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