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