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/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.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 /**
 386  * bond_dev_queue_xmit - Prepare skb for xmit.
 387  *
 388  * @bond: bond device that got this skb for tx.
 389  * @skb: hw accel VLAN tagged skb to transmit
 390  * @slave_dev: slave that is supposed to xmit this skbuff
 391  */
 392 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 393                         struct net_device *slave_dev)
 394 {
 395         skb->dev = slave_dev;
 396
 397         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 398                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 399         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 400
 401         if (unlikely(netpoll_tx_running(slave_dev)))
 402                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 403         else
 404                 dev_queue_xmit(skb);
 405
 406         return 0;
 407 }
 408
 409 /*
 410  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 411  * We don't protect the slave list iteration with a lock because:
 412  * a. This operation is performed in IOCTL context,
 413  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 414  * c. Holding a lock with BH disabled while directly calling a base driver
 415  *    entry point is generally a BAD idea.
 416  *
 417  * The design of synchronization/protection for this operation in the 8021q
 418  * module is good for one or more VLAN devices over a single physical device
 419  * and cannot be extended for a teaming solution like bonding, so there is a
 420  * potential race condition here where a net device from the vlan group might
 421  * be referenced (either by a base driver or the 8021q code) while it is being
 422  * removed from the system. However, it turns out we're not making matters
 423  * worse, and if it works for regular VLAN usage it will work here too.
 424 */
 425
 426 /**
 427  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 428  * @bond_dev: bonding net device that got called
 429  * @vid: vlan id being added
 430  */
 431 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 432 {
 433         struct bonding *bond = netdev_priv(bond_dev);
 434         struct slave *slave, *stop_at;
 435         int i, res;
 436
 437         bond_for_each_slave(bond, slave, i) {
 438                 res = vlan_vid_add(slave->dev, vid);
 439                 if (res)
 440                         goto unwind;
 441         }
 442
 443         res = bond_add_vlan(bond, vid);
 444         if (res) {
 445                 pr_err("%s: Error: Failed to add vlan id %d\n",
 446                        bond_dev->name, vid);
 447                 return res;
 448         }
 449
 450         return 0;
 451
 452 unwind:
 453         /* unwind from head to the slave that failed */
 454         stop_at = slave;
 455         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
 456                 vlan_vid_del(slave->dev, vid);
 457
 458         return res;
 459 }
 460
 461 /**
 462  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 463  * @bond_dev: bonding net device that got called
 464  * @vid: vlan id being removed
 465  */
 466 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 467 {
 468         struct bonding *bond = netdev_priv(bond_dev);
 469         struct slave *slave;
 470         int i, res;
 471
 472         bond_for_each_slave(bond, slave, i)
 473                 vlan_vid_del(slave->dev, vid);
 474
 475         res = bond_del_vlan(bond, vid);
 476         if (res) {
 477                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 478                        bond_dev->name, vid);
 479                 return res;
 480         }
 481
 482         return 0;
 483 }
 484
 485 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 486 {
 487         struct vlan_entry *vlan;
 488         int res;
 489
 490         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 491                 res = vlan_vid_add(slave_dev, vlan->vlan_id);
 492                 if (res)
 493                         pr_warning("%s: Failed to add vlan id %d to device %s\n",
 494                                    bond->dev->name, vlan->vlan_id,
 495                                    slave_dev->name);
 496         }
 497 }
 498
 499 static void bond_del_vlans_from_slave(struct bonding *bond,
 500                                       struct net_device *slave_dev)
 501 {
 502         struct vlan_entry *vlan;
 503
 504         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 505                 if (!vlan->vlan_id)
 506                         continue;
 507                 vlan_vid_del(slave_dev, vlan->vlan_id);
 508         }
 509 }
 510
 511 /*------------------------------- Link status -------------------------------*/
 512
 513 /*
 514  * Set the carrier state for the master according to the state of its
 515  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 516  * do special 802.3ad magic.
 517  *
 518  * Returns zero if carrier state does not change, nonzero if it does.
 519  */
 520 static int bond_set_carrier(struct bonding *bond)
 521 {
 522         struct slave *slave;
 523         int i;
 524
 525         if (bond->slave_cnt == 0)
 526                 goto down;
 527
 528         if (bond->params.mode == BOND_MODE_8023AD)
 529                 return bond_3ad_set_carrier(bond);
 530
 531         bond_for_each_slave(bond, slave, i) {
 532                 if (slave->link == BOND_LINK_UP) {
 533                         if (!netif_carrier_ok(bond->dev)) {
 534                                 netif_carrier_on(bond->dev);
 535                                 return 1;
 536                         }
 537                         return 0;
 538                 }
 539         }
 540
 541 down:
 542         if (netif_carrier_ok(bond->dev)) {
 543                 netif_carrier_off(bond->dev);
 544                 return 1;
 545         }
 546         return 0;
 547 }
 548
 549 /*
 550  * Get link speed and duplex from the slave's base driver
 551  * using ethtool. If for some reason the call fails or the
 552  * values are invalid, set speed and duplex to -1,
 553  * and return.
 554  */
 555 static void bond_update_speed_duplex(struct slave *slave)
 556 {
 557         struct net_device *slave_dev = slave->dev;
 558         struct ethtool_cmd ecmd;
 559         u32 slave_speed;
 560         int res;
 561
 562         slave->speed = SPEED_UNKNOWN;
 563         slave->duplex = DUPLEX_UNKNOWN;
 564
 565         res = __ethtool_get_settings(slave_dev, &ecmd);
 566         if (res < 0)
 567                 return;
 568
 569         slave_speed = ethtool_cmd_speed(&ecmd);
 570         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 571                 return;
 572
 573         switch (ecmd.duplex) {
 574         case DUPLEX_FULL:
 575         case DUPLEX_HALF:
 576                 break;
 577         default:
 578                 return;
 579         }
 580
 581         slave->speed = slave_speed;
 582         slave->duplex = ecmd.duplex;
 583
 584         return;
 585 }
 586
 587 /*
 588  * if <dev> supports MII link status reporting, check its link status.
 589  *
 590  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 591  * depending upon the setting of the use_carrier parameter.
 592  *
 593  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 594  * can't tell and just pretend it is), or 0, meaning that the link is
 595  * down.
 596  *
 597  * If reporting is non-zero, instead of faking link up, return -1 if
 598  * both ETHTOOL and MII ioctls fail (meaning the device does not
 599  * support them).  If use_carrier is set, return whatever it says.
 600  * It'd be nice if there was a good way to tell if a driver supports
 601  * netif_carrier, but there really isn't.
 602  */
 603 static int bond_check_dev_link(struct bonding *bond,
 604                                struct net_device *slave_dev, int reporting)
 605 {
 606         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 607         int (*ioctl)(struct net_device *, struct ifreq *, int);
 608         struct ifreq ifr;
 609         struct mii_ioctl_data *mii;
 610
 611         if (!reporting && !netif_running(slave_dev))
 612                 return 0;
 613
 614         if (bond->params.use_carrier)
 615                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 616
 617         /* Try to get link status using Ethtool first. */
 618         if (slave_dev->ethtool_ops) {
 619                 if (slave_dev->ethtool_ops->get_link) {
 620                         u32 link;
 621
 622                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 623
 624                         return link ? BMSR_LSTATUS : 0;
 625                 }
 626         }
 627
 628         /* Ethtool can't be used, fallback to MII ioctls. */
 629         ioctl = slave_ops->ndo_do_ioctl;
 630         if (ioctl) {
 631                 /* TODO: set pointer to correct ioctl on a per team member */
 632                 /*       bases to make this more efficient. that is, once  */
 633                 /*       we determine the correct ioctl, we will always    */
 634                 /*       call it and not the others for that team          */
 635                 /*       member.                                           */
 636
 637                 /*
 638                  * We cannot assume that SIOCGMIIPHY will also read a
 639                  * register; not all network drivers (e.g., e100)
 640                  * support that.
 641                  */
 642
 643                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 644                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 645                 mii = if_mii(&ifr);
 646                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 647                         mii->reg_num = MII_BMSR;
 648                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 649                                 return mii->val_out & BMSR_LSTATUS;
 650                 }
 651         }
 652
 653         /*
 654          * If reporting, report that either there's no dev->do_ioctl,
 655          * or both SIOCGMIIREG and get_link failed (meaning that we
 656          * cannot report link status).  If not reporting, pretend
 657          * we're ok.
 658          */
 659         return reporting ? -1 : BMSR_LSTATUS;
 660 }
 661
 662 /*----------------------------- Multicast list ------------------------------*/
 663
 664 /*
 665  * Push the promiscuity flag down to appropriate slaves
 666  */
 667 static int bond_set_promiscuity(struct bonding *bond, int inc)
 668 {
 669         int err = 0;
 670         if (USES_PRIMARY(bond->params.mode)) {
 671                 /* write lock already acquired */
 672                 if (bond->curr_active_slave) {
 673                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 674                                                   inc);
 675                 }
 676         } else {
 677                 struct slave *slave;
 678                 int i;
 679                 bond_for_each_slave(bond, slave, i) {
 680                         err = dev_set_promiscuity(slave->dev, inc);
 681                         if (err)
 682                                 return err;
 683                 }
 684         }
 685         return err;
 686 }
 687
 688 /*
 689  * Push the allmulti flag down to all slaves
 690  */
 691 static int bond_set_allmulti(struct bonding *bond, int inc)
 692 {
 693         int err = 0;
 694         if (USES_PRIMARY(bond->params.mode)) {
 695                 /* write lock already acquired */
 696                 if (bond->curr_active_slave) {
 697                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 698                                                inc);
 699                 }
 700         } else {
 701                 struct slave *slave;
 702                 int i;
 703                 bond_for_each_slave(bond, slave, i) {
 704                         err = dev_set_allmulti(slave->dev, inc);
 705                         if (err)
 706                                 return err;
 707                 }
 708         }
 709         return err;
 710 }
 711
 712 /*
 713  * Add a Multicast address to slaves
 714  * according to mode
 715  */
 716 static void bond_mc_add(struct bonding *bond, void *addr)
 717 {
 718         if (USES_PRIMARY(bond->params.mode)) {
 719                 /* write lock already acquired */
 720                 if (bond->curr_active_slave)
 721                         dev_mc_add(bond->curr_active_slave->dev, addr);
 722         } else {
 723                 struct slave *slave;
 724                 int i;
 725
 726                 bond_for_each_slave(bond, slave, i)
 727                         dev_mc_add(slave->dev, addr);
 728         }
 729 }
 730
 731 /*
 732  * Remove a multicast address from slave
 733  * according to mode
 734  */
 735 static void bond_mc_del(struct bonding *bond, void *addr)
 736 {
 737         if (USES_PRIMARY(bond->params.mode)) {
 738                 /* write lock already acquired */
 739                 if (bond->curr_active_slave)
 740                         dev_mc_del(bond->curr_active_slave->dev, addr);
 741         } else {
 742                 struct slave *slave;
 743                 int i;
 744                 bond_for_each_slave(bond, slave, i) {
 745                         dev_mc_del(slave->dev, addr);
 746                 }
 747         }
 748 }
 749
 750
 751 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 752 {
 753         struct in_device *in_dev;
 754
 755         rcu_read_lock();
 756         in_dev = __in_dev_get_rcu(dev);
 757         if (in_dev)
 758                 ip_mc_rejoin_groups(in_dev);
 759         rcu_read_unlock();
 760 }
 761
 762 /*
 763  * Retrieve the list of registered multicast addresses for the bonding
 764  * device and retransmit an IGMP JOIN request to the current active
 765  * slave.
 766  */
 767 static void bond_resend_igmp_join_requests(struct bonding *bond)
 768 {
 769         struct net_device *bond_dev, *vlan_dev, *master_dev;
 770         struct vlan_entry *vlan;
 771
 772         read_lock(&bond->lock);
 773
 774         bond_dev = bond->dev;
 775
 776         /* rejoin all groups on bond device */
 777         __bond_resend_igmp_join_requests(bond_dev);
 778
 779         /*
 780          * if bond is enslaved to a bridge,
 781          * then rejoin all groups on its master
 782          */
 783         master_dev = bond_dev->master;
 784         if (master_dev)
 785                 if ((master_dev->priv_flags & IFF_EBRIDGE)
 786                         && (bond_dev->priv_flags & IFF_BRIDGE_PORT))
 787                         __bond_resend_igmp_join_requests(master_dev);
 788
 789         /* rejoin all groups on vlan devices */
 790         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 791                 rcu_read_lock();
 792                 vlan_dev = __vlan_find_dev_deep(bond_dev,
 793                                                 vlan->vlan_id);
 794                 rcu_read_unlock();
 795                 if (vlan_dev)
 796                         __bond_resend_igmp_join_requests(vlan_dev);
 797         }
 798
 799         if (--bond->igmp_retrans > 0)
 800                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 801
 802         read_unlock(&bond->lock);
 803 }
 804
 805 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 806 {
 807         struct bonding *bond = container_of(work, struct bonding,
 808                                             mcast_work.work);
 809         bond_resend_igmp_join_requests(bond);
 810 }
 811
 812 /*
 813  * flush all members of flush->mc_list from device dev->mc_list
 814  */
 815 static void bond_mc_list_flush(struct net_device *bond_dev,
 816                                struct net_device *slave_dev)
 817 {
 818         struct bonding *bond = netdev_priv(bond_dev);
 819         struct netdev_hw_addr *ha;
 820
 821         netdev_for_each_mc_addr(ha, bond_dev)
 822                 dev_mc_del(slave_dev, ha->addr);
 823
 824         if (bond->params.mode == BOND_MODE_8023AD) {
 825                 /* del lacpdu mc addr from mc list */
 826                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 827
 828                 dev_mc_del(slave_dev, lacpdu_multicast);
 829         }
 830 }
 831
 832 /*--------------------------- Active slave change ---------------------------*/
 833
 834 /*
 835  * Update the mc list and multicast-related flags for the new and
 836  * old active slaves (if any) according to the multicast mode, and
 837  * promiscuous flags unconditionally.
 838  */
 839 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 840                          struct slave *old_active)
 841 {
 842         struct netdev_hw_addr *ha;
 843
 844         if (!USES_PRIMARY(bond->params.mode))
 845                 /* nothing to do -  mc list is already up-to-date on
 846                  * all slaves
 847                  */
 848                 return;
 849
 850         if (old_active) {
 851                 if (bond->dev->flags & IFF_PROMISC)
 852                         dev_set_promiscuity(old_active->dev, -1);
 853
 854                 if (bond->dev->flags & IFF_ALLMULTI)
 855                         dev_set_allmulti(old_active->dev, -1);
 856
 857                 netdev_for_each_mc_addr(ha, bond->dev)
 858                         dev_mc_del(old_active->dev, ha->addr);
 859         }
 860
 861         if (new_active) {
 862                 /* FIXME: Signal errors upstream. */
 863                 if (bond->dev->flags & IFF_PROMISC)
 864                         dev_set_promiscuity(new_active->dev, 1);
 865
 866                 if (bond->dev->flags & IFF_ALLMULTI)
 867                         dev_set_allmulti(new_active->dev, 1);
 868
 869                 netdev_for_each_mc_addr(ha, bond->dev)
 870                         dev_mc_add(new_active->dev, ha->addr);
 871         }
 872 }
 873
 874 /*
 875  * bond_do_fail_over_mac
 876  *
 877  * Perform special MAC address swapping for fail_over_mac settings
 878  *
 879  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 880  */
 881 static void bond_do_fail_over_mac(struct bonding *bond,
 882                                   struct slave *new_active,
 883                                   struct slave *old_active)
 884         __releases(&bond->curr_slave_lock)
 885         __releases(&bond->lock)
 886         __acquires(&bond->lock)
 887         __acquires(&bond->curr_slave_lock)
 888 {
 889         u8 tmp_mac[ETH_ALEN];
 890         struct sockaddr saddr;
 891         int rv;
 892
 893         switch (bond->params.fail_over_mac) {
 894         case BOND_FOM_ACTIVE:
 895                 if (new_active) {
 896                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 897                                new_active->dev->addr_len);
 898                         write_unlock_bh(&bond->curr_slave_lock);
 899                         read_unlock(&bond->lock);
 900                         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
 901                         read_lock(&bond->lock);
 902                         write_lock_bh(&bond->curr_slave_lock);
 903                 }
 904                 break;
 905         case BOND_FOM_FOLLOW:
 906                 /*
 907                  * if new_active && old_active, swap them
 908                  * if just old_active, do nothing (going to no active slave)
 909                  * if just new_active, set new_active to bond's MAC
 910                  */
 911                 if (!new_active)
 912                         return;
 913
 914                 write_unlock_bh(&bond->curr_slave_lock);
 915                 read_unlock(&bond->lock);
 916
 917                 if (old_active) {
 918                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 919                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 920                                ETH_ALEN);
 921                         saddr.sa_family = new_active->dev->type;
 922                 } else {
 923                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 924                         saddr.sa_family = bond->dev->type;
 925                 }
 926
 927                 rv = dev_set_mac_address(new_active->dev, &saddr);
 928                 if (rv) {
 929                         pr_err("%s: Error %d setting MAC of slave %s\n",
 930                                bond->dev->name, -rv, new_active->dev->name);
 931                         goto out;
 932                 }
 933
 934                 if (!old_active)
 935                         goto out;
 936
 937                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 938                 saddr.sa_family = old_active->dev->type;
 939
 940                 rv = dev_set_mac_address(old_active->dev, &saddr);
 941                 if (rv)
 942                         pr_err("%s: Error %d setting MAC of slave %s\n",
 943                                bond->dev->name, -rv, new_active->dev->name);
 944 out:
 945                 read_lock(&bond->lock);
 946                 write_lock_bh(&bond->curr_slave_lock);
 947                 break;
 948         default:
 949                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 950                        bond->dev->name, bond->params.fail_over_mac);
 951                 break;
 952         }
 953
 954 }
 955
 956 static bool bond_should_change_active(struct bonding *bond)
 957 {
 958         struct slave *prim = bond->primary_slave;
 959         struct slave *curr = bond->curr_active_slave;
 960
 961         if (!prim || !curr || curr->link != BOND_LINK_UP)
 962                 return true;
 963         if (bond->force_primary) {
 964                 bond->force_primary = false;
 965                 return true;
 966         }
 967         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 968             (prim->speed < curr->speed ||
 969              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 970                 return false;
 971         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 972                 return false;
 973         return true;
 974 }
 975
 976 /**
 977  * find_best_interface - select the best available slave to be the active one
 978  * @bond: our bonding struct
 979  *
 980  * Warning: Caller must hold curr_slave_lock for writing.
 981  */
 982 static struct slave *bond_find_best_slave(struct bonding *bond)
 983 {
 984         struct slave *new_active, *old_active;
 985         struct slave *bestslave = NULL;
 986         int mintime = bond->params.updelay;
 987         int i;
 988
 989         new_active = bond->curr_active_slave;
 990
 991         if (!new_active) { /* there were no active slaves left */
 992                 if (bond->slave_cnt > 0)   /* found one slave */
 993                         new_active = bond->first_slave;
 994                 else
 995                         return NULL; /* still no slave, return NULL */
 996         }
 997
 998         if ((bond->primary_slave) &&
 999             bond->primary_slave->link == BOND_LINK_UP &&
1000             bond_should_change_active(bond)) {
1001                 new_active = bond->primary_slave;
1002         }
1003
1004         /* remember where to stop iterating over the slaves */
1005         old_active = new_active;
1006
1007         bond_for_each_slave_from(bond, new_active, i, old_active) {
1008                 if (new_active->link == BOND_LINK_UP) {
1009                         return new_active;
1010                 } else if (new_active->link == BOND_LINK_BACK &&
1011                            IS_UP(new_active->dev)) {
1012                         /* link up, but waiting for stabilization */
1013                         if (new_active->delay < mintime) {
1014                                 mintime = new_active->delay;
1015                                 bestslave = new_active;
1016                         }
1017                 }
1018         }
1019
1020         return bestslave;
1021 }
1022
1023 static bool bond_should_notify_peers(struct bonding *bond)
1024 {
1025         struct slave *slave = bond->curr_active_slave;
1026
1027         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1028                  bond->dev->name, slave ? slave->dev->name : "NULL");
1029
1030         if (!slave || !bond->send_peer_notif ||
1031             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1032                 return false;
1033
1034         bond->send_peer_notif--;
1035         return true;
1036 }
1037
1038 /**
1039  * change_active_interface - change the active slave into the specified one
1040  * @bond: our bonding struct
1041  * @new: the new slave to make the active one
1042  *
1043  * Set the new slave to the bond's settings and unset them on the old
1044  * curr_active_slave.
1045  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1046  *
1047  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1048  * because it is apparently the best available slave we have, even though its
1049  * updelay hasn't timed out yet.
1050  *
1051  * If new_active is not NULL, caller must hold bond->lock for read and
1052  * curr_slave_lock for write_bh.
1053  */
1054 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1055 {
1056         struct slave *old_active = bond->curr_active_slave;
1057
1058         if (old_active == new_active)
1059                 return;
1060
1061         if (new_active) {
1062                 new_active->jiffies = jiffies;
1063
1064                 if (new_active->link == BOND_LINK_BACK) {
1065                         if (USES_PRIMARY(bond->params.mode)) {
1066                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1067                                         bond->dev->name, new_active->dev->name,
1068                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1069                         }
1070
1071                         new_active->delay = 0;
1072                         new_active->link = BOND_LINK_UP;
1073
1074                         if (bond->params.mode == BOND_MODE_8023AD)
1075                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1076
1077                         if (bond_is_lb(bond))
1078                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1079                 } else {
1080                         if (USES_PRIMARY(bond->params.mode)) {
1081                                 pr_info("%s: making interface %s the new active one.\n",
1082                                         bond->dev->name, new_active->dev->name);
1083                         }
1084                 }
1085         }
1086
1087         if (USES_PRIMARY(bond->params.mode))
1088                 bond_mc_swap(bond, new_active, old_active);
1089
1090         if (bond_is_lb(bond)) {
1091                 bond_alb_handle_active_change(bond, new_active);
1092                 if (old_active)
1093                         bond_set_slave_inactive_flags(old_active);
1094                 if (new_active)
1095                         bond_set_slave_active_flags(new_active);
1096         } else {
1097                 bond->curr_active_slave = new_active;
1098         }
1099
1100         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1101                 if (old_active)
1102                         bond_set_slave_inactive_flags(old_active);
1103
1104                 if (new_active) {
1105                         bool should_notify_peers = false;
1106
1107                         bond_set_slave_active_flags(new_active);
1108
1109                         if (bond->params.fail_over_mac)
1110                                 bond_do_fail_over_mac(bond, new_active,
1111                                                       old_active);
1112
1113                         if (netif_running(bond->dev)) {
1114                                 bond->send_peer_notif =
1115                                         bond->params.num_peer_notif;
1116                                 should_notify_peers =
1117                                         bond_should_notify_peers(bond);
1118                         }
1119
1120                         write_unlock_bh(&bond->curr_slave_lock);
1121                         read_unlock(&bond->lock);
1122
1123                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1124                         if (should_notify_peers)
1125                                 netdev_bonding_change(bond->dev,
1126                                                       NETDEV_NOTIFY_PEERS);
1127
1128                         read_lock(&bond->lock);
1129                         write_lock_bh(&bond->curr_slave_lock);
1130                 }
1131         }
1132
1133         /* resend IGMP joins since active slave has changed or
1134          * all were sent on curr_active_slave.
1135          * resend only if bond is brought up with the affected
1136          * bonding modes and the retransmission is enabled */
1137         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1138             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1139              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1140                 bond->igmp_retrans = bond->params.resend_igmp;
1141                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1142         }
1143 }
1144
1145 /**
1146  * bond_select_active_slave - select a new active slave, if needed
1147  * @bond: our bonding struct
1148  *
1149  * This functions should be called when one of the following occurs:
1150  * - The old curr_active_slave has been released or lost its link.
1151  * - The primary_slave has got its link back.
1152  * - A slave has got its link back and there's no old curr_active_slave.
1153  *
1154  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1155  */
1156 void bond_select_active_slave(struct bonding *bond)
1157 {
1158         struct slave *best_slave;
1159         int rv;
1160
1161         best_slave = bond_find_best_slave(bond);
1162         if (best_slave != bond->curr_active_slave) {
1163                 bond_change_active_slave(bond, best_slave);
1164                 rv = bond_set_carrier(bond);
1165                 if (!rv)
1166                         return;
1167
1168                 if (netif_carrier_ok(bond->dev)) {
1169                         pr_info("%s: first active interface up!\n",
1170                                 bond->dev->name);
1171                 } else {
1172                         pr_info("%s: now running without any active interface !\n",
1173                                 bond->dev->name);
1174                 }
1175         }
1176 }
1177
1178 /*--------------------------- slave list handling ---------------------------*/
1179
1180 /*
1181  * This function attaches the slave to the end of list.
1182  *
1183  * bond->lock held for writing by caller.
1184  */
1185 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1186 {
1187         if (bond->first_slave == NULL) { /* attaching the first slave */
1188                 new_slave->next = new_slave;
1189                 new_slave->prev = new_slave;
1190                 bond->first_slave = new_slave;
1191         } else {
1192                 new_slave->next = bond->first_slave;
1193                 new_slave->prev = bond->first_slave->prev;
1194                 new_slave->next->prev = new_slave;
1195                 new_slave->prev->next = new_slave;
1196         }
1197
1198         bond->slave_cnt++;
1199 }
1200
1201 /*
1202  * This function detaches the slave from the list.
1203  * WARNING: no check is made to verify if the slave effectively
1204  * belongs to <bond>.
1205  * Nothing is freed on return, structures are just unchained.
1206  * If any slave pointer in bond was pointing to <slave>,
1207  * it should be changed by the calling function.
1208  *
1209  * bond->lock held for writing by caller.
1210  */
1211 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1212 {
1213         if (slave->next)
1214                 slave->next->prev = slave->prev;
1215
1216         if (slave->prev)
1217                 slave->prev->next = slave->next;
1218
1219         if (bond->first_slave == slave) { /* slave is the first slave */
1220                 if (bond->slave_cnt > 1) { /* there are more slave */
1221                         bond->first_slave = slave->next;
1222                 } else {
1223                         bond->first_slave = NULL; /* slave was the last one */
1224                 }
1225         }
1226
1227         slave->next = NULL;
1228         slave->prev = NULL;
1229         bond->slave_cnt--;
1230 }
1231
1232 #ifdef CONFIG_NET_POLL_CONTROLLER
1233 static inline int slave_enable_netpoll(struct slave *slave)
1234 {
1235         struct netpoll *np;
1236         int err = 0;
1237
1238         np = kzalloc(sizeof(*np), GFP_KERNEL);
1239         err = -ENOMEM;
1240         if (!np)
1241                 goto out;
1242
1243         err = __netpoll_setup(np, slave->dev);
1244         if (err) {
1245                 kfree(np);
1246                 goto out;
1247         }
1248         slave->np = np;
1249 out:
1250         return err;
1251 }
1252 static inline void slave_disable_netpoll(struct slave *slave)
1253 {
1254         struct netpoll *np = slave->np;
1255
1256         if (!np)
1257                 return;
1258
1259         slave->np = NULL;
1260         synchronize_rcu_bh();
1261         __netpoll_cleanup(np);
1262         kfree(np);
1263 }
1264 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1265 {
1266         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1267                 return false;
1268         if (!slave_dev->netdev_ops->ndo_poll_controller)
1269                 return false;
1270         return true;
1271 }
1272
1273 static void bond_poll_controller(struct net_device *bond_dev)
1274 {
1275 }
1276
1277 static void __bond_netpoll_cleanup(struct bonding *bond)
1278 {
1279         struct slave *slave;
1280         int i;
1281
1282         bond_for_each_slave(bond, slave, i)
1283                 if (IS_UP(slave->dev))
1284                         slave_disable_netpoll(slave);
1285 }
1286 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1287 {
1288         struct bonding *bond = netdev_priv(bond_dev);
1289
1290         read_lock(&bond->lock);
1291         __bond_netpoll_cleanup(bond);
1292         read_unlock(&bond->lock);
1293 }
1294
1295 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1296 {
1297         struct bonding *bond = netdev_priv(dev);
1298         struct slave *slave;
1299         int i, err = 0;
1300
1301         read_lock(&bond->lock);
1302         bond_for_each_slave(bond, slave, i) {
1303                 err = slave_enable_netpoll(slave);
1304                 if (err) {
1305                         __bond_netpoll_cleanup(bond);
1306                         break;
1307                 }
1308         }
1309         read_unlock(&bond->lock);
1310         return err;
1311 }
1312
1313 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1314 {
1315         return bond->dev->npinfo;
1316 }
1317
1318 #else
1319 static inline int slave_enable_netpoll(struct slave *slave)
1320 {
1321         return 0;
1322 }
1323 static inline void slave_disable_netpoll(struct slave *slave)
1324 {
1325 }
1326 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1327 {
1328 }
1329 #endif
1330
1331 /*---------------------------------- IOCTL ----------------------------------*/
1332
1333 static int bond_sethwaddr(struct net_device *bond_dev,
1334                           struct net_device *slave_dev)
1335 {
1336         pr_debug("bond_dev=%p\n", bond_dev);
1337         pr_debug("slave_dev=%p\n", slave_dev);
1338         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1339         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1340         return 0;
1341 }
1342
1343 static netdev_features_t bond_fix_features(struct net_device *dev,
1344         netdev_features_t features)
1345 {
1346         struct slave *slave;
1347         struct bonding *bond = netdev_priv(dev);
1348         netdev_features_t mask;
1349         int i;
1350
1351         read_lock(&bond->lock);
1352
1353         if (!bond->first_slave) {
1354                 /* Disable adding VLANs to empty bond. But why? --mq */
1355                 features |= NETIF_F_VLAN_CHALLENGED;
1356                 goto out;
1357         }
1358
1359         mask = features;
1360         features &= ~NETIF_F_ONE_FOR_ALL;
1361         features |= NETIF_F_ALL_FOR_ALL;
1362
1363         bond_for_each_slave(bond, slave, i) {
1364                 features = netdev_increment_features(features,
1365                                                      slave->dev->features,
1366                                                      mask);
1367         }
1368
1369 out:
1370         read_unlock(&bond->lock);
1371         return features;
1372 }
1373
1374 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1375                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1376                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1377
1378 static void bond_compute_features(struct bonding *bond)
1379 {
1380         struct slave *slave;
1381         struct net_device *bond_dev = bond->dev;
1382         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1383         unsigned short max_hard_header_len = ETH_HLEN;
1384         int i;
1385         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1386
1387         read_lock(&bond->lock);
1388
1389         if (!bond->first_slave)
1390                 goto done;
1391
1392         bond_for_each_slave(bond, slave, i) {
1393                 vlan_features = netdev_increment_features(vlan_features,
1394                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1395
1396                 dst_release_flag &= slave->dev->priv_flags;
1397                 if (slave->dev->hard_header_len > max_hard_header_len)
1398                         max_hard_header_len = slave->dev->hard_header_len;
1399         }
1400
1401 done:
1402         bond_dev->vlan_features = vlan_features;
1403         bond_dev->hard_header_len = max_hard_header_len;
1404
1405         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1406         bond_dev->priv_flags = flags | dst_release_flag;
1407
1408         read_unlock(&bond->lock);
1409
1410         netdev_change_features(bond_dev);
1411 }
1412
1413 static void bond_setup_by_slave(struct net_device *bond_dev,
1414                                 struct net_device *slave_dev)
1415 {
1416         struct bonding *bond = netdev_priv(bond_dev);
1417
1418         bond_dev->header_ops        = slave_dev->header_ops;
1419
1420         bond_dev->type              = slave_dev->type;
1421         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1422         bond_dev->addr_len          = slave_dev->addr_len;
1423
1424         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1425                 slave_dev->addr_len);
1426         bond->setup_by_slave = 1;
1427 }
1428
1429 /* On bonding slaves other than the currently active slave, suppress
1430  * duplicates except for alb non-mcast/bcast.
1431  */
1432 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1433                                             struct slave *slave,
1434                                             struct bonding *bond)
1435 {
1436         if (bond_is_slave_inactive(slave)) {
1437                 if (bond->params.mode == BOND_MODE_ALB &&
1438                     skb->pkt_type != PACKET_BROADCAST &&
1439                     skb->pkt_type != PACKET_MULTICAST)
1440                         return false;
1441                 return true;
1442         }
1443         return false;
1444 }
1445
1446 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1447 {
1448         struct sk_buff *skb = *pskb;
1449         struct slave *slave;
1450         struct bonding *bond;
1451         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1452                           struct slave *);
1453         int ret = RX_HANDLER_ANOTHER;
1454
1455         skb = skb_share_check(skb, GFP_ATOMIC);
1456         if (unlikely(!skb))
1457                 return RX_HANDLER_CONSUMED;
1458
1459         *pskb = skb;
1460
1461         slave = bond_slave_get_rcu(skb->dev);
1462         bond = slave->bond;
1463
1464         if (bond->params.arp_interval)
1465                 slave->dev->last_rx = jiffies;
1466
1467         recv_probe = ACCESS_ONCE(bond->recv_probe);
1468         if (recv_probe) {
1469                 ret = recv_probe(skb, bond, slave);
1470                 if (ret == RX_HANDLER_CONSUMED) {
1471                         consume_skb(skb);
1472                         return ret;
1473                 }
1474         }
1475
1476         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1477                 return RX_HANDLER_EXACT;
1478         }
1479
1480         skb->dev = bond->dev;
1481
1482         if (bond->params.mode == BOND_MODE_ALB &&
1483             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1484             skb->pkt_type == PACKET_HOST) {
1485
1486                 if (unlikely(skb_cow_head(skb,
1487                                           skb->data - skb_mac_header(skb)))) {
1488                         kfree_skb(skb);
1489                         return RX_HANDLER_CONSUMED;
1490                 }
1491                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1492         }
1493
1494         return ret;
1495 }
1496
1497 /* enslave device <slave> to bond device <master> */
1498 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1499 {
1500         struct bonding *bond = netdev_priv(bond_dev);
1501         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1502         struct slave *new_slave = NULL;
1503         struct netdev_hw_addr *ha;
1504         struct sockaddr addr;
1505         int link_reporting;
1506         int res = 0;
1507
1508         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1509                 slave_ops->ndo_do_ioctl == NULL) {
1510                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1511                            bond_dev->name, slave_dev->name);
1512         }
1513
1514         /* already enslaved */
1515         if (slave_dev->flags & IFF_SLAVE) {
1516                 pr_debug("Error, Device was already enslaved\n");
1517                 return -EBUSY;
1518         }
1519
1520         /* vlan challenged mutual exclusion */
1521         /* no need to lock since we're protected by rtnl_lock */
1522         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1523                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1524                 if (bond_vlan_used(bond)) {
1525                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1526                                bond_dev->name, slave_dev->name, bond_dev->name);
1527                         return -EPERM;
1528                 } else {
1529                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1530                                    bond_dev->name, slave_dev->name,
1531                                    slave_dev->name, bond_dev->name);
1532                 }
1533         } else {
1534                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1535         }
1536
1537         /*
1538          * Old ifenslave binaries are no longer supported.  These can
1539          * be identified with moderate accuracy by the state of the slave:
1540          * the current ifenslave will set the interface down prior to
1541          * enslaving it; the old ifenslave will not.
1542          */
1543         if ((slave_dev->flags & IFF_UP)) {
1544                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1545                        slave_dev->name);
1546                 res = -EPERM;
1547                 goto err_undo_flags;
1548         }
1549
1550         /* set bonding device ether type by slave - bonding netdevices are
1551          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1552          * there is a need to override some of the type dependent attribs/funcs.
1553          *
1554          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1555          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1556          */
1557         if (bond->slave_cnt == 0) {
1558                 if (bond_dev->type != slave_dev->type) {
1559                         pr_debug("%s: change device type from %d to %d\n",
1560                                  bond_dev->name,
1561                                  bond_dev->type, slave_dev->type);
1562
1563                         res = netdev_bonding_change(bond_dev,
1564                                                     NETDEV_PRE_TYPE_CHANGE);
1565                         res = notifier_to_errno(res);
1566                         if (res) {
1567                                 pr_err("%s: refused to change device type\n",
1568                                        bond_dev->name);
1569                                 res = -EBUSY;
1570                                 goto err_undo_flags;
1571                         }
1572
1573                         /* Flush unicast and multicast addresses */
1574                         dev_uc_flush(bond_dev);
1575                         dev_mc_flush(bond_dev);
1576
1577                         if (slave_dev->type != ARPHRD_ETHER)
1578                                 bond_setup_by_slave(bond_dev, slave_dev);
1579                         else {
1580                                 ether_setup(bond_dev);
1581                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1582                         }
1583
1584                         netdev_bonding_change(bond_dev,
1585                                               NETDEV_POST_TYPE_CHANGE);
1586                 }
1587         } else if (bond_dev->type != slave_dev->type) {
1588                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1589                        slave_dev->name,
1590                        slave_dev->type, bond_dev->type);
1591                 res = -EINVAL;
1592                 goto err_undo_flags;
1593         }
1594
1595         if (slave_ops->ndo_set_mac_address == NULL) {
1596                 if (bond->slave_cnt == 0) {
1597                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1598                                    bond_dev->name);
1599                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1600                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1601                         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",
1602                                bond_dev->name);
1603                         res = -EOPNOTSUPP;
1604                         goto err_undo_flags;
1605                 }
1606         }
1607
1608         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1609
1610         /* If this is the first slave, then we need to set the master's hardware
1611          * address to be the same as the slave's. */
1612         if (is_zero_ether_addr(bond->dev->dev_addr))
1613                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1614                        slave_dev->addr_len);
1615
1616
1617         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1618         if (!new_slave) {
1619                 res = -ENOMEM;
1620                 goto err_undo_flags;
1621         }
1622
1623         /*
1624          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1625          * is set via sysfs or module option if desired.
1626          */
1627         new_slave->queue_id = 0;
1628
1629         /* Save slave's original mtu and then set it to match the bond */
1630         new_slave->original_mtu = slave_dev->mtu;
1631         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1632         if (res) {
1633                 pr_debug("Error %d calling dev_set_mtu\n", res);
1634                 goto err_free;
1635         }
1636
1637         /*
1638          * Save slave's original ("permanent") mac address for modes
1639          * that need it, and for restoring it upon release, and then
1640          * set it to the master's address
1641          */
1642         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1643
1644         if (!bond->params.fail_over_mac) {
1645                 /*
1646                  * Set slave to master's mac address.  The application already
1647                  * set the master's mac address to that of the first slave
1648                  */
1649                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1650                 addr.sa_family = slave_dev->type;
1651                 res = dev_set_mac_address(slave_dev, &addr);
1652                 if (res) {
1653                         pr_debug("Error %d calling set_mac_address\n", res);
1654                         goto err_restore_mtu;
1655                 }
1656         }
1657
1658         res = netdev_set_bond_master(slave_dev, bond_dev);
1659         if (res) {
1660                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1661                 goto err_restore_mac;
1662         }
1663
1664         /* open the slave since the application closed it */
1665         res = dev_open(slave_dev);
1666         if (res) {
1667                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1668                 goto err_unset_master;
1669         }
1670
1671         new_slave->bond = bond;
1672         new_slave->dev = slave_dev;
1673         slave_dev->priv_flags |= IFF_BONDING;
1674
1675         if (bond_is_lb(bond)) {
1676                 /* bond_alb_init_slave() must be called before all other stages since
1677                  * it might fail and we do not want to have to undo everything
1678                  */
1679                 res = bond_alb_init_slave(bond, new_slave);
1680                 if (res)
1681                         goto err_close;
1682         }
1683
1684         /* If the mode USES_PRIMARY, then the new slave gets the
1685          * master's promisc (and mc) settings only if it becomes the
1686          * curr_active_slave, and that is taken care of later when calling
1687          * bond_change_active()
1688          */
1689         if (!USES_PRIMARY(bond->params.mode)) {
1690                 /* set promiscuity level to new slave */
1691                 if (bond_dev->flags & IFF_PROMISC) {
1692                         res = dev_set_promiscuity(slave_dev, 1);
1693                         if (res)
1694                                 goto err_close;
1695                 }
1696
1697                 /* set allmulti level to new slave */
1698                 if (bond_dev->flags & IFF_ALLMULTI) {
1699                         res = dev_set_allmulti(slave_dev, 1);
1700                         if (res)
1701                                 goto err_close;
1702                 }
1703
1704                 netif_addr_lock_bh(bond_dev);
1705                 /* upload master's mc_list to new slave */
1706                 netdev_for_each_mc_addr(ha, bond_dev)
1707                         dev_mc_add(slave_dev, ha->addr);
1708                 netif_addr_unlock_bh(bond_dev);
1709         }
1710
1711         if (bond->params.mode == BOND_MODE_8023AD) {
1712                 /* add lacpdu mc addr to mc list */
1713                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1714
1715                 dev_mc_add(slave_dev, lacpdu_multicast);
1716         }
1717
1718         bond_add_vlans_on_slave(bond, slave_dev);
1719
1720         write_lock_bh(&bond->lock);
1721
1722         bond_attach_slave(bond, new_slave);
1723
1724         new_slave->delay = 0;
1725         new_slave->link_failure_count = 0;
1726
1727         write_unlock_bh(&bond->lock);
1728
1729         bond_compute_features(bond);
1730
1731         read_lock(&bond->lock);
1732
1733         new_slave->last_arp_rx = jiffies -
1734                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1735
1736         if (bond->params.miimon && !bond->params.use_carrier) {
1737                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1738
1739                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1740                         /*
1741                          * miimon is set but a bonded network driver
1742                          * does not support ETHTOOL/MII and
1743                          * arp_interval is not set.  Note: if
1744                          * use_carrier is enabled, we will never go
1745                          * here (because netif_carrier is always
1746                          * supported); thus, we don't need to change
1747                          * the messages for netif_carrier.
1748                          */
1749                         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",
1750                                bond_dev->name, slave_dev->name);
1751                 } else if (link_reporting == -1) {
1752                         /* unable get link status using mii/ethtool */
1753                         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",
1754                                    bond_dev->name, slave_dev->name);
1755                 }
1756         }
1757
1758         /* check for initial state */
1759         if (bond->params.miimon) {
1760                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1761                         if (bond->params.updelay) {
1762                                 new_slave->link = BOND_LINK_BACK;
1763                                 new_slave->delay = bond->params.updelay;
1764                         } else {
1765                                 new_slave->link = BOND_LINK_UP;
1766                         }
1767                 } else {
1768                         new_slave->link = BOND_LINK_DOWN;
1769                 }
1770         } else if (bond->params.arp_interval) {
1771                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1772                         BOND_LINK_UP : BOND_LINK_DOWN);
1773         } else {
1774                 new_slave->link = BOND_LINK_UP;
1775         }
1776
1777         if (new_slave->link != BOND_LINK_DOWN)
1778                 new_slave->jiffies = jiffies;
1779         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1780                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1781                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1782
1783         bond_update_speed_duplex(new_slave);
1784
1785         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1786                 /* if there is a primary slave, remember it */
1787                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1788                         bond->primary_slave = new_slave;
1789                         bond->force_primary = true;
1790                 }
1791         }
1792
1793         write_lock_bh(&bond->curr_slave_lock);
1794
1795         switch (bond->params.mode) {
1796         case BOND_MODE_ACTIVEBACKUP:
1797                 bond_set_slave_inactive_flags(new_slave);
1798                 bond_select_active_slave(bond);
1799                 break;
1800         case BOND_MODE_8023AD:
1801                 /* in 802.3ad mode, the internal mechanism
1802                  * will activate the slaves in the selected
1803                  * aggregator
1804                  */
1805                 bond_set_slave_inactive_flags(new_slave);
1806                 /* if this is the first slave */
1807                 if (bond->slave_cnt == 1) {
1808                         SLAVE_AD_INFO(new_slave).id = 1;
1809                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1810                          * can be called only after the mac address of the bond is set
1811                          */
1812                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1813                 } else {
1814                         SLAVE_AD_INFO(new_slave).id =
1815                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1816                 }
1817
1818                 bond_3ad_bind_slave(new_slave);
1819                 break;
1820         case BOND_MODE_TLB:
1821         case BOND_MODE_ALB:
1822                 bond_set_active_slave(new_slave);
1823                 bond_set_slave_inactive_flags(new_slave);
1824                 bond_select_active_slave(bond);
1825                 break;
1826         default:
1827                 pr_debug("This slave is always active in trunk mode\n");
1828
1829                 /* always active in trunk mode */
1830                 bond_set_active_slave(new_slave);
1831
1832                 /* In trunking mode there is little meaning to curr_active_slave
1833                  * anyway (it holds no special properties of the bond device),
1834                  * so we can change it without calling change_active_interface()
1835                  */
1836                 if (!bond->curr_active_slave)
1837                         bond->curr_active_slave = new_slave;
1838
1839                 break;
1840         } /* switch(bond_mode) */
1841
1842         write_unlock_bh(&bond->curr_slave_lock);
1843
1844         bond_set_carrier(bond);
1845
1846 #ifdef CONFIG_NET_POLL_CONTROLLER
1847         slave_dev->npinfo = bond_netpoll_info(bond);
1848         if (slave_dev->npinfo) {
1849                 if (slave_enable_netpoll(new_slave)) {
1850                         read_unlock(&bond->lock);
1851                         pr_info("Error, %s: master_dev is using netpoll, "
1852                                  "but new slave device does not support netpoll.\n",
1853                                  bond_dev->name);
1854                         res = -EBUSY;
1855                         goto err_detach;
1856                 }
1857         }
1858 #endif
1859
1860         read_unlock(&bond->lock);
1861
1862         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1863         if (res)
1864                 goto err_detach;
1865
1866         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1867                                          new_slave);
1868         if (res) {
1869                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1870                 goto err_dest_symlinks;
1871         }
1872
1873         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1874                 bond_dev->name, slave_dev->name,
1875                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1876                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1877
1878         /* enslave is successful */
1879         return 0;
1880
1881 /* Undo stages on error */
1882 err_dest_symlinks:
1883         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1884
1885 err_detach:
1886         write_lock_bh(&bond->lock);
1887         bond_detach_slave(bond, new_slave);
1888         write_unlock_bh(&bond->lock);
1889
1890 err_close:
1891         dev_close(slave_dev);
1892
1893 err_unset_master:
1894         netdev_set_bond_master(slave_dev, NULL);
1895
1896 err_restore_mac:
1897         if (!bond->params.fail_over_mac) {
1898                 /* XXX TODO - fom follow mode needs to change master's
1899                  * MAC if this slave's MAC is in use by the bond, or at
1900                  * least print a warning.
1901                  */
1902                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1903                 addr.sa_family = slave_dev->type;
1904                 dev_set_mac_address(slave_dev, &addr);
1905         }
1906
1907 err_restore_mtu:
1908         dev_set_mtu(slave_dev, new_slave->original_mtu);
1909
1910 err_free:
1911         kfree(new_slave);
1912
1913 err_undo_flags:
1914         bond_compute_features(bond);
1915
1916         return res;
1917 }
1918
1919 /*
1920  * Try to release the slave device <slave> from the bond device <master>
1921  * It is legal to access curr_active_slave without a lock because all the function
1922  * is write-locked.
1923  *
1924  * The rules for slave state should be:
1925  *   for Active/Backup:
1926  *     Active stays on all backups go down
1927  *   for Bonded connections:
1928  *     The first up interface should be left on and all others downed.
1929  */
1930 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1931 {
1932         struct bonding *bond = netdev_priv(bond_dev);
1933         struct slave *slave, *oldcurrent;
1934         struct sockaddr addr;
1935         netdev_features_t old_features = bond_dev->features;
1936
1937         /* slave is not a slave or master is not master of this slave */
1938         if (!(slave_dev->flags & IFF_SLAVE) ||
1939             (slave_dev->master != bond_dev)) {
1940                 pr_err("%s: Error: cannot release %s.\n",
1941                        bond_dev->name, slave_dev->name);
1942                 return -EINVAL;
1943         }
1944
1945         block_netpoll_tx();
1946         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1947         write_lock_bh(&bond->lock);
1948
1949         slave = bond_get_slave_by_dev(bond, slave_dev);
1950         if (!slave) {
1951                 /* not a slave of this bond */
1952                 pr_info("%s: %s not enslaved\n",
1953                         bond_dev->name, slave_dev->name);
1954                 write_unlock_bh(&bond->lock);
1955                 unblock_netpoll_tx();
1956                 return -EINVAL;
1957         }
1958
1959         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1960          * for this slave anymore.
1961          */
1962         netdev_rx_handler_unregister(slave_dev);
1963         write_unlock_bh(&bond->lock);
1964         synchronize_net();
1965         write_lock_bh(&bond->lock);
1966
1967         if (!bond->params.fail_over_mac) {
1968                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1969                     bond->slave_cnt > 1)
1970                         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",
1971                                    bond_dev->name, slave_dev->name,
1972                                    slave->perm_hwaddr,
1973                                    bond_dev->name, slave_dev->name);
1974         }
1975
1976         /* Inform AD package of unbinding of slave. */
1977         if (bond->params.mode == BOND_MODE_8023AD) {
1978                 /* must be called before the slave is
1979                  * detached from the list
1980                  */
1981                 bond_3ad_unbind_slave(slave);
1982         }
1983
1984         pr_info("%s: releasing %s interface %s\n",
1985                 bond_dev->name,
1986                 bond_is_active_slave(slave) ? "active" : "backup",
1987                 slave_dev->name);
1988
1989         oldcurrent = bond->curr_active_slave;
1990
1991         bond->current_arp_slave = NULL;
1992
1993         /* release the slave from its bond */
1994         bond_detach_slave(bond, slave);
1995
1996         if (bond->primary_slave == slave)
1997                 bond->primary_slave = NULL;
1998
1999         if (oldcurrent == slave)
2000                 bond_change_active_slave(bond, NULL);
2001
2002         if (bond_is_lb(bond)) {
2003                 /* Must be called only after the slave has been
2004                  * detached from the list and the curr_active_slave
2005                  * has been cleared (if our_slave == old_current),
2006                  * but before a new active slave is selected.
2007                  */
2008                 write_unlock_bh(&bond->lock);
2009                 bond_alb_deinit_slave(bond, slave);
2010                 write_lock_bh(&bond->lock);
2011         }
2012
2013         if (oldcurrent == slave) {
2014                 /*
2015                  * Note that we hold RTNL over this sequence, so there
2016                  * is no concern that another slave add/remove event
2017                  * will interfere.
2018                  */
2019                 write_unlock_bh(&bond->lock);
2020                 read_lock(&bond->lock);
2021                 write_lock_bh(&bond->curr_slave_lock);
2022
2023                 bond_select_active_slave(bond);
2024
2025                 write_unlock_bh(&bond->curr_slave_lock);
2026                 read_unlock(&bond->lock);
2027                 write_lock_bh(&bond->lock);
2028         }
2029
2030         if (bond->slave_cnt == 0) {
2031                 bond_set_carrier(bond);
2032
2033                 /* if the last slave was removed, zero the mac address
2034                  * of the master so it will be set by the application
2035                  * to the mac address of the first slave
2036                  */
2037                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2038
2039                 if (bond_vlan_used(bond)) {
2040                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2041                                    bond_dev->name, bond_dev->name);
2042                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2043                                    bond_dev->name);
2044                 }
2045         }
2046
2047         write_unlock_bh(&bond->lock);
2048         unblock_netpoll_tx();
2049
2050         if (bond->slave_cnt == 0)
2051                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2052
2053         bond_compute_features(bond);
2054         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2055             (old_features & NETIF_F_VLAN_CHALLENGED))
2056                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2057                         bond_dev->name, slave_dev->name, bond_dev->name);
2058
2059         /* must do this from outside any spinlocks */
2060         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2061
2062         bond_del_vlans_from_slave(bond, slave_dev);
2063
2064         /* If the mode USES_PRIMARY, then we should only remove its
2065          * promisc and mc settings if it was the curr_active_slave, but that was
2066          * already taken care of above when we detached the slave
2067          */
2068         if (!USES_PRIMARY(bond->params.mode)) {
2069                 /* unset promiscuity level from slave */
2070                 if (bond_dev->flags & IFF_PROMISC)
2071                         dev_set_promiscuity(slave_dev, -1);
2072
2073                 /* unset allmulti level from slave */
2074                 if (bond_dev->flags & IFF_ALLMULTI)
2075                         dev_set_allmulti(slave_dev, -1);
2076
2077                 /* flush master's mc_list from slave */
2078                 netif_addr_lock_bh(bond_dev);
2079                 bond_mc_list_flush(bond_dev, slave_dev);
2080                 netif_addr_unlock_bh(bond_dev);
2081         }
2082
2083         netdev_set_bond_master(slave_dev, NULL);
2084
2085         slave_disable_netpoll(slave);
2086
2087         /* close slave before restoring its mac address */
2088         dev_close(slave_dev);
2089
2090         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2091                 /* restore original ("permanent") mac address */
2092                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2093                 addr.sa_family = slave_dev->type;
2094                 dev_set_mac_address(slave_dev, &addr);
2095         }
2096
2097         dev_set_mtu(slave_dev, slave->original_mtu);
2098
2099         slave_dev->priv_flags &= ~IFF_BONDING;
2100
2101         kfree(slave);
2102
2103         return 0;  /* deletion OK */
2104 }
2105
2106 /*
2107 * First release a slave and then destroy the bond if no more slaves are left.
2108 * Must be under rtnl_lock when this function is called.
2109 */
2110 static int  bond_release_and_destroy(struct net_device *bond_dev,
2111                                      struct net_device *slave_dev)
2112 {
2113         struct bonding *bond = netdev_priv(bond_dev);
2114         int ret;
2115
2116         ret = bond_release(bond_dev, slave_dev);
2117         if ((ret == 0) && (bond->slave_cnt == 0)) {
2118                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2119                 pr_info("%s: destroying bond %s.\n",
2120                         bond_dev->name, bond_dev->name);
2121                 unregister_netdevice(bond_dev);
2122         }
2123         return ret;
2124 }
2125
2126 /*
2127  * This function releases all slaves.
2128  */
2129 static int bond_release_all(struct net_device *bond_dev)
2130 {
2131         struct bonding *bond = netdev_priv(bond_dev);
2132         struct slave *slave;
2133         struct net_device *slave_dev;
2134         struct sockaddr addr;
2135
2136         write_lock_bh(&bond->lock);
2137
2138         netif_carrier_off(bond_dev);
2139
2140         if (bond->slave_cnt == 0)
2141                 goto out;
2142
2143         bond->current_arp_slave = NULL;
2144         bond->primary_slave = NULL;
2145         bond_change_active_slave(bond, NULL);
2146
2147         while ((slave = bond->first_slave) != NULL) {
2148                 /* Inform AD package of unbinding of slave
2149                  * before slave is detached from the list.
2150                  */
2151                 if (bond->params.mode == BOND_MODE_8023AD)
2152                         bond_3ad_unbind_slave(slave);
2153
2154                 slave_dev = slave->dev;
2155                 bond_detach_slave(bond, slave);
2156
2157                 /* now that the slave is detached, unlock and perform
2158                  * all the undo steps that should not be called from
2159                  * within a lock.
2160                  */
2161                 write_unlock_bh(&bond->lock);
2162
2163                 /* unregister rx_handler early so bond_handle_frame wouldn't
2164                  * be called for this slave anymore.
2165                  */
2166                 netdev_rx_handler_unregister(slave_dev);
2167                 synchronize_net();
2168
2169                 if (bond_is_lb(bond)) {
2170                         /* must be called only after the slave
2171                          * has been detached from the list
2172                          */
2173                         bond_alb_deinit_slave(bond, slave);
2174                 }
2175
2176                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2177                 bond_del_vlans_from_slave(bond, slave_dev);
2178
2179                 /* If the mode USES_PRIMARY, then we should only remove its
2180                  * promisc and mc settings if it was the curr_active_slave, but that was
2181                  * already taken care of above when we detached the slave
2182                  */
2183                 if (!USES_PRIMARY(bond->params.mode)) {
2184                         /* unset promiscuity level from slave */
2185                         if (bond_dev->flags & IFF_PROMISC)
2186                                 dev_set_promiscuity(slave_dev, -1);
2187
2188                         /* unset allmulti level from slave */
2189                         if (bond_dev->flags & IFF_ALLMULTI)
2190                                 dev_set_allmulti(slave_dev, -1);
2191
2192                         /* flush master's mc_list from slave */
2193                         netif_addr_lock_bh(bond_dev);
2194                         bond_mc_list_flush(bond_dev, slave_dev);
2195                         netif_addr_unlock_bh(bond_dev);
2196                 }
2197
2198                 netdev_set_bond_master(slave_dev, NULL);
2199
2200                 slave_disable_netpoll(slave);
2201
2202                 /* close slave before restoring its mac address */
2203                 dev_close(slave_dev);
2204
2205                 if (!bond->params.fail_over_mac) {
2206                         /* restore original ("permanent") mac address*/
2207                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2208                         addr.sa_family = slave_dev->type;
2209                         dev_set_mac_address(slave_dev, &addr);
2210                 }
2211
2212                 kfree(slave);
2213
2214                 /* re-acquire the lock before getting the next slave */
2215                 write_lock_bh(&bond->lock);
2216         }
2217
2218         /* zero the mac address of the master so it will be
2219          * set by the application to the mac address of the
2220          * first slave
2221          */
2222         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2223
2224         if (bond_vlan_used(bond)) {
2225                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2226                            bond_dev->name, bond_dev->name);
2227                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2228                            bond_dev->name);
2229         }
2230
2231         pr_info("%s: released all slaves\n", bond_dev->name);
2232
2233 out:
2234         write_unlock_bh(&bond->lock);
2235
2236         bond_compute_features(bond);
2237
2238         return 0;
2239 }
2240
2241 /*
2242  * This function changes the active slave to slave <slave_dev>.
2243  * It returns -EINVAL in the following cases.
2244  *  - <slave_dev> is not found in the list.
2245  *  - There is not active slave now.
2246  *  - <slave_dev> is already active.
2247  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2248  *  - <slave_dev> is not running.
2249  * In these cases, this function does nothing.
2250  * In the other cases, current_slave pointer is changed and 0 is returned.
2251  */
2252 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2253 {
2254         struct bonding *bond = netdev_priv(bond_dev);
2255         struct slave *old_active = NULL;
2256         struct slave *new_active = NULL;
2257         int res = 0;
2258
2259         if (!USES_PRIMARY(bond->params.mode))
2260                 return -EINVAL;
2261
2262         /* Verify that master_dev is indeed the master of slave_dev */
2263         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2264                 return -EINVAL;
2265
2266         read_lock(&bond->lock);
2267
2268         read_lock(&bond->curr_slave_lock);
2269         old_active = bond->curr_active_slave;
2270         read_unlock(&bond->curr_slave_lock);
2271
2272         new_active = bond_get_slave_by_dev(bond, slave_dev);
2273
2274         /*
2275          * Changing to the current active: do nothing; return success.
2276          */
2277         if (new_active && (new_active == old_active)) {
2278                 read_unlock(&bond->lock);
2279                 return 0;
2280         }
2281
2282         if ((new_active) &&
2283             (old_active) &&
2284             (new_active->link == BOND_LINK_UP) &&
2285             IS_UP(new_active->dev)) {
2286                 block_netpoll_tx();
2287                 write_lock_bh(&bond->curr_slave_lock);
2288                 bond_change_active_slave(bond, new_active);
2289                 write_unlock_bh(&bond->curr_slave_lock);
2290                 unblock_netpoll_tx();
2291         } else
2292                 res = -EINVAL;
2293
2294         read_unlock(&bond->lock);
2295
2296         return res;
2297 }
2298
2299 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2300 {
2301         struct bonding *bond = netdev_priv(bond_dev);
2302
2303         info->bond_mode = bond->params.mode;
2304         info->miimon = bond->params.miimon;
2305
2306         read_lock(&bond->lock);
2307         info->num_slaves = bond->slave_cnt;
2308         read_unlock(&bond->lock);
2309
2310         return 0;
2311 }
2312
2313 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2314 {
2315         struct bonding *bond = netdev_priv(bond_dev);
2316         struct slave *slave;
2317         int i, res = -ENODEV;
2318
2319         read_lock(&bond->lock);
2320
2321         bond_for_each_slave(bond, slave, i) {
2322                 if (i == (int)info->slave_id) {
2323                         res = 0;
2324                         strcpy(info->slave_name, slave->dev->name);
2325                         info->link = slave->link;
2326                         info->state = bond_slave_state(slave);
2327                         info->link_failure_count = slave->link_failure_count;
2328                         break;
2329                 }
2330         }
2331
2332         read_unlock(&bond->lock);
2333
2334         return res;
2335 }
2336
2337 /*-------------------------------- Monitoring -------------------------------*/
2338
2339
2340 static int bond_miimon_inspect(struct bonding *bond)
2341 {
2342         struct slave *slave;
2343         int i, link_state, commit = 0;
2344         bool ignore_updelay;
2345
2346         ignore_updelay = !bond->curr_active_slave ? true : false;
2347
2348         bond_for_each_slave(bond, slave, i) {
2349                 slave->new_link = BOND_LINK_NOCHANGE;
2350
2351                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2352
2353                 switch (slave->link) {
2354                 case BOND_LINK_UP:
2355                         if (link_state)
2356                                 continue;
2357
2358                         slave->link = BOND_LINK_FAIL;
2359                         slave->delay = bond->params.downdelay;
2360                         if (slave->delay) {
2361                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2362                                         bond->dev->name,
2363                                         (bond->params.mode ==
2364                                          BOND_MODE_ACTIVEBACKUP) ?
2365                                         (bond_is_active_slave(slave) ?
2366                                          "active " : "backup ") : "",
2367                                         slave->dev->name,
2368                                         bond->params.downdelay * bond->params.miimon);
2369                         }
2370                         /*FALLTHRU*/
2371                 case BOND_LINK_FAIL:
2372                         if (link_state) {
2373                                 /*
2374                                  * recovered before downdelay expired
2375                                  */
2376                                 slave->link = BOND_LINK_UP;
2377                                 slave->jiffies = jiffies;
2378                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2379                                         bond->dev->name,
2380                                         (bond->params.downdelay - slave->delay) *
2381                                         bond->params.miimon,
2382                                         slave->dev->name);
2383                                 continue;
2384                         }
2385
2386                         if (slave->delay <= 0) {
2387                                 slave->new_link = BOND_LINK_DOWN;
2388                                 commit++;
2389                                 continue;
2390                         }
2391
2392                         slave->delay--;
2393                         break;
2394
2395                 case BOND_LINK_DOWN:
2396                         if (!link_state)
2397                                 continue;
2398
2399                         slave->link = BOND_LINK_BACK;
2400                         slave->delay = bond->params.updelay;
2401
2402                         if (slave->delay) {
2403                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2404                                         bond->dev->name, slave->dev->name,
2405                                         ignore_updelay ? 0 :
2406                                         bond->params.updelay *
2407                                         bond->params.miimon);
2408                         }
2409                         /*FALLTHRU*/
2410                 case BOND_LINK_BACK:
2411                         if (!link_state) {
2412                                 slave->link = BOND_LINK_DOWN;
2413                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2414                                         bond->dev->name,
2415                                         (bond->params.updelay - slave->delay) *
2416                                         bond->params.miimon,
2417                                         slave->dev->name);
2418
2419                                 continue;
2420                         }
2421
2422                         if (ignore_updelay)
2423                                 slave->delay = 0;
2424
2425                         if (slave->delay <= 0) {
2426                                 slave->new_link = BOND_LINK_UP;
2427                                 commit++;
2428                                 ignore_updelay = false;
2429                                 continue;
2430                         }
2431
2432                         slave->delay--;
2433                         break;
2434                 }
2435         }
2436
2437         return commit;
2438 }
2439
2440 static void bond_miimon_commit(struct bonding *bond)
2441 {
2442         struct slave *slave;
2443         int i;
2444
2445         bond_for_each_slave(bond, slave, i) {
2446                 switch (slave->new_link) {
2447                 case BOND_LINK_NOCHANGE:
2448                         continue;
2449
2450                 case BOND_LINK_UP:
2451                         slave->link = BOND_LINK_UP;
2452                         slave->jiffies = jiffies;
2453
2454                         if (bond->params.mode == BOND_MODE_8023AD) {
2455                                 /* prevent it from being the active one */
2456                                 bond_set_backup_slave(slave);
2457                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2458                                 /* make it immediately active */
2459                                 bond_set_active_slave(slave);
2460                         } else if (slave != bond->primary_slave) {
2461                                 /* prevent it from being the active one */
2462                                 bond_set_backup_slave(slave);
2463                         }
2464
2465                         bond_update_speed_duplex(slave);
2466
2467                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2468                                 bond->dev->name, slave->dev->name,
2469                                 slave->speed, slave->duplex ? "full" : "half");
2470
2471                         /* notify ad that the link status has changed */
2472                         if (bond->params.mode == BOND_MODE_8023AD)
2473                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2474
2475                         if (bond_is_lb(bond))
2476                                 bond_alb_handle_link_change(bond, slave,
2477                                                             BOND_LINK_UP);
2478
2479                         if (!bond->curr_active_slave ||
2480                             (slave == bond->primary_slave))
2481                                 goto do_failover;
2482
2483                         continue;
2484
2485                 case BOND_LINK_DOWN:
2486                         if (slave->link_failure_count < UINT_MAX)
2487                                 slave->link_failure_count++;
2488
2489                         slave->link = BOND_LINK_DOWN;
2490
2491                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2492                             bond->params.mode == BOND_MODE_8023AD)
2493                                 bond_set_slave_inactive_flags(slave);
2494
2495                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2496                                 bond->dev->name, slave->dev->name);
2497
2498                         if (bond->params.mode == BOND_MODE_8023AD)
2499                                 bond_3ad_handle_link_change(slave,
2500                                                             BOND_LINK_DOWN);
2501
2502                         if (bond_is_lb(bond))
2503                                 bond_alb_handle_link_change(bond, slave,
2504                                                             BOND_LINK_DOWN);
2505
2506                         if (slave == bond->curr_active_slave)
2507                                 goto do_failover;
2508
2509                         continue;
2510
2511                 default:
2512                         pr_err("%s: invalid new link %d on slave %s\n",
2513                                bond->dev->name, slave->new_link,
2514                                slave->dev->name);
2515                         slave->new_link = BOND_LINK_NOCHANGE;
2516
2517                         continue;
2518                 }
2519
2520 do_failover:
2521                 ASSERT_RTNL();
2522                 block_netpoll_tx();
2523                 write_lock_bh(&bond->curr_slave_lock);
2524                 bond_select_active_slave(bond);
2525                 write_unlock_bh(&bond->curr_slave_lock);
2526                 unblock_netpoll_tx();
2527         }
2528
2529         bond_set_carrier(bond);
2530 }
2531
2532 /*
2533  * bond_mii_monitor
2534  *
2535  * Really a wrapper that splits the mii monitor into two phases: an
2536  * inspection, then (if inspection indicates something needs to be done)
2537  * an acquisition of appropriate locks followed by a commit phase to
2538  * implement whatever link state changes are indicated.
2539  */
2540 void bond_mii_monitor(struct work_struct *work)
2541 {
2542         struct bonding *bond = container_of(work, struct bonding,
2543                                             mii_work.work);
2544         bool should_notify_peers = false;
2545         unsigned long delay;
2546
2547         read_lock(&bond->lock);
2548
2549         delay = msecs_to_jiffies(bond->params.miimon);
2550
2551         if (bond->slave_cnt == 0)
2552                 goto re_arm;
2553
2554         should_notify_peers = bond_should_notify_peers(bond);
2555
2556         if (bond_miimon_inspect(bond)) {
2557                 read_unlock(&bond->lock);
2558
2559                 /* Race avoidance with bond_close cancel of workqueue */
2560                 if (!rtnl_trylock()) {
2561                         read_lock(&bond->lock);
2562                         delay = 1;
2563                         should_notify_peers = false;
2564                         goto re_arm;
2565                 }
2566
2567                 read_lock(&bond->lock);
2568
2569                 bond_miimon_commit(bond);
2570
2571                 read_unlock(&bond->lock);
2572                 rtnl_unlock();  /* might sleep, hold no other locks */
2573                 read_lock(&bond->lock);
2574         }
2575
2576 re_arm:
2577         if (bond->params.miimon)
2578                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2579
2580         read_unlock(&bond->lock);
2581
2582         if (should_notify_peers) {
2583                 if (!rtnl_trylock()) {
2584                         read_lock(&bond->lock);
2585                         bond->send_peer_notif++;
2586                         read_unlock(&bond->lock);
2587                         return;
2588                 }
2589                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2590                 rtnl_unlock();
2591         }
2592 }
2593
2594 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2595 {
2596         struct vlan_entry *vlan;
2597         struct net_device *vlan_dev;
2598
2599         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2600                 return 1;
2601
2602         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2603                 rcu_read_lock();
2604                 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2605                 rcu_read_unlock();
2606                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2607                         return 1;
2608         }
2609
2610         return 0;
2611 }
2612
2613 /*
2614  * We go to the (large) trouble of VLAN tagging ARP frames because
2615  * switches in VLAN mode (especially if ports are configured as
2616  * "native" to a VLAN) might not pass non-tagged frames.
2617  */
2618 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2619 {
2620         struct sk_buff *skb;
2621
2622         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2623                  slave_dev->name, dest_ip, src_ip, vlan_id);
2624
2625         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2626                          NULL, slave_dev->dev_addr, NULL);
2627
2628         if (!skb) {
2629                 pr_err("ARP packet allocation failed\n");
2630                 return;
2631         }
2632         if (vlan_id) {
2633                 skb = vlan_put_tag(skb, vlan_id);
2634                 if (!skb) {
2635                         pr_err("failed to insert VLAN tag\n");
2636                         return;
2637                 }
2638         }
2639         arp_xmit(skb);
2640 }
2641
2642
2643 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2644 {
2645         int i, vlan_id;
2646         __be32 *targets = bond->params.arp_targets;
2647         struct vlan_entry *vlan;
2648         struct net_device *vlan_dev = NULL;
2649         struct rtable *rt;
2650
2651         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2652                 __be32 addr;
2653                 if (!targets[i])
2654                         break;
2655                 pr_debug("basa: target %x\n", targets[i]);
2656                 if (!bond_vlan_used(bond)) {
2657                         pr_debug("basa: empty vlan: arp_send\n");
2658                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2659                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2660                                       addr, 0);
2661                         continue;
2662                 }
2663
2664                 /*
2665                  * If VLANs are configured, we do a route lookup to
2666                  * determine which VLAN interface would be used, so we
2667                  * can tag the ARP with the proper VLAN tag.
2668                  */
2669                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2670                                      RTO_ONLINK, 0);
2671                 if (IS_ERR(rt)) {
2672                         if (net_ratelimit()) {
2673                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2674                                            bond->dev->name, &targets[i]);
2675                         }
2676                         continue;
2677                 }
2678
2679                 /*
2680                  * This target is not on a VLAN
2681                  */
2682                 if (rt->dst.dev == bond->dev) {
2683                         ip_rt_put(rt);
2684                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2685                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2686                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2687                                       addr, 0);
2688                         continue;
2689                 }
2690
2691                 vlan_id = 0;
2692                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2693                         rcu_read_lock();
2694                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2695                                                         vlan->vlan_id);
2696                         rcu_read_unlock();
2697                         if (vlan_dev == rt->dst.dev) {
2698                                 vlan_id = vlan->vlan_id;
2699                                 pr_debug("basa: vlan match on %s %d\n",
2700                                        vlan_dev->name, vlan_id);
2701                                 break;
2702                         }
2703                 }
2704
2705                 if (vlan_id && vlan_dev) {
2706                         ip_rt_put(rt);
2707                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2708                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2709                                       addr, vlan_id);
2710                         continue;
2711                 }
2712
2713                 if (net_ratelimit()) {
2714                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2715                                    bond->dev->name, &targets[i],
2716                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2717                 }
2718                 ip_rt_put(rt);
2719         }
2720 }
2721
2722 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2723 {
2724         int i;
2725         __be32 *targets = bond->params.arp_targets;
2726
2727         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2728                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2729                          &sip, &tip, i, &targets[i],
2730                          bond_has_this_ip(bond, tip));
2731                 if (sip == targets[i]) {
2732                         if (bond_has_this_ip(bond, tip))
2733                                 slave->last_arp_rx = jiffies;
2734                         return;
2735                 }
2736         }
2737 }
2738
2739 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2740                         struct slave *slave)
2741 {
2742         struct arphdr *arp = (struct arphdr *)skb->data;
2743         unsigned char *arp_ptr;
2744         __be32 sip, tip;
2745         int alen;
2746
2747         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2748                 return RX_HANDLER_ANOTHER;
2749
2750         read_lock(&bond->lock);
2751         alen = arp_hdr_len(bond->dev);
2752
2753         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2754                  bond->dev->name, skb->dev->name);
2755
2756         if (alen > skb_headlen(skb)) {
2757                 arp = kmalloc(alen, GFP_ATOMIC);
2758                 if (!arp)
2759                         goto out_unlock;
2760                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2761                         goto out_unlock;
2762         }
2763
2764         if (arp->ar_hln != bond->dev->addr_len ||
2765             skb->pkt_type == PACKET_OTHERHOST ||
2766             skb->pkt_type == PACKET_LOOPBACK ||
2767             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2768             arp->ar_pro != htons(ETH_P_IP) ||
2769             arp->ar_pln != 4)
2770                 goto out_unlock;
2771
2772         arp_ptr = (unsigned char *)(arp + 1);
2773         arp_ptr += bond->dev->addr_len;
2774         memcpy(&sip, arp_ptr, 4);
2775         arp_ptr += 4 + bond->dev->addr_len;
2776         memcpy(&tip, arp_ptr, 4);
2777
2778         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2779                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2780                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2781                  &sip, &tip);
2782
2783         /*
2784          * Backup slaves won't see the ARP reply, but do come through
2785          * here for each ARP probe (so we swap the sip/tip to validate
2786          * the probe).  In a "redundant switch, common router" type of
2787          * configuration, the ARP probe will (hopefully) travel from
2788          * the active, through one switch, the router, then the other
2789          * switch before reaching the backup.
2790          */
2791         if (bond_is_active_slave(slave))
2792                 bond_validate_arp(bond, slave, sip, tip);
2793         else
2794                 bond_validate_arp(bond, slave, tip, sip);
2795
2796 out_unlock:
2797         read_unlock(&bond->lock);
2798         if (arp != (struct arphdr *)skb->data)
2799                 kfree(arp);
2800         return RX_HANDLER_ANOTHER;
2801 }
2802
2803 /*
2804  * this function is called regularly to monitor each slave's link
2805  * ensuring that traffic is being sent and received when arp monitoring
2806  * is used in load-balancing mode. if the adapter has been dormant, then an
2807  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2808  * arp monitoring in active backup mode.
2809  */
2810 void bond_loadbalance_arp_mon(struct work_struct *work)
2811 {
2812         struct bonding *bond = container_of(work, struct bonding,
2813                                             arp_work.work);
2814         struct slave *slave, *oldcurrent;
2815         int do_failover = 0;
2816         int delta_in_ticks;
2817         int i;
2818
2819         read_lock(&bond->lock);
2820
2821         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2822
2823         if (bond->slave_cnt == 0)
2824                 goto re_arm;
2825
2826         read_lock(&bond->curr_slave_lock);
2827         oldcurrent = bond->curr_active_slave;
2828         read_unlock(&bond->curr_slave_lock);
2829
2830         /* see if any of the previous devices are up now (i.e. they have
2831          * xmt and rcv traffic). the curr_active_slave does not come into
2832          * the picture unless it is null. also, slave->jiffies is not needed
2833          * here because we send an arp on each slave and give a slave as
2834          * long as it needs to get the tx/rx within the delta.
2835          * TODO: what about up/down delay in arp mode? it wasn't here before
2836          *       so it can wait
2837          */
2838         bond_for_each_slave(bond, slave, i) {
2839                 unsigned long trans_start = dev_trans_start(slave->dev);
2840
2841                 if (slave->link != BOND_LINK_UP) {
2842                         if (time_in_range(jiffies,
2843                                 trans_start - delta_in_ticks,
2844                                 trans_start + delta_in_ticks) &&
2845                             time_in_range(jiffies,
2846                                 slave->dev->last_rx - delta_in_ticks,
2847                                 slave->dev->last_rx + delta_in_ticks)) {
2848
2849                                 slave->link  = BOND_LINK_UP;
2850                                 bond_set_active_slave(slave);
2851
2852                                 /* primary_slave has no meaning in round-robin
2853                                  * mode. the window of a slave being up and
2854                                  * curr_active_slave being null after enslaving
2855                                  * is closed.
2856                                  */
2857                                 if (!oldcurrent) {
2858                                         pr_info("%s: link status definitely up for interface %s, ",
2859                                                 bond->dev->name,
2860                                                 slave->dev->name);
2861                                         do_failover = 1;
2862                                 } else {
2863                                         pr_info("%s: interface %s is now up\n",
2864                                                 bond->dev->name,
2865                                                 slave->dev->name);
2866                                 }
2867                         }
2868                 } else {
2869                         /* slave->link == BOND_LINK_UP */
2870
2871                         /* not all switches will respond to an arp request
2872                          * when the source ip is 0, so don't take the link down
2873                          * if we don't know our ip yet
2874                          */
2875                         if (!time_in_range(jiffies,
2876                                 trans_start - delta_in_ticks,
2877                                 trans_start + 2 * delta_in_ticks) ||
2878                             !time_in_range(jiffies,
2879                                 slave->dev->last_rx - delta_in_ticks,
2880                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2881
2882                                 slave->link  = BOND_LINK_DOWN;
2883                                 bond_set_backup_slave(slave);
2884
2885                                 if (slave->link_failure_count < UINT_MAX)
2886                                         slave->link_failure_count++;
2887
2888                                 pr_info("%s: interface %s is now down.\n",
2889                                         bond->dev->name,
2890                                         slave->dev->name);
2891
2892                                 if (slave == oldcurrent)
2893                                         do_failover = 1;
2894                         }
2895                 }
2896
2897                 /* note: if switch is in round-robin mode, all links
2898                  * must tx arp to ensure all links rx an arp - otherwise
2899                  * links may oscillate or not come up at all; if switch is
2900                  * in something like xor mode, there is nothing we can
2901                  * do - all replies will be rx'ed on same link causing slaves
2902                  * to be unstable during low/no traffic periods
2903                  */
2904                 if (IS_UP(slave->dev))
2905                         bond_arp_send_all(bond, slave);
2906         }
2907
2908         if (do_failover) {
2909                 block_netpoll_tx();
2910                 write_lock_bh(&bond->curr_slave_lock);
2911
2912                 bond_select_active_slave(bond);
2913
2914                 write_unlock_bh(&bond->curr_slave_lock);
2915                 unblock_netpoll_tx();
2916         }
2917
2918 re_arm:
2919         if (bond->params.arp_interval)
2920                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2921
2922         read_unlock(&bond->lock);
2923 }
2924
2925 /*
2926  * Called to inspect slaves for active-backup mode ARP monitor link state
2927  * changes.  Sets new_link in slaves to specify what action should take
2928  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2929  * to link states must be committed.
2930  *
2931  * Called with bond->lock held for read.
2932  */
2933 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2934 {
2935         struct slave *slave;
2936         int i, commit = 0;
2937         unsigned long trans_start;
2938
2939         bond_for_each_slave(bond, slave, i) {
2940                 slave->new_link = BOND_LINK_NOCHANGE;
2941
2942                 if (slave->link != BOND_LINK_UP) {
2943                         if (time_in_range(jiffies,
2944                                 slave_last_rx(bond, slave) - delta_in_ticks,
2945                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2946
2947                                 slave->new_link = BOND_LINK_UP;
2948                                 commit++;
2949                         }
2950
2951                         continue;
2952                 }
2953
2954                 /*
2955                  * Give slaves 2*delta after being enslaved or made
2956                  * active.  This avoids bouncing, as the last receive
2957                  * times need a full ARP monitor cycle to be updated.
2958                  */
2959                 if (time_in_range(jiffies,
2960                                   slave->jiffies - delta_in_ticks,
2961                                   slave->jiffies + 2 * delta_in_ticks))
2962                         continue;
2963
2964                 /*
2965                  * Backup slave is down if:
2966                  * - No current_arp_slave AND
2967                  * - more than 3*delta since last receive AND
2968                  * - the bond has an IP address
2969                  *
2970                  * Note: a non-null current_arp_slave indicates
2971                  * the curr_active_slave went down and we are
2972                  * searching for a new one; under this condition
2973                  * we only take the curr_active_slave down - this
2974                  * gives each slave a chance to tx/rx traffic
2975                  * before being taken out
2976                  */
2977                 if (!bond_is_active_slave(slave) &&
2978                     !bond->current_arp_slave &&
2979                     !time_in_range(jiffies,
2980                         slave_last_rx(bond, slave) - delta_in_ticks,
2981                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2982
2983                         slave->new_link = BOND_LINK_DOWN;
2984                         commit++;
2985                 }
2986
2987                 /*
2988                  * Active slave is down if:
2989                  * - more than 2*delta since transmitting OR
2990                  * - (more than 2*delta since receive AND
2991                  *    the bond has an IP address)
2992                  */
2993                 trans_start = dev_trans_start(slave->dev);
2994                 if (bond_is_active_slave(slave) &&
2995                     (!time_in_range(jiffies,
2996                         trans_start - delta_in_ticks,
2997                         trans_start + 2 * delta_in_ticks) ||
2998                      !time_in_range(jiffies,
2999                         slave_last_rx(bond, slave) - delta_in_ticks,
3000                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3001
3002                         slave->new_link = BOND_LINK_DOWN;
3003                         commit++;
3004                 }
3005         }
3006
3007         return commit;
3008 }
3009
3010 /*
3011  * Called to commit link state changes noted by inspection step of
3012  * active-backup mode ARP monitor.
3013  *
3014  * Called with RTNL and bond->lock for read.
3015  */
3016 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3017 {
3018         struct slave *slave;
3019         int i;
3020         unsigned long trans_start;
3021
3022         bond_for_each_slave(bond, slave, i) {
3023                 switch (slave->new_link) {
3024                 case BOND_LINK_NOCHANGE:
3025                         continue;
3026
3027                 case BOND_LINK_UP:
3028                         trans_start = dev_trans_start(slave->dev);
3029                         if ((!bond->curr_active_slave &&
3030                              time_in_range(jiffies,
3031                                            trans_start - delta_in_ticks,
3032                                            trans_start + delta_in_ticks)) ||
3033                             bond->curr_active_slave != slave) {
3034                                 slave->link = BOND_LINK_UP;
3035                                 if (bond->current_arp_slave) {
3036                                         bond_set_slave_inactive_flags(
3037                                                 bond->current_arp_slave);
3038                                         bond->current_arp_slave = NULL;
3039                                 }
3040
3041                                 pr_info("%s: link status definitely up for interface %s.\n",
3042                                         bond->dev->name, slave->dev->name);
3043
3044                                 if (!bond->curr_active_slave ||
3045                                     (slave == bond->primary_slave))
3046                                         goto do_failover;
3047
3048                         }
3049
3050                         continue;
3051
3052                 case BOND_LINK_DOWN:
3053                         if (slave->link_failure_count < UINT_MAX)
3054                                 slave->link_failure_count++;
3055
3056                         slave->link = BOND_LINK_DOWN;
3057                         bond_set_slave_inactive_flags(slave);
3058
3059                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3060                                 bond->dev->name, slave->dev->name);
3061
3062                         if (slave == bond->curr_active_slave) {
3063                                 bond->current_arp_slave = NULL;
3064                                 goto do_failover;
3065                         }
3066
3067                         continue;
3068
3069                 default:
3070                         pr_err("%s: impossible: new_link %d on slave %s\n",
3071                                bond->dev->name, slave->new_link,
3072                                slave->dev->name);
3073                         continue;
3074                 }
3075
3076 do_failover:
3077                 ASSERT_RTNL();
3078                 block_netpoll_tx();
3079                 write_lock_bh(&bond->curr_slave_lock);
3080                 bond_select_active_slave(bond);
3081                 write_unlock_bh(&bond->curr_slave_lock);
3082                 unblock_netpoll_tx();
3083         }
3084
3085         bond_set_carrier(bond);
3086 }
3087
3088 /*
3089  * Send ARP probes for active-backup mode ARP monitor.
3090  *
3091  * Called with bond->lock held for read.
3092  */
3093 static void bond_ab_arp_probe(struct bonding *bond)
3094 {
3095         struct slave *slave;
3096         int i;
3097
3098         read_lock(&bond->curr_slave_lock);
3099
3100         if (bond->current_arp_slave && bond->curr_active_slave)
3101                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3102                         bond->current_arp_slave->dev->name,
3103                         bond->curr_active_slave->dev->name);
3104
3105         if (bond->curr_active_slave) {
3106                 bond_arp_send_all(bond, bond->curr_active_slave);
3107                 read_unlock(&bond->curr_slave_lock);
3108                 return;
3109         }
3110
3111         read_unlock(&bond->curr_slave_lock);
3112
3113         /* if we don't have a curr_active_slave, search for the next available
3114          * backup slave from the current_arp_slave and make it the candidate
3115          * for becoming the curr_active_slave
3116          */
3117
3118         if (!bond->current_arp_slave) {
3119                 bond->current_arp_slave = bond->first_slave;
3120                 if (!bond->current_arp_slave)
3121                         return;
3122         }
3123
3124         bond_set_slave_inactive_flags(bond->current_arp_slave);
3125
3126         /* search for next candidate */
3127         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3128                 if (IS_UP(slave->dev)) {
3129                         slave->link = BOND_LINK_BACK;
3130                         bond_set_slave_active_flags(slave);
3131                         bond_arp_send_all(bond, slave);
3132                         slave->jiffies = jiffies;
3133                         bond->current_arp_slave = slave;
3134                         break;
3135                 }
3136
3137                 /* if the link state is up at this point, we
3138                  * mark it down - this can happen if we have
3139                  * simultaneous link failures and
3140                  * reselect_active_interface doesn't make this
3141                  * one the current slave so it is still marked
3142                  * up when it is actually down
3143                  */
3144                 if (slave->link == BOND_LINK_UP) {
3145                         slave->link = BOND_LINK_DOWN;
3146                         if (slave->link_failure_count < UINT_MAX)
3147                                 slave->link_failure_count++;
3148
3149                         bond_set_slave_inactive_flags(slave);
3150
3151                         pr_info("%s: backup interface %s is now down.\n",
3152                                 bond->dev->name, slave->dev->name);
3153                 }
3154         }
3155 }
3156
3157 void bond_activebackup_arp_mon(struct work_struct *work)
3158 {
3159         struct bonding *bond = container_of(work, struct bonding,
3160                                             arp_work.work);
3161         bool should_notify_peers = false;
3162         int delta_in_ticks;
3163
3164         read_lock(&bond->lock);
3165
3166         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3167
3168         if (bond->slave_cnt == 0)
3169                 goto re_arm;
3170
3171         should_notify_peers = bond_should_notify_peers(bond);
3172
3173         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3174                 read_unlock(&bond->lock);
3175
3176                 /* Race avoidance with bond_close flush of workqueue */
3177                 if (!rtnl_trylock()) {
3178                         read_lock(&bond->lock);
3179                         delta_in_ticks = 1;
3180                         should_notify_peers = false;
3181                         goto re_arm;
3182                 }
3183
3184                 read_lock(&bond->lock);
3185
3186                 bond_ab_arp_commit(bond, delta_in_ticks);
3187
3188                 read_unlock(&bond->lock);
3189                 rtnl_unlock();
3190                 read_lock(&bond->lock);
3191         }
3192
3193         bond_ab_arp_probe(bond);
3194
3195 re_arm:
3196         if (bond->params.arp_interval)
3197                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3198
3199         read_unlock(&bond->lock);
3200
3201         if (should_notify_peers) {
3202                 if (!rtnl_trylock()) {
3203                         read_lock(&bond->lock);
3204                         bond->send_peer_notif++;
3205                         read_unlock(&bond->lock);
3206                         return;
3207                 }
3208                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3209                 rtnl_unlock();
3210         }
3211 }
3212
3213 /*-------------------------- netdev event handling --------------------------*/
3214
3215 /*
3216  * Change device name
3217  */
3218 static int bond_event_changename(struct bonding *bond)
3219 {
3220         bond_remove_proc_entry(bond);
3221         bond_create_proc_entry(bond);
3222
3223         bond_debug_reregister(bond);
3224
3225         return NOTIFY_DONE;
3226 }
3227
3228 static int bond_master_netdev_event(unsigned long event,
3229                                     struct net_device *bond_dev)
3230 {
3231         struct bonding *event_bond = netdev_priv(bond_dev);
3232
3233         switch (event) {
3234         case NETDEV_CHANGENAME:
3235                 return bond_event_changename(event_bond);
3236         case NETDEV_UNREGISTER:
3237                 bond_remove_proc_entry(event_bond);
3238                 break;
3239         case NETDEV_REGISTER:
3240                 bond_create_proc_entry(event_bond);
3241                 break;
3242         default:
3243                 break;
3244         }
3245
3246         return NOTIFY_DONE;
3247 }
3248
3249 static int bond_slave_netdev_event(unsigned long event,
3250                                    struct net_device *slave_dev)
3251 {
3252         struct net_device *bond_dev = slave_dev->master;
3253         struct bonding *bond = netdev_priv(bond_dev);
3254         struct slave *slave = NULL;
3255
3256         switch (event) {
3257         case NETDEV_UNREGISTER:
3258                 if (bond_dev) {
3259                         if (bond->setup_by_slave)
3260                                 bond_release_and_destroy(bond_dev, slave_dev);
3261                         else
3262                                 bond_release(bond_dev, slave_dev);
3263                 }
3264                 break;
3265         case NETDEV_UP:
3266         case NETDEV_CHANGE:
3267                 slave = bond_get_slave_by_dev(bond, slave_dev);
3268                 if (slave) {
3269                         u32 old_speed = slave->speed;
3270                         u8  old_duplex = slave->duplex;
3271
3272                         bond_update_speed_duplex(slave);
3273
3274                         if (bond->params.mode == BOND_MODE_8023AD) {
3275                                 if (old_speed != slave->speed)
3276                                         bond_3ad_adapter_speed_changed(slave);
3277                                 if (old_duplex != slave->duplex)
3278                                         bond_3ad_adapter_duplex_changed(slave);
3279                         }
3280                 }
3281
3282                 break;
3283         case NETDEV_DOWN:
3284                 /*
3285                  * ... Or is it this?
3286                  */
3287                 break;
3288         case NETDEV_CHANGEMTU:
3289                 /*
3290                  * TODO: Should slaves be allowed to
3291                  * independently alter their MTU?  For
3292                  * an active-backup bond, slaves need
3293                  * not be the same type of device, so
3294                  * MTUs may vary.  For other modes,
3295                  * slaves arguably should have the
3296                  * same MTUs. To do this, we'd need to
3297                  * take over the slave's change_mtu
3298                  * function for the duration of their
3299                  * servitude.
3300                  */
3301                 break;
3302         case NETDEV_CHANGENAME:
3303                 /*
3304                  * TODO: handle changing the primary's name
3305                  */
3306                 break;
3307         case NETDEV_FEAT_CHANGE:
3308                 bond_compute_features(bond);
3309                 break;
3310         default:
3311                 break;
3312         }
3313
3314         return NOTIFY_DONE;
3315 }
3316
3317 /*
3318  * bond_netdev_event: handle netdev notifier chain events.
3319  *
3320  * This function receives events for the netdev chain.  The caller (an
3321  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3322  * locks for us to safely manipulate the slave devices (RTNL lock,
3323  * dev_probe_lock).
3324  */
3325 static int bond_netdev_event(struct notifier_block *this,
3326                              unsigned long event, void *ptr)
3327 {
3328         struct net_device *event_dev = (struct net_device *)ptr;
3329
3330         pr_debug("event_dev: %s, event: %lx\n",
3331                  event_dev ? event_dev->name : "None",
3332                  event);
3333
3334         if (!(event_dev->priv_flags & IFF_BONDING))
3335                 return NOTIFY_DONE;
3336
3337         if (event_dev->flags & IFF_MASTER) {
3338                 pr_debug("IFF_MASTER\n");
3339                 return bond_master_netdev_event(event, event_dev);
3340         }
3341
3342         if (event_dev->flags & IFF_SLAVE) {
3343                 pr_debug("IFF_SLAVE\n");
3344                 return bond_slave_netdev_event(event, event_dev);
3345         }
3346
3347         return NOTIFY_DONE;
3348 }
3349
3350 static struct notifier_block bond_netdev_notifier = {
3351         .notifier_call = bond_netdev_event,
3352 };
3353
3354 /*---------------------------- Hashing Policies -----------------------------*/
3355
3356 /*
3357  * Hash for the output device based upon layer 2 and layer 3 data. If
3358  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3359  */
3360 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3361 {
3362         struct ethhdr *data = (struct ethhdr *)skb->data;
3363         struct iphdr *iph = ip_hdr(skb);
3364
3365         if (skb->protocol == htons(ETH_P_IP)) {
3366                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3367                         (data->h_dest[5] ^ data->h_source[5])) % count;
3368         }
3369
3370         return (data->h_dest[5] ^ data->h_source[5]) % count;
3371 }
3372
3373 /*
3374  * Hash for the output device based upon layer 3 and layer 4 data. If
3375  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3376  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3377  */
3378 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3379 {
3380         struct ethhdr *data = (struct ethhdr *)skb->data;
3381         struct iphdr *iph = ip_hdr(skb);
3382         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3383         int layer4_xor = 0;
3384
3385         if (skb->protocol == htons(ETH_P_IP)) {
3386                 if (!ip_is_fragment(iph) &&
3387                     (iph->protocol == IPPROTO_TCP ||
3388                      iph->protocol == IPPROTO_UDP)) {
3389                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3390                 }
3391                 return (layer4_xor ^
3392                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3393
3394         }
3395
3396         return (data->h_dest[5] ^ data->h_source[5]) % count;
3397 }
3398
3399 /*
3400  * Hash for the output device based upon layer 2 data
3401  */
3402 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3403 {
3404         struct ethhdr *data = (struct ethhdr *)skb->data;
3405
3406         return (data->h_dest[5] ^ data->h_source[5]) % count;
3407 }
3408
3409 /*-------------------------- Device entry points ----------------------------*/
3410
3411 static int bond_open(struct net_device *bond_dev)
3412 {
3413         struct bonding *bond = netdev_priv(bond_dev);
3414         struct slave *slave;
3415         int i;
3416
3417         /* reset slave->backup and slave->inactive */
3418         read_lock(&bond->lock);
3419         if (bond->slave_cnt > 0) {
3420                 read_lock(&bond->curr_slave_lock);
3421                 bond_for_each_slave(bond, slave, i) {
3422                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3423                                 && (slave != bond->curr_active_slave)) {
3424                                 bond_set_slave_inactive_flags(slave);
3425                         } else {
3426                                 bond_set_slave_active_flags(slave);
3427                         }
3428                 }
3429                 read_unlock(&bond->curr_slave_lock);
3430         }
3431         read_unlock(&bond->lock);
3432
3433         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3434
3435         if (bond_is_lb(bond)) {
3436                 /* bond_alb_initialize must be called before the timer
3437                  * is started.
3438                  */
3439                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3440                         /* something went wrong - fail the open operation */
3441                         return -ENOMEM;
3442                 }
3443
3444                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3445                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3446         }
3447
3448         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3449                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3450                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3451         }
3452
3453         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3454                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3455                         INIT_DELAYED_WORK(&bond->arp_work,
3456                                           bond_activebackup_arp_mon);
3457                 else
3458                         INIT_DELAYED_WORK(&bond->arp_work,
3459                                           bond_loadbalance_arp_mon);
3460
3461                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3462                 if (bond->params.arp_validate)
3463                         bond->recv_probe = bond_arp_rcv;
3464         }
3465
3466         if (bond->params.mode == BOND_MODE_8023AD) {
3467                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3468                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3469                 /* register to receive LACPDUs */
3470                 bond->recv_probe = bond_3ad_lacpdu_recv;
3471                 bond_3ad_initiate_agg_selection(bond, 1);
3472         }
3473
3474         return 0;
3475 }
3476
3477 static int bond_close(struct net_device *bond_dev)
3478 {
3479         struct bonding *bond = netdev_priv(bond_dev);
3480
3481         write_lock_bh(&bond->lock);
3482
3483         bond->send_peer_notif = 0;
3484
3485         write_unlock_bh(&bond->lock);
3486
3487         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3488                 cancel_delayed_work_sync(&bond->mii_work);
3489         }
3490
3491         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3492                 cancel_delayed_work_sync(&bond->arp_work);
3493         }
3494
3495         switch (bond->params.mode) {
3496         case BOND_MODE_8023AD:
3497                 cancel_delayed_work_sync(&bond->ad_work);
3498                 break;
3499         case BOND_MODE_TLB:
3500         case BOND_MODE_ALB:
3501                 cancel_delayed_work_sync(&bond->alb_work);
3502                 break;
3503         default:
3504                 break;
3505         }
3506
3507         if (delayed_work_pending(&bond->mcast_work))
3508                 cancel_delayed_work_sync(&bond->mcast_work);
3509
3510         if (bond_is_lb(bond)) {
3511                 /* Must be called only after all
3512                  * slaves have been released
3513                  */
3514                 bond_alb_deinitialize(bond);
3515         }
3516         bond->recv_probe = NULL;
3517
3518         return 0;
3519 }
3520
3521 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3522                                                 struct rtnl_link_stats64 *stats)
3523 {
3524         struct bonding *bond = netdev_priv(bond_dev);
3525         struct rtnl_link_stats64 temp;
3526         struct slave *slave;
3527         int i;
3528
3529         memset(stats, 0, sizeof(*stats));
3530
3531         read_lock_bh(&bond->lock);
3532
3533         bond_for_each_slave(bond, slave, i) {
3534                 const struct rtnl_link_stats64 *sstats =
3535                         dev_get_stats(slave->dev, &temp);
3536
3537                 stats->rx_packets += sstats->rx_packets;
3538                 stats->rx_bytes += sstats->rx_bytes;
3539                 stats->rx_errors += sstats->rx_errors;
3540                 stats->rx_dropped += sstats->rx_dropped;
3541
3542                 stats->tx_packets += sstats->tx_packets;
3543                 stats->tx_bytes += sstats->tx_bytes;
3544                 stats->tx_errors += sstats->tx_errors;
3545                 stats->tx_dropped += sstats->tx_dropped;
3546
3547                 stats->multicast += sstats->multicast;
3548                 stats->collisions += sstats->collisions;
3549
3550                 stats->rx_length_errors += sstats->rx_length_errors;
3551                 stats->rx_over_errors += sstats->rx_over_errors;
3552                 stats->rx_crc_errors += sstats->rx_crc_errors;
3553                 stats->rx_frame_errors += sstats->rx_frame_errors;
3554                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3555                 stats->rx_missed_errors += sstats->rx_missed_errors;
3556
3557                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3558                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3559                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3560                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3561                 stats->tx_window_errors += sstats->tx_window_errors;
3562         }
3563
3564         read_unlock_bh(&bond->lock);
3565
3566         return stats;
3567 }
3568
3569 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3570 {
3571         struct net_device *slave_dev = NULL;
3572         struct ifbond k_binfo;
3573         struct ifbond __user *u_binfo = NULL;
3574         struct ifslave k_sinfo;
3575         struct ifslave __user *u_sinfo = NULL;
3576         struct mii_ioctl_data *mii = NULL;
3577         int res = 0;
3578
3579         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3580
3581         switch (cmd) {
3582         case SIOCGMIIPHY:
3583                 mii = if_mii(ifr);
3584                 if (!mii)
3585                         return -EINVAL;
3586
3587                 mii->phy_id = 0;
3588                 /* Fall Through */
3589         case SIOCGMIIREG:
3590                 /*
3591                  * We do this again just in case we were called by SIOCGMIIREG
3592                  * instead of SIOCGMIIPHY.
3593                  */
3594                 mii = if_mii(ifr);
3595                 if (!mii)
3596                         return -EINVAL;
3597
3598
3599                 if (mii->reg_num == 1) {
3600                         struct bonding *bond = netdev_priv(bond_dev);
3601                         mii->val_out = 0;
3602                         read_lock(&bond->lock);
3603                         read_lock(&bond->curr_slave_lock);
3604                         if (netif_carrier_ok(bond->dev))
3605                                 mii->val_out = BMSR_LSTATUS;
3606
3607                         read_unlock(&bond->curr_slave_lock);
3608                         read_unlock(&bond->lock);
3609                 }
3610
3611                 return 0;
3612         case BOND_INFO_QUERY_OLD:
3613         case SIOCBONDINFOQUERY:
3614                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3615
3616                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3617                         return -EFAULT;
3618
3619                 res = bond_info_query(bond_dev, &k_binfo);
3620                 if (res == 0 &&
3621                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3622                         return -EFAULT;
3623
3624                 return res;
3625         case BOND_SLAVE_INFO_QUERY_OLD:
3626         case SIOCBONDSLAVEINFOQUERY:
3627                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3628
3629                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3630                         return -EFAULT;
3631
3632                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3633                 if (res == 0 &&
3634                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3635                         return -EFAULT;
3636
3637                 return res;
3638         default:
3639                 /* Go on */
3640                 break;
3641         }
3642
3643         if (!capable(CAP_NET_ADMIN))
3644                 return -EPERM;
3645
3646         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3647
3648         pr_debug("slave_dev=%p:\n", slave_dev);
3649
3650         if (!slave_dev)
3651                 res = -ENODEV;
3652         else {
3653                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3654                 switch (cmd) {
3655                 case BOND_ENSLAVE_OLD:
3656                 case SIOCBONDENSLAVE:
3657                         res = bond_enslave(bond_dev, slave_dev);
3658                         break;
3659                 case BOND_RELEASE_OLD:
3660                 case SIOCBONDRELEASE:
3661                         res = bond_release(bond_dev, slave_dev);
3662                         break;
3663                 case BOND_SETHWADDR_OLD:
3664                 case SIOCBONDSETHWADDR:
3665                         res = bond_sethwaddr(bond_dev, slave_dev);
3666                         break;
3667                 case BOND_CHANGE_ACTIVE_OLD:
3668                 case SIOCBONDCHANGEACTIVE:
3669                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3670                         break;
3671                 default:
3672                         res = -EOPNOTSUPP;
3673                 }
3674
3675                 dev_put(slave_dev);
3676         }
3677
3678         return res;
3679 }
3680
3681 static bool bond_addr_in_mc_list(unsigned char *addr,
3682                                  struct netdev_hw_addr_list *list,
3683                                  int addrlen)
3684 {
3685         struct netdev_hw_addr *ha;
3686
3687         netdev_hw_addr_list_for_each(ha, list)
3688                 if (!memcmp(ha->addr, addr, addrlen))
3689                         return true;
3690
3691         return false;
3692 }
3693
3694 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3695 {
3696         struct bonding *bond = netdev_priv(bond_dev);
3697
3698         if (change & IFF_PROMISC)
3699                 bond_set_promiscuity(bond,
3700                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3701
3702         if (change & IFF_ALLMULTI)
3703                 bond_set_allmulti(bond,
3704                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3705 }
3706
3707 static void bond_set_multicast_list(struct net_device *bond_dev)
3708 {
3709         struct bonding *bond = netdev_priv(bond_dev);
3710         struct netdev_hw_addr *ha;
3711         bool found;
3712
3713         read_lock(&bond->lock);
3714
3715         /* looking for addresses to add to slaves' mc list */
3716         netdev_for_each_mc_addr(ha, bond_dev) {
3717                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3718                                              bond_dev->addr_len);
3719                 if (!found)
3720                         bond_mc_add(bond, ha->addr);
3721         }
3722
3723         /* looking for addresses to delete from slaves' list */
3724         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3725                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3726                                              bond_dev->addr_len);
3727                 if (!found)
3728                         bond_mc_del(bond, ha->addr);
3729         }
3730
3731         /* save master's multicast list */
3732         __hw_addr_flush(&bond->mc_list);
3733         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3734                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3735
3736         read_unlock(&bond->lock);
3737 }
3738
3739 static int bond_neigh_init(struct neighbour *n)
3740 {
3741         struct bonding *bond = netdev_priv(n->dev);
3742         struct slave *slave = bond->first_slave;
3743         const struct net_device_ops *slave_ops;
3744         struct neigh_parms parms;
3745         int ret;
3746
3747         if (!slave)
3748                 return 0;
3749
3750         slave_ops = slave->dev->netdev_ops;
3751
3752         if (!slave_ops->ndo_neigh_setup)
3753                 return 0;
3754
3755         parms.neigh_setup = NULL;
3756         parms.neigh_cleanup = NULL;
3757         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3758         if (ret)
3759                 return ret;
3760
3761         /*
3762          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3763          * after the last slave has been detached.  Assumes that all slaves
3764          * utilize the same neigh_cleanup (true at this writing as only user
3765          * is ipoib).
3766          */
3767         n->parms->neigh_cleanup = parms.neigh_cleanup;
3768
3769         if (!parms.neigh_setup)
3770                 return 0;
3771
3772         return parms.neigh_setup(n);
3773 }
3774
3775 /*
3776  * The bonding ndo_neigh_setup is called at init time beofre any
3777  * slave exists. So we must declare proxy setup function which will
3778  * be used at run time to resolve the actual slave neigh param setup.
3779  */
3780 static int bond_neigh_setup(struct net_device *dev,
3781                             struct neigh_parms *parms)
3782 {
3783         parms->neigh_setup   = bond_neigh_init;
3784
3785         return 0;
3786 }
3787
3788 /*
3789  * Change the MTU of all of a master's slaves to match the master
3790  */
3791 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3792 {
3793         struct bonding *bond = netdev_priv(bond_dev);
3794         struct slave *slave, *stop_at;
3795         int res = 0;
3796         int i;
3797
3798         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3799                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3800
3801         /* Can't hold bond->lock with bh disabled here since
3802          * some base drivers panic. On the other hand we can't
3803          * hold bond->lock without bh disabled because we'll
3804          * deadlock. The only solution is to rely on the fact
3805          * that we're under rtnl_lock here, and the slaves
3806          * list won't change. This doesn't solve the problem
3807          * of setting the slave's MTU while it is
3808          * transmitting, but the assumption is that the base
3809          * driver can handle that.
3810          *
3811          * TODO: figure out a way to safely iterate the slaves
3812          * list, but without holding a lock around the actual
3813          * call to the base driver.
3814          */
3815
3816         bond_for_each_slave(bond, slave, i) {
3817                 pr_debug("s %p s->p %p c_m %p\n",
3818                          slave,
3819                          slave->prev,
3820                          slave->dev->netdev_ops->ndo_change_mtu);
3821
3822                 res = dev_set_mtu(slave->dev, new_mtu);
3823
3824                 if (res) {
3825                         /* If we failed to set the slave's mtu to the new value
3826                          * we must abort the operation even in ACTIVE_BACKUP
3827                          * mode, because if we allow the backup slaves to have
3828                          * different mtu values than the active slave we'll
3829                          * need to change their mtu when doing a failover. That
3830                          * means changing their mtu from timer context, which
3831                          * is probably not a good idea.
3832                          */
3833                         pr_debug("err %d %s\n", res, slave->dev->name);
3834                         goto unwind;
3835                 }
3836         }
3837
3838         bond_dev->mtu = new_mtu;
3839
3840         return 0;
3841
3842 unwind:
3843         /* unwind from head to the slave that failed */
3844         stop_at = slave;
3845         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3846                 int tmp_res;
3847
3848                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3849                 if (tmp_res) {
3850                         pr_debug("unwind err %d dev %s\n",
3851                                  tmp_res, slave->dev->name);
3852                 }
3853         }
3854
3855         return res;
3856 }
3857
3858 /*
3859  * Change HW address
3860  *
3861  * Note that many devices must be down to change the HW address, and
3862  * downing the master releases all slaves.  We can make bonds full of
3863  * bonding devices to test this, however.
3864  */
3865 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3866 {
3867         struct bonding *bond = netdev_priv(bond_dev);
3868         struct sockaddr *sa = addr, tmp_sa;
3869         struct slave *slave, *stop_at;
3870         int res = 0;
3871         int i;
3872
3873         if (bond->params.mode == BOND_MODE_ALB)
3874                 return bond_alb_set_mac_address(bond_dev, addr);
3875
3876
3877         pr_debug("bond=%p, name=%s\n",
3878                  bond, bond_dev ? bond_dev->name : "None");
3879
3880         /*
3881          * If fail_over_mac is set to active, do nothing and return
3882          * success.  Returning an error causes ifenslave to fail.
3883          */
3884         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3885                 return 0;
3886
3887         if (!is_valid_ether_addr(sa->sa_data))
3888                 return -EADDRNOTAVAIL;
3889
3890         /* Can't hold bond->lock with bh disabled here since
3891          * some base drivers panic. On the other hand we can't
3892          * hold bond->lock without bh disabled because we'll
3893          * deadlock. The only solution is to rely on the fact
3894          * that we're under rtnl_lock here, and the slaves
3895          * list won't change. This doesn't solve the problem
3896          * of setting the slave's hw address while it is
3897          * transmitting, but the assumption is that the base
3898          * driver can handle that.
3899          *
3900          * TODO: figure out a way to safely iterate the slaves
3901          * list, but without holding a lock around the actual
3902          * call to the base driver.
3903          */
3904
3905         bond_for_each_slave(bond, slave, i) {
3906                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3907                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3908
3909                 if (slave_ops->ndo_set_mac_address == NULL) {
3910                         res = -EOPNOTSUPP;
3911                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3912                         goto unwind;
3913                 }
3914
3915                 res = dev_set_mac_address(slave->dev, addr);
3916                 if (res) {
3917                         /* TODO: consider downing the slave
3918                          * and retry ?
3919                          * User should expect communications
3920                          * breakage anyway until ARP finish
3921                          * updating, so...
3922                          */
3923                         pr_debug("err %d %s\n", res, slave->dev->name);
3924                         goto unwind;
3925                 }
3926         }
3927
3928         /* success */
3929         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3930         return 0;
3931
3932 unwind:
3933         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3934         tmp_sa.sa_family = bond_dev->type;
3935
3936         /* unwind from head to the slave that failed */
3937         stop_at = slave;
3938         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3939                 int tmp_res;
3940
3941                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3942                 if (tmp_res) {
3943                         pr_debug("unwind err %d dev %s\n",
3944                                  tmp_res, slave->dev->name);
3945                 }
3946         }
3947
3948         return res;
3949 }
3950
3951 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3952 {
3953         struct bonding *bond = netdev_priv(bond_dev);
3954         struct slave *slave, *start_at;
3955         int i, slave_no, res = 1;
3956         struct iphdr *iph = ip_hdr(skb);
3957
3958         /*
3959          * Start with the curr_active_slave that joined the bond as the
3960          * default for sending IGMP traffic.  For failover purposes one
3961          * needs to maintain some consistency for the interface that will
3962          * send the join/membership reports.  The curr_active_slave found
3963          * will send all of this type of traffic.
3964          */
3965         if ((iph->protocol == IPPROTO_IGMP) &&
3966             (skb->protocol == htons(ETH_P_IP))) {
3967
3968                 read_lock(&bond->curr_slave_lock);
3969                 slave = bond->curr_active_slave;
3970                 read_unlock(&bond->curr_slave_lock);
3971
3972                 if (!slave)
3973                         goto out;
3974         } else {
3975                 /*
3976                  * Concurrent TX may collide on rr_tx_counter; we accept
3977                  * that as being rare enough not to justify using an
3978                  * atomic op here.
3979                  */
3980                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3981
3982                 bond_for_each_slave(bond, slave, i) {
3983                         slave_no--;
3984                         if (slave_no < 0)
3985                                 break;
3986                 }
3987         }
3988
3989         start_at = slave;
3990         bond_for_each_slave_from(bond, slave, i, start_at) {
3991                 if (IS_UP(slave->dev) &&
3992                     (slave->link == BOND_LINK_UP) &&
3993                     bond_is_active_slave(slave)) {
3994                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3995                         break;
3996                 }
3997         }
3998
3999 out:
4000         if (res) {
4001                 /* no suitable interface, frame not sent */
4002                 kfree_skb(skb);
4003         }
4004
4005         return NETDEV_TX_OK;
4006 }
4007
4008
4009 /*
4010  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4011  * the bond has a usable interface.
4012  */
4013 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4014 {
4015         struct bonding *bond = netdev_priv(bond_dev);
4016         int res = 1;
4017
4018         read_lock(&bond->curr_slave_lock);
4019
4020         if (bond->curr_active_slave)
4021                 res = bond_dev_queue_xmit(bond, skb,
4022                         bond->curr_active_slave->dev);
4023
4024         read_unlock(&bond->curr_slave_lock);
4025
4026         if (res)
4027                 /* no suitable interface, frame not sent */
4028                 kfree_skb(skb);
4029
4030         return NETDEV_TX_OK;
4031 }
4032
4033 /*
4034  * In bond_xmit_xor() , we determine the output device by using a pre-
4035  * determined xmit_hash_policy(), If the selected device is not enabled,
4036  * find the next active slave.
4037  */
4038 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4039 {
4040         struct bonding *bond = netdev_priv(bond_dev);
4041         struct slave *slave, *start_at;
4042         int slave_no;
4043         int i;
4044         int res = 1;
4045
4046         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4047
4048         bond_for_each_slave(bond, slave, i) {
4049                 slave_no--;
4050                 if (slave_no < 0)
4051                         break;
4052         }
4053
4054         start_at = slave;
4055
4056         bond_for_each_slave_from(bond, slave, i, start_at) {
4057                 if (IS_UP(slave->dev) &&
4058                     (slave->link == BOND_LINK_UP) &&
4059                     bond_is_active_slave(slave)) {
4060                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4061                         break;
4062                 }
4063         }
4064
4065         if (res) {
4066                 /* no suitable interface, frame not sent */
4067                 kfree_skb(skb);
4068         }
4069
4070         return NETDEV_TX_OK;
4071 }
4072
4073 /*
4074  * in broadcast mode, we send everything to all usable interfaces.
4075  */
4076 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4077 {
4078         struct bonding *bond = netdev_priv(bond_dev);
4079         struct slave *slave, *start_at;
4080         struct net_device *tx_dev = NULL;
4081         int i;
4082         int res = 1;
4083
4084         read_lock(&bond->curr_slave_lock);
4085         start_at = bond->curr_active_slave;
4086         read_unlock(&bond->curr_slave_lock);
4087
4088         if (!start_at)
4089                 goto out;
4090
4091         bond_for_each_slave_from(bond, slave, i, start_at) {
4092                 if (IS_UP(slave->dev) &&
4093                     (slave->link == BOND_LINK_UP) &&
4094                     bond_is_active_slave(slave)) {
4095                         if (tx_dev) {
4096                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4097                                 if (!skb2) {
4098                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4099                                                bond_dev->name);
4100                                         continue;
4101                                 }
4102
4103                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4104                                 if (res) {
4105                                         kfree_skb(skb2);
4106                                         continue;
4107                                 }
4108                         }
4109                         tx_dev = slave->dev;
4110                 }
4111         }
4112
4113         if (tx_dev)
4114                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4115
4116 out:
4117         if (res)
4118                 /* no suitable interface, frame not sent */
4119                 kfree_skb(skb);
4120
4121         /* frame sent to all suitable interfaces */
4122         return NETDEV_TX_OK;
4123 }
4124
4125 /*------------------------- Device initialization ---------------------------*/
4126
4127 static void bond_set_xmit_hash_policy(struct bonding *bond)
4128 {
4129         switch (bond->params.xmit_policy) {
4130         case BOND_XMIT_POLICY_LAYER23:
4131                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4132                 break;
4133         case BOND_XMIT_POLICY_LAYER34:
4134                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4135                 break;
4136         case BOND_XMIT_POLICY_LAYER2:
4137         default:
4138                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4139                 break;
4140         }
4141 }
4142
4143 /*
4144  * Lookup the slave that corresponds to a qid
4145  */
4146 static inline int bond_slave_override(struct bonding *bond,
4147                                       struct sk_buff *skb)
4148 {
4149         int i, res = 1;
4150         struct slave *slave = NULL;
4151         struct slave *check_slave;
4152
4153         if (!skb->queue_mapping)
4154                 return 1;
4155
4156         /* Find out if any slaves have the same mapping as this skb. */
4157         bond_for_each_slave(bond, check_slave, i) {
4158                 if (check_slave->queue_id == skb->queue_mapping) {
4159                         slave = check_slave;
4160                         break;
4161                 }
4162         }
4163
4164         /* If the slave isn't UP, use default transmit policy. */
4165         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4166             (slave->link == BOND_LINK_UP)) {
4167                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4168         }
4169
4170         return res;
4171 }
4172
4173
4174 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4175 {
4176         /*
4177          * This helper function exists to help dev_pick_tx get the correct
4178          * destination queue.  Using a helper function skips a call to
4179          * skb_tx_hash and will put the skbs in the queue we expect on their
4180          * way down to the bonding driver.
4181          */
4182         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4183
4184         /*
4185          * Save the original txq to restore before passing to the driver
4186          */
4187         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4188
4189         if (unlikely(txq >= dev->real_num_tx_queues)) {
4190                 do {
4191                         txq -= dev->real_num_tx_queues;
4192                 } while (txq >= dev->real_num_tx_queues);
4193         }
4194         return txq;
4195 }
4196
4197 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4198 {
4199         struct bonding *bond = netdev_priv(dev);
4200
4201         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4202                 if (!bond_slave_override(bond, skb))
4203                         return NETDEV_TX_OK;
4204         }
4205
4206         switch (bond->params.mode) {
4207         case BOND_MODE_ROUNDROBIN:
4208                 return bond_xmit_roundrobin(skb, dev);
4209         case BOND_MODE_ACTIVEBACKUP:
4210                 return bond_xmit_activebackup(skb, dev);
4211         case BOND_MODE_XOR:
4212                 return bond_xmit_xor(skb, dev);
4213         case BOND_MODE_BROADCAST:
4214                 return bond_xmit_broadcast(skb, dev);
4215         case BOND_MODE_8023AD:
4216                 return bond_3ad_xmit_xor(skb, dev);
4217         case BOND_MODE_ALB:
4218         case BOND_MODE_TLB:
4219                 return bond_alb_xmit(skb, dev);
4220         default:
4221                 /* Should never happen, mode already checked */
4222                 pr_err("%s: Error: Unknown bonding mode %d\n",
4223                        dev->name, bond->params.mode);
4224                 WARN_ON_ONCE(1);
4225                 kfree_skb(skb);
4226                 return NETDEV_TX_OK;
4227         }
4228 }
4229
4230 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4231 {
4232         struct bonding *bond = netdev_priv(dev);
4233         netdev_tx_t ret = NETDEV_TX_OK;
4234
4235         /*
4236          * If we risk deadlock from transmitting this in the
4237          * netpoll path, tell netpoll to queue the frame for later tx
4238          */
4239         if (is_netpoll_tx_blocked(dev))
4240                 return NETDEV_TX_BUSY;
4241
4242         read_lock(&bond->lock);
4243
4244         if (bond->slave_cnt)
4245                 ret = __bond_start_xmit(skb, dev);
4246         else
4247                 kfree_skb(skb);
4248
4249         read_unlock(&bond->lock);
4250
4251         return ret;
4252 }
4253
4254 /*
4255  * set bond mode specific net device operations
4256  */
4257 void bond_set_mode_ops(struct bonding *bond, int mode)
4258 {
4259         struct net_device *bond_dev = bond->dev;
4260
4261         switch (mode) {
4262         case BOND_MODE_ROUNDROBIN:
4263                 break;
4264         case BOND_MODE_ACTIVEBACKUP:
4265                 break;
4266         case BOND_MODE_XOR:
4267                 bond_set_xmit_hash_policy(bond);
4268                 break;
4269         case BOND_MODE_BROADCAST:
4270                 break;
4271         case BOND_MODE_8023AD:
4272                 bond_set_xmit_hash_policy(bond);
4273                 break;
4274         case BOND_MODE_ALB:
4275                 /* FALLTHRU */
4276         case BOND_MODE_TLB:
4277                 break;
4278         default:
4279                 /* Should never happen, mode already checked */
4280                 pr_err("%s: Error: Unknown bonding mode %d\n",
4281                        bond_dev->name, mode);
4282                 break;
4283         }
4284 }
4285
4286 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4287                                     struct ethtool_drvinfo *drvinfo)
4288 {
4289         strncpy(drvinfo->driver, DRV_NAME, 32);
4290         strncpy(drvinfo->version, DRV_VERSION, 32);
4291         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4292 }
4293
4294 static const struct ethtool_ops bond_ethtool_ops = {
4295         .get_drvinfo            = bond_ethtool_get_drvinfo,
4296         .get_link               = ethtool_op_get_link,
4297 };
4298
4299 static const struct net_device_ops bond_netdev_ops = {
4300         .ndo_init               = bond_init,
4301         .ndo_uninit             = bond_uninit,
4302         .ndo_open               = bond_open,
4303         .ndo_stop               = bond_close,
4304         .ndo_start_xmit         = bond_start_xmit,
4305         .ndo_select_queue       = bond_select_queue,
4306         .ndo_get_stats64        = bond_get_stats,
4307         .ndo_do_ioctl           = bond_do_ioctl,
4308         .ndo_change_rx_flags    = bond_change_rx_flags,
4309         .ndo_set_rx_mode        = 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_HW_CSUM);
4385         bond_dev->features |= bond_dev->hw_features;
4386 }
4387
4388 static void bond_work_cancel_all(struct bonding *bond)
4389 {
4390         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4391                 cancel_delayed_work_sync(&bond->mii_work);
4392
4393         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4394                 cancel_delayed_work_sync(&bond->arp_work);
4395
4396         if (bond->params.mode == BOND_MODE_ALB &&
4397             delayed_work_pending(&bond->alb_work))
4398                 cancel_delayed_work_sync(&bond->alb_work);
4399
4400         if (bond->params.mode == BOND_MODE_8023AD &&
4401             delayed_work_pending(&bond->ad_work))
4402                 cancel_delayed_work_sync(&bond->ad_work);
4403
4404         if (delayed_work_pending(&bond->mcast_work))
4405                 cancel_delayed_work_sync(&bond->mcast_work);
4406 }
4407
4408 /*
4409 * Destroy a bonding device.
4410 * Must be under rtnl_lock when this function is called.
4411 */
4412 static void bond_uninit(struct net_device *bond_dev)
4413 {
4414         struct bonding *bond = netdev_priv(bond_dev);
4415         struct vlan_entry *vlan, *tmp;
4416
4417         bond_netpoll_cleanup(bond_dev);
4418
4419         /* Release the bonded slaves */
4420         bond_release_all(bond_dev);
4421
4422         list_del(&bond->bond_list);
4423
4424         bond_work_cancel_all(bond);
4425
4426         bond_debug_unregister(bond);
4427
4428         __hw_addr_flush(&bond->mc_list);
4429
4430         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4431                 list_del(&vlan->vlan_list);
4432                 kfree(vlan);
4433         }
4434 }
4435
4436 /*------------------------- Module initialization ---------------------------*/
4437
4438 /*
4439  * Convert string input module parms.  Accept either the
4440  * number of the mode or its string name.  A bit complicated because
4441  * some mode names are substrings of other names, and calls from sysfs
4442  * may have whitespace in the name (trailing newlines, for example).
4443  */
4444 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4445 {
4446         int modeint = -1, i, rv;
4447         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4448
4449         for (p = (char *)buf; *p; p++)
4450                 if (!(isdigit(*p) || isspace(*p)))
4451                         break;
4452
4453         if (*p)
4454                 rv = sscanf(buf, "%20s", modestr);
4455         else
4456                 rv = sscanf(buf, "%d", &modeint);
4457
4458         if (!rv)
4459                 return -1;
4460
4461         for (i = 0; tbl[i].modename; i++) {
4462                 if (modeint == tbl[i].mode)
4463                         return tbl[i].mode;
4464                 if (strcmp(modestr, tbl[i].modename) == 0)
4465                         return tbl[i].mode;
4466         }
4467
4468         return -1;
4469 }
4470
4471 static int bond_check_params(struct bond_params *params)
4472 {
4473         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4474
4475         /*
4476          * Convert string parameters.
4477          */
4478         if (mode) {
4479                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4480                 if (bond_mode == -1) {
4481                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4482                                mode == NULL ? "NULL" : mode);
4483                         return -EINVAL;
4484                 }
4485         }
4486
4487         if (xmit_hash_policy) {
4488                 if ((bond_mode != BOND_MODE_XOR) &&
4489                     (bond_mode != BOND_MODE_8023AD)) {
4490                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4491                                bond_mode_name(bond_mode));
4492                 } else {
4493                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4494                                                         xmit_hashtype_tbl);
4495                         if (xmit_hashtype == -1) {
4496                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4497                                        xmit_hash_policy == NULL ? "NULL" :
4498                                        xmit_hash_policy);
4499                                 return -EINVAL;
4500                         }
4501                 }
4502         }
4503
4504         if (lacp_rate) {
4505                 if (bond_mode != BOND_MODE_8023AD) {
4506                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4507                                 bond_mode_name(bond_mode));
4508                 } else {
4509                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4510                         if (lacp_fast == -1) {
4511                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4512                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4513                                 return -EINVAL;
4514                         }
4515                 }
4516         }
4517
4518         if (ad_select) {
4519                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4520                 if (params->ad_select == -1) {
4521                         pr_err("Error: Invalid ad_select \"%s\"\n",
4522                                ad_select == NULL ? "NULL" : ad_select);
4523                         return -EINVAL;
4524                 }
4525
4526                 if (bond_mode != BOND_MODE_8023AD) {
4527                         pr_warning("ad_select param only affects 802.3ad mode\n");
4528                 }
4529         } else {
4530                 params->ad_select = BOND_AD_STABLE;
4531         }
4532
4533         if (max_bonds < 0) {
4534                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4535                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4536                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4537         }
4538
4539         if (miimon < 0) {
4540                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4541                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4542                 miimon = BOND_LINK_MON_INTERV;
4543         }
4544
4545         if (updelay < 0) {
4546                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4547                            updelay, INT_MAX);
4548                 updelay = 0;
4549         }
4550
4551         if (downdelay < 0) {
4552                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4553                            downdelay, INT_MAX);
4554                 downdelay = 0;
4555         }
4556
4557         if ((use_carrier != 0) && (use_carrier != 1)) {
4558                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4559                            use_carrier);
4560                 use_carrier = 1;
4561         }
4562
4563         if (num_peer_notif < 0 || num_peer_notif > 255) {
4564                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4565                            num_peer_notif);
4566                 num_peer_notif = 1;
4567         }
4568
4569         /* reset values for 802.3ad */
4570         if (bond_mode == BOND_MODE_8023AD) {
4571                 if (!miimon) {
4572                         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");
4573                         pr_warning("Forcing miimon to 100msec\n");
4574                         miimon = 100;
4575                 }
4576         }
4577
4578         if (tx_queues < 1 || tx_queues > 255) {
4579                 pr_warning("Warning: tx_queues (%d) should be between "
4580                            "1 and 255, resetting to %d\n",
4581                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4582                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4583         }
4584
4585         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4586                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4587                            "not of valid value (0/1), so it was set to "
4588                            "0\n", all_slaves_active);
4589                 all_slaves_active = 0;
4590         }
4591
4592         if (resend_igmp < 0 || resend_igmp > 255) {
4593                 pr_warning("Warning: resend_igmp (%d) should be between "
4594                            "0 and 255, resetting to %d\n",
4595                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4596                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4597         }
4598
4599         /* reset values for TLB/ALB */
4600         if ((bond_mode == BOND_MODE_TLB) ||
4601             (bond_mode == BOND_MODE_ALB)) {
4602                 if (!miimon) {
4603                         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");
4604                         pr_warning("Forcing miimon to 100msec\n");
4605                         miimon = 100;
4606                 }
4607         }
4608
4609         if (bond_mode == BOND_MODE_ALB) {
4610                 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",
4611                           updelay);
4612         }
4613
4614         if (!miimon) {
4615                 if (updelay || downdelay) {
4616                         /* just warn the user the up/down delay will have
4617                          * no effect since miimon is zero...
4618                          */
4619                         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",
4620                                    updelay, downdelay);
4621                 }
4622         } else {
4623                 /* don't allow arp monitoring */
4624                 if (arp_interval) {
4625                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4626                                    miimon, arp_interval);
4627                         arp_interval = 0;
4628                 }
4629
4630                 if ((updelay % miimon) != 0) {
4631                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4632                                    updelay, miimon,
4633                                    (updelay / miimon) * miimon);
4634                 }
4635
4636                 updelay /= miimon;
4637
4638                 if ((downdelay % miimon) != 0) {
4639                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4640                                    downdelay, miimon,
4641                                    (downdelay / miimon) * miimon);
4642                 }
4643
4644                 downdelay /= miimon;
4645         }
4646
4647         if (arp_interval < 0) {
4648                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4649                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4650                 arp_interval = BOND_LINK_ARP_INTERV;
4651         }
4652
4653         for (arp_ip_count = 0;
4654              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4655              arp_ip_count++) {
4656                 /* not complete check, but should be good enough to
4657                    catch mistakes */
4658                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4659                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4660                                    arp_ip_target[arp_ip_count]);
4661                         arp_interval = 0;
4662                 } else {
4663                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4664                         arp_target[arp_ip_count] = ip;
4665                 }
4666         }
4667
4668         if (arp_interval && !arp_ip_count) {
4669                 /* don't allow arping if no arp_ip_target given... */
4670                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4671                            arp_interval);
4672                 arp_interval = 0;
4673         }
4674
4675         if (arp_validate) {
4676                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4677                         pr_err("arp_validate only supported in active-backup mode\n");
4678                         return -EINVAL;
4679                 }
4680                 if (!arp_interval) {
4681                         pr_err("arp_validate requires arp_interval\n");
4682                         return -EINVAL;
4683                 }
4684
4685                 arp_validate_value = bond_parse_parm(arp_validate,
4686                                                      arp_validate_tbl);
4687                 if (arp_validate_value == -1) {
4688                         pr_err("Error: invalid arp_validate \"%s\"\n",
4689                                arp_validate == NULL ? "NULL" : arp_validate);
4690                         return -EINVAL;
4691                 }
4692         } else
4693                 arp_validate_value = 0;
4694
4695         if (miimon) {
4696                 pr_info("MII link monitoring set to %d ms\n", miimon);
4697         } else if (arp_interval) {
4698                 int i;
4699
4700                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4701                         arp_interval,
4702                         arp_validate_tbl[arp_validate_value].modename,
4703                         arp_ip_count);
4704
4705                 for (i = 0; i < arp_ip_count; i++)
4706                         pr_info(" %s", arp_ip_target[i]);
4707
4708                 pr_info("\n");
4709
4710         } else if (max_bonds) {
4711                 /* miimon and arp_interval not set, we need one so things
4712                  * work as expected, see bonding.txt for details
4713                  */
4714                 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");
4715         }
4716
4717         if (primary && !USES_PRIMARY(bond_mode)) {
4718                 /* currently, using a primary only makes sense
4719                  * in active backup, TLB or ALB modes
4720                  */
4721                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4722                            primary, bond_mode_name(bond_mode));
4723                 primary = NULL;
4724         }
4725
4726         if (primary && primary_reselect) {
4727                 primary_reselect_value = bond_parse_parm(primary_reselect,
4728                                                          pri_reselect_tbl);
4729                 if (primary_reselect_value == -1) {
4730                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4731                                primary_reselect ==
4732                                         NULL ? "NULL" : primary_reselect);
4733                         return -EINVAL;
4734                 }
4735         } else {
4736                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4737         }
4738
4739         if (fail_over_mac) {
4740                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4741                                                       fail_over_mac_tbl);
4742                 if (fail_over_mac_value == -1) {
4743                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4744                                arp_validate == NULL ? "NULL" : arp_validate);
4745                         return -EINVAL;
4746                 }
4747
4748                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4749                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4750         } else {
4751                 fail_over_mac_value = BOND_FOM_NONE;
4752         }
4753
4754         /* fill params struct with the proper values */
4755         params->mode = bond_mode;
4756         params->xmit_policy = xmit_hashtype;
4757         params->miimon = miimon;
4758         params->num_peer_notif = num_peer_notif;
4759         params->arp_interval = arp_interval;
4760         params->arp_validate = arp_validate_value;
4761         params->updelay = updelay;
4762         params->downdelay = downdelay;
4763         params->use_carrier = use_carrier;
4764         params->lacp_fast = lacp_fast;
4765         params->primary[0] = 0;
4766         params->primary_reselect = primary_reselect_value;
4767         params->fail_over_mac = fail_over_mac_value;
4768         params->tx_queues = tx_queues;
4769         params->all_slaves_active = all_slaves_active;
4770         params->resend_igmp = resend_igmp;
4771         params->min_links = min_links;
4772
4773         if (primary) {
4774                 strncpy(params->primary, primary, IFNAMSIZ);
4775                 params->primary[IFNAMSIZ - 1] = 0;
4776         }
4777
4778         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4779
4780         return 0;
4781 }
4782
4783 static struct lock_class_key bonding_netdev_xmit_lock_key;
4784 static struct lock_class_key bonding_netdev_addr_lock_key;
4785
4786 static void bond_set_lockdep_class_one(struct net_device *dev,
4787                                        struct netdev_queue *txq,
4788                                        void *_unused)
4789 {
4790         lockdep_set_class(&txq->_xmit_lock,
4791                           &bonding_netdev_xmit_lock_key);
4792 }
4793
4794 static void bond_set_lockdep_class(struct net_device *dev)
4795 {
4796         lockdep_set_class(&dev->addr_list_lock,
4797                           &bonding_netdev_addr_lock_key);
4798         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4799 }
4800
4801 /*
4802  * Called from registration process
4803  */
4804 static int bond_init(struct net_device *bond_dev)
4805 {
4806         struct bonding *bond = netdev_priv(bond_dev);
4807         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4808         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4809
4810         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4811
4812         /*
4813          * Initialize locks that may be required during
4814          * en/deslave operations.  All of the bond_open work
4815          * (of which this is part) should really be moved to
4816          * a phase prior to dev_open
4817          */
4818         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4819         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4820
4821         bond->wq = create_singlethread_workqueue(bond_dev->name);
4822         if (!bond->wq)
4823                 return -ENOMEM;
4824
4825         bond_set_lockdep_class(bond_dev);
4826
4827         list_add_tail(&bond->bond_list, &bn->dev_list);
4828
4829         bond_prepare_sysfs_group(bond);
4830
4831         bond_debug_register(bond);
4832
4833         __hw_addr_init(&bond->mc_list);
4834         return 0;
4835 }
4836
4837 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4838 {
4839         if (tb[IFLA_ADDRESS]) {
4840                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4841                         return -EINVAL;
4842                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4843                         return -EADDRNOTAVAIL;
4844         }
4845         return 0;
4846 }
4847
4848 static unsigned int bond_get_num_tx_queues(void)
4849 {
4850         return tx_queues;
4851 }
4852
4853 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4854         .kind                   = "bond",
4855         .priv_size              = sizeof(struct bonding),
4856         .setup                  = bond_setup,
4857         .validate               = bond_validate,
4858         .get_num_tx_queues      = bond_get_num_tx_queues,
4859         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4860                                                              as for TX queues */
4861 };
4862
4863 /* Create a new bond based on the specified name and bonding parameters.
4864  * If name is NULL, obtain a suitable "bond%d" name for us.
4865  * Caller must NOT hold rtnl_lock; we need to release it here before we
4866  * set up our sysfs entries.
4867  */
4868 int bond_create(struct net *net, const char *name)
4869 {
4870         struct net_device *bond_dev;
4871         int res;
4872
4873         rtnl_lock();
4874
4875         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4876                                    name ? name : "bond%d",
4877                                    bond_setup, tx_queues);
4878         if (!bond_dev) {
4879                 pr_err("%s: eek! can't alloc netdev!\n", name);
4880                 rtnl_unlock();
4881                 return -ENOMEM;
4882         }
4883
4884         dev_net_set(bond_dev, net);
4885         bond_dev->rtnl_link_ops = &bond_link_ops;
4886
4887         res = register_netdevice(bond_dev);
4888
4889         netif_carrier_off(bond_dev);
4890
4891         rtnl_unlock();
4892         if (res < 0)
4893                 bond_destructor(bond_dev);
4894         return res;
4895 }
4896
4897 static int __net_init bond_net_init(struct net *net)
4898 {
4899         struct bond_net *bn = net_generic(net, bond_net_id);
4900
4901         bn->net = net;
4902         INIT_LIST_HEAD(&bn->dev_list);
4903
4904         bond_create_proc_dir(bn);
4905         bond_create_sysfs(bn);
4906
4907         return 0;
4908 }
4909
4910 static void __net_exit bond_net_exit(struct net *net)
4911 {
4912         struct bond_net *bn = net_generic(net, bond_net_id);
4913
4914         bond_destroy_sysfs(bn);
4915         bond_destroy_proc_dir(bn);
4916 }
4917
4918 static struct pernet_operations bond_net_ops = {
4919         .init = bond_net_init,
4920         .exit = bond_net_exit,
4921         .id   = &bond_net_id,
4922         .size = sizeof(struct bond_net),
4923 };
4924
4925 static int __init bonding_init(void)
4926 {
4927         int i;
4928         int res;
4929
4930         pr_info("%s", bond_version);
4931
4932         res = bond_check_params(&bonding_defaults);
4933         if (res)
4934                 goto out;
4935
4936         res = register_pernet_subsys(&bond_net_ops);
4937         if (res)
4938                 goto out;
4939
4940         res = rtnl_link_register(&bond_link_ops);
4941         if (res)
4942                 goto err_link;
4943
4944         bond_create_debugfs();
4945
4946         for (i = 0; i < max_bonds; i++) {
4947                 res = bond_create(&init_net, NULL);
4948                 if (res)
4949                         goto err;
4950         }
4951
4952         register_netdevice_notifier(&bond_netdev_notifier);
4953 out:
4954         return res;
4955 err:
4956         rtnl_link_unregister(&bond_link_ops);
4957 err_link:
4958         unregister_pernet_subsys(&bond_net_ops);
4959         goto out;
4960
4961 }
4962
4963 static void __exit bonding_exit(void)
4964 {
4965         unregister_netdevice_notifier(&bond_netdev_notifier);
4966
4967         bond_destroy_debugfs();
4968
4969         rtnl_link_unregister(&bond_link_ops);
4970         unregister_pernet_subsys(&bond_net_ops);
4971
4972 #ifdef CONFIG_NET_POLL_CONTROLLER
4973         /*
4974          * Make sure we don't have an imbalance on our netpoll blocking
4975          */
4976         WARN_ON(atomic_read(&netpoll_block_tx));
4977 #endif
4978 }
4979
4980 module_init(bonding_init);
4981 module_exit(bonding_exit);
4982 MODULE_LICENSE("GPL");
4983 MODULE_VERSION(DRV_VERSION);
4984 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4985 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4986 MODULE_ALIAS_RTNL_LINK("bond");