drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/system.h>
  58 #include <asm/dma.h>
  59 #include <linux/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/smp.h>
  69 #include <linux/if_ether.h>
  70 #include <net/arp.h>
  71 #include <linux/mii.h>
  72 #include <linux/ethtool.h>
  73 #include <linux/if_vlan.h>
  74 #include <linux/if_bonding.h>
  75 #include <linux/jiffies.h>
  76 #include <linux/preempt.h>
  77 #include <net/route.h>
  78 #include <net/net_namespace.h>
  79 #include <net/netns/generic.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon       = BOND_LINK_MON_INTERV;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static int all_slaves_active = 0;
 109 static struct bond_params bonding_defaults;
 110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 111
 112 module_param(max_bonds, int, 0);
 113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 114 module_param(tx_queues, int, 0);
 115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 118                                "failover event (alias of num_unsol_na)");
 119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 121                                "failover event (alias of num_grat_arp)");
 122 module_param(miimon, int, 0);
 123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 124 module_param(updelay, int, 0);
 125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 126 module_param(downdelay, int, 0);
 127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 128                             "in milliseconds");
 129 module_param(use_carrier, int, 0);
 130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 131                               "0 for off, 1 for on (default)");
 132 module_param(mode, charp, 0);
 133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 134                        "1 for active-backup, 2 for balance-xor, "
 135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 136                        "6 for balance-alb");
 137 module_param(primary, charp, 0);
 138 MODULE_PARM_DESC(primary, "Primary network device to use");
 139 module_param(primary_reselect, charp, 0);
 140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 141                                    "once it comes up; "
 142                                    "0 for always (default), "
 143                                    "1 for only if speed of primary is "
 144                                    "better, "
 145                                    "2 for only on active slave "
 146                                    "failure");
 147 module_param(lacp_rate, charp, 0);
 148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 149                             "0 for slow, 1 for fast");
 150 module_param(ad_select, charp, 0);
 151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 152                             "0 for stable (default), 1 for bandwidth, "
 153                             "2 for count");
 154 module_param(min_links, int, 0);
 155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 156
 157 module_param(xmit_hash_policy, charp, 0);
 158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 159                                    "0 for layer 2 (default), 1 for layer 3+4, "
 160                                    "2 for layer 2+3");
 161 module_param(arp_interval, int, 0);
 162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 163 module_param_array(arp_ip_target, charp, NULL, 0);
 164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 165 module_param(arp_validate, charp, 0);
 166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 167                                "0 for none (default), 1 for active, "
 168                                "2 for backup, 3 for all");
 169 module_param(fail_over_mac, charp, 0);
 170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 171                                 "the same MAC; 0 for none (default), "
 172                                 "1 for active, 2 for follow");
 173 module_param(all_slaves_active, int, 0);
 174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 175                                      "by setting active flag for all slaves; "
 176                                      "0 for never (default), 1 for always.");
 177 module_param(resend_igmp, int, 0);
 178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 179                               "link failure");
 180
 181 /*----------------------------- Global variables ----------------------------*/
 182
 183 #ifdef CONFIG_NET_POLL_CONTROLLER
 184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 185 #endif
 186
 187 int bond_net_id __read_mostly;
 188
 189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 190 static int arp_ip_count;
 191 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 193 static int lacp_fast;
 194
 195 const struct bond_parm_tbl bond_lacp_tbl[] = {
 196 {       "slow",         AD_LACP_SLOW},
 197 {       "fast",         AD_LACP_FAST},
 198 {       NULL,           -1},
 199 };
 200
 201 const struct bond_parm_tbl bond_mode_tbl[] = {
 202 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 203 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 204 {       "balance-xor",          BOND_MODE_XOR},
 205 {       "broadcast",            BOND_MODE_BROADCAST},
 206 {       "802.3ad",              BOND_MODE_8023AD},
 207 {       "balance-tlb",          BOND_MODE_TLB},
 208 {       "balance-alb",          BOND_MODE_ALB},
 209 {       NULL,                   -1},
 210 };
 211
 212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 213 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 214 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 215 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 216 {       NULL,                   -1},
 217 };
 218
 219 const struct bond_parm_tbl arp_validate_tbl[] = {
 220 {       "none",                 BOND_ARP_VALIDATE_NONE},
 221 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 222 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 223 {       "all",                  BOND_ARP_VALIDATE_ALL},
 224 {       NULL,                   -1},
 225 };
 226
 227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 228 {       "none",                 BOND_FOM_NONE},
 229 {       "active",               BOND_FOM_ACTIVE},
 230 {       "follow",               BOND_FOM_FOLLOW},
 231 {       NULL,                   -1},
 232 };
 233
 234 const struct bond_parm_tbl pri_reselect_tbl[] = {
 235 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 236 {       "better",               BOND_PRI_RESELECT_BETTER},
 237 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 238 {       NULL,                   -1},
 239 };
 240
 241 struct bond_parm_tbl ad_select_tbl[] = {
 242 {       "stable",       BOND_AD_STABLE},
 243 {       "bandwidth",    BOND_AD_BANDWIDTH},
 244 {       "count",        BOND_AD_COUNT},
 245 {       NULL,           -1},
 246 };
 247
 248 /*-------------------------- Forward declarations ---------------------------*/
 249
 250 static int bond_init(struct net_device *bond_dev);
 251 static void bond_uninit(struct net_device *bond_dev);
 252
 253 /*---------------------------- General routines -----------------------------*/
 254
 255 const char *bond_mode_name(int mode)
 256 {
 257         static const char *names[] = {
 258                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 259                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 260                 [BOND_MODE_XOR] = "load balancing (xor)",
 261                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 262                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 263                 [BOND_MODE_TLB] = "transmit load balancing",
 264                 [BOND_MODE_ALB] = "adaptive load balancing",
 265         };
 266
 267         if (mode < 0 || mode > BOND_MODE_ALB)
 268                 return "unknown";
 269
 270         return names[mode];
 271 }
 272
 273 /*---------------------------------- VLAN -----------------------------------*/
 274
 275 /**
 276  * bond_add_vlan - add a new vlan id on bond
 277  * @bond: bond that got the notification
 278  * @vlan_id: the vlan id to add
 279  *
 280  * Returns -ENOMEM if allocation failed.
 281  */
 282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 283 {
 284         struct vlan_entry *vlan;
 285
 286         pr_debug("bond: %s, vlan id %d\n",
 287                  (bond ? bond->dev->name : "None"), vlan_id);
 288
 289         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 290         if (!vlan)
 291                 return -ENOMEM;
 292
 293         INIT_LIST_HEAD(&vlan->vlan_list);
 294         vlan->vlan_id = vlan_id;
 295
 296         write_lock_bh(&bond->lock);
 297
 298         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 299
 300         write_unlock_bh(&bond->lock);
 301
 302         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 303
 304         return 0;
 305 }
 306
 307 /**
 308  * bond_del_vlan - delete a vlan id from bond
 309  * @bond: bond that got the notification
 310  * @vlan_id: the vlan id to delete
 311  *
 312  * returns -ENODEV if @vlan_id was not found in @bond.
 313  */
 314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 315 {
 316         struct vlan_entry *vlan;
 317         int res = -ENODEV;
 318
 319         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 320
 321         block_netpoll_tx();
 322         write_lock_bh(&bond->lock);
 323
 324         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 325                 if (vlan->vlan_id == vlan_id) {
 326                         list_del(&vlan->vlan_list);
 327
 328                         if (bond_is_lb(bond))
 329                                 bond_alb_clear_vlan(bond, vlan_id);
 330
 331                         pr_debug("removed VLAN ID %d from bond %s\n",
 332                                  vlan_id, bond->dev->name);
 333
 334                         kfree(vlan);
 335
 336                         res = 0;
 337                         goto out;
 338                 }
 339         }
 340
 341         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 342                  vlan_id, bond->dev->name);
 343
 344 out:
 345         write_unlock_bh(&bond->lock);
 346         unblock_netpoll_tx();
 347         return res;
 348 }
 349
 350 /**
 351  * bond_next_vlan - safely skip to the next item in the vlans list.
 352  * @bond: the bond we're working on
 353  * @curr: item we're advancing from
 354  *
 355  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 356  * or @curr->next otherwise (even if it is @curr itself again).
 357  *
 358  * Caller must hold bond->lock
 359  */
 360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 361 {
 362         struct vlan_entry *next, *last;
 363
 364         if (list_empty(&bond->vlan_list))
 365                 return NULL;
 366
 367         if (!curr) {
 368                 next = list_entry(bond->vlan_list.next,
 369                                   struct vlan_entry, vlan_list);
 370         } else {
 371                 last = list_entry(bond->vlan_list.prev,
 372                                   struct vlan_entry, vlan_list);
 373                 if (last == curr) {
 374                         next = list_entry(bond->vlan_list.next,
 375                                           struct vlan_entry, vlan_list);
 376                 } else {
 377                         next = list_entry(curr->vlan_list.next,
 378                                           struct vlan_entry, vlan_list);
 379                 }
 380         }
 381
 382         return next;
 383 }
 384
 385 #define bond_queue_mapping(skb) (*(u16 *)((skb)->cb))
 386
 387 /**
 388  * bond_dev_queue_xmit - Prepare skb for xmit.
 389  *
 390  * @bond: bond device that got this skb for tx.
 391  * @skb: hw accel VLAN tagged skb to transmit
 392  * @slave_dev: slave that is supposed to xmit this skbuff
 393  */
 394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 395                         struct net_device *slave_dev)
 396 {
 397         skb->dev = slave_dev;
 398         skb->priority = 1;
 399
 400         skb->queue_mapping = bond_queue_mapping(skb);
 401
 402         if (unlikely(netpoll_tx_running(slave_dev)))
 403                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 404         else
 405                 dev_queue_xmit(skb);
 406
 407         return 0;
 408 }
 409
 410 /*
 411  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 412  * We don't protect the slave list iteration with a lock because:
 413  * a. This operation is performed in IOCTL context,
 414  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 415  * c. Holding a lock with BH disabled while directly calling a base driver
 416  *    entry point is generally a BAD idea.
 417  *
 418  * The design of synchronization/protection for this operation in the 8021q
 419  * module is good for one or more VLAN devices over a single physical device
 420  * and cannot be extended for a teaming solution like bonding, so there is a
 421  * potential race condition here where a net device from the vlan group might
 422  * be referenced (either by a base driver or the 8021q code) while it is being
 423  * removed from the system. However, it turns out we're not making matters
 424  * worse, and if it works for regular VLAN usage it will work here too.
 425 */
 426
 427 /**
 428  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 429  * @bond_dev: bonding net device that got called
 430  * @vid: vlan id being added
 431  */
 432 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 433 {
 434         struct bonding *bond = netdev_priv(bond_dev);
 435         struct slave *slave;
 436         int i, res;
 437
 438         bond_for_each_slave(bond, slave, i) {
 439                 struct net_device *slave_dev = slave->dev;
 440                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 441
 442                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 443                     slave_ops->ndo_vlan_rx_add_vid) {
 444                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
 445                 }
 446         }
 447
 448         res = bond_add_vlan(bond, vid);
 449         if (res) {
 450                 pr_err("%s: Error: Failed to add vlan id %d\n",
 451                        bond_dev->name, vid);
 452         }
 453 }
 454
 455 /**
 456  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 457  * @bond_dev: bonding net device that got called
 458  * @vid: vlan id being removed
 459  */
 460 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 461 {
 462         struct bonding *bond = netdev_priv(bond_dev);
 463         struct slave *slave;
 464         int i, res;
 465
 466         bond_for_each_slave(bond, slave, i) {
 467                 struct net_device *slave_dev = slave->dev;
 468                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 469
 470                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
 471                     slave_ops->ndo_vlan_rx_kill_vid) {
 472                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
 473                 }
 474         }
 475
 476         res = bond_del_vlan(bond, vid);
 477         if (res) {
 478                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 479                        bond_dev->name, vid);
 480         }
 481 }
 482
 483 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 484 {
 485         struct vlan_entry *vlan;
 486         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 487
 488         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 489             !(slave_ops->ndo_vlan_rx_add_vid))
 490                 return;
 491
 492         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
 493                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
 494 }
 495
 496 static void bond_del_vlans_from_slave(struct bonding *bond,
 497                                       struct net_device *slave_dev)
 498 {
 499         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 500         struct vlan_entry *vlan;
 501
 502         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
 503             !(slave_ops->ndo_vlan_rx_kill_vid))
 504                 return;
 505
 506         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 507                 if (!vlan->vlan_id)
 508                         continue;
 509                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
 510         }
 511 }
 512
 513 /*------------------------------- Link status -------------------------------*/
 514
 515 /*
 516  * Set the carrier state for the master according to the state of its
 517  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 518  * do special 802.3ad magic.
 519  *
 520  * Returns zero if carrier state does not change, nonzero if it does.
 521  */
 522 static int bond_set_carrier(struct bonding *bond)
 523 {
 524         struct slave *slave;
 525         int i;
 526
 527         if (bond->slave_cnt == 0)
 528                 goto down;
 529
 530         if (bond->params.mode == BOND_MODE_8023AD)
 531                 return bond_3ad_set_carrier(bond);
 532
 533         bond_for_each_slave(bond, slave, i) {
 534                 if (slave->link == BOND_LINK_UP) {
 535                         if (!netif_carrier_ok(bond->dev)) {
 536                                 netif_carrier_on(bond->dev);
 537                                 return 1;
 538                         }
 539                         return 0;
 540                 }
 541         }
 542
 543 down:
 544         if (netif_carrier_ok(bond->dev)) {
 545                 netif_carrier_off(bond->dev);
 546                 return 1;
 547         }
 548         return 0;
 549 }
 550
 551 /*
 552  * Get link speed and duplex from the slave's base driver
 553  * using ethtool. If for some reason the call fails or the
 554  * values are invalid, fake speed and duplex to 100/Full
 555  * and return error.
 556  */
 557 static int bond_update_speed_duplex(struct slave *slave)
 558 {
 559         struct net_device *slave_dev = slave->dev;
 560         struct ethtool_cmd etool = { .cmd = ETHTOOL_GSET };
 561         u32 slave_speed;
 562         int res;
 563
 564         /* Fake speed and duplex */
 565         slave->speed = SPEED_100;
 566         slave->duplex = DUPLEX_FULL;
 567
 568         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
 569                 return -1;
 570
 571         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
 572         if (res < 0)
 573                 return -1;
 574
 575         slave_speed = ethtool_cmd_speed(&etool);
 576         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 577                 return -1;
 578
 579         switch (etool.duplex) {
 580         case DUPLEX_FULL:
 581         case DUPLEX_HALF:
 582                 break;
 583         default:
 584                 return -1;
 585         }
 586
 587         slave->speed = slave_speed;
 588         slave->duplex = etool.duplex;
 589
 590         return 0;
 591 }
 592
 593 /*
 594  * if <dev> supports MII link status reporting, check its link status.
 595  *
 596  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 597  * depending upon the setting of the use_carrier parameter.
 598  *
 599  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 600  * can't tell and just pretend it is), or 0, meaning that the link is
 601  * down.
 602  *
 603  * If reporting is non-zero, instead of faking link up, return -1 if
 604  * both ETHTOOL and MII ioctls fail (meaning the device does not
 605  * support them).  If use_carrier is set, return whatever it says.
 606  * It'd be nice if there was a good way to tell if a driver supports
 607  * netif_carrier, but there really isn't.
 608  */
 609 static int bond_check_dev_link(struct bonding *bond,
 610                                struct net_device *slave_dev, int reporting)
 611 {
 612         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 613         int (*ioctl)(struct net_device *, struct ifreq *, int);
 614         struct ifreq ifr;
 615         struct mii_ioctl_data *mii;
 616
 617         if (!reporting && !netif_running(slave_dev))
 618                 return 0;
 619
 620         if (bond->params.use_carrier)
 621                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 622
 623         /* Try to get link status using Ethtool first. */
 624         if (slave_dev->ethtool_ops) {
 625                 if (slave_dev->ethtool_ops->get_link) {
 626                         u32 link;
 627
 628                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 629
 630                         return link ? BMSR_LSTATUS : 0;
 631                 }
 632         }
 633
 634         /* Ethtool can't be used, fallback to MII ioctls. */
 635         ioctl = slave_ops->ndo_do_ioctl;
 636         if (ioctl) {
 637                 /* TODO: set pointer to correct ioctl on a per team member */
 638                 /*       bases to make this more efficient. that is, once  */
 639                 /*       we determine the correct ioctl, we will always    */
 640                 /*       call it and not the others for that team          */
 641                 /*       member.                                           */
 642
 643                 /*
 644                  * We cannot assume that SIOCGMIIPHY will also read a
 645                  * register; not all network drivers (e.g., e100)
 646                  * support that.
 647                  */
 648
 649                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 650                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 651                 mii = if_mii(&ifr);
 652                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 653                         mii->reg_num = MII_BMSR;
 654                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 655                                 return mii->val_out & BMSR_LSTATUS;
 656                 }
 657         }
 658
 659         /*
 660          * If reporting, report that either there's no dev->do_ioctl,
 661          * or both SIOCGMIIREG and get_link failed (meaning that we
 662          * cannot report link status).  If not reporting, pretend
 663          * we're ok.
 664          */
 665         return reporting ? -1 : BMSR_LSTATUS;
 666 }
 667
 668 /*----------------------------- Multicast list ------------------------------*/
 669
 670 /*
 671  * Push the promiscuity flag down to appropriate slaves
 672  */
 673 static int bond_set_promiscuity(struct bonding *bond, int inc)
 674 {
 675         int err = 0;
 676         if (USES_PRIMARY(bond->params.mode)) {
 677                 /* write lock already acquired */
 678                 if (bond->curr_active_slave) {
 679                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 680                                                   inc);
 681                 }
 682         } else {
 683                 struct slave *slave;
 684                 int i;
 685                 bond_for_each_slave(bond, slave, i) {
 686                         err = dev_set_promiscuity(slave->dev, inc);
 687                         if (err)
 688                                 return err;
 689                 }
 690         }
 691         return err;
 692 }
 693
 694 /*
 695  * Push the allmulti flag down to all slaves
 696  */
 697 static int bond_set_allmulti(struct bonding *bond, int inc)
 698 {
 699         int err = 0;
 700         if (USES_PRIMARY(bond->params.mode)) {
 701                 /* write lock already acquired */
 702                 if (bond->curr_active_slave) {
 703                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 704                                                inc);
 705                 }
 706         } else {
 707                 struct slave *slave;
 708                 int i;
 709                 bond_for_each_slave(bond, slave, i) {
 710                         err = dev_set_allmulti(slave->dev, inc);
 711                         if (err)
 712                                 return err;
 713                 }
 714         }
 715         return err;
 716 }
 717
 718 /*
 719  * Add a Multicast address to slaves
 720  * according to mode
 721  */
 722 static void bond_mc_add(struct bonding *bond, void *addr)
 723 {
 724         if (USES_PRIMARY(bond->params.mode)) {
 725                 /* write lock already acquired */
 726                 if (bond->curr_active_slave)
 727                         dev_mc_add(bond->curr_active_slave->dev, addr);
 728         } else {
 729                 struct slave *slave;
 730                 int i;
 731
 732                 bond_for_each_slave(bond, slave, i)
 733                         dev_mc_add(slave->dev, addr);
 734         }
 735 }
 736
 737 /*
 738  * Remove a multicast address from slave
 739  * according to mode
 740  */
 741 static void bond_mc_del(struct bonding *bond, void *addr)
 742 {
 743         if (USES_PRIMARY(bond->params.mode)) {
 744                 /* write lock already acquired */
 745                 if (bond->curr_active_slave)
 746                         dev_mc_del(bond->curr_active_slave->dev, addr);
 747         } else {
 748                 struct slave *slave;
 749                 int i;
 750                 bond_for_each_slave(bond, slave, i) {
 751                         dev_mc_del(slave->dev, addr);
 752                 }
 753         }
 754 }
 755
 756
 757 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 758 {
 759         struct in_device *in_dev;
 760
 761         rcu_read_lock();
 762         in_dev = __in_dev_get_rcu(dev);
 763         if (in_dev)
 764                 ip_mc_rejoin_groups(in_dev);
 765         rcu_read_unlock();
 766 }
 767
 768 /*
 769  * Retrieve the list of registered multicast addresses for the bonding
 770  * device and retransmit an IGMP JOIN request to the current active
 771  * slave.
 772  */
 773 static void bond_resend_igmp_join_requests(struct bonding *bond)
 774 {
 775         struct net_device *vlan_dev;
 776         struct vlan_entry *vlan;
 777
 778         read_lock(&bond->lock);
 779
 780         /* rejoin all groups on bond device */
 781         __bond_resend_igmp_join_requests(bond->dev);
 782
 783         /* rejoin all groups on vlan devices */
 784         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 785                 rcu_read_lock();
 786                 vlan_dev = __vlan_find_dev_deep(bond->dev,
 787                                                 vlan->vlan_id);
 788                 rcu_read_unlock();
 789                 if (vlan_dev)
 790                         __bond_resend_igmp_join_requests(vlan_dev);
 791         }
 792
 793         if (--bond->igmp_retrans > 0)
 794                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 795
 796         read_unlock(&bond->lock);
 797 }
 798
 799 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 800 {
 801         struct bonding *bond = container_of(work, struct bonding,
 802                                             mcast_work.work);
 803         bond_resend_igmp_join_requests(bond);
 804 }
 805
 806 /*
 807  * flush all members of flush->mc_list from device dev->mc_list
 808  */
 809 static void bond_mc_list_flush(struct net_device *bond_dev,
 810                                struct net_device *slave_dev)
 811 {
 812         struct bonding *bond = netdev_priv(bond_dev);
 813         struct netdev_hw_addr *ha;
 814
 815         netdev_for_each_mc_addr(ha, bond_dev)
 816                 dev_mc_del(slave_dev, ha->addr);
 817
 818         if (bond->params.mode == BOND_MODE_8023AD) {
 819                 /* del lacpdu mc addr from mc list */
 820                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 821
 822                 dev_mc_del(slave_dev, lacpdu_multicast);
 823         }
 824 }
 825
 826 /*--------------------------- Active slave change ---------------------------*/
 827
 828 /*
 829  * Update the mc list and multicast-related flags for the new and
 830  * old active slaves (if any) according to the multicast mode, and
 831  * promiscuous flags unconditionally.
 832  */
 833 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 834                          struct slave *old_active)
 835 {
 836         struct netdev_hw_addr *ha;
 837
 838         if (!USES_PRIMARY(bond->params.mode))
 839                 /* nothing to do -  mc list is already up-to-date on
 840                  * all slaves
 841                  */
 842                 return;
 843
 844         if (old_active) {
 845                 if (bond->dev->flags & IFF_PROMISC)
 846                         dev_set_promiscuity(old_active->dev, -1);
 847
 848                 if (bond->dev->flags & IFF_ALLMULTI)
 849                         dev_set_allmulti(old_active->dev, -1);
 850
 851                 netdev_for_each_mc_addr(ha, bond->dev)
 852                         dev_mc_del(old_active->dev, ha->addr);
 853         }
 854
 855         if (new_active) {
 856                 /* FIXME: Signal errors upstream. */
 857                 if (bond->dev->flags & IFF_PROMISC)
 858                         dev_set_promiscuity(new_active->dev, 1);
 859
 860                 if (bond->dev->flags & IFF_ALLMULTI)
 861                         dev_set_allmulti(new_active->dev, 1);
 862
 863                 netdev_for_each_mc_addr(ha, bond->dev)
 864                         dev_mc_add(new_active->dev, ha->addr);
 865         }
 866 }
 867
 868 /*
 869  * bond_do_fail_over_mac
 870  *
 871  * Perform special MAC address swapping for fail_over_mac settings
 872  *
 873  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 874  */
 875 static void bond_do_fail_over_mac(struct bonding *bond,
 876                                   struct slave *new_active,
 877                                   struct slave *old_active)
 878         __releases(&bond->curr_slave_lock)
 879         __releases(&bond->lock)
 880         __acquires(&bond->lock)
 881         __acquires(&bond->curr_slave_lock)
 882 {
 883         u8 tmp_mac[ETH_ALEN];
 884         struct sockaddr saddr;
 885         int rv;
 886
 887         switch (bond->params.fail_over_mac) {
 888         case BOND_FOM_ACTIVE:
 889                 if (new_active)
 890                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 891                                new_active->dev->addr_len);
 892                 break;
 893         case BOND_FOM_FOLLOW:
 894                 /*
 895                  * if new_active && old_active, swap them
 896                  * if just old_active, do nothing (going to no active slave)
 897                  * if just new_active, set new_active to bond's MAC
 898                  */
 899                 if (!new_active)
 900                         return;
 901
 902                 write_unlock_bh(&bond->curr_slave_lock);
 903                 read_unlock(&bond->lock);
 904
 905                 if (old_active) {
 906                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 907                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 908                                ETH_ALEN);
 909                         saddr.sa_family = new_active->dev->type;
 910                 } else {
 911                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 912                         saddr.sa_family = bond->dev->type;
 913                 }
 914
 915                 rv = dev_set_mac_address(new_active->dev, &saddr);
 916                 if (rv) {
 917                         pr_err("%s: Error %d setting MAC of slave %s\n",
 918                                bond->dev->name, -rv, new_active->dev->name);
 919                         goto out;
 920                 }
 921
 922                 if (!old_active)
 923                         goto out;
 924
 925                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 926                 saddr.sa_family = old_active->dev->type;
 927
 928                 rv = dev_set_mac_address(old_active->dev, &saddr);
 929                 if (rv)
 930                         pr_err("%s: Error %d setting MAC of slave %s\n",
 931                                bond->dev->name, -rv, new_active->dev->name);
 932 out:
 933                 read_lock(&bond->lock);
 934                 write_lock_bh(&bond->curr_slave_lock);
 935                 break;
 936         default:
 937                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 938                        bond->dev->name, bond->params.fail_over_mac);
 939                 break;
 940         }
 941
 942 }
 943
 944 static bool bond_should_change_active(struct bonding *bond)
 945 {
 946         struct slave *prim = bond->primary_slave;
 947         struct slave *curr = bond->curr_active_slave;
 948
 949         if (!prim || !curr || curr->link != BOND_LINK_UP)
 950                 return true;
 951         if (bond->force_primary) {
 952                 bond->force_primary = false;
 953                 return true;
 954         }
 955         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 956             (prim->speed < curr->speed ||
 957              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 958                 return false;
 959         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 960                 return false;
 961         return true;
 962 }
 963
 964 /**
 965  * find_best_interface - select the best available slave to be the active one
 966  * @bond: our bonding struct
 967  *
 968  * Warning: Caller must hold curr_slave_lock for writing.
 969  */
 970 static struct slave *bond_find_best_slave(struct bonding *bond)
 971 {
 972         struct slave *new_active, *old_active;
 973         struct slave *bestslave = NULL;
 974         int mintime = bond->params.updelay;
 975         int i;
 976
 977         new_active = bond->curr_active_slave;
 978
 979         if (!new_active) { /* there were no active slaves left */
 980                 if (bond->slave_cnt > 0)   /* found one slave */
 981                         new_active = bond->first_slave;
 982                 else
 983                         return NULL; /* still no slave, return NULL */
 984         }
 985
 986         if ((bond->primary_slave) &&
 987             bond->primary_slave->link == BOND_LINK_UP &&
 988             bond_should_change_active(bond)) {
 989                 new_active = bond->primary_slave;
 990         }
 991
 992         /* remember where to stop iterating over the slaves */
 993         old_active = new_active;
 994
 995         bond_for_each_slave_from(bond, new_active, i, old_active) {
 996                 if (new_active->link == BOND_LINK_UP) {
 997                         return new_active;
 998                 } else if (new_active->link == BOND_LINK_BACK &&
 999                            IS_UP(new_active->dev)) {
1000                         /* link up, but waiting for stabilization */
1001                         if (new_active->delay < mintime) {
1002                                 mintime = new_active->delay;
1003                                 bestslave = new_active;
1004                         }
1005                 }
1006         }
1007
1008         return bestslave;
1009 }
1010
1011 static bool bond_should_notify_peers(struct bonding *bond)
1012 {
1013         struct slave *slave = bond->curr_active_slave;
1014
1015         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1016                  bond->dev->name, slave ? slave->dev->name : "NULL");
1017
1018         if (!slave || !bond->send_peer_notif ||
1019             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1020                 return false;
1021
1022         bond->send_peer_notif--;
1023         return true;
1024 }
1025
1026 /**
1027  * change_active_interface - change the active slave into the specified one
1028  * @bond: our bonding struct
1029  * @new: the new slave to make the active one
1030  *
1031  * Set the new slave to the bond's settings and unset them on the old
1032  * curr_active_slave.
1033  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1034  *
1035  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1036  * because it is apparently the best available slave we have, even though its
1037  * updelay hasn't timed out yet.
1038  *
1039  * If new_active is not NULL, caller must hold bond->lock for read and
1040  * curr_slave_lock for write_bh.
1041  */
1042 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1043 {
1044         struct slave *old_active = bond->curr_active_slave;
1045
1046         if (old_active == new_active)
1047                 return;
1048
1049         if (new_active) {
1050                 new_active->jiffies = jiffies;
1051
1052                 if (new_active->link == BOND_LINK_BACK) {
1053                         if (USES_PRIMARY(bond->params.mode)) {
1054                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1055                                         bond->dev->name, new_active->dev->name,
1056                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1057                         }
1058
1059                         new_active->delay = 0;
1060                         new_active->link = BOND_LINK_UP;
1061
1062                         if (bond->params.mode == BOND_MODE_8023AD)
1063                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1064
1065                         if (bond_is_lb(bond))
1066                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1067                 } else {
1068                         if (USES_PRIMARY(bond->params.mode)) {
1069                                 pr_info("%s: making interface %s the new active one.\n",
1070                                         bond->dev->name, new_active->dev->name);
1071                         }
1072                 }
1073         }
1074
1075         if (USES_PRIMARY(bond->params.mode))
1076                 bond_mc_swap(bond, new_active, old_active);
1077
1078         if (bond_is_lb(bond)) {
1079                 bond_alb_handle_active_change(bond, new_active);
1080                 if (old_active)
1081                         bond_set_slave_inactive_flags(old_active);
1082                 if (new_active)
1083                         bond_set_slave_active_flags(new_active);
1084         } else {
1085                 bond->curr_active_slave = new_active;
1086         }
1087
1088         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1089                 if (old_active)
1090                         bond_set_slave_inactive_flags(old_active);
1091
1092                 if (new_active) {
1093                         bool should_notify_peers = false;
1094
1095                         bond_set_slave_active_flags(new_active);
1096
1097                         if (bond->params.fail_over_mac)
1098                                 bond_do_fail_over_mac(bond, new_active,
1099                                                       old_active);
1100
1101                         if (netif_running(bond->dev)) {
1102                                 bond->send_peer_notif =
1103                                         bond->params.num_peer_notif;
1104                                 should_notify_peers =
1105                                         bond_should_notify_peers(bond);
1106                         }
1107
1108                         write_unlock_bh(&bond->curr_slave_lock);
1109                         read_unlock(&bond->lock);
1110
1111                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1112                         if (should_notify_peers)
1113                                 netdev_bonding_change(bond->dev,
1114                                                       NETDEV_NOTIFY_PEERS);
1115
1116                         read_lock(&bond->lock);
1117                         write_lock_bh(&bond->curr_slave_lock);
1118                 }
1119         }
1120
1121         /* resend IGMP joins since active slave has changed or
1122          * all were sent on curr_active_slave.
1123          * resend only if bond is brought up with the affected
1124          * bonding modes and the retransmission is enabled */
1125         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1126             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1127              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1128                 bond->igmp_retrans = bond->params.resend_igmp;
1129                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1130         }
1131 }
1132
1133 /**
1134  * bond_select_active_slave - select a new active slave, if needed
1135  * @bond: our bonding struct
1136  *
1137  * This functions should be called when one of the following occurs:
1138  * - The old curr_active_slave has been released or lost its link.
1139  * - The primary_slave has got its link back.
1140  * - A slave has got its link back and there's no old curr_active_slave.
1141  *
1142  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1143  */
1144 void bond_select_active_slave(struct bonding *bond)
1145 {
1146         struct slave *best_slave;
1147         int rv;
1148
1149         best_slave = bond_find_best_slave(bond);
1150         if (best_slave != bond->curr_active_slave) {
1151                 bond_change_active_slave(bond, best_slave);
1152                 rv = bond_set_carrier(bond);
1153                 if (!rv)
1154                         return;
1155
1156                 if (netif_carrier_ok(bond->dev)) {
1157                         pr_info("%s: first active interface up!\n",
1158                                 bond->dev->name);
1159                 } else {
1160                         pr_info("%s: now running without any active interface !\n",
1161                                 bond->dev->name);
1162                 }
1163         }
1164 }
1165
1166 /*--------------------------- slave list handling ---------------------------*/
1167
1168 /*
1169  * This function attaches the slave to the end of list.
1170  *
1171  * bond->lock held for writing by caller.
1172  */
1173 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1174 {
1175         if (bond->first_slave == NULL) { /* attaching the first slave */
1176                 new_slave->next = new_slave;
1177                 new_slave->prev = new_slave;
1178                 bond->first_slave = new_slave;
1179         } else {
1180                 new_slave->next = bond->first_slave;
1181                 new_slave->prev = bond->first_slave->prev;
1182                 new_slave->next->prev = new_slave;
1183                 new_slave->prev->next = new_slave;
1184         }
1185
1186         bond->slave_cnt++;
1187 }
1188
1189 /*
1190  * This function detaches the slave from the list.
1191  * WARNING: no check is made to verify if the slave effectively
1192  * belongs to <bond>.
1193  * Nothing is freed on return, structures are just unchained.
1194  * If any slave pointer in bond was pointing to <slave>,
1195  * it should be changed by the calling function.
1196  *
1197  * bond->lock held for writing by caller.
1198  */
1199 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1200 {
1201         if (slave->next)
1202                 slave->next->prev = slave->prev;
1203
1204         if (slave->prev)
1205                 slave->prev->next = slave->next;
1206
1207         if (bond->first_slave == slave) { /* slave is the first slave */
1208                 if (bond->slave_cnt > 1) { /* there are more slave */
1209                         bond->first_slave = slave->next;
1210                 } else {
1211                         bond->first_slave = NULL; /* slave was the last one */
1212                 }
1213         }
1214
1215         slave->next = NULL;
1216         slave->prev = NULL;
1217         bond->slave_cnt--;
1218 }
1219
1220 #ifdef CONFIG_NET_POLL_CONTROLLER
1221 static inline int slave_enable_netpoll(struct slave *slave)
1222 {
1223         struct netpoll *np;
1224         int err = 0;
1225
1226         np = kzalloc(sizeof(*np), GFP_KERNEL);
1227         err = -ENOMEM;
1228         if (!np)
1229                 goto out;
1230
1231         np->dev = slave->dev;
1232         strlcpy(np->dev_name, slave->dev->name, IFNAMSIZ);
1233         err = __netpoll_setup(np);
1234         if (err) {
1235                 kfree(np);
1236                 goto out;
1237         }
1238         slave->np = np;
1239 out:
1240         return err;
1241 }
1242 static inline void slave_disable_netpoll(struct slave *slave)
1243 {
1244         struct netpoll *np = slave->np;
1245
1246         if (!np)
1247                 return;
1248
1249         slave->np = NULL;
1250         synchronize_rcu_bh();
1251         __netpoll_cleanup(np);
1252         kfree(np);
1253 }
1254 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1255 {
1256         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1257                 return false;
1258         if (!slave_dev->netdev_ops->ndo_poll_controller)
1259                 return false;
1260         return true;
1261 }
1262
1263 static void bond_poll_controller(struct net_device *bond_dev)
1264 {
1265 }
1266
1267 static void __bond_netpoll_cleanup(struct bonding *bond)
1268 {
1269         struct slave *slave;
1270         int i;
1271
1272         bond_for_each_slave(bond, slave, i)
1273                 if (IS_UP(slave->dev))
1274                         slave_disable_netpoll(slave);
1275 }
1276 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1277 {
1278         struct bonding *bond = netdev_priv(bond_dev);
1279
1280         read_lock(&bond->lock);
1281         __bond_netpoll_cleanup(bond);
1282         read_unlock(&bond->lock);
1283 }
1284
1285 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1286 {
1287         struct bonding *bond = netdev_priv(dev);
1288         struct slave *slave;
1289         int i, err = 0;
1290
1291         read_lock(&bond->lock);
1292         bond_for_each_slave(bond, slave, i) {
1293                 err = slave_enable_netpoll(slave);
1294                 if (err) {
1295                         __bond_netpoll_cleanup(bond);
1296                         break;
1297                 }
1298         }
1299         read_unlock(&bond->lock);
1300         return err;
1301 }
1302
1303 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1304 {
1305         return bond->dev->npinfo;
1306 }
1307
1308 #else
1309 static inline int slave_enable_netpoll(struct slave *slave)
1310 {
1311         return 0;
1312 }
1313 static inline void slave_disable_netpoll(struct slave *slave)
1314 {
1315 }
1316 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1317 {
1318 }
1319 #endif
1320
1321 /*---------------------------------- IOCTL ----------------------------------*/
1322
1323 static int bond_sethwaddr(struct net_device *bond_dev,
1324                           struct net_device *slave_dev)
1325 {
1326         pr_debug("bond_dev=%p\n", bond_dev);
1327         pr_debug("slave_dev=%p\n", slave_dev);
1328         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1329         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1330         return 0;
1331 }
1332
1333 static u32 bond_fix_features(struct net_device *dev, u32 features)
1334 {
1335         struct slave *slave;
1336         struct bonding *bond = netdev_priv(dev);
1337         u32 mask;
1338         int i;
1339
1340         read_lock(&bond->lock);
1341
1342         if (!bond->first_slave) {
1343                 /* Disable adding VLANs to empty bond. But why? --mq */
1344                 features |= NETIF_F_VLAN_CHALLENGED;
1345                 goto out;
1346         }
1347
1348         mask = features;
1349         features &= ~NETIF_F_ONE_FOR_ALL;
1350         features |= NETIF_F_ALL_FOR_ALL;
1351
1352         bond_for_each_slave(bond, slave, i) {
1353                 features = netdev_increment_features(features,
1354                                                      slave->dev->features,
1355                                                      mask);
1356         }
1357
1358 out:
1359         read_unlock(&bond->lock);
1360         return features;
1361 }
1362
1363 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1364                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1365                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1366
1367 static void bond_compute_features(struct bonding *bond)
1368 {
1369         struct slave *slave;
1370         struct net_device *bond_dev = bond->dev;
1371         u32 vlan_features = BOND_VLAN_FEATURES;
1372         unsigned short max_hard_header_len = ETH_HLEN;
1373         int i;
1374
1375         read_lock(&bond->lock);
1376
1377         if (!bond->first_slave)
1378                 goto done;
1379
1380         bond_for_each_slave(bond, slave, i) {
1381                 vlan_features = netdev_increment_features(vlan_features,
1382                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1383
1384                 if (slave->dev->hard_header_len > max_hard_header_len)
1385                         max_hard_header_len = slave->dev->hard_header_len;
1386         }
1387
1388 done:
1389         bond_dev->vlan_features = vlan_features;
1390         bond_dev->hard_header_len = max_hard_header_len;
1391
1392         read_unlock(&bond->lock);
1393
1394         netdev_change_features(bond_dev);
1395 }
1396
1397 static void bond_setup_by_slave(struct net_device *bond_dev,
1398                                 struct net_device *slave_dev)
1399 {
1400         struct bonding *bond = netdev_priv(bond_dev);
1401
1402         bond_dev->header_ops        = slave_dev->header_ops;
1403
1404         bond_dev->type              = slave_dev->type;
1405         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1406         bond_dev->addr_len          = slave_dev->addr_len;
1407
1408         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1409                 slave_dev->addr_len);
1410         bond->setup_by_slave = 1;
1411 }
1412
1413 /* On bonding slaves other than the currently active slave, suppress
1414  * duplicates except for alb non-mcast/bcast.
1415  */
1416 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1417                                             struct slave *slave,
1418                                             struct bonding *bond)
1419 {
1420         if (bond_is_slave_inactive(slave)) {
1421                 if (bond->params.mode == BOND_MODE_ALB &&
1422                     skb->pkt_type != PACKET_BROADCAST &&
1423                     skb->pkt_type != PACKET_MULTICAST)
1424                         return false;
1425                 return true;
1426         }
1427         return false;
1428 }
1429
1430 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1431 {
1432         struct sk_buff *skb = *pskb;
1433         struct slave *slave;
1434         struct bonding *bond;
1435
1436         skb = skb_share_check(skb, GFP_ATOMIC);
1437         if (unlikely(!skb))
1438                 return RX_HANDLER_CONSUMED;
1439
1440         *pskb = skb;
1441
1442         slave = bond_slave_get_rcu(skb->dev);
1443         bond = slave->bond;
1444
1445         if (bond->params.arp_interval)
1446                 slave->dev->last_rx = jiffies;
1447
1448         if (bond->recv_probe) {
1449                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1450
1451                 if (likely(nskb)) {
1452                         bond->recv_probe(nskb, bond, slave);
1453                         dev_kfree_skb(nskb);
1454                 }
1455         }
1456
1457         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1458                 return RX_HANDLER_EXACT;
1459         }
1460
1461         skb->dev = bond->dev;
1462
1463         if (bond->params.mode == BOND_MODE_ALB &&
1464             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1465             skb->pkt_type == PACKET_HOST) {
1466
1467                 if (unlikely(skb_cow_head(skb,
1468                                           skb->data - skb_mac_header(skb)))) {
1469                         kfree_skb(skb);
1470                         return RX_HANDLER_CONSUMED;
1471                 }
1472                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1473         }
1474
1475         return RX_HANDLER_ANOTHER;
1476 }
1477
1478 /* enslave device <slave> to bond device <master> */
1479 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1480 {
1481         struct bonding *bond = netdev_priv(bond_dev);
1482         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1483         struct slave *new_slave = NULL;
1484         struct netdev_hw_addr *ha;
1485         struct sockaddr addr;
1486         int link_reporting;
1487         int res = 0;
1488
1489         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1490                 slave_ops->ndo_do_ioctl == NULL) {
1491                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1492                            bond_dev->name, slave_dev->name);
1493         }
1494
1495         /* already enslaved */
1496         if (slave_dev->flags & IFF_SLAVE) {
1497                 pr_debug("Error, Device was already enslaved\n");
1498                 return -EBUSY;
1499         }
1500
1501         /* vlan challenged mutual exclusion */
1502         /* no need to lock since we're protected by rtnl_lock */
1503         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1504                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1505                 if (bond_vlan_used(bond)) {
1506                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1507                                bond_dev->name, slave_dev->name, bond_dev->name);
1508                         return -EPERM;
1509                 } else {
1510                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1511                                    bond_dev->name, slave_dev->name,
1512                                    slave_dev->name, bond_dev->name);
1513                 }
1514         } else {
1515                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1516         }
1517
1518         /*
1519          * Old ifenslave binaries are no longer supported.  These can
1520          * be identified with moderate accuracy by the state of the slave:
1521          * the current ifenslave will set the interface down prior to
1522          * enslaving it; the old ifenslave will not.
1523          */
1524         if ((slave_dev->flags & IFF_UP)) {
1525                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1526                        slave_dev->name);
1527                 res = -EPERM;
1528                 goto err_undo_flags;
1529         }
1530
1531         /* set bonding device ether type by slave - bonding netdevices are
1532          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1533          * there is a need to override some of the type dependent attribs/funcs.
1534          *
1535          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1536          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1537          */
1538         if (bond->slave_cnt == 0) {
1539                 if (bond_dev->type != slave_dev->type) {
1540                         pr_debug("%s: change device type from %d to %d\n",
1541                                  bond_dev->name,
1542                                  bond_dev->type, slave_dev->type);
1543
1544                         res = netdev_bonding_change(bond_dev,
1545                                                     NETDEV_PRE_TYPE_CHANGE);
1546                         res = notifier_to_errno(res);
1547                         if (res) {
1548                                 pr_err("%s: refused to change device type\n",
1549                                        bond_dev->name);
1550                                 res = -EBUSY;
1551                                 goto err_undo_flags;
1552                         }
1553
1554                         /* Flush unicast and multicast addresses */
1555                         dev_uc_flush(bond_dev);
1556                         dev_mc_flush(bond_dev);
1557
1558                         if (slave_dev->type != ARPHRD_ETHER)
1559                                 bond_setup_by_slave(bond_dev, slave_dev);
1560                         else
1561                                 ether_setup(bond_dev);
1562
1563                         netdev_bonding_change(bond_dev,
1564                                               NETDEV_POST_TYPE_CHANGE);
1565                 }
1566         } else if (bond_dev->type != slave_dev->type) {
1567                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1568                        slave_dev->name,
1569                        slave_dev->type, bond_dev->type);
1570                 res = -EINVAL;
1571                 goto err_undo_flags;
1572         }
1573
1574         if (slave_ops->ndo_set_mac_address == NULL) {
1575                 if (bond->slave_cnt == 0) {
1576                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1577                                    bond_dev->name);
1578                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1579                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1580                         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",
1581                                bond_dev->name);
1582                         res = -EOPNOTSUPP;
1583                         goto err_undo_flags;
1584                 }
1585         }
1586
1587         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1588
1589         /* If this is the first slave, then we need to set the master's hardware
1590          * address to be the same as the slave's. */
1591         if (is_zero_ether_addr(bond->dev->dev_addr))
1592                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1593                        slave_dev->addr_len);
1594
1595
1596         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1597         if (!new_slave) {
1598                 res = -ENOMEM;
1599                 goto err_undo_flags;
1600         }
1601
1602         /*
1603          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1604          * is set via sysfs or module option if desired.
1605          */
1606         new_slave->queue_id = 0;
1607
1608         /* Save slave's original mtu and then set it to match the bond */
1609         new_slave->original_mtu = slave_dev->mtu;
1610         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1611         if (res) {
1612                 pr_debug("Error %d calling dev_set_mtu\n", res);
1613                 goto err_free;
1614         }
1615
1616         /*
1617          * Save slave's original ("permanent") mac address for modes
1618          * that need it, and for restoring it upon release, and then
1619          * set it to the master's address
1620          */
1621         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1622
1623         if (!bond->params.fail_over_mac) {
1624                 /*
1625                  * Set slave to master's mac address.  The application already
1626                  * set the master's mac address to that of the first slave
1627                  */
1628                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1629                 addr.sa_family = slave_dev->type;
1630                 res = dev_set_mac_address(slave_dev, &addr);
1631                 if (res) {
1632                         pr_debug("Error %d calling set_mac_address\n", res);
1633                         goto err_restore_mtu;
1634                 }
1635         }
1636
1637         res = netdev_set_bond_master(slave_dev, bond_dev);
1638         if (res) {
1639                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1640                 goto err_restore_mac;
1641         }
1642
1643         /* open the slave since the application closed it */
1644         res = dev_open(slave_dev);
1645         if (res) {
1646                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1647                 goto err_unset_master;
1648         }
1649
1650         new_slave->bond = bond;
1651         new_slave->dev = slave_dev;
1652         slave_dev->priv_flags |= IFF_BONDING;
1653
1654         if (bond_is_lb(bond)) {
1655                 /* bond_alb_init_slave() must be called before all other stages since
1656                  * it might fail and we do not want to have to undo everything
1657                  */
1658                 res = bond_alb_init_slave(bond, new_slave);
1659                 if (res)
1660                         goto err_close;
1661         }
1662
1663         /* If the mode USES_PRIMARY, then the new slave gets the
1664          * master's promisc (and mc) settings only if it becomes the
1665          * curr_active_slave, and that is taken care of later when calling
1666          * bond_change_active()
1667          */
1668         if (!USES_PRIMARY(bond->params.mode)) {
1669                 /* set promiscuity level to new slave */
1670                 if (bond_dev->flags & IFF_PROMISC) {
1671                         res = dev_set_promiscuity(slave_dev, 1);
1672                         if (res)
1673                                 goto err_close;
1674                 }
1675
1676                 /* set allmulti level to new slave */
1677                 if (bond_dev->flags & IFF_ALLMULTI) {
1678                         res = dev_set_allmulti(slave_dev, 1);
1679                         if (res)
1680                                 goto err_close;
1681                 }
1682
1683                 netif_addr_lock_bh(bond_dev);
1684                 /* upload master's mc_list to new slave */
1685                 netdev_for_each_mc_addr(ha, bond_dev)
1686                         dev_mc_add(slave_dev, ha->addr);
1687                 netif_addr_unlock_bh(bond_dev);
1688         }
1689
1690         if (bond->params.mode == BOND_MODE_8023AD) {
1691                 /* add lacpdu mc addr to mc list */
1692                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1693
1694                 dev_mc_add(slave_dev, lacpdu_multicast);
1695         }
1696
1697         bond_add_vlans_on_slave(bond, slave_dev);
1698
1699         write_lock_bh(&bond->lock);
1700
1701         bond_attach_slave(bond, new_slave);
1702
1703         new_slave->delay = 0;
1704         new_slave->link_failure_count = 0;
1705
1706         write_unlock_bh(&bond->lock);
1707
1708         bond_compute_features(bond);
1709
1710         read_lock(&bond->lock);
1711
1712         new_slave->last_arp_rx = jiffies;
1713
1714         if (bond->params.miimon && !bond->params.use_carrier) {
1715                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1716
1717                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1718                         /*
1719                          * miimon is set but a bonded network driver
1720                          * does not support ETHTOOL/MII and
1721                          * arp_interval is not set.  Note: if
1722                          * use_carrier is enabled, we will never go
1723                          * here (because netif_carrier is always
1724                          * supported); thus, we don't need to change
1725                          * the messages for netif_carrier.
1726                          */
1727                         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",
1728                                bond_dev->name, slave_dev->name);
1729                 } else if (link_reporting == -1) {
1730                         /* unable get link status using mii/ethtool */
1731                         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",
1732                                    bond_dev->name, slave_dev->name);
1733                 }
1734         }
1735
1736         /* check for initial state */
1737         if (!bond->params.miimon ||
1738             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1739                 if (bond->params.updelay) {
1740                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1741                         new_slave->link  = BOND_LINK_BACK;
1742                         new_slave->delay = bond->params.updelay;
1743                 } else {
1744                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1745                         new_slave->link  = BOND_LINK_UP;
1746                 }
1747                 new_slave->jiffies = jiffies;
1748         } else {
1749                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1750                 new_slave->link  = BOND_LINK_DOWN;
1751         }
1752
1753         if (bond_update_speed_duplex(new_slave) &&
1754             (new_slave->link != BOND_LINK_DOWN)) {
1755                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1756                            bond_dev->name, new_slave->dev->name);
1757
1758                 if (bond->params.mode == BOND_MODE_8023AD) {
1759                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1760                                    bond_dev->name);
1761                 }
1762         }
1763
1764         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1765                 /* if there is a primary slave, remember it */
1766                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1767                         bond->primary_slave = new_slave;
1768                         bond->force_primary = true;
1769                 }
1770         }
1771
1772         write_lock_bh(&bond->curr_slave_lock);
1773
1774         switch (bond->params.mode) {
1775         case BOND_MODE_ACTIVEBACKUP:
1776                 bond_set_slave_inactive_flags(new_slave);
1777                 bond_select_active_slave(bond);
1778                 break;
1779         case BOND_MODE_8023AD:
1780                 /* in 802.3ad mode, the internal mechanism
1781                  * will activate the slaves in the selected
1782                  * aggregator
1783                  */
1784                 bond_set_slave_inactive_flags(new_slave);
1785                 /* if this is the first slave */
1786                 if (bond->slave_cnt == 1) {
1787                         SLAVE_AD_INFO(new_slave).id = 1;
1788                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1789                          * can be called only after the mac address of the bond is set
1790                          */
1791                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1792                 } else {
1793                         SLAVE_AD_INFO(new_slave).id =
1794                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1795                 }
1796
1797                 bond_3ad_bind_slave(new_slave);
1798                 break;
1799         case BOND_MODE_TLB:
1800         case BOND_MODE_ALB:
1801                 bond_set_active_slave(new_slave);
1802                 bond_set_slave_inactive_flags(new_slave);
1803                 bond_select_active_slave(bond);
1804                 break;
1805         default:
1806                 pr_debug("This slave is always active in trunk mode\n");
1807
1808                 /* always active in trunk mode */
1809                 bond_set_active_slave(new_slave);
1810
1811                 /* In trunking mode there is little meaning to curr_active_slave
1812                  * anyway (it holds no special properties of the bond device),
1813                  * so we can change it without calling change_active_interface()
1814                  */
1815                 if (!bond->curr_active_slave)
1816                         bond->curr_active_slave = new_slave;
1817
1818                 break;
1819         } /* switch(bond_mode) */
1820
1821         write_unlock_bh(&bond->curr_slave_lock);
1822
1823         bond_set_carrier(bond);
1824
1825 #ifdef CONFIG_NET_POLL_CONTROLLER
1826         slave_dev->npinfo = bond_netpoll_info(bond);
1827         if (slave_dev->npinfo) {
1828                 if (slave_enable_netpoll(new_slave)) {
1829                         read_unlock(&bond->lock);
1830                         pr_info("Error, %s: master_dev is using netpoll, "
1831                                  "but new slave device does not support netpoll.\n",
1832                                  bond_dev->name);
1833                         res = -EBUSY;
1834                         goto err_close;
1835                 }
1836         }
1837 #endif
1838
1839         read_unlock(&bond->lock);
1840
1841         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1842         if (res)
1843                 goto err_close;
1844
1845         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1846                                          new_slave);
1847         if (res) {
1848                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1849                 goto err_dest_symlinks;
1850         }
1851
1852         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1853                 bond_dev->name, slave_dev->name,
1854                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1855                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1856
1857         /* enslave is successful */
1858         return 0;
1859
1860 /* Undo stages on error */
1861 err_dest_symlinks:
1862         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1863
1864 err_close:
1865         dev_close(slave_dev);
1866
1867 err_unset_master:
1868         netdev_set_bond_master(slave_dev, NULL);
1869
1870 err_restore_mac:
1871         if (!bond->params.fail_over_mac) {
1872                 /* XXX TODO - fom follow mode needs to change master's
1873                  * MAC if this slave's MAC is in use by the bond, or at
1874                  * least print a warning.
1875                  */
1876                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1877                 addr.sa_family = slave_dev->type;
1878                 dev_set_mac_address(slave_dev, &addr);
1879         }
1880
1881 err_restore_mtu:
1882         dev_set_mtu(slave_dev, new_slave->original_mtu);
1883
1884 err_free:
1885         kfree(new_slave);
1886
1887 err_undo_flags:
1888         bond_compute_features(bond);
1889
1890         return res;
1891 }
1892
1893 /*
1894  * Try to release the slave device <slave> from the bond device <master>
1895  * It is legal to access curr_active_slave without a lock because all the function
1896  * is write-locked.
1897  *
1898  * The rules for slave state should be:
1899  *   for Active/Backup:
1900  *     Active stays on all backups go down
1901  *   for Bonded connections:
1902  *     The first up interface should be left on and all others downed.
1903  */
1904 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1905 {
1906         struct bonding *bond = netdev_priv(bond_dev);
1907         struct slave *slave, *oldcurrent;
1908         struct sockaddr addr;
1909         u32 old_features = bond_dev->features;
1910
1911         /* slave is not a slave or master is not master of this slave */
1912         if (!(slave_dev->flags & IFF_SLAVE) ||
1913             (slave_dev->master != bond_dev)) {
1914                 pr_err("%s: Error: cannot release %s.\n",
1915                        bond_dev->name, slave_dev->name);
1916                 return -EINVAL;
1917         }
1918
1919         block_netpoll_tx();
1920         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1921         write_lock_bh(&bond->lock);
1922
1923         slave = bond_get_slave_by_dev(bond, slave_dev);
1924         if (!slave) {
1925                 /* not a slave of this bond */
1926                 pr_info("%s: %s not enslaved\n",
1927                         bond_dev->name, slave_dev->name);
1928                 write_unlock_bh(&bond->lock);
1929                 unblock_netpoll_tx();
1930                 return -EINVAL;
1931         }
1932
1933         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1934          * for this slave anymore.
1935          */
1936         netdev_rx_handler_unregister(slave_dev);
1937         write_unlock_bh(&bond->lock);
1938         synchronize_net();
1939         write_lock_bh(&bond->lock);
1940
1941         if (!bond->params.fail_over_mac) {
1942                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1943                     bond->slave_cnt > 1)
1944                         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",
1945                                    bond_dev->name, slave_dev->name,
1946                                    slave->perm_hwaddr,
1947                                    bond_dev->name, slave_dev->name);
1948         }
1949
1950         /* Inform AD package of unbinding of slave. */
1951         if (bond->params.mode == BOND_MODE_8023AD) {
1952                 /* must be called before the slave is
1953                  * detached from the list
1954                  */
1955                 bond_3ad_unbind_slave(slave);
1956         }
1957
1958         pr_info("%s: releasing %s interface %s\n",
1959                 bond_dev->name,
1960                 bond_is_active_slave(slave) ? "active" : "backup",
1961                 slave_dev->name);
1962
1963         oldcurrent = bond->curr_active_slave;
1964
1965         bond->current_arp_slave = NULL;
1966
1967         /* release the slave from its bond */
1968         bond_detach_slave(bond, slave);
1969
1970         if (bond->primary_slave == slave)
1971                 bond->primary_slave = NULL;
1972
1973         if (oldcurrent == slave)
1974                 bond_change_active_slave(bond, NULL);
1975
1976         if (bond_is_lb(bond)) {
1977                 /* Must be called only after the slave has been
1978                  * detached from the list and the curr_active_slave
1979                  * has been cleared (if our_slave == old_current),
1980                  * but before a new active slave is selected.
1981                  */
1982                 write_unlock_bh(&bond->lock);
1983                 bond_alb_deinit_slave(bond, slave);
1984                 write_lock_bh(&bond->lock);
1985         }
1986
1987         if (oldcurrent == slave) {
1988                 /*
1989                  * Note that we hold RTNL over this sequence, so there
1990                  * is no concern that another slave add/remove event
1991                  * will interfere.
1992                  */
1993                 write_unlock_bh(&bond->lock);
1994                 read_lock(&bond->lock);
1995                 write_lock_bh(&bond->curr_slave_lock);
1996
1997                 bond_select_active_slave(bond);
1998
1999                 write_unlock_bh(&bond->curr_slave_lock);
2000                 read_unlock(&bond->lock);
2001                 write_lock_bh(&bond->lock);
2002         }
2003
2004         if (bond->slave_cnt == 0) {
2005                 bond_set_carrier(bond);
2006
2007                 /* if the last slave was removed, zero the mac address
2008                  * of the master so it will be set by the application
2009                  * to the mac address of the first slave
2010                  */
2011                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2012
2013                 if (bond_vlan_used(bond)) {
2014                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2015                                    bond_dev->name, bond_dev->name);
2016                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2017                                    bond_dev->name);
2018                 }
2019         }
2020
2021         write_unlock_bh(&bond->lock);
2022         unblock_netpoll_tx();
2023
2024         bond_compute_features(bond);
2025         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2026             (old_features & NETIF_F_VLAN_CHALLENGED))
2027                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2028                         bond_dev->name, slave_dev->name, bond_dev->name);
2029
2030         /* must do this from outside any spinlocks */
2031         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2032
2033         bond_del_vlans_from_slave(bond, slave_dev);
2034
2035         /* If the mode USES_PRIMARY, then we should only remove its
2036          * promisc and mc settings if it was the curr_active_slave, but that was
2037          * already taken care of above when we detached the slave
2038          */
2039         if (!USES_PRIMARY(bond->params.mode)) {
2040                 /* unset promiscuity level from slave */
2041                 if (bond_dev->flags & IFF_PROMISC)
2042                         dev_set_promiscuity(slave_dev, -1);
2043
2044                 /* unset allmulti level from slave */
2045                 if (bond_dev->flags & IFF_ALLMULTI)
2046                         dev_set_allmulti(slave_dev, -1);
2047
2048                 /* flush master's mc_list from slave */
2049                 netif_addr_lock_bh(bond_dev);
2050                 bond_mc_list_flush(bond_dev, slave_dev);
2051                 netif_addr_unlock_bh(bond_dev);
2052         }
2053
2054         netdev_set_bond_master(slave_dev, NULL);
2055
2056         slave_disable_netpoll(slave);
2057
2058         /* close slave before restoring its mac address */
2059         dev_close(slave_dev);
2060
2061         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2062                 /* restore original ("permanent") mac address */
2063                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2064                 addr.sa_family = slave_dev->type;
2065                 dev_set_mac_address(slave_dev, &addr);
2066         }
2067
2068         dev_set_mtu(slave_dev, slave->original_mtu);
2069
2070         slave_dev->priv_flags &= ~IFF_BONDING;
2071
2072         kfree(slave);
2073
2074         return 0;  /* deletion OK */
2075 }
2076
2077 /*
2078 * First release a slave and then destroy the bond if no more slaves are left.
2079 * Must be under rtnl_lock when this function is called.
2080 */
2081 static int  bond_release_and_destroy(struct net_device *bond_dev,
2082                                      struct net_device *slave_dev)
2083 {
2084         struct bonding *bond = netdev_priv(bond_dev);
2085         int ret;
2086
2087         ret = bond_release(bond_dev, slave_dev);
2088         if ((ret == 0) && (bond->slave_cnt == 0)) {
2089                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2090                 pr_info("%s: destroying bond %s.\n",
2091                         bond_dev->name, bond_dev->name);
2092                 unregister_netdevice(bond_dev);
2093         }
2094         return ret;
2095 }
2096
2097 /*
2098  * This function releases all slaves.
2099  */
2100 static int bond_release_all(struct net_device *bond_dev)
2101 {
2102         struct bonding *bond = netdev_priv(bond_dev);
2103         struct slave *slave;
2104         struct net_device *slave_dev;
2105         struct sockaddr addr;
2106
2107         write_lock_bh(&bond->lock);
2108
2109         netif_carrier_off(bond_dev);
2110
2111         if (bond->slave_cnt == 0)
2112                 goto out;
2113
2114         bond->current_arp_slave = NULL;
2115         bond->primary_slave = NULL;
2116         bond_change_active_slave(bond, NULL);
2117
2118         while ((slave = bond->first_slave) != NULL) {
2119                 /* Inform AD package of unbinding of slave
2120                  * before slave is detached from the list.
2121                  */
2122                 if (bond->params.mode == BOND_MODE_8023AD)
2123                         bond_3ad_unbind_slave(slave);
2124
2125                 slave_dev = slave->dev;
2126                 bond_detach_slave(bond, slave);
2127
2128                 /* now that the slave is detached, unlock and perform
2129                  * all the undo steps that should not be called from
2130                  * within a lock.
2131                  */
2132                 write_unlock_bh(&bond->lock);
2133
2134                 /* unregister rx_handler early so bond_handle_frame wouldn't
2135                  * be called for this slave anymore.
2136                  */
2137                 netdev_rx_handler_unregister(slave_dev);
2138                 synchronize_net();
2139
2140                 if (bond_is_lb(bond)) {
2141                         /* must be called only after the slave
2142                          * has been detached from the list
2143                          */
2144                         bond_alb_deinit_slave(bond, slave);
2145                 }
2146
2147                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2148                 bond_del_vlans_from_slave(bond, slave_dev);
2149
2150                 /* If the mode USES_PRIMARY, then we should only remove its
2151                  * promisc and mc settings if it was the curr_active_slave, but that was
2152                  * already taken care of above when we detached the slave
2153                  */
2154                 if (!USES_PRIMARY(bond->params.mode)) {
2155                         /* unset promiscuity level from slave */
2156                         if (bond_dev->flags & IFF_PROMISC)
2157                                 dev_set_promiscuity(slave_dev, -1);
2158
2159                         /* unset allmulti level from slave */
2160                         if (bond_dev->flags & IFF_ALLMULTI)
2161                                 dev_set_allmulti(slave_dev, -1);
2162
2163                         /* flush master's mc_list from slave */
2164                         netif_addr_lock_bh(bond_dev);
2165                         bond_mc_list_flush(bond_dev, slave_dev);
2166                         netif_addr_unlock_bh(bond_dev);
2167                 }
2168
2169                 netdev_set_bond_master(slave_dev, NULL);
2170
2171                 slave_disable_netpoll(slave);
2172
2173                 /* close slave before restoring its mac address */
2174                 dev_close(slave_dev);
2175
2176                 if (!bond->params.fail_over_mac) {
2177                         /* restore original ("permanent") mac address*/
2178                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2179                         addr.sa_family = slave_dev->type;
2180                         dev_set_mac_address(slave_dev, &addr);
2181                 }
2182
2183                 kfree(slave);
2184
2185                 /* re-acquire the lock before getting the next slave */
2186                 write_lock_bh(&bond->lock);
2187         }
2188
2189         /* zero the mac address of the master so it will be
2190          * set by the application to the mac address of the
2191          * first slave
2192          */
2193         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2194
2195         if (bond_vlan_used(bond)) {
2196                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2197                            bond_dev->name, bond_dev->name);
2198                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2199                            bond_dev->name);
2200         }
2201
2202         pr_info("%s: released all slaves\n", bond_dev->name);
2203
2204 out:
2205         write_unlock_bh(&bond->lock);
2206
2207         bond_compute_features(bond);
2208
2209         return 0;
2210 }
2211
2212 /*
2213  * This function changes the active slave to slave <slave_dev>.
2214  * It returns -EINVAL in the following cases.
2215  *  - <slave_dev> is not found in the list.
2216  *  - There is not active slave now.
2217  *  - <slave_dev> is already active.
2218  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2219  *  - <slave_dev> is not running.
2220  * In these cases, this function does nothing.
2221  * In the other cases, current_slave pointer is changed and 0 is returned.
2222  */
2223 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2224 {
2225         struct bonding *bond = netdev_priv(bond_dev);
2226         struct slave *old_active = NULL;
2227         struct slave *new_active = NULL;
2228         int res = 0;
2229
2230         if (!USES_PRIMARY(bond->params.mode))
2231                 return -EINVAL;
2232
2233         /* Verify that master_dev is indeed the master of slave_dev */
2234         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2235                 return -EINVAL;
2236
2237         read_lock(&bond->lock);
2238
2239         read_lock(&bond->curr_slave_lock);
2240         old_active = bond->curr_active_slave;
2241         read_unlock(&bond->curr_slave_lock);
2242
2243         new_active = bond_get_slave_by_dev(bond, slave_dev);
2244
2245         /*
2246          * Changing to the current active: do nothing; return success.
2247          */
2248         if (new_active && (new_active == old_active)) {
2249                 read_unlock(&bond->lock);
2250                 return 0;
2251         }
2252
2253         if ((new_active) &&
2254             (old_active) &&
2255             (new_active->link == BOND_LINK_UP) &&
2256             IS_UP(new_active->dev)) {
2257                 block_netpoll_tx();
2258                 write_lock_bh(&bond->curr_slave_lock);
2259                 bond_change_active_slave(bond, new_active);
2260                 write_unlock_bh(&bond->curr_slave_lock);
2261                 unblock_netpoll_tx();
2262         } else
2263                 res = -EINVAL;
2264
2265         read_unlock(&bond->lock);
2266
2267         return res;
2268 }
2269
2270 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2271 {
2272         struct bonding *bond = netdev_priv(bond_dev);
2273
2274         info->bond_mode = bond->params.mode;
2275         info->miimon = bond->params.miimon;
2276
2277         read_lock(&bond->lock);
2278         info->num_slaves = bond->slave_cnt;
2279         read_unlock(&bond->lock);
2280
2281         return 0;
2282 }
2283
2284 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2285 {
2286         struct bonding *bond = netdev_priv(bond_dev);
2287         struct slave *slave;
2288         int i, res = -ENODEV;
2289
2290         read_lock(&bond->lock);
2291
2292         bond_for_each_slave(bond, slave, i) {
2293                 if (i == (int)info->slave_id) {
2294                         res = 0;
2295                         strcpy(info->slave_name, slave->dev->name);
2296                         info->link = slave->link;
2297                         info->state = bond_slave_state(slave);
2298                         info->link_failure_count = slave->link_failure_count;
2299                         break;
2300                 }
2301         }
2302
2303         read_unlock(&bond->lock);
2304
2305         return res;
2306 }
2307
2308 /*-------------------------------- Monitoring -------------------------------*/
2309
2310
2311 static int bond_miimon_inspect(struct bonding *bond)
2312 {
2313         struct slave *slave;
2314         int i, link_state, commit = 0;
2315         bool ignore_updelay;
2316
2317         ignore_updelay = !bond->curr_active_slave ? true : false;
2318
2319         bond_for_each_slave(bond, slave, i) {
2320                 slave->new_link = BOND_LINK_NOCHANGE;
2321
2322                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2323
2324                 switch (slave->link) {
2325                 case BOND_LINK_UP:
2326                         if (link_state)
2327                                 continue;
2328
2329                         slave->link = BOND_LINK_FAIL;
2330                         slave->delay = bond->params.downdelay;
2331                         if (slave->delay) {
2332                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2333                                         bond->dev->name,
2334                                         (bond->params.mode ==
2335                                          BOND_MODE_ACTIVEBACKUP) ?
2336                                         (bond_is_active_slave(slave) ?
2337                                          "active " : "backup ") : "",
2338                                         slave->dev->name,
2339                                         bond->params.downdelay * bond->params.miimon);
2340                         }
2341                         /*FALLTHRU*/
2342                 case BOND_LINK_FAIL:
2343                         if (link_state) {
2344                                 /*
2345                                  * recovered before downdelay expired
2346                                  */
2347                                 slave->link = BOND_LINK_UP;
2348                                 slave->jiffies = jiffies;
2349                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2350                                         bond->dev->name,
2351                                         (bond->params.downdelay - slave->delay) *
2352                                         bond->params.miimon,
2353                                         slave->dev->name);
2354                                 continue;
2355                         }
2356
2357                         if (slave->delay <= 0) {
2358                                 slave->new_link = BOND_LINK_DOWN;
2359                                 commit++;
2360                                 continue;
2361                         }
2362
2363                         slave->delay--;
2364                         break;
2365
2366                 case BOND_LINK_DOWN:
2367                         if (!link_state)
2368                                 continue;
2369
2370                         slave->link = BOND_LINK_BACK;
2371                         slave->delay = bond->params.updelay;
2372
2373                         if (slave->delay) {
2374                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2375                                         bond->dev->name, slave->dev->name,
2376                                         ignore_updelay ? 0 :
2377                                         bond->params.updelay *
2378                                         bond->params.miimon);
2379                         }
2380                         /*FALLTHRU*/
2381                 case BOND_LINK_BACK:
2382                         if (!link_state) {
2383                                 slave->link = BOND_LINK_DOWN;
2384                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2385                                         bond->dev->name,
2386                                         (bond->params.updelay - slave->delay) *
2387                                         bond->params.miimon,
2388                                         slave->dev->name);
2389
2390                                 continue;
2391                         }
2392
2393                         if (ignore_updelay)
2394                                 slave->delay = 0;
2395
2396                         if (slave->delay <= 0) {
2397                                 slave->new_link = BOND_LINK_UP;
2398                                 commit++;
2399                                 ignore_updelay = false;
2400                                 continue;
2401                         }
2402
2403                         slave->delay--;
2404                         break;
2405                 }
2406         }
2407
2408         return commit;
2409 }
2410
2411 static void bond_miimon_commit(struct bonding *bond)
2412 {
2413         struct slave *slave;
2414         int i;
2415
2416         bond_for_each_slave(bond, slave, i) {
2417                 switch (slave->new_link) {
2418                 case BOND_LINK_NOCHANGE:
2419                         continue;
2420
2421                 case BOND_LINK_UP:
2422                         slave->link = BOND_LINK_UP;
2423                         slave->jiffies = jiffies;
2424
2425                         if (bond->params.mode == BOND_MODE_8023AD) {
2426                                 /* prevent it from being the active one */
2427                                 bond_set_backup_slave(slave);
2428                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2429                                 /* make it immediately active */
2430                                 bond_set_active_slave(slave);
2431                         } else if (slave != bond->primary_slave) {
2432                                 /* prevent it from being the active one */
2433                                 bond_set_backup_slave(slave);
2434                         }
2435
2436                         bond_update_speed_duplex(slave);
2437
2438                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2439                                 bond->dev->name, slave->dev->name,
2440                                 slave->speed, slave->duplex ? "full" : "half");
2441
2442                         /* notify ad that the link status has changed */
2443                         if (bond->params.mode == BOND_MODE_8023AD)
2444                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2445
2446                         if (bond_is_lb(bond))
2447                                 bond_alb_handle_link_change(bond, slave,
2448                                                             BOND_LINK_UP);
2449
2450                         if (!bond->curr_active_slave ||
2451                             (slave == bond->primary_slave))
2452                                 goto do_failover;
2453
2454                         continue;
2455
2456                 case BOND_LINK_DOWN:
2457                         if (slave->link_failure_count < UINT_MAX)
2458                                 slave->link_failure_count++;
2459
2460                         slave->link = BOND_LINK_DOWN;
2461
2462                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2463                             bond->params.mode == BOND_MODE_8023AD)
2464                                 bond_set_slave_inactive_flags(slave);
2465
2466                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2467                                 bond->dev->name, slave->dev->name);
2468
2469                         if (bond->params.mode == BOND_MODE_8023AD)
2470                                 bond_3ad_handle_link_change(slave,
2471                                                             BOND_LINK_DOWN);
2472
2473                         if (bond_is_lb(bond))
2474                                 bond_alb_handle_link_change(bond, slave,
2475                                                             BOND_LINK_DOWN);
2476
2477                         if (slave == bond->curr_active_slave)
2478                                 goto do_failover;
2479
2480                         continue;
2481
2482                 default:
2483                         pr_err("%s: invalid new link %d on slave %s\n",
2484                                bond->dev->name, slave->new_link,
2485                                slave->dev->name);
2486                         slave->new_link = BOND_LINK_NOCHANGE;
2487
2488                         continue;
2489                 }
2490
2491 do_failover:
2492                 ASSERT_RTNL();
2493                 block_netpoll_tx();
2494                 write_lock_bh(&bond->curr_slave_lock);
2495                 bond_select_active_slave(bond);
2496                 write_unlock_bh(&bond->curr_slave_lock);
2497                 unblock_netpoll_tx();
2498         }
2499
2500         bond_set_carrier(bond);
2501 }
2502
2503 /*
2504  * bond_mii_monitor
2505  *
2506  * Really a wrapper that splits the mii monitor into two phases: an
2507  * inspection, then (if inspection indicates something needs to be done)
2508  * an acquisition of appropriate locks followed by a commit phase to
2509  * implement whatever link state changes are indicated.
2510  */
2511 void bond_mii_monitor(struct work_struct *work)
2512 {
2513         struct bonding *bond = container_of(work, struct bonding,
2514                                             mii_work.work);
2515         bool should_notify_peers = false;
2516
2517         read_lock(&bond->lock);
2518         if (bond->kill_timers)
2519                 goto out;
2520
2521         if (bond->slave_cnt == 0)
2522                 goto re_arm;
2523
2524         should_notify_peers = bond_should_notify_peers(bond);
2525
2526         if (bond_miimon_inspect(bond)) {
2527                 read_unlock(&bond->lock);
2528                 rtnl_lock();
2529                 read_lock(&bond->lock);
2530
2531                 bond_miimon_commit(bond);
2532
2533                 read_unlock(&bond->lock);
2534                 rtnl_unlock();  /* might sleep, hold no other locks */
2535                 read_lock(&bond->lock);
2536         }
2537
2538 re_arm:
2539         if (bond->params.miimon)
2540                 queue_delayed_work(bond->wq, &bond->mii_work,
2541                                    msecs_to_jiffies(bond->params.miimon));
2542 out:
2543         read_unlock(&bond->lock);
2544
2545         if (should_notify_peers) {
2546                 rtnl_lock();
2547                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2548                 rtnl_unlock();
2549         }
2550 }
2551
2552 static __be32 bond_glean_dev_ip(struct net_device *dev)
2553 {
2554         struct in_device *idev;
2555         struct in_ifaddr *ifa;
2556         __be32 addr = 0;
2557
2558         if (!dev)
2559                 return 0;
2560
2561         rcu_read_lock();
2562         idev = __in_dev_get_rcu(dev);
2563         if (!idev)
2564                 goto out;
2565
2566         ifa = idev->ifa_list;
2567         if (!ifa)
2568                 goto out;
2569
2570         addr = ifa->ifa_local;
2571 out:
2572         rcu_read_unlock();
2573         return addr;
2574 }
2575
2576 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2577 {
2578         struct vlan_entry *vlan;
2579
2580         if (ip == bond->master_ip)
2581                 return 1;
2582
2583         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2584                 if (ip == vlan->vlan_ip)
2585                         return 1;
2586         }
2587
2588         return 0;
2589 }
2590
2591 /*
2592  * We go to the (large) trouble of VLAN tagging ARP frames because
2593  * switches in VLAN mode (especially if ports are configured as
2594  * "native" to a VLAN) might not pass non-tagged frames.
2595  */
2596 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2597 {
2598         struct sk_buff *skb;
2599
2600         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2601                  slave_dev->name, dest_ip, src_ip, vlan_id);
2602
2603         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2604                          NULL, slave_dev->dev_addr, NULL);
2605
2606         if (!skb) {
2607                 pr_err("ARP packet allocation failed\n");
2608                 return;
2609         }
2610         if (vlan_id) {
2611                 skb = vlan_put_tag(skb, vlan_id);
2612                 if (!skb) {
2613                         pr_err("failed to insert VLAN tag\n");
2614                         return;
2615                 }
2616         }
2617         arp_xmit(skb);
2618 }
2619
2620
2621 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2622 {
2623         int i, vlan_id;
2624         __be32 *targets = bond->params.arp_targets;
2625         struct vlan_entry *vlan;
2626         struct net_device *vlan_dev;
2627         struct rtable *rt;
2628
2629         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2630                 if (!targets[i])
2631                         break;
2632                 pr_debug("basa: target %x\n", targets[i]);
2633                 if (!bond_vlan_used(bond)) {
2634                         pr_debug("basa: empty vlan: arp_send\n");
2635                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2636                                       bond->master_ip, 0);
2637                         continue;
2638                 }
2639
2640                 /*
2641                  * If VLANs are configured, we do a route lookup to
2642                  * determine which VLAN interface would be used, so we
2643                  * can tag the ARP with the proper VLAN tag.
2644                  */
2645                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2646                                      RTO_ONLINK, 0);
2647                 if (IS_ERR(rt)) {
2648                         if (net_ratelimit()) {
2649                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2650                                            bond->dev->name, &targets[i]);
2651                         }
2652                         continue;
2653                 }
2654
2655                 /*
2656                  * This target is not on a VLAN
2657                  */
2658                 if (rt->dst.dev == bond->dev) {
2659                         ip_rt_put(rt);
2660                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2661                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2662                                       bond->master_ip, 0);
2663                         continue;
2664                 }
2665
2666                 vlan_id = 0;
2667                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2668                         rcu_read_lock();
2669                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2670                                                         vlan->vlan_id);
2671                         rcu_read_unlock();
2672                         if (vlan_dev == rt->dst.dev) {
2673                                 vlan_id = vlan->vlan_id;
2674                                 pr_debug("basa: vlan match on %s %d\n",
2675                                        vlan_dev->name, vlan_id);
2676                                 break;
2677                         }
2678                 }
2679
2680                 if (vlan_id) {
2681                         ip_rt_put(rt);
2682                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2683                                       vlan->vlan_ip, vlan_id);
2684                         continue;
2685                 }
2686
2687                 if (net_ratelimit()) {
2688                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2689                                    bond->dev->name, &targets[i],
2690                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2691                 }
2692                 ip_rt_put(rt);
2693         }
2694 }
2695
2696 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2697 {
2698         int i;
2699         __be32 *targets = bond->params.arp_targets;
2700
2701         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2702                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2703                          &sip, &tip, i, &targets[i],
2704                          bond_has_this_ip(bond, tip));
2705                 if (sip == targets[i]) {
2706                         if (bond_has_this_ip(bond, tip))
2707                                 slave->last_arp_rx = jiffies;
2708                         return;
2709                 }
2710         }
2711 }
2712
2713 static void bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2714                          struct slave *slave)
2715 {
2716         struct arphdr *arp;
2717         unsigned char *arp_ptr;
2718         __be32 sip, tip;
2719
2720         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2721                 return;
2722
2723         read_lock(&bond->lock);
2724
2725         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2726                  bond->dev->name, skb->dev->name);
2727
2728         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2729                 goto out_unlock;
2730
2731         arp = arp_hdr(skb);
2732         if (arp->ar_hln != bond->dev->addr_len ||
2733             skb->pkt_type == PACKET_OTHERHOST ||
2734             skb->pkt_type == PACKET_LOOPBACK ||
2735             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2736             arp->ar_pro != htons(ETH_P_IP) ||
2737             arp->ar_pln != 4)
2738                 goto out_unlock;
2739
2740         arp_ptr = (unsigned char *)(arp + 1);
2741         arp_ptr += bond->dev->addr_len;
2742         memcpy(&sip, arp_ptr, 4);
2743         arp_ptr += 4 + bond->dev->addr_len;
2744         memcpy(&tip, arp_ptr, 4);
2745
2746         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2747                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2748                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2749                  &sip, &tip);
2750
2751         /*
2752          * Backup slaves won't see the ARP reply, but do come through
2753          * here for each ARP probe (so we swap the sip/tip to validate
2754          * the probe).  In a "redundant switch, common router" type of
2755          * configuration, the ARP probe will (hopefully) travel from
2756          * the active, through one switch, the router, then the other
2757          * switch before reaching the backup.
2758          */
2759         if (bond_is_active_slave(slave))
2760                 bond_validate_arp(bond, slave, sip, tip);
2761         else
2762                 bond_validate_arp(bond, slave, tip, sip);
2763
2764 out_unlock:
2765         read_unlock(&bond->lock);
2766 }
2767
2768 /*
2769  * this function is called regularly to monitor each slave's link
2770  * ensuring that traffic is being sent and received when arp monitoring
2771  * is used in load-balancing mode. if the adapter has been dormant, then an
2772  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2773  * arp monitoring in active backup mode.
2774  */
2775 void bond_loadbalance_arp_mon(struct work_struct *work)
2776 {
2777         struct bonding *bond = container_of(work, struct bonding,
2778                                             arp_work.work);
2779         struct slave *slave, *oldcurrent;
2780         int do_failover = 0;
2781         int delta_in_ticks;
2782         int i;
2783
2784         read_lock(&bond->lock);
2785
2786         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2787
2788         if (bond->kill_timers)
2789                 goto out;
2790
2791         if (bond->slave_cnt == 0)
2792                 goto re_arm;
2793
2794         read_lock(&bond->curr_slave_lock);
2795         oldcurrent = bond->curr_active_slave;
2796         read_unlock(&bond->curr_slave_lock);
2797
2798         /* see if any of the previous devices are up now (i.e. they have
2799          * xmt and rcv traffic). the curr_active_slave does not come into
2800          * the picture unless it is null. also, slave->jiffies is not needed
2801          * here because we send an arp on each slave and give a slave as
2802          * long as it needs to get the tx/rx within the delta.
2803          * TODO: what about up/down delay in arp mode? it wasn't here before
2804          *       so it can wait
2805          */
2806         bond_for_each_slave(bond, slave, i) {
2807                 unsigned long trans_start = dev_trans_start(slave->dev);
2808
2809                 if (slave->link != BOND_LINK_UP) {
2810                         if (time_in_range(jiffies,
2811                                 trans_start - delta_in_ticks,
2812                                 trans_start + delta_in_ticks) &&
2813                             time_in_range(jiffies,
2814                                 slave->dev->last_rx - delta_in_ticks,
2815                                 slave->dev->last_rx + delta_in_ticks)) {
2816
2817                                 slave->link  = BOND_LINK_UP;
2818                                 bond_set_active_slave(slave);
2819
2820                                 /* primary_slave has no meaning in round-robin
2821                                  * mode. the window of a slave being up and
2822                                  * curr_active_slave being null after enslaving
2823                                  * is closed.
2824                                  */
2825                                 if (!oldcurrent) {
2826                                         pr_info("%s: link status definitely up for interface %s, ",
2827                                                 bond->dev->name,
2828                                                 slave->dev->name);
2829                                         do_failover = 1;
2830                                 } else {
2831                                         pr_info("%s: interface %s is now up\n",
2832                                                 bond->dev->name,
2833                                                 slave->dev->name);
2834                                 }
2835                         }
2836                 } else {
2837                         /* slave->link == BOND_LINK_UP */
2838
2839                         /* not all switches will respond to an arp request
2840                          * when the source ip is 0, so don't take the link down
2841                          * if we don't know our ip yet
2842                          */
2843                         if (!time_in_range(jiffies,
2844                                 trans_start - delta_in_ticks,
2845                                 trans_start + 2 * delta_in_ticks) ||
2846                             !time_in_range(jiffies,
2847                                 slave->dev->last_rx - delta_in_ticks,
2848                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2849
2850                                 slave->link  = BOND_LINK_DOWN;
2851                                 bond_set_backup_slave(slave);
2852
2853                                 if (slave->link_failure_count < UINT_MAX)
2854                                         slave->link_failure_count++;
2855
2856                                 pr_info("%s: interface %s is now down.\n",
2857                                         bond->dev->name,
2858                                         slave->dev->name);
2859
2860                                 if (slave == oldcurrent)
2861                                         do_failover = 1;
2862                         }
2863                 }
2864
2865                 /* note: if switch is in round-robin mode, all links
2866                  * must tx arp to ensure all links rx an arp - otherwise
2867                  * links may oscillate or not come up at all; if switch is
2868                  * in something like xor mode, there is nothing we can
2869                  * do - all replies will be rx'ed on same link causing slaves
2870                  * to be unstable during low/no traffic periods
2871                  */
2872                 if (IS_UP(slave->dev))
2873                         bond_arp_send_all(bond, slave);
2874         }
2875
2876         if (do_failover) {
2877                 block_netpoll_tx();
2878                 write_lock_bh(&bond->curr_slave_lock);
2879
2880                 bond_select_active_slave(bond);
2881
2882                 write_unlock_bh(&bond->curr_slave_lock);
2883                 unblock_netpoll_tx();
2884         }
2885
2886 re_arm:
2887         if (bond->params.arp_interval)
2888                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2889 out:
2890         read_unlock(&bond->lock);
2891 }
2892
2893 /*
2894  * Called to inspect slaves for active-backup mode ARP monitor link state
2895  * changes.  Sets new_link in slaves to specify what action should take
2896  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2897  * to link states must be committed.
2898  *
2899  * Called with bond->lock held for read.
2900  */
2901 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2902 {
2903         struct slave *slave;
2904         int i, commit = 0;
2905         unsigned long trans_start;
2906
2907         bond_for_each_slave(bond, slave, i) {
2908                 slave->new_link = BOND_LINK_NOCHANGE;
2909
2910                 if (slave->link != BOND_LINK_UP) {
2911                         if (time_in_range(jiffies,
2912                                 slave_last_rx(bond, slave) - delta_in_ticks,
2913                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2914
2915                                 slave->new_link = BOND_LINK_UP;
2916                                 commit++;
2917                         }
2918
2919                         continue;
2920                 }
2921
2922                 /*
2923                  * Give slaves 2*delta after being enslaved or made
2924                  * active.  This avoids bouncing, as the last receive
2925                  * times need a full ARP monitor cycle to be updated.
2926                  */
2927                 if (time_in_range(jiffies,
2928                                   slave->jiffies - delta_in_ticks,
2929                                   slave->jiffies + 2 * delta_in_ticks))
2930                         continue;
2931
2932                 /*
2933                  * Backup slave is down if:
2934                  * - No current_arp_slave AND
2935                  * - more than 3*delta since last receive AND
2936                  * - the bond has an IP address
2937                  *
2938                  * Note: a non-null current_arp_slave indicates
2939                  * the curr_active_slave went down and we are
2940                  * searching for a new one; under this condition
2941                  * we only take the curr_active_slave down - this
2942                  * gives each slave a chance to tx/rx traffic
2943                  * before being taken out
2944                  */
2945                 if (!bond_is_active_slave(slave) &&
2946                     !bond->current_arp_slave &&
2947                     !time_in_range(jiffies,
2948                         slave_last_rx(bond, slave) - delta_in_ticks,
2949                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2950
2951                         slave->new_link = BOND_LINK_DOWN;
2952                         commit++;
2953                 }
2954
2955                 /*
2956                  * Active slave is down if:
2957                  * - more than 2*delta since transmitting OR
2958                  * - (more than 2*delta since receive AND
2959                  *    the bond has an IP address)
2960                  */
2961                 trans_start = dev_trans_start(slave->dev);
2962                 if (bond_is_active_slave(slave) &&
2963                     (!time_in_range(jiffies,
2964                         trans_start - delta_in_ticks,
2965                         trans_start + 2 * delta_in_ticks) ||
2966                      !time_in_range(jiffies,
2967                         slave_last_rx(bond, slave) - delta_in_ticks,
2968                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2969
2970                         slave->new_link = BOND_LINK_DOWN;
2971                         commit++;
2972                 }
2973         }
2974
2975         return commit;
2976 }
2977
2978 /*
2979  * Called to commit link state changes noted by inspection step of
2980  * active-backup mode ARP monitor.
2981  *
2982  * Called with RTNL and bond->lock for read.
2983  */
2984 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2985 {
2986         struct slave *slave;
2987         int i;
2988         unsigned long trans_start;
2989
2990         bond_for_each_slave(bond, slave, i) {
2991                 switch (slave->new_link) {
2992                 case BOND_LINK_NOCHANGE:
2993                         continue;
2994
2995                 case BOND_LINK_UP:
2996                         trans_start = dev_trans_start(slave->dev);
2997                         if ((!bond->curr_active_slave &&
2998                              time_in_range(jiffies,
2999                                            trans_start - delta_in_ticks,
3000                                            trans_start + delta_in_ticks)) ||
3001                             bond->curr_active_slave != slave) {
3002                                 slave->link = BOND_LINK_UP;
3003                                 bond->current_arp_slave = NULL;
3004
3005                                 pr_info("%s: link status definitely up for interface %s.\n",
3006                                         bond->dev->name, slave->dev->name);
3007
3008                                 if (!bond->curr_active_slave ||
3009                                     (slave == bond->primary_slave))
3010                                         goto do_failover;
3011
3012                         }
3013
3014                         continue;
3015
3016                 case BOND_LINK_DOWN:
3017                         if (slave->link_failure_count < UINT_MAX)
3018                                 slave->link_failure_count++;
3019
3020                         slave->link = BOND_LINK_DOWN;
3021                         bond_set_slave_inactive_flags(slave);
3022
3023                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3024                                 bond->dev->name, slave->dev->name);
3025
3026                         if (slave == bond->curr_active_slave) {
3027                                 bond->current_arp_slave = NULL;
3028                                 goto do_failover;
3029                         }
3030
3031                         continue;
3032
3033                 default:
3034                         pr_err("%s: impossible: new_link %d on slave %s\n",
3035                                bond->dev->name, slave->new_link,
3036                                slave->dev->name);
3037                         continue;
3038                 }
3039
3040 do_failover:
3041                 ASSERT_RTNL();
3042                 block_netpoll_tx();
3043                 write_lock_bh(&bond->curr_slave_lock);
3044                 bond_select_active_slave(bond);
3045                 write_unlock_bh(&bond->curr_slave_lock);
3046                 unblock_netpoll_tx();
3047         }
3048
3049         bond_set_carrier(bond);
3050 }
3051
3052 /*
3053  * Send ARP probes for active-backup mode ARP monitor.
3054  *
3055  * Called with bond->lock held for read.
3056  */
3057 static void bond_ab_arp_probe(struct bonding *bond)
3058 {
3059         struct slave *slave;
3060         int i;
3061
3062         read_lock(&bond->curr_slave_lock);
3063
3064         if (bond->current_arp_slave && bond->curr_active_slave)
3065                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3066                         bond->current_arp_slave->dev->name,
3067                         bond->curr_active_slave->dev->name);
3068
3069         if (bond->curr_active_slave) {
3070                 bond_arp_send_all(bond, bond->curr_active_slave);
3071                 read_unlock(&bond->curr_slave_lock);
3072                 return;
3073         }
3074
3075         read_unlock(&bond->curr_slave_lock);
3076
3077         /* if we don't have a curr_active_slave, search for the next available
3078          * backup slave from the current_arp_slave and make it the candidate
3079          * for becoming the curr_active_slave
3080          */
3081
3082         if (!bond->current_arp_slave) {
3083                 bond->current_arp_slave = bond->first_slave;
3084                 if (!bond->current_arp_slave)
3085                         return;
3086         }
3087
3088         bond_set_slave_inactive_flags(bond->current_arp_slave);
3089
3090         /* search for next candidate */
3091         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3092                 if (IS_UP(slave->dev)) {
3093                         slave->link = BOND_LINK_BACK;
3094                         bond_set_slave_active_flags(slave);
3095                         bond_arp_send_all(bond, slave);
3096                         slave->jiffies = jiffies;
3097                         bond->current_arp_slave = slave;
3098                         break;
3099                 }
3100
3101                 /* if the link state is up at this point, we
3102                  * mark it down - this can happen if we have
3103                  * simultaneous link failures and
3104                  * reselect_active_interface doesn't make this
3105                  * one the current slave so it is still marked
3106                  * up when it is actually down
3107                  */
3108                 if (slave->link == BOND_LINK_UP) {
3109                         slave->link = BOND_LINK_DOWN;
3110                         if (slave->link_failure_count < UINT_MAX)
3111                                 slave->link_failure_count++;
3112
3113                         bond_set_slave_inactive_flags(slave);
3114
3115                         pr_info("%s: backup interface %s is now down.\n",
3116                                 bond->dev->name, slave->dev->name);
3117                 }
3118         }
3119 }
3120
3121 void bond_activebackup_arp_mon(struct work_struct *work)
3122 {
3123         struct bonding *bond = container_of(work, struct bonding,
3124                                             arp_work.work);
3125         bool should_notify_peers = false;
3126         int delta_in_ticks;
3127
3128         read_lock(&bond->lock);
3129
3130         if (bond->kill_timers)
3131                 goto out;
3132
3133         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3134
3135         if (bond->slave_cnt == 0)
3136                 goto re_arm;
3137
3138         should_notify_peers = bond_should_notify_peers(bond);
3139
3140         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3141                 read_unlock(&bond->lock);
3142                 rtnl_lock();
3143                 read_lock(&bond->lock);
3144
3145                 bond_ab_arp_commit(bond, delta_in_ticks);
3146
3147                 read_unlock(&bond->lock);
3148                 rtnl_unlock();
3149                 read_lock(&bond->lock);
3150         }
3151
3152         bond_ab_arp_probe(bond);
3153
3154 re_arm:
3155         if (bond->params.arp_interval)
3156                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3157 out:
3158         read_unlock(&bond->lock);
3159
3160         if (should_notify_peers) {
3161                 rtnl_lock();
3162                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3163                 rtnl_unlock();
3164         }
3165 }
3166
3167 /*-------------------------- netdev event handling --------------------------*/
3168
3169 /*
3170  * Change device name
3171  */
3172 static int bond_event_changename(struct bonding *bond)
3173 {
3174         bond_remove_proc_entry(bond);
3175         bond_create_proc_entry(bond);
3176
3177         bond_debug_reregister(bond);
3178
3179         return NOTIFY_DONE;
3180 }
3181
3182 static int bond_master_netdev_event(unsigned long event,
3183                                     struct net_device *bond_dev)
3184 {
3185         struct bonding *event_bond = netdev_priv(bond_dev);
3186
3187         switch (event) {
3188         case NETDEV_CHANGENAME:
3189                 return bond_event_changename(event_bond);
3190         default:
3191                 break;
3192         }
3193
3194         return NOTIFY_DONE;
3195 }
3196
3197 static int bond_slave_netdev_event(unsigned long event,
3198                                    struct net_device *slave_dev)
3199 {
3200         struct net_device *bond_dev = slave_dev->master;
3201         struct bonding *bond = netdev_priv(bond_dev);
3202
3203         switch (event) {
3204         case NETDEV_UNREGISTER:
3205                 if (bond_dev) {
3206                         if (bond->setup_by_slave)
3207                                 bond_release_and_destroy(bond_dev, slave_dev);
3208                         else
3209                                 bond_release(bond_dev, slave_dev);
3210                 }
3211                 break;
3212         case NETDEV_CHANGE:
3213                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3214                         struct slave *slave;
3215
3216                         slave = bond_get_slave_by_dev(bond, slave_dev);
3217                         if (slave) {
3218                                 u32 old_speed = slave->speed;
3219                                 u8  old_duplex = slave->duplex;
3220
3221                                 bond_update_speed_duplex(slave);
3222
3223                                 if (bond_is_lb(bond))
3224                                         break;
3225
3226                                 if (old_speed != slave->speed)
3227                                         bond_3ad_adapter_speed_changed(slave);
3228                                 if (old_duplex != slave->duplex)
3229                                         bond_3ad_adapter_duplex_changed(slave);
3230                         }
3231                 }
3232
3233                 break;
3234         case NETDEV_DOWN:
3235                 /*
3236                  * ... Or is it this?
3237                  */
3238                 break;
3239         case NETDEV_CHANGEMTU:
3240                 /*
3241                  * TODO: Should slaves be allowed to
3242                  * independently alter their MTU?  For
3243                  * an active-backup bond, slaves need
3244                  * not be the same type of device, so
3245                  * MTUs may vary.  For other modes,
3246                  * slaves arguably should have the
3247                  * same MTUs. To do this, we'd need to
3248                  * take over the slave's change_mtu
3249                  * function for the duration of their
3250                  * servitude.
3251                  */
3252                 break;
3253         case NETDEV_CHANGENAME:
3254                 /*
3255                  * TODO: handle changing the primary's name
3256                  */
3257                 break;
3258         case NETDEV_FEAT_CHANGE:
3259                 bond_compute_features(bond);
3260                 break;
3261         default:
3262                 break;
3263         }
3264
3265         return NOTIFY_DONE;
3266 }
3267
3268 /*
3269  * bond_netdev_event: handle netdev notifier chain events.
3270  *
3271  * This function receives events for the netdev chain.  The caller (an
3272  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3273  * locks for us to safely manipulate the slave devices (RTNL lock,
3274  * dev_probe_lock).
3275  */
3276 static int bond_netdev_event(struct notifier_block *this,
3277                              unsigned long event, void *ptr)
3278 {
3279         struct net_device *event_dev = (struct net_device *)ptr;
3280
3281         pr_debug("event_dev: %s, event: %lx\n",
3282                  event_dev ? event_dev->name : "None",
3283                  event);
3284
3285         if (!(event_dev->priv_flags & IFF_BONDING))
3286                 return NOTIFY_DONE;
3287
3288         if (event_dev->flags & IFF_MASTER) {
3289                 pr_debug("IFF_MASTER\n");
3290                 return bond_master_netdev_event(event, event_dev);
3291         }
3292
3293         if (event_dev->flags & IFF_SLAVE) {
3294                 pr_debug("IFF_SLAVE\n");
3295                 return bond_slave_netdev_event(event, event_dev);
3296         }
3297
3298         return NOTIFY_DONE;
3299 }
3300
3301 /*
3302  * bond_inetaddr_event: handle inetaddr notifier chain events.
3303  *
3304  * We keep track of device IPs primarily to use as source addresses in
3305  * ARP monitor probes (rather than spewing out broadcasts all the time).
3306  *
3307  * We track one IP for the main device (if it has one), plus one per VLAN.
3308  */
3309 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3310 {
3311         struct in_ifaddr *ifa = ptr;
3312         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3313         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3314         struct bonding *bond;
3315         struct vlan_entry *vlan;
3316
3317         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3318                 if (bond->dev == event_dev) {
3319                         switch (event) {
3320                         case NETDEV_UP:
3321                                 bond->master_ip = ifa->ifa_local;
3322                                 return NOTIFY_OK;
3323                         case NETDEV_DOWN:
3324                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3325                                 return NOTIFY_OK;
3326                         default:
3327                                 return NOTIFY_DONE;
3328                         }
3329                 }
3330
3331                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3332                         vlan_dev = __vlan_find_dev_deep(bond->dev,
3333                                                         vlan->vlan_id);
3334                         if (vlan_dev == event_dev) {
3335                                 switch (event) {
3336                                 case NETDEV_UP:
3337                                         vlan->vlan_ip = ifa->ifa_local;
3338                                         return NOTIFY_OK;
3339                                 case NETDEV_DOWN:
3340                                         vlan->vlan_ip =
3341                                                 bond_glean_dev_ip(vlan_dev);
3342                                         return NOTIFY_OK;
3343                                 default:
3344                                         return NOTIFY_DONE;
3345                                 }
3346                         }
3347                 }
3348         }
3349         return NOTIFY_DONE;
3350 }
3351
3352 static struct notifier_block bond_netdev_notifier = {
3353         .notifier_call = bond_netdev_event,
3354 };
3355
3356 static struct notifier_block bond_inetaddr_notifier = {
3357         .notifier_call = bond_inetaddr_event,
3358 };
3359
3360 /*---------------------------- Hashing Policies -----------------------------*/
3361
3362 /*
3363  * Hash for the output device based upon layer 2 and layer 3 data. If
3364  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3365  */
3366 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3367 {
3368         struct ethhdr *data = (struct ethhdr *)skb->data;
3369         struct iphdr *iph = ip_hdr(skb);
3370
3371         if (skb->protocol == htons(ETH_P_IP)) {
3372                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3373                         (data->h_dest[5] ^ data->h_source[5])) % count;
3374         }
3375
3376         return (data->h_dest[5] ^ data->h_source[5]) % count;
3377 }
3378
3379 /*
3380  * Hash for the output device based upon layer 3 and layer 4 data. If
3381  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3382  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3383  */
3384 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3385 {
3386         struct ethhdr *data = (struct ethhdr *)skb->data;
3387         struct iphdr *iph = ip_hdr(skb);
3388         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3389         int layer4_xor = 0;
3390
3391         if (skb->protocol == htons(ETH_P_IP)) {
3392                 if (!ip_is_fragment(iph) &&
3393                     (iph->protocol == IPPROTO_TCP ||
3394                      iph->protocol == IPPROTO_UDP)) {
3395                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3396                 }
3397                 return (layer4_xor ^
3398                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3399
3400         }
3401
3402         return (data->h_dest[5] ^ data->h_source[5]) % count;
3403 }
3404
3405 /*
3406  * Hash for the output device based upon layer 2 data
3407  */
3408 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3409 {
3410         struct ethhdr *data = (struct ethhdr *)skb->data;
3411
3412         return (data->h_dest[5] ^ data->h_source[5]) % count;
3413 }
3414
3415 /*-------------------------- Device entry points ----------------------------*/
3416
3417 static int bond_open(struct net_device *bond_dev)
3418 {
3419         struct bonding *bond = netdev_priv(bond_dev);
3420
3421         bond->kill_timers = 0;
3422
3423         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3424
3425         if (bond_is_lb(bond)) {
3426                 /* bond_alb_initialize must be called before the timer
3427                  * is started.
3428                  */
3429                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3430                         /* something went wrong - fail the open operation */
3431                         return -ENOMEM;
3432                 }
3433
3434                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3435                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3436         }
3437
3438         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3439                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3440                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3441         }
3442
3443         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3444                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3445                         INIT_DELAYED_WORK(&bond->arp_work,
3446                                           bond_activebackup_arp_mon);
3447                 else
3448                         INIT_DELAYED_WORK(&bond->arp_work,
3449                                           bond_loadbalance_arp_mon);
3450
3451                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3452                 if (bond->params.arp_validate)
3453                         bond->recv_probe = bond_arp_rcv;
3454         }
3455
3456         if (bond->params.mode == BOND_MODE_8023AD) {
3457                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3458                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3459                 /* register to receive LACPDUs */
3460                 bond->recv_probe = bond_3ad_lacpdu_recv;
3461                 bond_3ad_initiate_agg_selection(bond, 1);
3462         }
3463
3464         return 0;
3465 }
3466
3467 static int bond_close(struct net_device *bond_dev)
3468 {
3469         struct bonding *bond = netdev_priv(bond_dev);
3470
3471         write_lock_bh(&bond->lock);
3472
3473         bond->send_peer_notif = 0;
3474
3475         /* signal timers not to re-arm */
3476         bond->kill_timers = 1;
3477
3478         write_unlock_bh(&bond->lock);
3479
3480         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3481                 cancel_delayed_work(&bond->mii_work);
3482         }
3483
3484         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3485                 cancel_delayed_work(&bond->arp_work);
3486         }
3487
3488         switch (bond->params.mode) {
3489         case BOND_MODE_8023AD:
3490                 cancel_delayed_work(&bond->ad_work);
3491                 break;
3492         case BOND_MODE_TLB:
3493         case BOND_MODE_ALB:
3494                 cancel_delayed_work(&bond->alb_work);
3495                 break;
3496         default:
3497                 break;
3498         }
3499
3500         if (delayed_work_pending(&bond->mcast_work))
3501                 cancel_delayed_work(&bond->mcast_work);
3502
3503         if (bond_is_lb(bond)) {
3504                 /* Must be called only after all
3505                  * slaves have been released
3506                  */
3507                 bond_alb_deinitialize(bond);
3508         }
3509         bond->recv_probe = NULL;
3510
3511         return 0;
3512 }
3513
3514 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3515                                                 struct rtnl_link_stats64 *stats)
3516 {
3517         struct bonding *bond = netdev_priv(bond_dev);
3518         struct rtnl_link_stats64 temp;
3519         struct slave *slave;
3520         int i;
3521
3522         memset(stats, 0, sizeof(*stats));
3523
3524         read_lock_bh(&bond->lock);
3525
3526         bond_for_each_slave(bond, slave, i) {
3527                 const struct rtnl_link_stats64 *sstats =
3528                         dev_get_stats(slave->dev, &temp);
3529
3530                 stats->rx_packets += sstats->rx_packets;
3531                 stats->rx_bytes += sstats->rx_bytes;
3532                 stats->rx_errors += sstats->rx_errors;
3533                 stats->rx_dropped += sstats->rx_dropped;
3534
3535                 stats->tx_packets += sstats->tx_packets;
3536                 stats->tx_bytes += sstats->tx_bytes;
3537                 stats->tx_errors += sstats->tx_errors;
3538                 stats->tx_dropped += sstats->tx_dropped;
3539
3540                 stats->multicast += sstats->multicast;
3541                 stats->collisions += sstats->collisions;
3542
3543                 stats->rx_length_errors += sstats->rx_length_errors;
3544                 stats->rx_over_errors += sstats->rx_over_errors;
3545                 stats->rx_crc_errors += sstats->rx_crc_errors;
3546                 stats->rx_frame_errors += sstats->rx_frame_errors;
3547                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3548                 stats->rx_missed_errors += sstats->rx_missed_errors;
3549
3550                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3551                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3552                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3553                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3554                 stats->tx_window_errors += sstats->tx_window_errors;
3555         }
3556
3557         read_unlock_bh(&bond->lock);
3558
3559         return stats;
3560 }
3561
3562 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3563 {
3564         struct net_device *slave_dev = NULL;
3565         struct ifbond k_binfo;
3566         struct ifbond __user *u_binfo = NULL;
3567         struct ifslave k_sinfo;
3568         struct ifslave __user *u_sinfo = NULL;
3569         struct mii_ioctl_data *mii = NULL;
3570         int res = 0;
3571
3572         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3573
3574         switch (cmd) {
3575         case SIOCGMIIPHY:
3576                 mii = if_mii(ifr);
3577                 if (!mii)
3578                         return -EINVAL;
3579
3580                 mii->phy_id = 0;
3581                 /* Fall Through */
3582         case SIOCGMIIREG:
3583                 /*
3584                  * We do this again just in case we were called by SIOCGMIIREG
3585                  * instead of SIOCGMIIPHY.
3586                  */
3587                 mii = if_mii(ifr);
3588                 if (!mii)
3589                         return -EINVAL;
3590
3591
3592                 if (mii->reg_num == 1) {
3593                         struct bonding *bond = netdev_priv(bond_dev);
3594                         mii->val_out = 0;
3595                         read_lock(&bond->lock);
3596                         read_lock(&bond->curr_slave_lock);
3597                         if (netif_carrier_ok(bond->dev))
3598                                 mii->val_out = BMSR_LSTATUS;
3599
3600                         read_unlock(&bond->curr_slave_lock);
3601                         read_unlock(&bond->lock);
3602                 }
3603
3604                 return 0;
3605         case BOND_INFO_QUERY_OLD:
3606         case SIOCBONDINFOQUERY:
3607                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3608
3609                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3610                         return -EFAULT;
3611
3612                 res = bond_info_query(bond_dev, &k_binfo);
3613                 if (res == 0 &&
3614                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3615                         return -EFAULT;
3616
3617                 return res;
3618         case BOND_SLAVE_INFO_QUERY_OLD:
3619         case SIOCBONDSLAVEINFOQUERY:
3620                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3621
3622                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3623                         return -EFAULT;
3624
3625                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3626                 if (res == 0 &&
3627                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3628                         return -EFAULT;
3629
3630                 return res;
3631         default:
3632                 /* Go on */
3633                 break;
3634         }
3635
3636         if (!capable(CAP_NET_ADMIN))
3637                 return -EPERM;
3638
3639         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3640
3641         pr_debug("slave_dev=%p:\n", slave_dev);
3642
3643         if (!slave_dev)
3644                 res = -ENODEV;
3645         else {
3646                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3647                 switch (cmd) {
3648                 case BOND_ENSLAVE_OLD:
3649                 case SIOCBONDENSLAVE:
3650                         res = bond_enslave(bond_dev, slave_dev);
3651                         break;
3652                 case BOND_RELEASE_OLD:
3653                 case SIOCBONDRELEASE:
3654                         res = bond_release(bond_dev, slave_dev);
3655                         break;
3656                 case BOND_SETHWADDR_OLD:
3657                 case SIOCBONDSETHWADDR:
3658                         res = bond_sethwaddr(bond_dev, slave_dev);
3659                         break;
3660                 case BOND_CHANGE_ACTIVE_OLD:
3661                 case SIOCBONDCHANGEACTIVE:
3662                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3663                         break;
3664                 default:
3665                         res = -EOPNOTSUPP;
3666                 }
3667
3668                 dev_put(slave_dev);
3669         }
3670
3671         return res;
3672 }
3673
3674 static bool bond_addr_in_mc_list(unsigned char *addr,
3675                                  struct netdev_hw_addr_list *list,
3676                                  int addrlen)
3677 {
3678         struct netdev_hw_addr *ha;
3679
3680         netdev_hw_addr_list_for_each(ha, list)
3681                 if (!memcmp(ha->addr, addr, addrlen))
3682                         return true;
3683
3684         return false;
3685 }
3686
3687 static void bond_set_multicast_list(struct net_device *bond_dev)
3688 {
3689         struct bonding *bond = netdev_priv(bond_dev);
3690         struct netdev_hw_addr *ha;
3691         bool found;
3692
3693         /*
3694          * Do promisc before checking multicast_mode
3695          */
3696         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3697                 /*
3698                  * FIXME: Need to handle the error when one of the multi-slaves
3699                  * encounters error.
3700                  */
3701                 bond_set_promiscuity(bond, 1);
3702
3703
3704         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3705                 bond_set_promiscuity(bond, -1);
3706
3707
3708         /* set allmulti flag to slaves */
3709         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3710                 /*
3711                  * FIXME: Need to handle the error when one of the multi-slaves
3712                  * encounters error.
3713                  */
3714                 bond_set_allmulti(bond, 1);
3715
3716
3717         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3718                 bond_set_allmulti(bond, -1);
3719
3720
3721         read_lock(&bond->lock);
3722
3723         bond->flags = bond_dev->flags;
3724
3725         /* looking for addresses to add to slaves' mc list */
3726         netdev_for_each_mc_addr(ha, bond_dev) {
3727                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3728                                              bond_dev->addr_len);
3729                 if (!found)
3730                         bond_mc_add(bond, ha->addr);
3731         }
3732
3733         /* looking for addresses to delete from slaves' list */
3734         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3735                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3736                                              bond_dev->addr_len);
3737                 if (!found)
3738                         bond_mc_del(bond, ha->addr);
3739         }
3740
3741         /* save master's multicast list */
3742         __hw_addr_flush(&bond->mc_list);
3743         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3744                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3745
3746         read_unlock(&bond->lock);
3747 }
3748
3749 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3750 {
3751         struct bonding *bond = netdev_priv(dev);
3752         struct slave *slave = bond->first_slave;
3753
3754         if (slave) {
3755                 const struct net_device_ops *slave_ops
3756                         = slave->dev->netdev_ops;
3757                 if (slave_ops->ndo_neigh_setup)
3758                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3759         }
3760         return 0;
3761 }
3762
3763 /*
3764  * Change the MTU of all of a master's slaves to match the master
3765  */
3766 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3767 {
3768         struct bonding *bond = netdev_priv(bond_dev);
3769         struct slave *slave, *stop_at;
3770         int res = 0;
3771         int i;
3772
3773         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3774                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3775
3776         /* Can't hold bond->lock with bh disabled here since
3777          * some base drivers panic. On the other hand we can't
3778          * hold bond->lock without bh disabled because we'll
3779          * deadlock. The only solution is to rely on the fact
3780          * that we're under rtnl_lock here, and the slaves
3781          * list won't change. This doesn't solve the problem
3782          * of setting the slave's MTU while it is
3783          * transmitting, but the assumption is that the base
3784          * driver can handle that.
3785          *
3786          * TODO: figure out a way to safely iterate the slaves
3787          * list, but without holding a lock around the actual
3788          * call to the base driver.
3789          */
3790
3791         bond_for_each_slave(bond, slave, i) {
3792                 pr_debug("s %p s->p %p c_m %p\n",
3793                          slave,
3794                          slave->prev,
3795                          slave->dev->netdev_ops->ndo_change_mtu);
3796
3797                 res = dev_set_mtu(slave->dev, new_mtu);
3798
3799                 if (res) {
3800                         /* If we failed to set the slave's mtu to the new value
3801                          * we must abort the operation even in ACTIVE_BACKUP
3802                          * mode, because if we allow the backup slaves to have
3803                          * different mtu values than the active slave we'll
3804                          * need to change their mtu when doing a failover. That
3805                          * means changing their mtu from timer context, which
3806                          * is probably not a good idea.
3807                          */
3808                         pr_debug("err %d %s\n", res, slave->dev->name);
3809                         goto unwind;
3810                 }
3811         }
3812
3813         bond_dev->mtu = new_mtu;
3814
3815         return 0;
3816
3817 unwind:
3818         /* unwind from head to the slave that failed */
3819         stop_at = slave;
3820         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3821                 int tmp_res;
3822
3823                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3824                 if (tmp_res) {
3825                         pr_debug("unwind err %d dev %s\n",
3826                                  tmp_res, slave->dev->name);
3827                 }
3828         }
3829
3830         return res;
3831 }
3832
3833 /*
3834  * Change HW address
3835  *
3836  * Note that many devices must be down to change the HW address, and
3837  * downing the master releases all slaves.  We can make bonds full of
3838  * bonding devices to test this, however.
3839  */
3840 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3841 {
3842         struct bonding *bond = netdev_priv(bond_dev);
3843         struct sockaddr *sa = addr, tmp_sa;
3844         struct slave *slave, *stop_at;
3845         int res = 0;
3846         int i;
3847
3848         if (bond->params.mode == BOND_MODE_ALB)
3849                 return bond_alb_set_mac_address(bond_dev, addr);
3850
3851
3852         pr_debug("bond=%p, name=%s\n",
3853                  bond, bond_dev ? bond_dev->name : "None");
3854
3855         /*
3856          * If fail_over_mac is set to active, do nothing and return
3857          * success.  Returning an error causes ifenslave to fail.
3858          */
3859         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3860                 return 0;
3861
3862         if (!is_valid_ether_addr(sa->sa_data))
3863                 return -EADDRNOTAVAIL;
3864
3865         /* Can't hold bond->lock with bh disabled here since
3866          * some base drivers panic. On the other hand we can't
3867          * hold bond->lock without bh disabled because we'll
3868          * deadlock. The only solution is to rely on the fact
3869          * that we're under rtnl_lock here, and the slaves
3870          * list won't change. This doesn't solve the problem
3871          * of setting the slave's hw address while it is
3872          * transmitting, but the assumption is that the base
3873          * driver can handle that.
3874          *
3875          * TODO: figure out a way to safely iterate the slaves
3876          * list, but without holding a lock around the actual
3877          * call to the base driver.
3878          */
3879
3880         bond_for_each_slave(bond, slave, i) {
3881                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3882                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3883
3884                 if (slave_ops->ndo_set_mac_address == NULL) {
3885                         res = -EOPNOTSUPP;
3886                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3887                         goto unwind;
3888                 }
3889
3890                 res = dev_set_mac_address(slave->dev, addr);
3891                 if (res) {
3892                         /* TODO: consider downing the slave
3893                          * and retry ?
3894                          * User should expect communications
3895                          * breakage anyway until ARP finish
3896                          * updating, so...
3897                          */
3898                         pr_debug("err %d %s\n", res, slave->dev->name);
3899                         goto unwind;
3900                 }
3901         }
3902
3903         /* success */
3904         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3905         return 0;
3906
3907 unwind:
3908         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3909         tmp_sa.sa_family = bond_dev->type;
3910
3911         /* unwind from head to the slave that failed */
3912         stop_at = slave;
3913         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3914                 int tmp_res;
3915
3916                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3917                 if (tmp_res) {
3918                         pr_debug("unwind err %d dev %s\n",
3919                                  tmp_res, slave->dev->name);
3920                 }
3921         }
3922
3923         return res;
3924 }
3925
3926 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3927 {
3928         struct bonding *bond = netdev_priv(bond_dev);
3929         struct slave *slave, *start_at;
3930         int i, slave_no, res = 1;
3931         struct iphdr *iph = ip_hdr(skb);
3932
3933         /*
3934          * Start with the curr_active_slave that joined the bond as the
3935          * default for sending IGMP traffic.  For failover purposes one
3936          * needs to maintain some consistency for the interface that will
3937          * send the join/membership reports.  The curr_active_slave found
3938          * will send all of this type of traffic.
3939          */
3940         if ((iph->protocol == IPPROTO_IGMP) &&
3941             (skb->protocol == htons(ETH_P_IP))) {
3942
3943                 read_lock(&bond->curr_slave_lock);
3944                 slave = bond->curr_active_slave;
3945                 read_unlock(&bond->curr_slave_lock);
3946
3947                 if (!slave)
3948                         goto out;
3949         } else {
3950                 /*
3951                  * Concurrent TX may collide on rr_tx_counter; we accept
3952                  * that as being rare enough not to justify using an
3953                  * atomic op here.
3954                  */
3955                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3956
3957                 bond_for_each_slave(bond, slave, i) {
3958                         slave_no--;
3959                         if (slave_no < 0)
3960                                 break;
3961                 }
3962         }
3963
3964         start_at = slave;
3965         bond_for_each_slave_from(bond, slave, i, start_at) {
3966                 if (IS_UP(slave->dev) &&
3967                     (slave->link == BOND_LINK_UP) &&
3968                     bond_is_active_slave(slave)) {
3969                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3970                         break;
3971                 }
3972         }
3973
3974 out:
3975         if (res) {
3976                 /* no suitable interface, frame not sent */
3977                 dev_kfree_skb(skb);
3978         }
3979
3980         return NETDEV_TX_OK;
3981 }
3982
3983
3984 /*
3985  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3986  * the bond has a usable interface.
3987  */
3988 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3989 {
3990         struct bonding *bond = netdev_priv(bond_dev);
3991         int res = 1;
3992
3993         read_lock(&bond->curr_slave_lock);
3994
3995         if (bond->curr_active_slave)
3996                 res = bond_dev_queue_xmit(bond, skb,
3997                         bond->curr_active_slave->dev);
3998
3999         if (res)
4000                 /* no suitable interface, frame not sent */
4001                 dev_kfree_skb(skb);
4002
4003         read_unlock(&bond->curr_slave_lock);
4004
4005         return NETDEV_TX_OK;
4006 }
4007
4008 /*
4009  * In bond_xmit_xor() , we determine the output device by using a pre-
4010  * determined xmit_hash_policy(), If the selected device is not enabled,
4011  * find the next active slave.
4012  */
4013 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4014 {
4015         struct bonding *bond = netdev_priv(bond_dev);
4016         struct slave *slave, *start_at;
4017         int slave_no;
4018         int i;
4019         int res = 1;
4020
4021         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4022
4023         bond_for_each_slave(bond, slave, i) {
4024                 slave_no--;
4025                 if (slave_no < 0)
4026                         break;
4027         }
4028
4029         start_at = slave;
4030
4031         bond_for_each_slave_from(bond, slave, i, start_at) {
4032                 if (IS_UP(slave->dev) &&
4033                     (slave->link == BOND_LINK_UP) &&
4034                     bond_is_active_slave(slave)) {
4035                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4036                         break;
4037                 }
4038         }
4039
4040         if (res) {
4041                 /* no suitable interface, frame not sent */
4042                 dev_kfree_skb(skb);
4043         }
4044
4045         return NETDEV_TX_OK;
4046 }
4047
4048 /*
4049  * in broadcast mode, we send everything to all usable interfaces.
4050  */
4051 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4052 {
4053         struct bonding *bond = netdev_priv(bond_dev);
4054         struct slave *slave, *start_at;
4055         struct net_device *tx_dev = NULL;
4056         int i;
4057         int res = 1;
4058
4059         read_lock(&bond->curr_slave_lock);
4060         start_at = bond->curr_active_slave;
4061         read_unlock(&bond->curr_slave_lock);
4062
4063         if (!start_at)
4064                 goto out;
4065
4066         bond_for_each_slave_from(bond, slave, i, start_at) {
4067                 if (IS_UP(slave->dev) &&
4068                     (slave->link == BOND_LINK_UP) &&
4069                     bond_is_active_slave(slave)) {
4070                         if (tx_dev) {
4071                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4072                                 if (!skb2) {
4073                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4074                                                bond_dev->name);
4075                                         continue;
4076                                 }
4077
4078                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4079                                 if (res) {
4080                                         dev_kfree_skb(skb2);
4081                                         continue;
4082                                 }
4083                         }
4084                         tx_dev = slave->dev;
4085                 }
4086         }
4087
4088         if (tx_dev)
4089                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4090
4091 out:
4092         if (res)
4093                 /* no suitable interface, frame not sent */
4094                 dev_kfree_skb(skb);
4095
4096         /* frame sent to all suitable interfaces */
4097         return NETDEV_TX_OK;
4098 }
4099
4100 /*------------------------- Device initialization ---------------------------*/
4101
4102 static void bond_set_xmit_hash_policy(struct bonding *bond)
4103 {
4104         switch (bond->params.xmit_policy) {
4105         case BOND_XMIT_POLICY_LAYER23:
4106                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4107                 break;
4108         case BOND_XMIT_POLICY_LAYER34:
4109                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4110                 break;
4111         case BOND_XMIT_POLICY_LAYER2:
4112         default:
4113                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4114                 break;
4115         }
4116 }
4117
4118 /*
4119  * Lookup the slave that corresponds to a qid
4120  */
4121 static inline int bond_slave_override(struct bonding *bond,
4122                                       struct sk_buff *skb)
4123 {
4124         int i, res = 1;
4125         struct slave *slave = NULL;
4126         struct slave *check_slave;
4127
4128         if (!skb->queue_mapping)
4129                 return 1;
4130
4131         /* Find out if any slaves have the same mapping as this skb. */
4132         bond_for_each_slave(bond, check_slave, i) {
4133                 if (check_slave->queue_id == skb->queue_mapping) {
4134                         slave = check_slave;
4135                         break;
4136                 }
4137         }
4138
4139         /* If the slave isn't UP, use default transmit policy. */
4140         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4141             (slave->link == BOND_LINK_UP)) {
4142                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4143         }
4144
4145         return res;
4146 }
4147
4148
4149 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4150 {
4151         /*
4152          * This helper function exists to help dev_pick_tx get the correct
4153          * destination queue.  Using a helper function skips a call to
4154          * skb_tx_hash and will put the skbs in the queue we expect on their
4155          * way down to the bonding driver.
4156          */
4157         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4158
4159         /*
4160          * Save the original txq to restore before passing to the driver
4161          */
4162         bond_queue_mapping(skb) = skb->queue_mapping;
4163
4164         if (unlikely(txq >= dev->real_num_tx_queues)) {
4165                 do {
4166                         txq -= dev->real_num_tx_queues;
4167                 } while (txq >= dev->real_num_tx_queues);
4168         }
4169         return txq;
4170 }
4171
4172 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4173 {
4174         struct bonding *bond = netdev_priv(dev);
4175
4176         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4177                 if (!bond_slave_override(bond, skb))
4178                         return NETDEV_TX_OK;
4179         }
4180
4181         switch (bond->params.mode) {
4182         case BOND_MODE_ROUNDROBIN:
4183                 return bond_xmit_roundrobin(skb, dev);
4184         case BOND_MODE_ACTIVEBACKUP:
4185                 return bond_xmit_activebackup(skb, dev);
4186         case BOND_MODE_XOR:
4187                 return bond_xmit_xor(skb, dev);
4188         case BOND_MODE_BROADCAST:
4189                 return bond_xmit_broadcast(skb, dev);
4190         case BOND_MODE_8023AD:
4191                 return bond_3ad_xmit_xor(skb, dev);
4192         case BOND_MODE_ALB:
4193         case BOND_MODE_TLB:
4194                 return bond_alb_xmit(skb, dev);
4195         default:
4196                 /* Should never happen, mode already checked */
4197                 pr_err("%s: Error: Unknown bonding mode %d\n",
4198                        dev->name, bond->params.mode);
4199                 WARN_ON_ONCE(1);
4200                 dev_kfree_skb(skb);
4201                 return NETDEV_TX_OK;
4202         }
4203 }
4204
4205 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4206 {
4207         struct bonding *bond = netdev_priv(dev);
4208         netdev_tx_t ret = NETDEV_TX_OK;
4209
4210         /*
4211          * If we risk deadlock from transmitting this in the
4212          * netpoll path, tell netpoll to queue the frame for later tx
4213          */
4214         if (is_netpoll_tx_blocked(dev))
4215                 return NETDEV_TX_BUSY;
4216
4217         read_lock(&bond->lock);
4218
4219         if (bond->slave_cnt)
4220                 ret = __bond_start_xmit(skb, dev);
4221         else
4222                 dev_kfree_skb(skb);
4223
4224         read_unlock(&bond->lock);
4225
4226         return ret;
4227 }
4228
4229 /*
4230  * set bond mode specific net device operations
4231  */
4232 void bond_set_mode_ops(struct bonding *bond, int mode)
4233 {
4234         struct net_device *bond_dev = bond->dev;
4235
4236         switch (mode) {
4237         case BOND_MODE_ROUNDROBIN:
4238                 break;
4239         case BOND_MODE_ACTIVEBACKUP:
4240                 break;
4241         case BOND_MODE_XOR:
4242                 bond_set_xmit_hash_policy(bond);
4243                 break;
4244         case BOND_MODE_BROADCAST:
4245                 break;
4246         case BOND_MODE_8023AD:
4247                 bond_set_xmit_hash_policy(bond);
4248                 break;
4249         case BOND_MODE_ALB:
4250                 /* FALLTHRU */
4251         case BOND_MODE_TLB:
4252                 break;
4253         default:
4254                 /* Should never happen, mode already checked */
4255                 pr_err("%s: Error: Unknown bonding mode %d\n",
4256                        bond_dev->name, mode);
4257                 break;
4258         }
4259 }
4260
4261 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4262                                     struct ethtool_drvinfo *drvinfo)
4263 {
4264         strncpy(drvinfo->driver, DRV_NAME, 32);
4265         strncpy(drvinfo->version, DRV_VERSION, 32);
4266         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4267 }
4268
4269 static const struct ethtool_ops bond_ethtool_ops = {
4270         .get_drvinfo            = bond_ethtool_get_drvinfo,
4271         .get_link               = ethtool_op_get_link,
4272 };
4273
4274 static const struct net_device_ops bond_netdev_ops = {
4275         .ndo_init               = bond_init,
4276         .ndo_uninit             = bond_uninit,
4277         .ndo_open               = bond_open,
4278         .ndo_stop               = bond_close,
4279         .ndo_start_xmit         = bond_start_xmit,
4280         .ndo_select_queue       = bond_select_queue,
4281         .ndo_get_stats64        = bond_get_stats,
4282         .ndo_do_ioctl           = bond_do_ioctl,
4283         .ndo_set_multicast_list = bond_set_multicast_list,
4284         .ndo_change_mtu         = bond_change_mtu,
4285         .ndo_set_mac_address    = bond_set_mac_address,
4286         .ndo_neigh_setup        = bond_neigh_setup,
4287         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4288         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4289 #ifdef CONFIG_NET_POLL_CONTROLLER
4290         .ndo_netpoll_setup      = bond_netpoll_setup,
4291         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4292         .ndo_poll_controller    = bond_poll_controller,
4293 #endif
4294         .ndo_add_slave          = bond_enslave,
4295         .ndo_del_slave          = bond_release,
4296         .ndo_fix_features       = bond_fix_features,
4297 };
4298
4299 static void bond_destructor(struct net_device *bond_dev)
4300 {
4301         struct bonding *bond = netdev_priv(bond_dev);
4302         if (bond->wq)
4303                 destroy_workqueue(bond->wq);
4304         free_netdev(bond_dev);
4305 }
4306
4307 static void bond_setup(struct net_device *bond_dev)
4308 {
4309         struct bonding *bond = netdev_priv(bond_dev);
4310
4311         /* initialize rwlocks */
4312         rwlock_init(&bond->lock);
4313         rwlock_init(&bond->curr_slave_lock);
4314
4315         bond->params = bonding_defaults;
4316
4317         /* Initialize pointers */
4318         bond->dev = bond_dev;
4319         INIT_LIST_HEAD(&bond->vlan_list);
4320
4321         /* Initialize the device entry points */
4322         ether_setup(bond_dev);
4323         bond_dev->netdev_ops = &bond_netdev_ops;
4324         bond_dev->ethtool_ops = &bond_ethtool_ops;
4325         bond_set_mode_ops(bond, bond->params.mode);
4326
4327         bond_dev->destructor = bond_destructor;
4328
4329         /* Initialize the device options */
4330         bond_dev->tx_queue_len = 0;
4331         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4332         bond_dev->priv_flags |= IFF_BONDING;
4333         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4334
4335         /* At first, we block adding VLANs. That's the only way to
4336          * prevent problems that occur when adding VLANs over an
4337          * empty bond. The block will be removed once non-challenged
4338          * slaves are enslaved.
4339          */
4340         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4341
4342         /* don't acquire bond device's netif_tx_lock when
4343          * transmitting */
4344         bond_dev->features |= NETIF_F_LLTX;
4345
4346         /* By default, we declare the bond to be fully
4347          * VLAN hardware accelerated capable. Special
4348          * care is taken in the various xmit functions
4349          * when there are slaves that are not hw accel
4350          * capable
4351          */
4352
4353         bond_dev->hw_features = BOND_VLAN_FEATURES |
4354                                 NETIF_F_HW_VLAN_TX |
4355                                 NETIF_F_HW_VLAN_RX |
4356                                 NETIF_F_HW_VLAN_FILTER;
4357
4358         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_NO_CSUM);
4359         bond_dev->features |= bond_dev->hw_features;
4360 }
4361
4362 static void bond_work_cancel_all(struct bonding *bond)
4363 {
4364         write_lock_bh(&bond->lock);
4365         bond->kill_timers = 1;
4366         write_unlock_bh(&bond->lock);
4367
4368         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4369                 cancel_delayed_work(&bond->mii_work);
4370
4371         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4372                 cancel_delayed_work(&bond->arp_work);
4373
4374         if (bond->params.mode == BOND_MODE_ALB &&
4375             delayed_work_pending(&bond->alb_work))
4376                 cancel_delayed_work(&bond->alb_work);
4377
4378         if (bond->params.mode == BOND_MODE_8023AD &&
4379             delayed_work_pending(&bond->ad_work))
4380                 cancel_delayed_work(&bond->ad_work);
4381
4382         if (delayed_work_pending(&bond->mcast_work))
4383                 cancel_delayed_work(&bond->mcast_work);
4384 }
4385
4386 /*
4387 * Destroy a bonding device.
4388 * Must be under rtnl_lock when this function is called.
4389 */
4390 static void bond_uninit(struct net_device *bond_dev)
4391 {
4392         struct bonding *bond = netdev_priv(bond_dev);
4393         struct vlan_entry *vlan, *tmp;
4394
4395         bond_netpoll_cleanup(bond_dev);
4396
4397         /* Release the bonded slaves */
4398         bond_release_all(bond_dev);
4399
4400         list_del(&bond->bond_list);
4401
4402         bond_work_cancel_all(bond);
4403
4404         bond_remove_proc_entry(bond);
4405
4406         bond_debug_unregister(bond);
4407
4408         __hw_addr_flush(&bond->mc_list);
4409
4410         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4411                 list_del(&vlan->vlan_list);
4412                 kfree(vlan);
4413         }
4414 }
4415
4416 /*------------------------- Module initialization ---------------------------*/
4417
4418 /*
4419  * Convert string input module parms.  Accept either the
4420  * number of the mode or its string name.  A bit complicated because
4421  * some mode names are substrings of other names, and calls from sysfs
4422  * may have whitespace in the name (trailing newlines, for example).
4423  */
4424 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4425 {
4426         int modeint = -1, i, rv;
4427         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4428
4429         for (p = (char *)buf; *p; p++)
4430                 if (!(isdigit(*p) || isspace(*p)))
4431                         break;
4432
4433         if (*p)
4434                 rv = sscanf(buf, "%20s", modestr);
4435         else
4436                 rv = sscanf(buf, "%d", &modeint);
4437
4438         if (!rv)
4439                 return -1;
4440
4441         for (i = 0; tbl[i].modename; i++) {
4442                 if (modeint == tbl[i].mode)
4443                         return tbl[i].mode;
4444                 if (strcmp(modestr, tbl[i].modename) == 0)
4445                         return tbl[i].mode;
4446         }
4447
4448         return -1;
4449 }
4450
4451 static int bond_check_params(struct bond_params *params)
4452 {
4453         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4454
4455         /*
4456          * Convert string parameters.
4457          */
4458         if (mode) {
4459                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4460                 if (bond_mode == -1) {
4461                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4462                                mode == NULL ? "NULL" : mode);
4463                         return -EINVAL;
4464                 }
4465         }
4466
4467         if (xmit_hash_policy) {
4468                 if ((bond_mode != BOND_MODE_XOR) &&
4469                     (bond_mode != BOND_MODE_8023AD)) {
4470                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4471                                bond_mode_name(bond_mode));
4472                 } else {
4473                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4474                                                         xmit_hashtype_tbl);
4475                         if (xmit_hashtype == -1) {
4476                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4477                                        xmit_hash_policy == NULL ? "NULL" :
4478                                        xmit_hash_policy);
4479                                 return -EINVAL;
4480                         }
4481                 }
4482         }
4483
4484         if (lacp_rate) {
4485                 if (bond_mode != BOND_MODE_8023AD) {
4486                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4487                                 bond_mode_name(bond_mode));
4488                 } else {
4489                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4490                         if (lacp_fast == -1) {
4491                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4492                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4493                                 return -EINVAL;
4494                         }
4495                 }
4496         }
4497
4498         if (ad_select) {
4499                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4500                 if (params->ad_select == -1) {
4501                         pr_err("Error: Invalid ad_select \"%s\"\n",
4502                                ad_select == NULL ? "NULL" : ad_select);
4503                         return -EINVAL;
4504                 }
4505
4506                 if (bond_mode != BOND_MODE_8023AD) {
4507                         pr_warning("ad_select param only affects 802.3ad mode\n");
4508                 }
4509         } else {
4510                 params->ad_select = BOND_AD_STABLE;
4511         }
4512
4513         if (max_bonds < 0) {
4514                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4515                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4516                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4517         }
4518
4519         if (miimon < 0) {
4520                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4521                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4522                 miimon = BOND_LINK_MON_INTERV;
4523         }
4524
4525         if (updelay < 0) {
4526                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4527                            updelay, INT_MAX);
4528                 updelay = 0;
4529         }
4530
4531         if (downdelay < 0) {
4532                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4533                            downdelay, INT_MAX);
4534                 downdelay = 0;
4535         }
4536
4537         if ((use_carrier != 0) && (use_carrier != 1)) {
4538                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4539                            use_carrier);
4540                 use_carrier = 1;
4541         }
4542
4543         if (num_peer_notif < 0 || num_peer_notif > 255) {
4544                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4545                            num_peer_notif);
4546                 num_peer_notif = 1;
4547         }
4548
4549         /* reset values for 802.3ad */
4550         if (bond_mode == BOND_MODE_8023AD) {
4551                 if (!miimon) {
4552                         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");
4553                         pr_warning("Forcing miimon to 100msec\n");
4554                         miimon = 100;
4555                 }
4556         }
4557
4558         if (tx_queues < 1 || tx_queues > 255) {
4559                 pr_warning("Warning: tx_queues (%d) should be between "
4560                            "1 and 255, resetting to %d\n",
4561                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4562                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4563         }
4564
4565         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4566                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4567                            "not of valid value (0/1), so it was set to "
4568                            "0\n", all_slaves_active);
4569                 all_slaves_active = 0;
4570         }
4571
4572         if (resend_igmp < 0 || resend_igmp > 255) {
4573                 pr_warning("Warning: resend_igmp (%d) should be between "
4574                            "0 and 255, resetting to %d\n",
4575                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4576                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4577         }
4578
4579         /* reset values for TLB/ALB */
4580         if ((bond_mode == BOND_MODE_TLB) ||
4581             (bond_mode == BOND_MODE_ALB)) {
4582                 if (!miimon) {
4583                         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");
4584                         pr_warning("Forcing miimon to 100msec\n");
4585                         miimon = 100;
4586                 }
4587         }
4588
4589         if (bond_mode == BOND_MODE_ALB) {
4590                 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",
4591                           updelay);
4592         }
4593
4594         if (!miimon) {
4595                 if (updelay || downdelay) {
4596                         /* just warn the user the up/down delay will have
4597                          * no effect since miimon is zero...
4598                          */
4599                         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",
4600                                    updelay, downdelay);
4601                 }
4602         } else {
4603                 /* don't allow arp monitoring */
4604                 if (arp_interval) {
4605                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4606                                    miimon, arp_interval);
4607                         arp_interval = 0;
4608                 }
4609
4610                 if ((updelay % miimon) != 0) {
4611                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4612                                    updelay, miimon,
4613                                    (updelay / miimon) * miimon);
4614                 }
4615
4616                 updelay /= miimon;
4617
4618                 if ((downdelay % miimon) != 0) {
4619                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4620                                    downdelay, miimon,
4621                                    (downdelay / miimon) * miimon);
4622                 }
4623
4624                 downdelay /= miimon;
4625         }
4626
4627         if (arp_interval < 0) {
4628                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4629                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4630                 arp_interval = BOND_LINK_ARP_INTERV;
4631         }
4632
4633         for (arp_ip_count = 0;
4634              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4635              arp_ip_count++) {
4636                 /* not complete check, but should be good enough to
4637                    catch mistakes */
4638                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4639                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4640                                    arp_ip_target[arp_ip_count]);
4641                         arp_interval = 0;
4642                 } else {
4643                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4644                         arp_target[arp_ip_count] = ip;
4645                 }
4646         }
4647
4648         if (arp_interval && !arp_ip_count) {
4649                 /* don't allow arping if no arp_ip_target given... */
4650                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4651                            arp_interval);
4652                 arp_interval = 0;
4653         }
4654
4655         if (arp_validate) {
4656                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4657                         pr_err("arp_validate only supported in active-backup mode\n");
4658                         return -EINVAL;
4659                 }
4660                 if (!arp_interval) {
4661                         pr_err("arp_validate requires arp_interval\n");
4662                         return -EINVAL;
4663                 }
4664
4665                 arp_validate_value = bond_parse_parm(arp_validate,
4666                                                      arp_validate_tbl);
4667                 if (arp_validate_value == -1) {
4668                         pr_err("Error: invalid arp_validate \"%s\"\n",
4669                                arp_validate == NULL ? "NULL" : arp_validate);
4670                         return -EINVAL;
4671                 }
4672         } else
4673                 arp_validate_value = 0;
4674
4675         if (miimon) {
4676                 pr_info("MII link monitoring set to %d ms\n", miimon);
4677         } else if (arp_interval) {
4678                 int i;
4679
4680                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4681                         arp_interval,
4682                         arp_validate_tbl[arp_validate_value].modename,
4683                         arp_ip_count);
4684
4685                 for (i = 0; i < arp_ip_count; i++)
4686                         pr_info(" %s", arp_ip_target[i]);
4687
4688                 pr_info("\n");
4689
4690         } else if (max_bonds) {
4691                 /* miimon and arp_interval not set, we need one so things
4692                  * work as expected, see bonding.txt for details
4693                  */
4694                 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");
4695         }
4696
4697         if (primary && !USES_PRIMARY(bond_mode)) {
4698                 /* currently, using a primary only makes sense
4699                  * in active backup, TLB or ALB modes
4700                  */
4701                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4702                            primary, bond_mode_name(bond_mode));
4703                 primary = NULL;
4704         }
4705
4706         if (primary && primary_reselect) {
4707                 primary_reselect_value = bond_parse_parm(primary_reselect,
4708                                                          pri_reselect_tbl);
4709                 if (primary_reselect_value == -1) {
4710                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4711                                primary_reselect ==
4712                                         NULL ? "NULL" : primary_reselect);
4713                         return -EINVAL;
4714                 }
4715         } else {
4716                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4717         }
4718
4719         if (fail_over_mac) {
4720                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4721                                                       fail_over_mac_tbl);
4722                 if (fail_over_mac_value == -1) {
4723                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4724                                arp_validate == NULL ? "NULL" : arp_validate);
4725                         return -EINVAL;
4726                 }
4727
4728                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4729                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4730         } else {
4731                 fail_over_mac_value = BOND_FOM_NONE;
4732         }
4733
4734         /* fill params struct with the proper values */
4735         params->mode = bond_mode;
4736         params->xmit_policy = xmit_hashtype;
4737         params->miimon = miimon;
4738         params->num_peer_notif = num_peer_notif;
4739         params->arp_interval = arp_interval;
4740         params->arp_validate = arp_validate_value;
4741         params->updelay = updelay;
4742         params->downdelay = downdelay;
4743         params->use_carrier = use_carrier;
4744         params->lacp_fast = lacp_fast;
4745         params->primary[0] = 0;
4746         params->primary_reselect = primary_reselect_value;
4747         params->fail_over_mac = fail_over_mac_value;
4748         params->tx_queues = tx_queues;
4749         params->all_slaves_active = all_slaves_active;
4750         params->resend_igmp = resend_igmp;
4751         params->min_links = min_links;
4752
4753         if (primary) {
4754                 strncpy(params->primary, primary, IFNAMSIZ);
4755                 params->primary[IFNAMSIZ - 1] = 0;
4756         }
4757
4758         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4759
4760         return 0;
4761 }
4762
4763 static struct lock_class_key bonding_netdev_xmit_lock_key;
4764 static struct lock_class_key bonding_netdev_addr_lock_key;
4765
4766 static void bond_set_lockdep_class_one(struct net_device *dev,
4767                                        struct netdev_queue *txq,
4768                                        void *_unused)
4769 {
4770         lockdep_set_class(&txq->_xmit_lock,
4771                           &bonding_netdev_xmit_lock_key);
4772 }
4773
4774 static void bond_set_lockdep_class(struct net_device *dev)
4775 {
4776         lockdep_set_class(&dev->addr_list_lock,
4777                           &bonding_netdev_addr_lock_key);
4778         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4779 }
4780
4781 /*
4782  * Called from registration process
4783  */
4784 static int bond_init(struct net_device *bond_dev)
4785 {
4786         struct bonding *bond = netdev_priv(bond_dev);
4787         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4788         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4789
4790         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4791
4792         /*
4793          * Initialize locks that may be required during
4794          * en/deslave operations.  All of the bond_open work
4795          * (of which this is part) should really be moved to
4796          * a phase prior to dev_open
4797          */
4798         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4799         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4800
4801         bond->wq = create_singlethread_workqueue(bond_dev->name);
4802         if (!bond->wq)
4803                 return -ENOMEM;
4804
4805         bond_set_lockdep_class(bond_dev);
4806
4807         bond_create_proc_entry(bond);
4808         list_add_tail(&bond->bond_list, &bn->dev_list);
4809
4810         bond_prepare_sysfs_group(bond);
4811
4812         bond_debug_register(bond);
4813
4814         __hw_addr_init(&bond->mc_list);
4815         return 0;
4816 }
4817
4818 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4819 {
4820         if (tb[IFLA_ADDRESS]) {
4821                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4822                         return -EINVAL;
4823                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4824                         return -EADDRNOTAVAIL;
4825         }
4826         return 0;
4827 }
4828
4829 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4830         .kind           = "bond",
4831         .priv_size      = sizeof(struct bonding),
4832         .setup          = bond_setup,
4833         .validate       = bond_validate,
4834 };
4835
4836 /* Create a new bond based on the specified name and bonding parameters.
4837  * If name is NULL, obtain a suitable "bond%d" name for us.
4838  * Caller must NOT hold rtnl_lock; we need to release it here before we
4839  * set up our sysfs entries.
4840  */
4841 int bond_create(struct net *net, const char *name)
4842 {
4843         struct net_device *bond_dev;
4844         int res;
4845
4846         rtnl_lock();
4847
4848         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4849                                    name ? name : "bond%d",
4850                                    bond_setup, tx_queues);
4851         if (!bond_dev) {
4852                 pr_err("%s: eek! can't alloc netdev!\n", name);
4853                 rtnl_unlock();
4854                 return -ENOMEM;
4855         }
4856
4857         dev_net_set(bond_dev, net);
4858         bond_dev->rtnl_link_ops = &bond_link_ops;
4859
4860         res = register_netdevice(bond_dev);
4861
4862         netif_carrier_off(bond_dev);
4863
4864         rtnl_unlock();
4865         if (res < 0)
4866                 bond_destructor(bond_dev);
4867         return res;
4868 }
4869
4870 static int __net_init bond_net_init(struct net *net)
4871 {
4872         struct bond_net *bn = net_generic(net, bond_net_id);
4873
4874         bn->net = net;
4875         INIT_LIST_HEAD(&bn->dev_list);
4876
4877         bond_create_proc_dir(bn);
4878
4879         return 0;
4880 }
4881
4882 static void __net_exit bond_net_exit(struct net *net)
4883 {
4884         struct bond_net *bn = net_generic(net, bond_net_id);
4885
4886         bond_destroy_proc_dir(bn);
4887 }
4888
4889 static struct pernet_operations bond_net_ops = {
4890         .init = bond_net_init,
4891         .exit = bond_net_exit,
4892         .id   = &bond_net_id,
4893         .size = sizeof(struct bond_net),
4894 };
4895
4896 static int __init bonding_init(void)
4897 {
4898         int i;
4899         int res;
4900
4901         pr_info("%s", bond_version);
4902
4903         res = bond_check_params(&bonding_defaults);
4904         if (res)
4905                 goto out;
4906
4907         res = register_pernet_subsys(&bond_net_ops);
4908         if (res)
4909                 goto out;
4910
4911         res = rtnl_link_register(&bond_link_ops);
4912         if (res)
4913                 goto err_link;
4914
4915         bond_create_debugfs();
4916
4917         for (i = 0; i < max_bonds; i++) {
4918                 res = bond_create(&init_net, NULL);
4919                 if (res)
4920                         goto err;
4921         }
4922
4923         res = bond_create_sysfs();
4924         if (res)
4925                 goto err;
4926
4927         register_netdevice_notifier(&bond_netdev_notifier);
4928         register_inetaddr_notifier(&bond_inetaddr_notifier);
4929 out:
4930         return res;
4931 err:
4932         rtnl_link_unregister(&bond_link_ops);
4933 err_link:
4934         unregister_pernet_subsys(&bond_net_ops);
4935         goto out;
4936
4937 }
4938
4939 static void __exit bonding_exit(void)
4940 {
4941         unregister_netdevice_notifier(&bond_netdev_notifier);
4942         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4943
4944         bond_destroy_sysfs();
4945         bond_destroy_debugfs();
4946
4947         rtnl_link_unregister(&bond_link_ops);
4948         unregister_pernet_subsys(&bond_net_ops);
4949
4950 #ifdef CONFIG_NET_POLL_CONTROLLER
4951         /*
4952          * Make sure we don't have an imbalance on our netpoll blocking
4953          */
4954         WARN_ON(atomic_read(&netpoll_block_tx));
4955 #endif
4956 }
4957
4958 module_init(bonding_init);
4959 module_exit(bonding_exit);
4960 MODULE_LICENSE("GPL");
4961 MODULE_VERSION(DRV_VERSION);
4962 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4963 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4964 MODULE_ALIAS_RTNL_LINK("bond");