bond_net_id,
5064         .size = sizeof(struct bond_net),
5065 };
5066
5067 static int __init bonding_init(void)
5068 {
5069         int i;
5070         int res;
5071
5072         pr_info("%s", version);
5073
5074         res = bond_check_params(&bonding_defaults);
5075         if (res)
5076                 goto out;
5077
5078         res = register_pernet_subsys(&bond_net_ops);
5079         if (res)
5080                 goto out;
5081
5082         res = rtnl_link_register(&bond_link_ops);
5083         if (res)
5084                 goto err_link;
5085
5086         for (i = 0; i < max_bonds; i++) {
5087                 res = bond_create(&init_net, NULL);
5088                 if (res)
5089                         goto err;
5090         }
5091
5092         res = bond_create_sysfs();
5093         if (res)
5094                 goto err;
5095
5096         register_netdevice_notifier(&bond_netdev_notifier);
5097         register_inetaddr_notifier(&bond_inetaddr_notifier);
5098         bond_register_ipv6_notifier();
5099 out:
5100         return res;
5101 err:
5102         rtnl_link_unregister(&bond_link_ops);
5103 err_link:
5104         unregister_pernet_subsys(&bond_net_ops);
5105         goto out;
5106
5107 }
5108
5109 static void __exit bonding_exit(void)
5110 {
5111         unregister_netdevice_notifier(&bond_netdev_notifier);
5112         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5113         bond_unregister_ipv6_notifier();
5114
5115         bond_destroy_sysfs();
5116
5117         rtnl_link_unregister(&bond_link_ops);
5118         unregister_pernet_subsys(&bond_net_ops);
5119 }
5120
5121 module_init(bonding_init);
5122 module_exit(bonding_exit);
5123 MODULE_LICENSE("GPL");
5124 MODULE_VERSION(DRV_VERSION);
5125 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5126 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5127 MODULE_ALIAS_RTNL_LINK("bond");