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.
 553  */
 554 static void 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;
 567
 568         slave_speed = ethtool_cmd_speed(&ecmd);
 569         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 570                 return;
 571
 572         switch (ecmd.duplex) {
 573         case DUPLEX_FULL:
 574         case DUPLEX_HALF:
 575                 break;
 576         default:
 577                 return;
 578         }
 579
 580         slave->speed = slave_speed;
 581         slave->duplex = ecmd.duplex;
 582
 583         return;
 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         int (*recv_probe)(struct sk_buff *, struct bonding *,
1448                                 struct slave *);
1449         int ret = RX_HANDLER_ANOTHER;
1450
1451         skb = skb_share_check(skb, GFP_ATOMIC);
1452         if (unlikely(!skb))
1453                 return RX_HANDLER_CONSUMED;
1454
1455         *pskb = skb;
1456
1457         slave = bond_slave_get_rcu(skb->dev);
1458         bond = slave->bond;
1459
1460         if (bond->params.arp_interval)
1461                 slave->dev->last_rx = jiffies;
1462
1463         recv_probe = ACCESS_ONCE(bond->recv_probe);
1464         if (recv_probe) {
1465                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1466
1467                 if (likely(nskb)) {
1468                         ret = recv_probe(nskb, bond, slave);
1469                         dev_kfree_skb(nskb);
1470                         if (ret == RX_HANDLER_CONSUMED) {
1471                                 consume_skb(skb);
1472                                 return ret;
1473                         }
1474                 }
1475         }
1476
1477         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1478                 return RX_HANDLER_EXACT;
1479         }
1480
1481         skb->dev = bond->dev;
1482
1483         if (bond->params.mode == BOND_MODE_ALB &&
1484             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1485             skb->pkt_type == PACKET_HOST) {
1486
1487                 if (unlikely(skb_cow_head(skb,
1488                                           skb->data - skb_mac_header(skb)))) {
1489                         kfree_skb(skb);
1490                         return RX_HANDLER_CONSUMED;
1491                 }
1492                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1493         }
1494
1495         return ret;
1496 }
1497
1498 /* enslave device <slave> to bond device <master> */
1499 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1500 {
1501         struct bonding *bond = netdev_priv(bond_dev);
1502         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1503         struct slave *new_slave = NULL;
1504         struct netdev_hw_addr *ha;
1505         struct sockaddr addr;
1506         int link_reporting;
1507         int res = 0;
1508
1509         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1510                 slave_ops->ndo_do_ioctl == NULL) {
1511                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1512                            bond_dev->name, slave_dev->name);
1513         }
1514
1515         /* already enslaved */
1516         if (slave_dev->flags & IFF_SLAVE) {
1517                 pr_debug("Error, Device was already enslaved\n");
1518                 return -EBUSY;
1519         }
1520
1521         /* vlan challenged mutual exclusion */
1522         /* no need to lock since we're protected by rtnl_lock */
1523         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1524                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1525                 if (bond_vlan_used(bond)) {
1526                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1527                                bond_dev->name, slave_dev->name, bond_dev->name);
1528                         return -EPERM;
1529                 } else {
1530                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1531                                    bond_dev->name, slave_dev->name,
1532                                    slave_dev->name, bond_dev->name);
1533                 }
1534         } else {
1535                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1536         }
1537
1538         /*
1539          * Old ifenslave binaries are no longer supported.  These can
1540          * be identified with moderate accuracy by the state of the slave:
1541          * the current ifenslave will set the interface down prior to
1542          * enslaving it; the old ifenslave will not.
1543          */
1544         if ((slave_dev->flags & IFF_UP)) {
1545                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1546                        slave_dev->name);
1547                 res = -EPERM;
1548                 goto err_undo_flags;
1549         }
1550
1551         /* set bonding device ether type by slave - bonding netdevices are
1552          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1553          * there is a need to override some of the type dependent attribs/funcs.
1554          *
1555          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1556          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1557          */
1558         if (bond->slave_cnt == 0) {
1559                 if (bond_dev->type != slave_dev->type) {
1560                         pr_debug("%s: change device type from %d to %d\n",
1561                                  bond_dev->name,
1562                                  bond_dev->type, slave_dev->type);
1563
1564                         res = netdev_bonding_change(bond_dev,
1565                                                     NETDEV_PRE_TYPE_CHANGE);
1566                         res = notifier_to_errno(res);
1567                         if (res) {
1568                                 pr_err("%s: refused to change device type\n",
1569                                        bond_dev->name);
1570                                 res = -EBUSY;
1571                                 goto err_undo_flags;
1572                         }
1573
1574                         /* Flush unicast and multicast addresses */
1575                         dev_uc_flush(bond_dev);
1576                         dev_mc_flush(bond_dev);
1577
1578                         if (slave_dev->type != ARPHRD_ETHER)
1579                                 bond_setup_by_slave(bond_dev, slave_dev);
1580                         else {
1581                                 ether_setup(bond_dev);
1582                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1583                         }
1584
1585                         netdev_bonding_change(bond_dev,
1586                                               NETDEV_POST_TYPE_CHANGE);
1587                 }
1588         } else if (bond_dev->type != slave_dev->type) {
1589                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1590                        slave_dev->name,
1591                        slave_dev->type, bond_dev->type);
1592                 res = -EINVAL;
1593                 goto err_undo_flags;
1594         }
1595
1596         if (slave_ops->ndo_set_mac_address == NULL) {
1597                 if (bond->slave_cnt == 0) {
1598                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1599                                    bond_dev->name);
1600                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1601                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1602                         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",
1603                                bond_dev->name);
1604                         res = -EOPNOTSUPP;
1605                         goto err_undo_flags;
1606                 }
1607         }
1608
1609         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1610
1611         /* If this is the first slave, then we need to set the master's hardware
1612          * address to be the same as the slave's. */
1613         if (is_zero_ether_addr(bond->dev->dev_addr))
1614                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1615                        slave_dev->addr_len);
1616
1617
1618         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1619         if (!new_slave) {
1620                 res = -ENOMEM;
1621                 goto err_undo_flags;
1622         }
1623
1624         /*
1625          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1626          * is set via sysfs or module option if desired.
1627          */
1628         new_slave->queue_id = 0;
1629
1630         /* Save slave's original mtu and then set it to match the bond */
1631         new_slave->original_mtu = slave_dev->mtu;
1632         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1633         if (res) {
1634                 pr_debug("Error %d calling dev_set_mtu\n", res);
1635                 goto err_free;
1636         }
1637
1638         /*
1639          * Save slave's original ("permanent") mac address for modes
1640          * that need it, and for restoring it upon release, and then
1641          * set it to the master's address
1642          */
1643         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1644
1645         if (!bond->params.fail_over_mac) {
1646                 /*
1647                  * Set slave to master's mac address.  The application already
1648                  * set the master's mac address to that of the first slave
1649                  */
1650                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1651                 addr.sa_family = slave_dev->type;
1652                 res = dev_set_mac_address(slave_dev, &addr);
1653                 if (res) {
1654                         pr_debug("Error %d calling set_mac_address\n", res);
1655                         goto err_restore_mtu;
1656                 }
1657         }
1658
1659         res = netdev_set_bond_master(slave_dev, bond_dev);
1660         if (res) {
1661                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1662                 goto err_restore_mac;
1663         }
1664
1665         /* open the slave since the application closed it */
1666         res = dev_open(slave_dev);
1667         if (res) {
1668                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1669                 goto err_unset_master;
1670         }
1671
1672         new_slave->bond = bond;
1673         new_slave->dev = slave_dev;
1674         slave_dev->priv_flags |= IFF_BONDING;
1675
1676         if (bond_is_lb(bond)) {
1677                 /* bond_alb_init_slave() must be called before all other stages since
1678                  * it might fail and we do not want to have to undo everything
1679                  */
1680                 res = bond_alb_init_slave(bond, new_slave);
1681                 if (res)
1682                         goto err_close;
1683         }
1684
1685         /* If the mode USES_PRIMARY, then the new slave gets the
1686          * master's promisc (and mc) settings only if it becomes the
1687          * curr_active_slave, and that is taken care of later when calling
1688          * bond_change_active()
1689          */
1690         if (!USES_PRIMARY(bond->params.mode)) {
1691                 /* set promiscuity level to new slave */
1692                 if (bond_dev->flags & IFF_PROMISC) {
1693                         res = dev_set_promiscuity(slave_dev, 1);
1694                         if (res)
1695                                 goto err_close;
1696                 }
1697
1698                 /* set allmulti level to new slave */
1699                 if (bond_dev->flags & IFF_ALLMULTI) {
1700                         res = dev_set_allmulti(slave_dev, 1);
1701                         if (res)
1702                                 goto err_close;
1703                 }
1704
1705                 netif_addr_lock_bh(bond_dev);
1706                 /* upload master's mc_list to new slave */
1707                 netdev_for_each_mc_addr(ha, bond_dev)
1708                         dev_mc_add(slave_dev, ha->addr);
1709                 netif_addr_unlock_bh(bond_dev);
1710         }
1711
1712         if (bond->params.mode == BOND_MODE_8023AD) {
1713                 /* add lacpdu mc addr to mc list */
1714                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1715
1716                 dev_mc_add(slave_dev, lacpdu_multicast);
1717         }
1718
1719         bond_add_vlans_on_slave(bond, slave_dev);
1720
1721         write_lock_bh(&bond->lock);
1722
1723         bond_attach_slave(bond, new_slave);
1724
1725         new_slave->delay = 0;
1726         new_slave->link_failure_count = 0;
1727
1728         write_unlock_bh(&bond->lock);
1729
1730         bond_compute_features(bond);
1731
1732         read_lock(&bond->lock);
1733
1734         new_slave->last_arp_rx = jiffies -
1735                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1736
1737         if (bond->params.miimon && !bond->params.use_carrier) {
1738                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1739
1740                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1741                         /*
1742                          * miimon is set but a bonded network driver
1743                          * does not support ETHTOOL/MII and
1744                          * arp_interval is not set.  Note: if
1745                          * use_carrier is enabled, we will never go
1746                          * here (because netif_carrier is always
1747                          * supported); thus, we don't need to change
1748                          * the messages for netif_carrier.
1749                          */
1750                         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",
1751                                bond_dev->name, slave_dev->name);
1752                 } else if (link_reporting == -1) {
1753                         /* unable get link status using mii/ethtool */
1754                         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",
1755                                    bond_dev->name, slave_dev->name);
1756                 }
1757         }
1758
1759         /* check for initial state */
1760         if (bond->params.miimon) {
1761                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1762                         if (bond->params.updelay) {
1763                                 new_slave->link = BOND_LINK_BACK;
1764                                 new_slave->delay = bond->params.updelay;
1765                         } else {
1766                                 new_slave->link = BOND_LINK_UP;
1767                         }
1768                 } else {
1769                         new_slave->link = BOND_LINK_DOWN;
1770                 }
1771         } else if (bond->params.arp_interval) {
1772                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1773                         BOND_LINK_UP : BOND_LINK_DOWN);
1774         } else {
1775                 new_slave->link = BOND_LINK_UP;
1776         }
1777
1778         if (new_slave->link != BOND_LINK_DOWN)
1779                 new_slave->jiffies = jiffies;
1780         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1781                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1782                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1783
1784         bond_update_speed_duplex(new_slave);
1785
1786         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1787                 /* if there is a primary slave, remember it */
1788                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1789                         bond->primary_slave = new_slave;
1790                         bond->force_primary = true;
1791                 }
1792         }
1793
1794         write_lock_bh(&bond->curr_slave_lock);
1795
1796         switch (bond->params.mode) {
1797         case BOND_MODE_ACTIVEBACKUP:
1798                 bond_set_slave_inactive_flags(new_slave);
1799                 bond_select_active_slave(bond);
1800                 break;
1801         case BOND_MODE_8023AD:
1802                 /* in 802.3ad mode, the internal mechanism
1803                  * will activate the slaves in the selected
1804                  * aggregator
1805                  */
1806                 bond_set_slave_inactive_flags(new_slave);
1807                 /* if this is the first slave */
1808                 if (bond->slave_cnt == 1) {
1809                         SLAVE_AD_INFO(new_slave).id = 1;
1810                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1811                          * can be called only after the mac address of the bond is set
1812                          */
1813                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1814                 } else {
1815                         SLAVE_AD_INFO(new_slave).id =
1816                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1817                 }
1818
1819                 bond_3ad_bind_slave(new_slave);
1820                 break;
1821         case BOND_MODE_TLB:
1822         case BOND_MODE_ALB:
1823                 bond_set_active_slave(new_slave);
1824                 bond_set_slave_inactive_flags(new_slave);
1825                 bond_select_active_slave(bond);
1826                 break;
1827         default:
1828                 pr_debug("This slave is always active in trunk mode\n");
1829
1830                 /* always active in trunk mode */
1831                 bond_set_active_slave(new_slave);
1832
1833                 /* In trunking mode there is little meaning to curr_active_slave
1834                  * anyway (it holds no special properties of the bond device),
1835                  * so we can change it without calling change_active_interface()
1836                  */
1837                 if (!bond->curr_active_slave)
1838                         bond->curr_active_slave = new_slave;
1839
1840                 break;
1841         } /* switch(bond_mode) */
1842
1843         write_unlock_bh(&bond->curr_slave_lock);
1844
1845         bond_set_carrier(bond);
1846
1847 #ifdef CONFIG_NET_POLL_CONTROLLER
1848         slave_dev->npinfo = bond_netpoll_info(bond);
1849         if (slave_dev->npinfo) {
1850                 if (slave_enable_netpoll(new_slave)) {
1851                         read_unlock(&bond->lock);
1852                         pr_info("Error, %s: master_dev is using netpoll, "
1853                                  "but new slave device does not support netpoll.\n",
1854                                  bond_dev->name);
1855                         res = -EBUSY;
1856                         goto err_detach;
1857                 }
1858         }
1859 #endif
1860
1861         read_unlock(&bond->lock);
1862
1863         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1864         if (res)
1865                 goto err_detach;
1866
1867         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1868                                          new_slave);
1869         if (res) {
1870                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1871                 goto err_dest_symlinks;
1872         }
1873
1874         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1875                 bond_dev->name, slave_dev->name,
1876                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1877                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1878
1879         /* enslave is successful */
1880         return 0;
1881
1882 /* Undo stages on error */
1883 err_dest_symlinks:
1884         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1885
1886 err_detach:
1887         write_lock_bh(&bond->lock);
1888         bond_detach_slave(bond, new_slave);
1889         write_unlock_bh(&bond->lock);
1890
1891 err_close:
1892         dev_close(slave_dev);
1893
1894 err_unset_master:
1895         netdev_set_bond_master(slave_dev, NULL);
1896
1897 err_restore_mac:
1898         if (!bond->params.fail_over_mac) {
1899                 /* XXX TODO - fom follow mode needs to change master's
1900                  * MAC if this slave's MAC is in use by the bond, or at
1901                  * least print a warning.
1902                  */
1903                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1904                 addr.sa_family = slave_dev->type;
1905                 dev_set_mac_address(slave_dev, &addr);
1906         }
1907
1908 err_restore_mtu:
1909         dev_set_mtu(slave_dev, new_slave->original_mtu);
1910
1911 err_free:
1912         kfree(new_slave);
1913
1914 err_undo_flags:
1915         bond_compute_features(bond);
1916
1917         return res;
1918 }
1919
1920 /*
1921  * Try to release the slave device <slave> from the bond device <master>
1922  * It is legal to access curr_active_slave without a lock because all the function
1923  * is write-locked.
1924  *
1925  * The rules for slave state should be:
1926  *   for Active/Backup:
1927  *     Active stays on all backups go down
1928  *   for Bonded connections:
1929  *     The first up interface should be left on and all others downed.
1930  */
1931 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1932 {
1933         struct bonding *bond = netdev_priv(bond_dev);
1934         struct slave *slave, *oldcurrent;
1935         struct sockaddr addr;
1936         netdev_features_t old_features = bond_dev->features;
1937
1938         /* slave is not a slave or master is not master of this slave */
1939         if (!(slave_dev->flags & IFF_SLAVE) ||
1940             (slave_dev->master != bond_dev)) {
1941                 pr_err("%s: Error: cannot release %s.\n",
1942                        bond_dev->name, slave_dev->name);
1943                 return -EINVAL;
1944         }
1945
1946         block_netpoll_tx();
1947         netdev_bonding_change(bond_dev, NETDEV_RELEASE);
1948         write_lock_bh(&bond->lock);
1949
1950         slave = bond_get_slave_by_dev(bond, slave_dev);
1951         if (!slave) {
1952                 /* not a slave of this bond */
1953                 pr_info("%s: %s not enslaved\n",
1954                         bond_dev->name, slave_dev->name);
1955                 write_unlock_bh(&bond->lock);
1956                 unblock_netpoll_tx();
1957                 return -EINVAL;
1958         }
1959
1960         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1961          * for this slave anymore.
1962          */
1963         netdev_rx_handler_unregister(slave_dev);
1964         write_unlock_bh(&bond->lock);
1965         synchronize_net();
1966         write_lock_bh(&bond->lock);
1967
1968         if (!bond->params.fail_over_mac) {
1969                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1970                     bond->slave_cnt > 1)
1971                         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",
1972                                    bond_dev->name, slave_dev->name,
1973                                    slave->perm_hwaddr,
1974                                    bond_dev->name, slave_dev->name);
1975         }
1976
1977         /* Inform AD package of unbinding of slave. */
1978         if (bond->params.mode == BOND_MODE_8023AD) {
1979                 /* must be called before the slave is
1980                  * detached from the list
1981                  */
1982                 bond_3ad_unbind_slave(slave);
1983         }
1984
1985         pr_info("%s: releasing %s interface %s\n",
1986                 bond_dev->name,
1987                 bond_is_active_slave(slave) ? "active" : "backup",
1988                 slave_dev->name);
1989
1990         oldcurrent = bond->curr_active_slave;
1991
1992         bond->current_arp_slave = NULL;
1993
1994         /* release the slave from its bond */
1995         bond_detach_slave(bond, slave);
1996
1997         if (bond->primary_slave == slave)
1998                 bond->primary_slave = NULL;
1999
2000         if (oldcurrent == slave)
2001                 bond_change_active_slave(bond, NULL);
2002
2003         if (bond_is_lb(bond)) {
2004                 /* Must be called only after the slave has been
2005                  * detached from the list and the curr_active_slave
2006                  * has been cleared (if our_slave == old_current),
2007                  * but before a new active slave is selected.
2008                  */
2009                 write_unlock_bh(&bond->lock);
2010                 bond_alb_deinit_slave(bond, slave);
2011                 write_lock_bh(&bond->lock);
2012         }
2013
2014         if (oldcurrent == slave) {
2015                 /*
2016                  * Note that we hold RTNL over this sequence, so there
2017                  * is no concern that another slave add/remove event
2018                  * will interfere.
2019                  */
2020                 write_unlock_bh(&bond->lock);
2021                 read_lock(&bond->lock);
2022                 write_lock_bh(&bond->curr_slave_lock);
2023
2024                 bond_select_active_slave(bond);
2025
2026                 write_unlock_bh(&bond->curr_slave_lock);
2027                 read_unlock(&bond->lock);
2028                 write_lock_bh(&bond->lock);
2029         }
2030
2031         if (bond->slave_cnt == 0) {
2032                 bond_set_carrier(bond);
2033
2034                 /* if the last slave was removed, zero the mac address
2035                  * of the master so it will be set by the application
2036                  * to the mac address of the first slave
2037                  */
2038                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2039
2040                 if (bond_vlan_used(bond)) {
2041                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2042                                    bond_dev->name, bond_dev->name);
2043                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2044                                    bond_dev->name);
2045                 }
2046         }
2047
2048         write_unlock_bh(&bond->lock);
2049         unblock_netpoll_tx();
2050
2051         if (bond->slave_cnt == 0)
2052                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2053
2054         bond_compute_features(bond);
2055         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2056             (old_features & NETIF_F_VLAN_CHALLENGED))
2057                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2058                         bond_dev->name, slave_dev->name, bond_dev->name);
2059
2060         /* must do this from outside any spinlocks */
2061         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2062
2063         bond_del_vlans_from_slave(bond, slave_dev);
2064
2065         /* If the mode USES_PRIMARY, then we should only remove its
2066          * promisc and mc settings if it was the curr_active_slave, but that was
2067          * already taken care of above when we detached the slave
2068          */
2069         if (!USES_PRIMARY(bond->params.mode)) {
2070                 /* unset promiscuity level from slave */
2071                 if (bond_dev->flags & IFF_PROMISC)
2072                         dev_set_promiscuity(slave_dev, -1);
2073
2074                 /* unset allmulti level from slave */
2075                 if (bond_dev->flags & IFF_ALLMULTI)
2076                         dev_set_allmulti(slave_dev, -1);
2077
2078                 /* flush master's mc_list from slave */
2079                 netif_addr_lock_bh(bond_dev);
2080                 bond_mc_list_flush(bond_dev, slave_dev);
2081                 netif_addr_unlock_bh(bond_dev);
2082         }
2083
2084         netdev_set_bond_master(slave_dev, NULL);
2085
2086         slave_disable_netpoll(slave);
2087
2088         /* close slave before restoring its mac address */
2089         dev_close(slave_dev);
2090
2091         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2092                 /* restore original ("permanent") mac address */
2093                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2094                 addr.sa_family = slave_dev->type;
2095                 dev_set_mac_address(slave_dev, &addr);
2096         }
2097
2098         dev_set_mtu(slave_dev, slave->original_mtu);
2099
2100         slave_dev->priv_flags &= ~IFF_BONDING;
2101
2102         kfree(slave);
2103
2104         return 0;  /* deletion OK */
2105 }
2106
2107 /*
2108 * First release a slave and then destroy the bond if no more slaves are left.
2109 * Must be under rtnl_lock when this function is called.
2110 */
2111 static int  bond_release_and_destroy(struct net_device *bond_dev,
2112                                      struct net_device *slave_dev)
2113 {
2114         struct bonding *bond = netdev_priv(bond_dev);
2115         int ret;
2116
2117         ret = bond_release(bond_dev, slave_dev);
2118         if ((ret == 0) && (bond->slave_cnt == 0)) {
2119                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2120                 pr_info("%s: destroying bond %s.\n",
2121                         bond_dev->name, bond_dev->name);
2122                 unregister_netdevice(bond_dev);
2123         }
2124         return ret;
2125 }
2126
2127 /*
2128  * This function releases all slaves.
2129  */
2130 static int bond_release_all(struct net_device *bond_dev)
2131 {
2132         struct bonding *bond = netdev_priv(bond_dev);
2133         struct slave *slave;
2134         struct net_device *slave_dev;
2135         struct sockaddr addr;
2136
2137         write_lock_bh(&bond->lock);
2138
2139         netif_carrier_off(bond_dev);
2140
2141         if (bond->slave_cnt == 0)
2142                 goto out;
2143
2144         bond->current_arp_slave = NULL;
2145         bond->primary_slave = NULL;
2146         bond_change_active_slave(bond, NULL);
2147
2148         while ((slave = bond->first_slave) != NULL) {
2149                 /* Inform AD package of unbinding of slave
2150                  * before slave is detached from the list.
2151                  */
2152                 if (bond->params.mode == BOND_MODE_8023AD)
2153                         bond_3ad_unbind_slave(slave);
2154
2155                 slave_dev = slave->dev;
2156                 bond_detach_slave(bond, slave);
2157
2158                 /* now that the slave is detached, unlock and perform
2159                  * all the undo steps that should not be called from
2160                  * within a lock.
2161                  */
2162                 write_unlock_bh(&bond->lock);
2163
2164                 /* unregister rx_handler early so bond_handle_frame wouldn't
2165                  * be called for this slave anymore.
2166                  */
2167                 netdev_rx_handler_unregister(slave_dev);
2168                 synchronize_net();
2169
2170                 if (bond_is_lb(bond)) {
2171                         /* must be called only after the slave
2172                          * has been detached from the list
2173                          */
2174                         bond_alb_deinit_slave(bond, slave);
2175                 }
2176
2177                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2178                 bond_del_vlans_from_slave(bond, slave_dev);
2179
2180                 /* If the mode USES_PRIMARY, then we should only remove its
2181                  * promisc and mc settings if it was the curr_active_slave, but that was
2182                  * already taken care of above when we detached the slave
2183                  */
2184                 if (!USES_PRIMARY(bond->params.mode)) {
2185                         /* unset promiscuity level from slave */
2186                         if (bond_dev->flags & IFF_PROMISC)
2187                                 dev_set_promiscuity(slave_dev, -1);
2188
2189                         /* unset allmulti level from slave */
2190                         if (bond_dev->flags & IFF_ALLMULTI)
2191                                 dev_set_allmulti(slave_dev, -1);
2192
2193                         /* flush master's mc_list from slave */
2194                         netif_addr_lock_bh(bond_dev);
2195                         bond_mc_list_flush(bond_dev, slave_dev);
2196                         netif_addr_unlock_bh(bond_dev);
2197                 }
2198
2199                 netdev_set_bond_master(slave_dev, NULL);
2200
2201                 slave_disable_netpoll(slave);
2202
2203                 /* close slave before restoring its mac address */
2204                 dev_close(slave_dev);
2205
2206                 if (!bond->params.fail_over_mac) {
2207                         /* restore original ("permanent") mac address*/
2208                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2209                         addr.sa_family = slave_dev->type;
2210                         dev_set_mac_address(slave_dev, &addr);
2211                 }
2212
2213                 kfree(slave);
2214
2215                 /* re-acquire the lock before getting the next slave */
2216                 write_lock_bh(&bond->lock);
2217         }
2218
2219         /* zero the mac address of the master so it will be
2220          * set by the application to the mac address of the
2221          * first slave
2222          */
2223         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2224
2225         if (bond_vlan_used(bond)) {
2226                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2227                            bond_dev->name, bond_dev->name);
2228                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2229                            bond_dev->name);
2230         }
2231
2232         pr_info("%s: released all slaves\n", bond_dev->name);
2233
2234 out:
2235         write_unlock_bh(&bond->lock);
2236
2237         bond_compute_features(bond);
2238
2239         return 0;
2240 }
2241
2242 /*
2243  * This function changes the active slave to slave <slave_dev>.
2244  * It returns -EINVAL in the following cases.
2245  *  - <slave_dev> is not found in the list.
2246  *  - There is not active slave now.
2247  *  - <slave_dev> is already active.
2248  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2249  *  - <slave_dev> is not running.
2250  * In these cases, this function does nothing.
2251  * In the other cases, current_slave pointer is changed and 0 is returned.
2252  */
2253 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2254 {
2255         struct bonding *bond = netdev_priv(bond_dev);
2256         struct slave *old_active = NULL;
2257         struct slave *new_active = NULL;
2258         int res = 0;
2259
2260         if (!USES_PRIMARY(bond->params.mode))
2261                 return -EINVAL;
2262
2263         /* Verify that master_dev is indeed the master of slave_dev */
2264         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2265                 return -EINVAL;
2266
2267         read_lock(&bond->lock);
2268
2269         read_lock(&bond->curr_slave_lock);
2270         old_active = bond->curr_active_slave;
2271         read_unlock(&bond->curr_slave_lock);
2272
2273         new_active = bond_get_slave_by_dev(bond, slave_dev);
2274
2275         /*
2276          * Changing to the current active: do nothing; return success.
2277          */
2278         if (new_active && (new_active == old_active)) {
2279                 read_unlock(&bond->lock);
2280                 return 0;
2281         }
2282
2283         if ((new_active) &&
2284             (old_active) &&
2285             (new_active->link == BOND_LINK_UP) &&
2286             IS_UP(new_active->dev)) {
2287                 block_netpoll_tx();
2288                 write_lock_bh(&bond->curr_slave_lock);
2289                 bond_change_active_slave(bond, new_active);
2290                 write_unlock_bh(&bond->curr_slave_lock);
2291                 unblock_netpoll_tx();
2292         } else
2293                 res = -EINVAL;
2294
2295         read_unlock(&bond->lock);
2296
2297         return res;
2298 }
2299
2300 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2301 {
2302         struct bonding *bond = netdev_priv(bond_dev);
2303
2304         info->bond_mode = bond->params.mode;
2305         info->miimon = bond->params.miimon;
2306
2307         read_lock(&bond->lock);
2308         info->num_slaves = bond->slave_cnt;
2309         read_unlock(&bond->lock);
2310
2311         return 0;
2312 }
2313
2314 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2315 {
2316         struct bonding *bond = netdev_priv(bond_dev);
2317         struct slave *slave;
2318         int i, res = -ENODEV;
2319
2320         read_lock(&bond->lock);
2321
2322         bond_for_each_slave(bond, slave, i) {
2323                 if (i == (int)info->slave_id) {
2324                         res = 0;
2325                         strcpy(info->slave_name, slave->dev->name);
2326                         info->link = slave->link;
2327                         info->state = bond_slave_state(slave);
2328                         info->link_failure_count = slave->link_failure_count;
2329                         break;
2330                 }
2331         }
2332
2333         read_unlock(&bond->lock);
2334
2335         return res;
2336 }
2337
2338 /*-------------------------------- Monitoring -------------------------------*/
2339
2340
2341 static int bond_miimon_inspect(struct bonding *bond)
2342 {
2343         struct slave *slave;
2344         int i, link_state, commit = 0;
2345         bool ignore_updelay;
2346
2347         ignore_updelay = !bond->curr_active_slave ? true : false;
2348
2349         bond_for_each_slave(bond, slave, i) {
2350                 slave->new_link = BOND_LINK_NOCHANGE;
2351
2352                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2353
2354                 switch (slave->link) {
2355                 case BOND_LINK_UP:
2356                         if (link_state)
2357                                 continue;
2358
2359                         slave->link = BOND_LINK_FAIL;
2360                         slave->delay = bond->params.downdelay;
2361                         if (slave->delay) {
2362                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2363                                         bond->dev->name,
2364                                         (bond->params.mode ==
2365                                          BOND_MODE_ACTIVEBACKUP) ?
2366                                         (bond_is_active_slave(slave) ?
2367                                          "active " : "backup ") : "",
2368                                         slave->dev->name,
2369                                         bond->params.downdelay * bond->params.miimon);
2370                         }
2371                         /*FALLTHRU*/
2372                 case BOND_LINK_FAIL:
2373                         if (link_state) {
2374                                 /*
2375                                  * recovered before downdelay expired
2376                                  */
2377                                 slave->link = BOND_LINK_UP;
2378                                 slave->jiffies = jiffies;
2379                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2380                                         bond->dev->name,
2381                                         (bond->params.downdelay - slave->delay) *
2382                                         bond->params.miimon,
2383                                         slave->dev->name);
2384                                 continue;
2385                         }
2386
2387                         if (slave->delay <= 0) {
2388                                 slave->new_link = BOND_LINK_DOWN;
2389                                 commit++;
2390                                 continue;
2391                         }
2392
2393                         slave->delay--;
2394                         break;
2395
2396                 case BOND_LINK_DOWN:
2397                         if (!link_state)
2398                                 continue;
2399
2400                         slave->link = BOND_LINK_BACK;
2401                         slave->delay = bond->params.updelay;
2402
2403                         if (slave->delay) {
2404                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2405                                         bond->dev->name, slave->dev->name,
2406                                         ignore_updelay ? 0 :
2407                                         bond->params.updelay *
2408                                         bond->params.miimon);
2409                         }
2410                         /*FALLTHRU*/
2411                 case BOND_LINK_BACK:
2412                         if (!link_state) {
2413                                 slave->link = BOND_LINK_DOWN;
2414                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2415                                         bond->dev->name,
2416                                         (bond->params.updelay - slave->delay) *
2417                                         bond->params.miimon,
2418                                         slave->dev->name);
2419
2420                                 continue;
2421                         }
2422
2423                         if (ignore_updelay)
2424                                 slave->delay = 0;
2425
2426                         if (slave->delay <= 0) {
2427                                 slave->new_link = BOND_LINK_UP;
2428                                 commit++;
2429                                 ignore_updelay = false;
2430                                 continue;
2431                         }
2432
2433                         slave->delay--;
2434                         break;
2435                 }
2436         }
2437
2438         return commit;
2439 }
2440
2441 static void bond_miimon_commit(struct bonding *bond)
2442 {
2443         struct slave *slave;
2444         int i;
2445
2446         bond_for_each_slave(bond, slave, i) {
2447                 switch (slave->new_link) {
2448                 case BOND_LINK_NOCHANGE:
2449                         continue;
2450
2451                 case BOND_LINK_UP:
2452                         slave->link = BOND_LINK_UP;
2453                         slave->jiffies = jiffies;
2454
2455                         if (bond->params.mode == BOND_MODE_8023AD) {
2456                                 /* prevent it from being the active one */
2457                                 bond_set_backup_slave(slave);
2458                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2459                                 /* make it immediately active */
2460                                 bond_set_active_slave(slave);
2461                         } else if (slave != bond->primary_slave) {
2462                                 /* prevent it from being the active one */
2463                                 bond_set_backup_slave(slave);
2464                         }
2465
2466                         bond_update_speed_duplex(slave);
2467
2468                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2469                                 bond->dev->name, slave->dev->name,
2470                                 slave->speed, slave->duplex ? "full" : "half");
2471
2472                         /* notify ad that the link status has changed */
2473                         if (bond->params.mode == BOND_MODE_8023AD)
2474                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2475
2476                         if (bond_is_lb(bond))
2477                                 bond_alb_handle_link_change(bond, slave,
2478                                                             BOND_LINK_UP);
2479
2480                         if (!bond->curr_active_slave ||
2481                             (slave == bond->primary_slave))
2482                                 goto do_failover;
2483
2484                         continue;
2485
2486                 case BOND_LINK_DOWN:
2487                         if (slave->link_failure_count < UINT_MAX)
2488                                 slave->link_failure_count++;
2489
2490                         slave->link = BOND_LINK_DOWN;
2491
2492                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2493                             bond->params.mode == BOND_MODE_8023AD)
2494                                 bond_set_slave_inactive_flags(slave);
2495
2496                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2497                                 bond->dev->name, slave->dev->name);
2498
2499                         if (bond->params.mode == BOND_MODE_8023AD)
2500                                 bond_3ad_handle_link_change(slave,
2501                                                             BOND_LINK_DOWN);
2502
2503                         if (bond_is_lb(bond))
2504                                 bond_alb_handle_link_change(bond, slave,
2505                                                             BOND_LINK_DOWN);
2506
2507                         if (slave == bond->curr_active_slave)
2508                                 goto do_failover;
2509
2510                         continue;
2511
2512                 default:
2513                         pr_err("%s: invalid new link %d on slave %s\n",
2514                                bond->dev->name, slave->new_link,
2515                                slave->dev->name);
2516                         slave->new_link = BOND_LINK_NOCHANGE;
2517
2518                         continue;
2519                 }
2520
2521 do_failover:
2522                 ASSERT_RTNL();
2523                 block_netpoll_tx();
2524                 write_lock_bh(&bond->curr_slave_lock);
2525                 bond_select_active_slave(bond);
2526                 write_unlock_bh(&bond->curr_slave_lock);
2527                 unblock_netpoll_tx();
2528         }
2529
2530         bond_set_carrier(bond);
2531 }
2532
2533 /*
2534  * bond_mii_monitor
2535  *
2536  * Really a wrapper that splits the mii monitor into two phases: an
2537  * inspection, then (if inspection indicates something needs to be done)
2538  * an acquisition of appropriate locks followed by a commit phase to
2539  * implement whatever link state changes are indicated.
2540  */
2541 void bond_mii_monitor(struct work_struct *work)
2542 {
2543         struct bonding *bond = container_of(work, struct bonding,
2544                                             mii_work.work);
2545         bool should_notify_peers = false;
2546         unsigned long delay;
2547
2548         read_lock(&bond->lock);
2549
2550         delay = msecs_to_jiffies(bond->params.miimon);
2551
2552         if (bond->slave_cnt == 0)
2553                 goto re_arm;
2554
2555         should_notify_peers = bond_should_notify_peers(bond);
2556
2557         if (bond_miimon_inspect(bond)) {
2558                 read_unlock(&bond->lock);
2559
2560                 /* Race avoidance with bond_close cancel of workqueue */
2561                 if (!rtnl_trylock()) {
2562                         read_lock(&bond->lock);
2563                         delay = 1;
2564                         should_notify_peers = false;
2565                         goto re_arm;
2566                 }
2567
2568                 read_lock(&bond->lock);
2569
2570                 bond_miimon_commit(bond);
2571
2572                 read_unlock(&bond->lock);
2573                 rtnl_unlock();  /* might sleep, hold no other locks */
2574                 read_lock(&bond->lock);
2575         }
2576
2577 re_arm:
2578         if (bond->params.miimon)
2579                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2580
2581         read_unlock(&bond->lock);
2582
2583         if (should_notify_peers) {
2584                 if (!rtnl_trylock()) {
2585                         read_lock(&bond->lock);
2586                         bond->send_peer_notif++;
2587                         read_unlock(&bond->lock);
2588                         return;
2589                 }
2590                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
2591                 rtnl_unlock();
2592         }
2593 }
2594
2595 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2596 {
2597         struct vlan_entry *vlan;
2598         struct net_device *vlan_dev;
2599
2600         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2601                 return 1;
2602
2603         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2604                 rcu_read_lock();
2605                 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2606                 rcu_read_unlock();
2607                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2608                         return 1;
2609         }
2610
2611         return 0;
2612 }
2613
2614 /*
2615  * We go to the (large) trouble of VLAN tagging ARP frames because
2616  * switches in VLAN mode (especially if ports are configured as
2617  * "native" to a VLAN) might not pass non-tagged frames.
2618  */
2619 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2620 {
2621         struct sk_buff *skb;
2622
2623         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2624                  slave_dev->name, dest_ip, src_ip, vlan_id);
2625
2626         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2627                          NULL, slave_dev->dev_addr, NULL);
2628
2629         if (!skb) {
2630                 pr_err("ARP packet allocation failed\n");
2631                 return;
2632         }
2633         if (vlan_id) {
2634                 skb = vlan_put_tag(skb, vlan_id);
2635                 if (!skb) {
2636                         pr_err("failed to insert VLAN tag\n");
2637                         return;
2638                 }
2639         }
2640         arp_xmit(skb);
2641 }
2642
2643
2644 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2645 {
2646         int i, vlan_id;
2647         __be32 *targets = bond->params.arp_targets;
2648         struct vlan_entry *vlan;
2649         struct net_device *vlan_dev = NULL;
2650         struct rtable *rt;
2651
2652         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2653                 __be32 addr;
2654                 if (!targets[i])
2655                         break;
2656                 pr_debug("basa: target %x\n", targets[i]);
2657                 if (!bond_vlan_used(bond)) {
2658                         pr_debug("basa: empty vlan: arp_send\n");
2659                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2660                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2661                                       addr, 0);
2662                         continue;
2663                 }
2664
2665                 /*
2666                  * If VLANs are configured, we do a route lookup to
2667                  * determine which VLAN interface would be used, so we
2668                  * can tag the ARP with the proper VLAN tag.
2669                  */
2670                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2671                                      RTO_ONLINK, 0);
2672                 if (IS_ERR(rt)) {
2673                         if (net_ratelimit()) {
2674                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2675                                            bond->dev->name, &targets[i]);
2676                         }
2677                         continue;
2678                 }
2679
2680                 /*
2681                  * This target is not on a VLAN
2682                  */
2683                 if (rt->dst.dev == bond->dev) {
2684                         ip_rt_put(rt);
2685                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2686                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2687                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2688                                       addr, 0);
2689                         continue;
2690                 }
2691
2692                 vlan_id = 0;
2693                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2694                         rcu_read_lock();
2695                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2696                                                         vlan->vlan_id);
2697                         rcu_read_unlock();
2698                         if (vlan_dev == rt->dst.dev) {
2699                                 vlan_id = vlan->vlan_id;
2700                                 pr_debug("basa: vlan match on %s %d\n",
2701                                        vlan_dev->name, vlan_id);
2702                                 break;
2703                         }
2704                 }
2705
2706                 if (vlan_id && vlan_dev) {
2707                         ip_rt_put(rt);
2708                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2709                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2710                                       addr, vlan_id);
2711                         continue;
2712                 }
2713
2714                 if (net_ratelimit()) {
2715                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2716                                    bond->dev->name, &targets[i],
2717                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2718                 }
2719                 ip_rt_put(rt);
2720         }
2721 }
2722
2723 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2724 {
2725         int i;
2726         __be32 *targets = bond->params.arp_targets;
2727
2728         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2729                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2730                          &sip, &tip, i, &targets[i],
2731                          bond_has_this_ip(bond, tip));
2732                 if (sip == targets[i]) {
2733                         if (bond_has_this_ip(bond, tip))
2734                                 slave->last_arp_rx = jiffies;
2735                         return;
2736                 }
2737         }
2738 }
2739
2740 static int bond_arp_rcv(struct sk_buff *skb, struct bonding *bond,
2741                          struct slave *slave)
2742 {
2743         struct arphdr *arp;
2744         unsigned char *arp_ptr;
2745         __be32 sip, tip;
2746
2747         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2748                 return RX_HANDLER_ANOTHER;
2749
2750         read_lock(&bond->lock);
2751
2752         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2753                  bond->dev->name, skb->dev->name);
2754
2755         if (!pskb_may_pull(skb, arp_hdr_len(bond->dev)))
2756                 goto out_unlock;
2757
2758         arp = arp_hdr(skb);
2759         if (arp->ar_hln != bond->dev->addr_len ||
2760             skb->pkt_type == PACKET_OTHERHOST ||
2761             skb->pkt_type == PACKET_LOOPBACK ||
2762             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2763             arp->ar_pro != htons(ETH_P_IP) ||
2764             arp->ar_pln != 4)
2765                 goto out_unlock;
2766
2767         arp_ptr = (unsigned char *)(arp + 1);
2768         arp_ptr += bond->dev->addr_len;
2769         memcpy(&sip, arp_ptr, 4);
2770         arp_ptr += 4 + bond->dev->addr_len;
2771         memcpy(&tip, arp_ptr, 4);
2772
2773         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2774                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2775                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2776                  &sip, &tip);
2777
2778         /*
2779          * Backup slaves won't see the ARP reply, but do come through
2780          * here for each ARP probe (so we swap the sip/tip to validate
2781          * the probe).  In a "redundant switch, common router" type of
2782          * configuration, the ARP probe will (hopefully) travel from
2783          * the active, through one switch, the router, then the other
2784          * switch before reaching the backup.
2785          */
2786         if (bond_is_active_slave(slave))
2787                 bond_validate_arp(bond, slave, sip, tip);
2788         else
2789                 bond_validate_arp(bond, slave, tip, sip);
2790
2791 out_unlock:
2792         read_unlock(&bond->lock);
2793         return RX_HANDLER_ANOTHER;
2794 }
2795
2796 /*
2797  * this function is called regularly to monitor each slave's link
2798  * ensuring that traffic is being sent and received when arp monitoring
2799  * is used in load-balancing mode. if the adapter has been dormant, then an
2800  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2801  * arp monitoring in active backup mode.
2802  */
2803 void bond_loadbalance_arp_mon(struct work_struct *work)
2804 {
2805         struct bonding *bond = container_of(work, struct bonding,
2806                                             arp_work.work);
2807         struct slave *slave, *oldcurrent;
2808         int do_failover = 0;
2809         int delta_in_ticks;
2810         int i;
2811
2812         read_lock(&bond->lock);
2813
2814         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2815
2816         if (bond->slave_cnt == 0)
2817                 goto re_arm;
2818
2819         read_lock(&bond->curr_slave_lock);
2820         oldcurrent = bond->curr_active_slave;
2821         read_unlock(&bond->curr_slave_lock);
2822
2823         /* see if any of the previous devices are up now (i.e. they have
2824          * xmt and rcv traffic). the curr_active_slave does not come into
2825          * the picture unless it is null. also, slave->jiffies is not needed
2826          * here because we send an arp on each slave and give a slave as
2827          * long as it needs to get the tx/rx within the delta.
2828          * TODO: what about up/down delay in arp mode? it wasn't here before
2829          *       so it can wait
2830          */
2831         bond_for_each_slave(bond, slave, i) {
2832                 unsigned long trans_start = dev_trans_start(slave->dev);
2833
2834                 if (slave->link != BOND_LINK_UP) {
2835                         if (time_in_range(jiffies,
2836                                 trans_start - delta_in_ticks,
2837                                 trans_start + delta_in_ticks) &&
2838                             time_in_range(jiffies,
2839                                 slave->dev->last_rx - delta_in_ticks,
2840                                 slave->dev->last_rx + delta_in_ticks)) {
2841
2842                                 slave->link  = BOND_LINK_UP;
2843                                 bond_set_active_slave(slave);
2844
2845                                 /* primary_slave has no meaning in round-robin
2846                                  * mode. the window of a slave being up and
2847                                  * curr_active_slave being null after enslaving
2848                                  * is closed.
2849                                  */
2850                                 if (!oldcurrent) {
2851                                         pr_info("%s: link status definitely up for interface %s, ",
2852                                                 bond->dev->name,
2853                                                 slave->dev->name);
2854                                         do_failover = 1;
2855                                 } else {
2856                                         pr_info("%s: interface %s is now up\n",
2857                                                 bond->dev->name,
2858                                                 slave->dev->name);
2859                                 }
2860                         }
2861                 } else {
2862                         /* slave->link == BOND_LINK_UP */
2863
2864                         /* not all switches will respond to an arp request
2865                          * when the source ip is 0, so don't take the link down
2866                          * if we don't know our ip yet
2867                          */
2868                         if (!time_in_range(jiffies,
2869                                 trans_start - delta_in_ticks,
2870                                 trans_start + 2 * delta_in_ticks) ||
2871                             !time_in_range(jiffies,
2872                                 slave->dev->last_rx - delta_in_ticks,
2873                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2874
2875                                 slave->link  = BOND_LINK_DOWN;
2876                                 bond_set_backup_slave(slave);
2877
2878                                 if (slave->link_failure_count < UINT_MAX)
2879                                         slave->link_failure_count++;
2880
2881                                 pr_info("%s: interface %s is now down.\n",
2882                                         bond->dev->name,
2883                                         slave->dev->name);
2884
2885                                 if (slave == oldcurrent)
2886                                         do_failover = 1;
2887                         }
2888                 }
2889
2890                 /* note: if switch is in round-robin mode, all links
2891                  * must tx arp to ensure all links rx an arp - otherwise
2892                  * links may oscillate or not come up at all; if switch is
2893                  * in something like xor mode, there is nothing we can
2894                  * do - all replies will be rx'ed on same link causing slaves
2895                  * to be unstable during low/no traffic periods
2896                  */
2897                 if (IS_UP(slave->dev))
2898                         bond_arp_send_all(bond, slave);
2899         }
2900
2901         if (do_failover) {
2902                 block_netpoll_tx();
2903                 write_lock_bh(&bond->curr_slave_lock);
2904
2905                 bond_select_active_slave(bond);
2906
2907                 write_unlock_bh(&bond->curr_slave_lock);
2908                 unblock_netpoll_tx();
2909         }
2910
2911 re_arm:
2912         if (bond->params.arp_interval)
2913                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2914
2915         read_unlock(&bond->lock);
2916 }
2917
2918 /*
2919  * Called to inspect slaves for active-backup mode ARP monitor link state
2920  * changes.  Sets new_link in slaves to specify what action should take
2921  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2922  * to link states must be committed.
2923  *
2924  * Called with bond->lock held for read.
2925  */
2926 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2927 {
2928         struct slave *slave;
2929         int i, commit = 0;
2930         unsigned long trans_start;
2931
2932         bond_for_each_slave(bond, slave, i) {
2933                 slave->new_link = BOND_LINK_NOCHANGE;
2934
2935                 if (slave->link != BOND_LINK_UP) {
2936                         if (time_in_range(jiffies,
2937                                 slave_last_rx(bond, slave) - delta_in_ticks,
2938                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2939
2940                                 slave->new_link = BOND_LINK_UP;
2941                                 commit++;
2942                         }
2943
2944                         continue;
2945                 }
2946
2947                 /*
2948                  * Give slaves 2*delta after being enslaved or made
2949                  * active.  This avoids bouncing, as the last receive
2950                  * times need a full ARP monitor cycle to be updated.
2951                  */
2952                 if (time_in_range(jiffies,
2953                                   slave->jiffies - delta_in_ticks,
2954                                   slave->jiffies + 2 * delta_in_ticks))
2955                         continue;
2956
2957                 /*
2958                  * Backup slave is down if:
2959                  * - No current_arp_slave AND
2960                  * - more than 3*delta since last receive AND
2961                  * - the bond has an IP address
2962                  *
2963                  * Note: a non-null current_arp_slave indicates
2964                  * the curr_active_slave went down and we are
2965                  * searching for a new one; under this condition
2966                  * we only take the curr_active_slave down - this
2967                  * gives each slave a chance to tx/rx traffic
2968                  * before being taken out
2969                  */
2970                 if (!bond_is_active_slave(slave) &&
2971                     !bond->current_arp_slave &&
2972                     !time_in_range(jiffies,
2973                         slave_last_rx(bond, slave) - delta_in_ticks,
2974                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2975
2976                         slave->new_link = BOND_LINK_DOWN;
2977                         commit++;
2978                 }
2979
2980                 /*
2981                  * Active slave is down if:
2982                  * - more than 2*delta since transmitting OR
2983                  * - (more than 2*delta since receive AND
2984                  *    the bond has an IP address)
2985                  */
2986                 trans_start = dev_trans_start(slave->dev);
2987                 if (bond_is_active_slave(slave) &&
2988                     (!time_in_range(jiffies,
2989                         trans_start - delta_in_ticks,
2990                         trans_start + 2 * delta_in_ticks) ||
2991                      !time_in_range(jiffies,
2992                         slave_last_rx(bond, slave) - delta_in_ticks,
2993                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2994
2995                         slave->new_link = BOND_LINK_DOWN;
2996                         commit++;
2997                 }
2998         }
2999
3000         return commit;
3001 }
3002
3003 /*
3004  * Called to commit link state changes noted by inspection step of
3005  * active-backup mode ARP monitor.
3006  *
3007  * Called with RTNL and bond->lock for read.
3008  */
3009 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3010 {
3011         struct slave *slave;
3012         int i;
3013         unsigned long trans_start;
3014
3015         bond_for_each_slave(bond, slave, i) {
3016                 switch (slave->new_link) {
3017                 case BOND_LINK_NOCHANGE:
3018                         continue;
3019
3020                 case BOND_LINK_UP:
3021                         trans_start = dev_trans_start(slave->dev);
3022                         if ((!bond->curr_active_slave &&
3023                              time_in_range(jiffies,
3024                                            trans_start - delta_in_ticks,
3025                                            trans_start + delta_in_ticks)) ||
3026                             bond->curr_active_slave != slave) {
3027                                 slave->link = BOND_LINK_UP;
3028                                 if (bond->current_arp_slave) {
3029                                         bond_set_slave_inactive_flags(
3030                                                 bond->current_arp_slave);
3031                                         bond->current_arp_slave = NULL;
3032                                 }
3033
3034                                 pr_info("%s: link status definitely up for interface %s.\n",
3035                                         bond->dev->name, slave->dev->name);
3036
3037                                 if (!bond->curr_active_slave ||
3038                                     (slave == bond->primary_slave))
3039                                         goto do_failover;
3040
3041                         }
3042
3043                         continue;
3044
3045                 case BOND_LINK_DOWN:
3046                         if (slave->link_failure_count < UINT_MAX)
3047                                 slave->link_failure_count++;
3048
3049                         slave->link = BOND_LINK_DOWN;
3050                         bond_set_slave_inactive_flags(slave);
3051
3052                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3053                                 bond->dev->name, slave->dev->name);
3054
3055                         if (slave == bond->curr_active_slave) {
3056                                 bond->current_arp_slave = NULL;
3057                                 goto do_failover;
3058                         }
3059
3060                         continue;
3061
3062                 default:
3063                         pr_err("%s: impossible: new_link %d on slave %s\n",
3064                                bond->dev->name, slave->new_link,
3065                                slave->dev->name);
3066                         continue;
3067                 }
3068
3069 do_failover:
3070                 ASSERT_RTNL();
3071                 block_netpoll_tx();
3072                 write_lock_bh(&bond->curr_slave_lock);
3073                 bond_select_active_slave(bond);
3074                 write_unlock_bh(&bond->curr_slave_lock);
3075                 unblock_netpoll_tx();
3076         }
3077
3078         bond_set_carrier(bond);
3079 }
3080
3081 /*
3082  * Send ARP probes for active-backup mode ARP monitor.
3083  *
3084  * Called with bond->lock held for read.
3085  */
3086 static void bond_ab_arp_probe(struct bonding *bond)
3087 {
3088         struct slave *slave;
3089         int i;
3090
3091         read_lock(&bond->curr_slave_lock);
3092
3093         if (bond->current_arp_slave && bond->curr_active_slave)
3094                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3095                         bond->current_arp_slave->dev->name,
3096                         bond->curr_active_slave->dev->name);
3097
3098         if (bond->curr_active_slave) {
3099                 bond_arp_send_all(bond, bond->curr_active_slave);
3100                 read_unlock(&bond->curr_slave_lock);
3101                 return;
3102         }
3103
3104         read_unlock(&bond->curr_slave_lock);
3105
3106         /* if we don't have a curr_active_slave, search for the next available
3107          * backup slave from the current_arp_slave and make it the candidate
3108          * for becoming the curr_active_slave
3109          */
3110
3111         if (!bond->current_arp_slave) {
3112                 bond->current_arp_slave = bond->first_slave;
3113                 if (!bond->current_arp_slave)
3114                         return;
3115         }
3116
3117         bond_set_slave_inactive_flags(bond->current_arp_slave);
3118
3119         /* search for next candidate */
3120         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3121                 if (IS_UP(slave->dev)) {
3122                         slave->link = BOND_LINK_BACK;
3123                         bond_set_slave_active_flags(slave);
3124                         bond_arp_send_all(bond, slave);
3125                         slave->jiffies = jiffies;
3126                         bond->current_arp_slave = slave;
3127                         break;
3128                 }
3129
3130                 /* if the link state is up at this point, we
3131                  * mark it down - this can happen if we have
3132                  * simultaneous link failures and
3133                  * reselect_active_interface doesn't make this
3134                  * one the current slave so it is still marked
3135                  * up when it is actually down
3136                  */
3137                 if (slave->link == BOND_LINK_UP) {
3138                         slave->link = BOND_LINK_DOWN;
3139                         if (slave->link_failure_count < UINT_MAX)
3140                                 slave->link_failure_count++;
3141
3142                         bond_set_slave_inactive_flags(slave);
3143
3144                         pr_info("%s: backup interface %s is now down.\n",
3145                                 bond->dev->name, slave->dev->name);
3146                 }
3147         }
3148 }
3149
3150 void bond_activebackup_arp_mon(struct work_struct *work)
3151 {
3152         struct bonding *bond = container_of(work, struct bonding,
3153                                             arp_work.work);
3154         bool should_notify_peers = false;
3155         int delta_in_ticks;
3156
3157         read_lock(&bond->lock);
3158
3159         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3160
3161         if (bond->slave_cnt == 0)
3162                 goto re_arm;
3163
3164         should_notify_peers = bond_should_notify_peers(bond);
3165
3166         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3167                 read_unlock(&bond->lock);
3168
3169                 /* Race avoidance with bond_close flush of workqueue */
3170                 if (!rtnl_trylock()) {
3171                         read_lock(&bond->lock);
3172                         delta_in_ticks = 1;
3173                         should_notify_peers = false;
3174                         goto re_arm;
3175                 }
3176
3177                 read_lock(&bond->lock);
3178
3179                 bond_ab_arp_commit(bond, delta_in_ticks);
3180
3181                 read_unlock(&bond->lock);
3182                 rtnl_unlock();
3183                 read_lock(&bond->lock);
3184         }
3185
3186         bond_ab_arp_probe(bond);
3187
3188 re_arm:
3189         if (bond->params.arp_interval)
3190                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3191
3192         read_unlock(&bond->lock);
3193
3194         if (should_notify_peers) {
3195                 if (!rtnl_trylock()) {
3196                         read_lock(&bond->lock);
3197                         bond->send_peer_notif++;
3198                         read_unlock(&bond->lock);
3199                         return;
3200                 }
3201                 netdev_bonding_change(bond->dev, NETDEV_NOTIFY_PEERS);
3202                 rtnl_unlock();
3203         }
3204 }
3205
3206 /*-------------------------- netdev event handling --------------------------*/
3207
3208 /*
3209  * Change device name
3210  */
3211 static int bond_event_changename(struct bonding *bond)
3212 {
3213         bond_remove_proc_entry(bond);
3214         bond_create_proc_entry(bond);
3215
3216         bond_debug_reregister(bond);
3217
3218         return NOTIFY_DONE;
3219 }
3220
3221 static int bond_master_netdev_event(unsigned long event,
3222                                     struct net_device *bond_dev)
3223 {
3224         struct bonding *event_bond = netdev_priv(bond_dev);
3225
3226         switch (event) {
3227         case NETDEV_CHANGENAME:
3228                 return bond_event_changename(event_bond);
3229         default:
3230                 break;
3231         }
3232
3233         return NOTIFY_DONE;
3234 }
3235
3236 static int bond_slave_netdev_event(unsigned long event,
3237                                    struct net_device *slave_dev)
3238 {
3239         struct net_device *bond_dev = slave_dev->master;
3240         struct bonding *bond = netdev_priv(bond_dev);
3241         struct slave *slave = NULL;
3242
3243         switch (event) {
3244         case NETDEV_UNREGISTER:
3245                 if (bond_dev) {
3246                         if (bond->setup_by_slave)
3247                                 bond_release_and_destroy(bond_dev, slave_dev);
3248                         else
3249                                 bond_release(bond_dev, slave_dev);
3250                 }
3251                 break;
3252         case NETDEV_UP:
3253         case NETDEV_CHANGE:
3254                 slave = bond_get_slave_by_dev(bond, slave_dev);
3255                 if (slave) {
3256                         u32 old_speed = slave->speed;
3257                         u8  old_duplex = slave->duplex;
3258
3259                         bond_update_speed_duplex(slave);
3260
3261                         if (bond->params.mode == BOND_MODE_8023AD) {
3262                                 if (old_speed != slave->speed)
3263                                         bond_3ad_adapter_speed_changed(slave);
3264                                 if (old_duplex != slave->duplex)
3265                                         bond_3ad_adapter_duplex_changed(slave);
3266                         }
3267                 }
3268
3269                 break;
3270         case NETDEV_DOWN:
3271                 /*
3272                  * ... Or is it this?
3273                  */
3274                 break;
3275         case NETDEV_CHANGEMTU:
3276                 /*
3277                  * TODO: Should slaves be allowed to
3278                  * independently alter their MTU?  For
3279                  * an active-backup bond, slaves need
3280                  * not be the same type of device, so
3281                  * MTUs may vary.  For other modes,
3282                  * slaves arguably should have the
3283                  * same MTUs. To do this, we'd need to
3284                  * take over the slave's change_mtu
3285                  * function for the duration of their
3286                  * servitude.
3287                  */
3288                 break;
3289         case NETDEV_CHANGENAME:
3290                 /*
3291                  * TODO: handle changing the primary's name
3292                  */
3293                 break;
3294         case NETDEV_FEAT_CHANGE:
3295                 bond_compute_features(bond);
3296                 break;
3297         default:
3298                 break;
3299         }
3300
3301         return NOTIFY_DONE;
3302 }
3303
3304 /*
3305  * bond_netdev_event: handle netdev notifier chain events.
3306  *
3307  * This function receives events for the netdev chain.  The caller (an
3308  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3309  * locks for us to safely manipulate the slave devices (RTNL lock,
3310  * dev_probe_lock).
3311  */
3312 static int bond_netdev_event(struct notifier_block *this,
3313                              unsigned long event, void *ptr)
3314 {
3315         struct net_device *event_dev = (struct net_device *)ptr;
3316
3317         pr_debug("event_dev: %s, event: %lx\n",
3318                  event_dev ? event_dev->name : "None",
3319                  event);
3320
3321         if (!(event_dev->priv_flags & IFF_BONDING))
3322                 return NOTIFY_DONE;
3323
3324         if (event_dev->flags & IFF_MASTER) {
3325                 pr_debug("IFF_MASTER\n");
3326                 return bond_master_netdev_event(event, event_dev);
3327         }
3328
3329         if (event_dev->flags & IFF_SLAVE) {
3330                 pr_debug("IFF_SLAVE\n");
3331                 return bond_slave_netdev_event(event, event_dev);
3332         }
3333
3334         return NOTIFY_DONE;
3335 }
3336
3337 static struct notifier_block bond_netdev_notifier = {
3338         .notifier_call = bond_netdev_event,
3339 };
3340
3341 /*---------------------------- Hashing Policies -----------------------------*/
3342
3343 /*
3344  * Hash for the output device based upon layer 2 and layer 3 data. If
3345  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3346  */
3347 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3348 {
3349         struct ethhdr *data = (struct ethhdr *)skb->data;
3350         struct iphdr *iph = ip_hdr(skb);
3351
3352         if (skb->protocol == htons(ETH_P_IP)) {
3353                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3354                         (data->h_dest[5] ^ data->h_source[5])) % count;
3355         }
3356
3357         return (data->h_dest[5] ^ data->h_source[5]) % count;
3358 }
3359
3360 /*
3361  * Hash for the output device based upon layer 3 and layer 4 data. If
3362  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3363  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3364  */
3365 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3366 {
3367         struct ethhdr *data = (struct ethhdr *)skb->data;
3368         struct iphdr *iph = ip_hdr(skb);
3369         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3370         int layer4_xor = 0;
3371
3372         if (skb->protocol == htons(ETH_P_IP)) {
3373                 if (!ip_is_fragment(iph) &&
3374                     (iph->protocol == IPPROTO_TCP ||
3375                      iph->protocol == IPPROTO_UDP)) {
3376                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3377                 }
3378                 return (layer4_xor ^
3379                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3380
3381         }
3382
3383         return (data->h_dest[5] ^ data->h_source[5]) % count;
3384 }
3385
3386 /*
3387  * Hash for the output device based upon layer 2 data
3388  */
3389 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3390 {
3391         struct ethhdr *data = (struct ethhdr *)skb->data;
3392
3393         return (data->h_dest[5] ^ data->h_source[5]) % count;
3394 }
3395
3396 /*-------------------------- Device entry points ----------------------------*/
3397
3398 static int bond_open(struct net_device *bond_dev)
3399 {
3400         struct bonding *bond = netdev_priv(bond_dev);
3401         struct slave *slave;
3402         int i;
3403
3404         /* reset slave->backup and slave->inactive */
3405         read_lock(&bond->lock);
3406         if (bond->slave_cnt > 0) {
3407                 read_lock(&bond->curr_slave_lock);
3408                 bond_for_each_slave(bond, slave, i) {
3409                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3410                                 && (slave != bond->curr_active_slave)) {
3411                                 bond_set_slave_inactive_flags(slave);
3412                         } else {
3413                                 bond_set_slave_active_flags(slave);
3414                         }
3415                 }
3416                 read_unlock(&bond->curr_slave_lock);
3417         }
3418         read_unlock(&bond->lock);
3419
3420         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3421
3422         if (bond_is_lb(bond)) {
3423                 /* bond_alb_initialize must be called before the timer
3424                  * is started.
3425                  */
3426                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3427                         /* something went wrong - fail the open operation */
3428                         return -ENOMEM;
3429                 }
3430
3431                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3432                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3433         }
3434
3435         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3436                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3437                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3438         }
3439
3440         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3441                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3442                         INIT_DELAYED_WORK(&bond->arp_work,
3443                                           bond_activebackup_arp_mon);
3444                 else
3445                         INIT_DELAYED_WORK(&bond->arp_work,
3446                                           bond_loadbalance_arp_mon);
3447
3448                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3449                 if (bond->params.arp_validate)
3450                         bond->recv_probe = bond_arp_rcv;
3451         }
3452
3453         if (bond->params.mode == BOND_MODE_8023AD) {
3454                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3455                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3456                 /* register to receive LACPDUs */
3457                 bond->recv_probe = bond_3ad_lacpdu_recv;
3458                 bond_3ad_initiate_agg_selection(bond, 1);
3459         }
3460
3461         return 0;
3462 }
3463
3464 static int bond_close(struct net_device *bond_dev)
3465 {
3466         struct bonding *bond = netdev_priv(bond_dev);
3467
3468         write_lock_bh(&bond->lock);
3469
3470         bond->send_peer_notif = 0;
3471
3472         write_unlock_bh(&bond->lock);
3473
3474         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3475                 cancel_delayed_work_sync(&bond->mii_work);
3476         }
3477
3478         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3479                 cancel_delayed_work_sync(&bond->arp_work);
3480         }
3481
3482         switch (bond->params.mode) {
3483         case BOND_MODE_8023AD:
3484                 cancel_delayed_work_sync(&bond->ad_work);
3485                 break;
3486         case BOND_MODE_TLB:
3487         case BOND_MODE_ALB:
3488                 cancel_delayed_work_sync(&bond->alb_work);
3489                 break;
3490         default:
3491                 break;
3492         }
3493
3494         if (delayed_work_pending(&bond->mcast_work))
3495                 cancel_delayed_work_sync(&bond->mcast_work);
3496
3497         if (bond_is_lb(bond)) {
3498                 /* Must be called only after all
3499                  * slaves have been released
3500                  */
3501                 bond_alb_deinitialize(bond);
3502         }
3503         bond->recv_probe = NULL;
3504
3505         return 0;
3506 }
3507
3508 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3509                                                 struct rtnl_link_stats64 *stats)
3510 {
3511         struct bonding *bond = netdev_priv(bond_dev);
3512         struct rtnl_link_stats64 temp;
3513         struct slave *slave;
3514         int i;
3515
3516         memset(stats, 0, sizeof(*stats));
3517
3518         read_lock_bh(&bond->lock);
3519
3520         bond_for_each_slave(bond, slave, i) {
3521                 const struct rtnl_link_stats64 *sstats =
3522                         dev_get_stats(slave->dev, &temp);
3523
3524                 stats->rx_packets += sstats->rx_packets;
3525                 stats->rx_bytes += sstats->rx_bytes;
3526                 stats->rx_errors += sstats->rx_errors;
3527                 stats->rx_dropped += sstats->rx_dropped;
3528
3529                 stats->tx_packets += sstats->tx_packets;
3530                 stats->tx_bytes += sstats->tx_bytes;
3531                 stats->tx_errors += sstats->tx_errors;
3532                 stats->tx_dropped += sstats->tx_dropped;
3533
3534                 stats->multicast += sstats->multicast;
3535                 stats->collisions += sstats->collisions;
3536
3537                 stats->rx_length_errors += sstats->rx_length_errors;
3538                 stats->rx_over_errors += sstats->rx_over_errors;
3539                 stats->rx_crc_errors += sstats->rx_crc_errors;
3540                 stats->rx_frame_errors += sstats->rx_frame_errors;
3541                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3542                 stats->rx_missed_errors += sstats->rx_missed_errors;
3543
3544                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3545                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3546                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3547                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3548                 stats->tx_window_errors += sstats->tx_window_errors;
3549         }
3550
3551         read_unlock_bh(&bond->lock);
3552
3553         return stats;
3554 }
3555
3556 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3557 {
3558         struct net_device *slave_dev = NULL;
3559         struct ifbond k_binfo;
3560         struct ifbond __user *u_binfo = NULL;
3561         struct ifslave k_sinfo;
3562         struct ifslave __user *u_sinfo = NULL;
3563         struct mii_ioctl_data *mii = NULL;
3564         int res = 0;
3565
3566         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3567
3568         switch (cmd) {
3569         case SIOCGMIIPHY:
3570                 mii = if_mii(ifr);
3571                 if (!mii)
3572                         return -EINVAL;
3573
3574                 mii->phy_id = 0;
3575                 /* Fall Through */
3576         case SIOCGMIIREG:
3577                 /*
3578                  * We do this again just in case we were called by SIOCGMIIREG
3579                  * instead of SIOCGMIIPHY.
3580                  */
3581                 mii = if_mii(ifr);
3582                 if (!mii)
3583                         return -EINVAL;
3584
3585
3586                 if (mii->reg_num == 1) {
3587                         struct bonding *bond = netdev_priv(bond_dev);
3588                         mii->val_out = 0;
3589                         read_lock(&bond->lock);
3590                         read_lock(&bond->curr_slave_lock);
3591                         if (netif_carrier_ok(bond->dev))
3592                                 mii->val_out = BMSR_LSTATUS;
3593
3594                         read_unlock(&bond->curr_slave_lock);
3595                         read_unlock(&bond->lock);
3596                 }
3597
3598                 return 0;
3599         case BOND_INFO_QUERY_OLD:
3600         case SIOCBONDINFOQUERY:
3601                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3602
3603                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3604                         return -EFAULT;
3605
3606                 res = bond_info_query(bond_dev, &k_binfo);
3607                 if (res == 0 &&
3608                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3609                         return -EFAULT;
3610
3611                 return res;
3612         case BOND_SLAVE_INFO_QUERY_OLD:
3613         case SIOCBONDSLAVEINFOQUERY:
3614                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3615
3616                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3617                         return -EFAULT;
3618
3619                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3620                 if (res == 0 &&
3621                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3622                         return -EFAULT;
3623
3624                 return res;
3625         default:
3626                 /* Go on */
3627                 break;
3628         }
3629
3630         if (!capable(CAP_NET_ADMIN))
3631                 return -EPERM;
3632
3633         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3634
3635         pr_debug("slave_dev=%p:\n", slave_dev);
3636
3637         if (!slave_dev)
3638                 res = -ENODEV;
3639         else {
3640                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3641                 switch (cmd) {
3642                 case BOND_ENSLAVE_OLD:
3643                 case SIOCBONDENSLAVE:
3644                         res = bond_enslave(bond_dev, slave_dev);
3645                         break;
3646                 case BOND_RELEASE_OLD:
3647                 case SIOCBONDRELEASE:
3648                         res = bond_release(bond_dev, slave_dev);
3649                         break;
3650                 case BOND_SETHWADDR_OLD:
3651                 case SIOCBONDSETHWADDR:
3652                         res = bond_sethwaddr(bond_dev, slave_dev);
3653                         break;
3654                 case BOND_CHANGE_ACTIVE_OLD:
3655                 case SIOCBONDCHANGEACTIVE:
3656                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3657                         break;
3658                 default:
3659                         res = -EOPNOTSUPP;
3660                 }
3661
3662                 dev_put(slave_dev);
3663         }
3664
3665         return res;
3666 }
3667
3668 static bool bond_addr_in_mc_list(unsigned char *addr,
3669                                  struct netdev_hw_addr_list *list,
3670                                  int addrlen)
3671 {
3672         struct netdev_hw_addr *ha;
3673
3674         netdev_hw_addr_list_for_each(ha, list)
3675                 if (!memcmp(ha->addr, addr, addrlen))
3676                         return true;
3677
3678         return false;
3679 }
3680
3681 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3682 {
3683         struct bonding *bond = netdev_priv(bond_dev);
3684
3685         if (change & IFF_PROMISC)
3686                 bond_set_promiscuity(bond,
3687                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3688
3689         if (change & IFF_ALLMULTI)
3690                 bond_set_allmulti(bond,
3691                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3692 }
3693
3694 static void bond_set_multicast_list(struct net_device *bond_dev)
3695 {
3696         struct bonding *bond = netdev_priv(bond_dev);
3697         struct netdev_hw_addr *ha;
3698         bool found;
3699
3700         read_lock(&bond->lock);
3701
3702         /* looking for addresses to add to slaves' mc list */
3703         netdev_for_each_mc_addr(ha, bond_dev) {
3704                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3705                                              bond_dev->addr_len);
3706                 if (!found)
3707                         bond_mc_add(bond, ha->addr);
3708         }
3709
3710         /* looking for addresses to delete from slaves' list */
3711         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3712                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3713                                              bond_dev->addr_len);
3714                 if (!found)
3715                         bond_mc_del(bond, ha->addr);
3716         }
3717
3718         /* save master's multicast list */
3719         __hw_addr_flush(&bond->mc_list);
3720         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3721                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3722
3723         read_unlock(&bond->lock);
3724 }
3725
3726 static int bond_neigh_init(struct neighbour *n)
3727 {
3728         struct bonding *bond = netdev_priv(n->dev);
3729         struct slave *slave = bond->first_slave;
3730         const struct net_device_ops *slave_ops;
3731         struct neigh_parms parms;
3732         int ret;
3733
3734         if (!slave)
3735                 return 0;
3736
3737         slave_ops = slave->dev->netdev_ops;
3738
3739         if (!slave_ops->ndo_neigh_setup)
3740                 return 0;
3741
3742         parms.neigh_setup = NULL;
3743         parms.neigh_cleanup = NULL;
3744         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3745         if (ret)
3746                 return ret;
3747
3748         /*
3749          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3750          * after the last slave has been detached.  Assumes that all slaves
3751          * utilize the same neigh_cleanup (true at this writing as only user
3752          * is ipoib).
3753          */
3754         n->parms->neigh_cleanup = parms.neigh_cleanup;
3755
3756         if (!parms.neigh_setup)
3757                 return 0;
3758
3759         return parms.neigh_setup(n);
3760 }
3761
3762 /*
3763  * The bonding ndo_neigh_setup is called at init time beofre any
3764  * slave exists. So we must declare proxy setup function which will
3765  * be used at run time to resolve the actual slave neigh param setup.
3766  */
3767 static int bond_neigh_setup(struct net_device *dev,
3768                             struct neigh_parms *parms)
3769 {
3770         parms->neigh_setup   = bond_neigh_init;
3771
3772         return 0;
3773 }
3774
3775 /*
3776  * Change the MTU of all of a master's slaves to match the master
3777  */
3778 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3779 {
3780         struct bonding *bond = netdev_priv(bond_dev);
3781         struct slave *slave, *stop_at;
3782         int res = 0;
3783         int i;
3784
3785         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3786                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3787
3788         /* Can't hold bond->lock with bh disabled here since
3789          * some base drivers panic. On the other hand we can't
3790          * hold bond->lock without bh disabled because we'll
3791          * deadlock. The only solution is to rely on the fact
3792          * that we're under rtnl_lock here, and the slaves
3793          * list won't change. This doesn't solve the problem
3794          * of setting the slave's MTU while it is
3795          * transmitting, but the assumption is that the base
3796          * driver can handle that.
3797          *
3798          * TODO: figure out a way to safely iterate the slaves
3799          * list, but without holding a lock around the actual
3800          * call to the base driver.
3801          */
3802
3803         bond_for_each_slave(bond, slave, i) {
3804                 pr_debug("s %p s->p %p c_m %p\n",
3805                          slave,
3806                          slave->prev,
3807                          slave->dev->netdev_ops->ndo_change_mtu);
3808
3809                 res = dev_set_mtu(slave->dev, new_mtu);
3810
3811                 if (res) {
3812                         /* If we failed to set the slave's mtu to the new value
3813                          * we must abort the operation even in ACTIVE_BACKUP
3814                          * mode, because if we allow the backup slaves to have
3815                          * different mtu values than the active slave we'll
3816                          * need to change their mtu when doing a failover. That
3817                          * means changing their mtu from timer context, which
3818                          * is probably not a good idea.
3819                          */
3820                         pr_debug("err %d %s\n", res, slave->dev->name);
3821                         goto unwind;
3822                 }
3823         }
3824
3825         bond_dev->mtu = new_mtu;
3826
3827         return 0;
3828
3829 unwind:
3830         /* unwind from head to the slave that failed */
3831         stop_at = slave;
3832         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3833                 int tmp_res;
3834
3835                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3836                 if (tmp_res) {
3837                         pr_debug("unwind err %d dev %s\n",
3838                                  tmp_res, slave->dev->name);
3839                 }
3840         }
3841
3842         return res;
3843 }
3844
3845 /*
3846  * Change HW address
3847  *
3848  * Note that many devices must be down to change the HW address, and
3849  * downing the master releases all slaves.  We can make bonds full of
3850  * bonding devices to test this, however.
3851  */
3852 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3853 {
3854         struct bonding *bond = netdev_priv(bond_dev);
3855         struct sockaddr *sa = addr, tmp_sa;
3856         struct slave *slave, *stop_at;
3857         int res = 0;
3858         int i;
3859
3860         if (bond->params.mode == BOND_MODE_ALB)
3861                 return bond_alb_set_mac_address(bond_dev, addr);
3862
3863
3864         pr_debug("bond=%p, name=%s\n",
3865                  bond, bond_dev ? bond_dev->name : "None");
3866
3867         /*
3868          * If fail_over_mac is set to active, do nothing and return
3869          * success.  Returning an error causes ifenslave to fail.
3870          */
3871         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3872                 return 0;
3873
3874         if (!is_valid_ether_addr(sa->sa_data))
3875                 return -EADDRNOTAVAIL;
3876
3877         /* Can't hold bond->lock with bh disabled here since
3878          * some base drivers panic. On the other hand we can't
3879          * hold bond->lock without bh disabled because we'll
3880          * deadlock. The only solution is to rely on the fact
3881          * that we're under rtnl_lock here, and the slaves
3882          * list won't change. This doesn't solve the problem
3883          * of setting the slave's hw address while it is
3884          * transmitting, but the assumption is that the base
3885          * driver can handle that.
3886          *
3887          * TODO: figure out a way to safely iterate the slaves
3888          * list, but without holding a lock around the actual
3889          * call to the base driver.
3890          */
3891
3892         bond_for_each_slave(bond, slave, i) {
3893                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3894                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3895
3896                 if (slave_ops->ndo_set_mac_address == NULL) {
3897                         res = -EOPNOTSUPP;
3898                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3899                         goto unwind;
3900                 }
3901
3902                 res = dev_set_mac_address(slave->dev, addr);
3903                 if (res) {
3904                         /* TODO: consider downing the slave
3905                          * and retry ?
3906                          * User should expect communications
3907                          * breakage anyway until ARP finish
3908                          * updating, so...
3909                          */
3910                         pr_debug("err %d %s\n", res, slave->dev->name);
3911                         goto unwind;
3912                 }
3913         }
3914
3915         /* success */
3916         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3917         return 0;
3918
3919 unwind:
3920         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3921         tmp_sa.sa_family = bond_dev->type;
3922
3923         /* unwind from head to the slave that failed */
3924         stop_at = slave;
3925         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3926                 int tmp_res;
3927
3928                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3929                 if (tmp_res) {
3930                         pr_debug("unwind err %d dev %s\n",
3931                                  tmp_res, slave->dev->name);
3932                 }
3933         }
3934
3935         return res;
3936 }
3937
3938 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3939 {
3940         struct bonding *bond = netdev_priv(bond_dev);
3941         struct slave *slave, *start_at;
3942         int i, slave_no, res = 1;
3943         struct iphdr *iph = ip_hdr(skb);
3944
3945         /*
3946          * Start with the curr_active_slave that joined the bond as the
3947          * default for sending IGMP traffic.  For failover purposes one
3948          * needs to maintain some consistency for the interface that will
3949          * send the join/membership reports.  The curr_active_slave found
3950          * will send all of this type of traffic.
3951          */
3952         if ((iph->protocol == IPPROTO_IGMP) &&
3953             (skb->protocol == htons(ETH_P_IP))) {
3954
3955                 read_lock(&bond->curr_slave_lock);
3956                 slave = bond->curr_active_slave;
3957                 read_unlock(&bond->curr_slave_lock);
3958
3959                 if (!slave)
3960                         goto out;
3961         } else {
3962                 /*
3963                  * Concurrent TX may collide on rr_tx_counter; we accept
3964                  * that as being rare enough not to justify using an
3965                  * atomic op here.
3966                  */
3967                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3968
3969                 bond_for_each_slave(bond, slave, i) {
3970                         slave_no--;
3971                         if (slave_no < 0)
3972                                 break;
3973                 }
3974         }
3975
3976         start_at = slave;
3977         bond_for_each_slave_from(bond, slave, i, start_at) {
3978                 if (IS_UP(slave->dev) &&
3979                     (slave->link == BOND_LINK_UP) &&
3980                     bond_is_active_slave(slave)) {
3981                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3982                         break;
3983                 }
3984         }
3985
3986 out:
3987         if (res) {
3988                 /* no suitable interface, frame not sent */
3989                 dev_kfree_skb(skb);
3990         }
3991
3992         return NETDEV_TX_OK;
3993 }
3994
3995
3996 /*
3997  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3998  * the bond has a usable interface.
3999  */
4000 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4001 {
4002         struct bonding *bond = netdev_priv(bond_dev);
4003         int res = 1;
4004
4005         read_lock(&bond->curr_slave_lock);
4006
4007         if (bond->curr_active_slave)
4008                 res = bond_dev_queue_xmit(bond, skb,
4009                         bond->curr_active_slave->dev);
4010
4011         if (res)
4012                 /* no suitable interface, frame not sent */
4013                 dev_kfree_skb(skb);
4014
4015         read_unlock(&bond->curr_slave_lock);
4016
4017         return NETDEV_TX_OK;
4018 }
4019
4020 /*
4021  * In bond_xmit_xor() , we determine the output device by using a pre-
4022  * determined xmit_hash_policy(), If the selected device is not enabled,
4023  * find the next active slave.
4024  */
4025 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4026 {
4027         struct bonding *bond = netdev_priv(bond_dev);
4028         struct slave *slave, *start_at;
4029         int slave_no;
4030         int i;
4031         int res = 1;
4032
4033         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4034
4035         bond_for_each_slave(bond, slave, i) {
4036                 slave_no--;
4037                 if (slave_no < 0)
4038                         break;
4039         }
4040
4041         start_at = slave;
4042
4043         bond_for_each_slave_from(bond, slave, i, start_at) {
4044                 if (IS_UP(slave->dev) &&
4045                     (slave->link == BOND_LINK_UP) &&
4046                     bond_is_active_slave(slave)) {
4047                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4048                         break;
4049                 }
4050         }
4051
4052         if (res) {
4053                 /* no suitable interface, frame not sent */
4054                 dev_kfree_skb(skb);
4055         }
4056
4057         return NETDEV_TX_OK;
4058 }
4059
4060 /*
4061  * in broadcast mode, we send everything to all usable interfaces.
4062  */
4063 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4064 {
4065         struct bonding *bond = netdev_priv(bond_dev);
4066         struct slave *slave, *start_at;
4067         struct net_device *tx_dev = NULL;
4068         int i;
4069         int res = 1;
4070
4071         read_lock(&bond->curr_slave_lock);
4072         start_at = bond->curr_active_slave;
4073         read_unlock(&bond->curr_slave_lock);
4074
4075         if (!start_at)
4076                 goto out;
4077
4078         bond_for_each_slave_from(bond, slave, i, start_at) {
4079                 if (IS_UP(slave->dev) &&
4080                     (slave->link == BOND_LINK_UP) &&
4081                     bond_is_active_slave(slave)) {
4082                         if (tx_dev) {
4083                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4084                                 if (!skb2) {
4085                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4086                                                bond_dev->name);
4087                                         continue;
4088                                 }
4089
4090                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4091                                 if (res) {
4092                                         dev_kfree_skb(skb2);
4093                                         continue;
4094                                 }
4095                         }
4096                         tx_dev = slave->dev;
4097                 }
4098         }
4099
4100         if (tx_dev)
4101                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4102
4103 out:
4104         if (res)
4105                 /* no suitable interface, frame not sent */
4106                 dev_kfree_skb(skb);
4107
4108         /* frame sent to all suitable interfaces */
4109         return NETDEV_TX_OK;
4110 }
4111
4112 /*------------------------- Device initialization ---------------------------*/
4113
4114 static void bond_set_xmit_hash_policy(struct bonding *bond)
4115 {
4116         switch (bond->params.xmit_policy) {
4117         case BOND_XMIT_POLICY_LAYER23:
4118                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4119                 break;
4120         case BOND_XMIT_POLICY_LAYER34:
4121                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4122                 break;
4123         case BOND_XMIT_POLICY_LAYER2:
4124         default:
4125                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4126                 break;
4127         }
4128 }
4129
4130 /*
4131  * Lookup the slave that corresponds to a qid
4132  */
4133 static inline int bond_slave_override(struct bonding *bond,
4134                                       struct sk_buff *skb)
4135 {
4136         int i, res = 1;
4137         struct slave *slave = NULL;
4138         struct slave *check_slave;
4139
4140         if (!skb->queue_mapping)
4141                 return 1;
4142
4143         /* Find out if any slaves have the same mapping as this skb. */
4144         bond_for_each_slave(bond, check_slave, i) {
4145                 if (check_slave->queue_id == skb->queue_mapping) {
4146                         slave = check_slave;
4147                         break;
4148                 }
4149         }
4150
4151         /* If the slave isn't UP, use default transmit policy. */
4152         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4153             (slave->link == BOND_LINK_UP)) {
4154                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4155         }
4156
4157         return res;
4158 }
4159
4160
4161 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4162 {
4163         /*
4164          * This helper function exists to help dev_pick_tx get the correct
4165          * destination queue.  Using a helper function skips a call to
4166          * skb_tx_hash and will put the skbs in the queue we expect on their
4167          * way down to the bonding driver.
4168          */
4169         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4170
4171         /*
4172          * Save the original txq to restore before passing to the driver
4173          */
4174         bond_queue_mapping(skb) = skb->queue_mapping;
4175
4176         if (unlikely(txq >= dev->real_num_tx_queues)) {
4177                 do {
4178                         txq -= dev->real_num_tx_queues;
4179                 } while (txq >= dev->real_num_tx_queues);
4180         }
4181         return txq;
4182 }
4183
4184 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4185 {
4186         struct bonding *bond = netdev_priv(dev);
4187
4188         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4189                 if (!bond_slave_override(bond, skb))
4190                         return NETDEV_TX_OK;
4191         }
4192
4193         switch (bond->params.mode) {
4194         case BOND_MODE_ROUNDROBIN:
4195                 return bond_xmit_roundrobin(skb, dev);
4196         case BOND_MODE_ACTIVEBACKUP:
4197                 return bond_xmit_activebackup(skb, dev);
4198         case BOND_MODE_XOR:
4199                 return bond_xmit_xor(skb, dev);
4200         case BOND_MODE_BROADCAST:
4201                 return bond_xmit_broadcast(skb, dev);
4202         case BOND_MODE_8023AD:
4203                 return bond_3ad_xmit_xor(skb, dev);
4204         case BOND_MODE_ALB:
4205         case BOND_MODE_TLB:
4206                 return bond_alb_xmit(skb, dev);
4207         default:
4208                 /* Should never happen, mode already checked */
4209                 pr_err("%s: Error: Unknown bonding mode %d\n",
4210                        dev->name, bond->params.mode);
4211                 WARN_ON_ONCE(1);
4212                 dev_kfree_skb(skb);
4213                 return NETDEV_TX_OK;
4214         }
4215 }
4216
4217 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4218 {
4219         struct bonding *bond = netdev_priv(dev);
4220         netdev_tx_t ret = NETDEV_TX_OK;
4221
4222         /*
4223          * If we risk deadlock from transmitting this in the
4224          * netpoll path, tell netpoll to queue the frame for later tx
4225          */
4226         if (is_netpoll_tx_blocked(dev))
4227                 return NETDEV_TX_BUSY;
4228
4229         read_lock(&bond->lock);
4230
4231         if (bond->slave_cnt)
4232                 ret = __bond_start_xmit(skb, dev);
4233         else
4234                 dev_kfree_skb(skb);
4235
4236         read_unlock(&bond->lock);
4237
4238         return ret;
4239 }
4240
4241 /*
4242  * set bond mode specific net device operations
4243  */
4244 void bond_set_mode_ops(struct bonding *bond, int mode)
4245 {
4246         struct net_device *bond_dev = bond->dev;
4247
4248         switch (mode) {
4249         case BOND_MODE_ROUNDROBIN:
4250                 break;
4251         case BOND_MODE_ACTIVEBACKUP:
4252                 break;
4253         case BOND_MODE_XOR:
4254                 bond_set_xmit_hash_policy(bond);
4255                 break;
4256         case BOND_MODE_BROADCAST:
4257                 break;
4258         case BOND_MODE_8023AD:
4259                 bond_set_xmit_hash_policy(bond);
4260                 break;
4261         case BOND_MODE_ALB:
4262                 /* FALLTHRU */
4263         case BOND_MODE_TLB:
4264                 break;
4265         default:
4266                 /* Should never happen, mode already checked */
4267                 pr_err("%s: Error: Unknown bonding mode %d\n",
4268                        bond_dev->name, mode);
4269                 break;
4270         }
4271 }
4272
4273 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4274                                     struct ethtool_drvinfo *drvinfo)
4275 {
4276         strncpy(drvinfo->driver, DRV_NAME, 32);
4277         strncpy(drvinfo->version, DRV_VERSION, 32);
4278         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4279 }
4280
4281 static const struct ethtool_ops bond_ethtool_ops = {
4282         .get_drvinfo            = bond_ethtool_get_drvinfo,
4283         .get_link               = ethtool_op_get_link,
4284 };
4285
4286 static const struct net_device_ops bond_netdev_ops = {
4287         .ndo_init               = bond_init,
4288         .ndo_uninit             = bond_uninit,
4289         .ndo_open               = bond_open,
4290         .ndo_stop               = bond_close,
4291         .ndo_start_xmit         = bond_start_xmit,
4292         .ndo_select_queue       = bond_select_queue,
4293         .ndo_get_stats64        = bond_get_stats,
4294         .ndo_do_ioctl           = bond_do_ioctl,
4295         .ndo_change_rx_flags    = bond_change_rx_flags,
4296         .ndo_set_rx_mode        = bond_set_multicast_list,
4297         .ndo_change_mtu         = bond_change_mtu,
4298         .ndo_set_mac_address    = bond_set_mac_address,
4299         .ndo_neigh_setup        = bond_neigh_setup,
4300         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4301         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4302 #ifdef CONFIG_NET_POLL_CONTROLLER
4303         .ndo_netpoll_setup      = bond_netpoll_setup,
4304         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4305         .ndo_poll_controller    = bond_poll_controller,
4306 #endif
4307         .ndo_add_slave          = bond_enslave,
4308         .ndo_del_slave          = bond_release,
4309         .ndo_fix_features       = bond_fix_features,
4310 };
4311
4312 static void bond_destructor(struct net_device *bond_dev)
4313 {
4314         struct bonding *bond = netdev_priv(bond_dev);
4315         if (bond->wq)
4316                 destroy_workqueue(bond->wq);
4317         free_netdev(bond_dev);
4318 }
4319
4320 static void bond_setup(struct net_device *bond_dev)
4321 {
4322         struct bonding *bond = netdev_priv(bond_dev);
4323
4324         /* initialize rwlocks */
4325         rwlock_init(&bond->lock);
4326         rwlock_init(&bond->curr_slave_lock);
4327
4328         bond->params = bonding_defaults;
4329
4330         /* Initialize pointers */
4331         bond->dev = bond_dev;
4332         INIT_LIST_HEAD(&bond->vlan_list);
4333
4334         /* Initialize the device entry points */
4335         ether_setup(bond_dev);
4336         bond_dev->netdev_ops = &bond_netdev_ops;
4337         bond_dev->ethtool_ops = &bond_ethtool_ops;
4338         bond_set_mode_ops(bond, bond->params.mode);
4339
4340         bond_dev->destructor = bond_destructor;
4341
4342         /* Initialize the device options */
4343         bond_dev->tx_queue_len = 0;
4344         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4345         bond_dev->priv_flags |= IFF_BONDING;
4346         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4347
4348         /* At first, we block adding VLANs. That's the only way to
4349          * prevent problems that occur when adding VLANs over an
4350          * empty bond. The block will be removed once non-challenged
4351          * slaves are enslaved.
4352          */
4353         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4354
4355         /* don't acquire bond device's netif_tx_lock when
4356          * transmitting */
4357         bond_dev->features |= NETIF_F_LLTX;
4358
4359         /* By default, we declare the bond to be fully
4360          * VLAN hardware accelerated capable. Special
4361          * care is taken in the various xmit functions
4362          * when there are slaves that are not hw accel
4363          * capable
4364          */
4365
4366         bond_dev->hw_features = BOND_VLAN_FEATURES |
4367                                 NETIF_F_HW_VLAN_TX |
4368                                 NETIF_F_HW_VLAN_RX |
4369                                 NETIF_F_HW_VLAN_FILTER;
4370
4371         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4372         bond_dev->features |= bond_dev->hw_features;
4373 }
4374
4375 static void bond_work_cancel_all(struct bonding *bond)
4376 {
4377         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4378                 cancel_delayed_work_sync(&bond->mii_work);
4379
4380         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4381                 cancel_delayed_work_sync(&bond->arp_work);
4382
4383         if (bond->params.mode == BOND_MODE_ALB &&
4384             delayed_work_pending(&bond->alb_work))
4385                 cancel_delayed_work_sync(&bond->alb_work);
4386
4387         if (bond->params.mode == BOND_MODE_8023AD &&
4388             delayed_work_pending(&bond->ad_work))
4389                 cancel_delayed_work_sync(&bond->ad_work);
4390
4391         if (delayed_work_pending(&bond->mcast_work))
4392                 cancel_delayed_work_sync(&bond->mcast_work);
4393 }
4394
4395 /*
4396 * Destroy a bonding device.
4397 * Must be under rtnl_lock when this function is called.
4398 */
4399 static void bond_uninit(struct net_device *bond_dev)
4400 {
4401         struct bonding *bond = netdev_priv(bond_dev);
4402         struct vlan_entry *vlan, *tmp;
4403
4404         bond_netpoll_cleanup(bond_dev);
4405
4406         /* Release the bonded slaves */
4407         bond_release_all(bond_dev);
4408
4409         list_del(&bond->bond_list);
4410
4411         bond_work_cancel_all(bond);
4412
4413         bond_remove_proc_entry(bond);
4414
4415         bond_debug_unregister(bond);
4416
4417         __hw_addr_flush(&bond->mc_list);
4418
4419         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4420                 list_del(&vlan->vlan_list);
4421                 kfree(vlan);
4422         }
4423 }
4424
4425 /*------------------------- Module initialization ---------------------------*/
4426
4427 /*
4428  * Convert string input module parms.  Accept either the
4429  * number of the mode or its string name.  A bit complicated because
4430  * some mode names are substrings of other names, and calls from sysfs
4431  * may have whitespace in the name (trailing newlines, for example).
4432  */
4433 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4434 {
4435         int modeint = -1, i, rv;
4436         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4437
4438         for (p = (char *)buf; *p; p++)
4439                 if (!(isdigit(*p) || isspace(*p)))
4440                         break;
4441
4442         if (*p)
4443                 rv = sscanf(buf, "%20s", modestr);
4444         else
4445                 rv = sscanf(buf, "%d", &modeint);
4446
4447         if (!rv)
4448                 return -1;
4449
4450         for (i = 0; tbl[i].modename; i++) {
4451                 if (modeint == tbl[i].mode)
4452                         return tbl[i].mode;
4453                 if (strcmp(modestr, tbl[i].modename) == 0)
4454                         return tbl[i].mode;
4455         }
4456
4457         return -1;
4458 }
4459
4460 static int bond_check_params(struct bond_params *params)
4461 {
4462         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4463
4464         /*
4465          * Convert string parameters.
4466          */
4467         if (mode) {
4468                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4469                 if (bond_mode == -1) {
4470                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4471                                mode == NULL ? "NULL" : mode);
4472                         return -EINVAL;
4473                 }
4474         }
4475
4476         if (xmit_hash_policy) {
4477                 if ((bond_mode != BOND_MODE_XOR) &&
4478                     (bond_mode != BOND_MODE_8023AD)) {
4479                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4480                                bond_mode_name(bond_mode));
4481                 } else {
4482                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4483                                                         xmit_hashtype_tbl);
4484                         if (xmit_hashtype == -1) {
4485                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4486                                        xmit_hash_policy == NULL ? "NULL" :
4487                                        xmit_hash_policy);
4488                                 return -EINVAL;
4489                         }
4490                 }
4491         }
4492
4493         if (lacp_rate) {
4494                 if (bond_mode != BOND_MODE_8023AD) {
4495                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4496                                 bond_mode_name(bond_mode));
4497                 } else {
4498                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4499                         if (lacp_fast == -1) {
4500                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4501                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4502                                 return -EINVAL;
4503                         }
4504                 }
4505         }
4506
4507         if (ad_select) {
4508                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4509                 if (params->ad_select == -1) {
4510                         pr_err("Error: Invalid ad_select \"%s\"\n",
4511                                ad_select == NULL ? "NULL" : ad_select);
4512                         return -EINVAL;
4513                 }
4514
4515                 if (bond_mode != BOND_MODE_8023AD) {
4516                         pr_warning("ad_select param only affects 802.3ad mode\n");
4517                 }
4518         } else {
4519                 params->ad_select = BOND_AD_STABLE;
4520         }
4521
4522         if (max_bonds < 0) {
4523                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4524                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4525                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4526         }
4527
4528         if (miimon < 0) {
4529                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4530                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4531                 miimon = BOND_LINK_MON_INTERV;
4532         }
4533
4534         if (updelay < 0) {
4535                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4536                            updelay, INT_MAX);
4537                 updelay = 0;
4538         }
4539
4540         if (downdelay < 0) {
4541                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4542                            downdelay, INT_MAX);
4543                 downdelay = 0;
4544         }
4545
4546         if ((use_carrier != 0) && (use_carrier != 1)) {
4547                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4548                            use_carrier);
4549                 use_carrier = 1;
4550         }
4551
4552         if (num_peer_notif < 0 || num_peer_notif > 255) {
4553                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4554                            num_peer_notif);
4555                 num_peer_notif = 1;
4556         }
4557
4558         /* reset values for 802.3ad */
4559         if (bond_mode == BOND_MODE_8023AD) {
4560                 if (!miimon) {
4561                         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");
4562                         pr_warning("Forcing miimon to 100msec\n");
4563                         miimon = 100;
4564                 }
4565         }
4566
4567         if (tx_queues < 1 || tx_queues > 255) {
4568                 pr_warning("Warning: tx_queues (%d) should be between "
4569                            "1 and 255, resetting to %d\n",
4570                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4571                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4572         }
4573
4574         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4575                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4576                            "not of valid value (0/1), so it was set to "
4577                            "0\n", all_slaves_active);
4578                 all_slaves_active = 0;
4579         }
4580
4581         if (resend_igmp < 0 || resend_igmp > 255) {
4582                 pr_warning("Warning: resend_igmp (%d) should be between "
4583                            "0 and 255, resetting to %d\n",
4584                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4585                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4586         }
4587
4588         /* reset values for TLB/ALB */
4589         if ((bond_mode == BOND_MODE_TLB) ||
4590             (bond_mode == BOND_MODE_ALB)) {
4591                 if (!miimon) {
4592                         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");
4593                         pr_warning("Forcing miimon to 100msec\n");
4594                         miimon = 100;
4595                 }
4596         }
4597
4598         if (bond_mode == BOND_MODE_ALB) {
4599                 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",
4600                           updelay);
4601         }
4602
4603         if (!miimon) {
4604                 if (updelay || downdelay) {
4605                         /* just warn the user the up/down delay will have
4606                          * no effect since miimon is zero...
4607                          */
4608                         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",
4609                                    updelay, downdelay);
4610                 }
4611         } else {
4612                 /* don't allow arp monitoring */
4613                 if (arp_interval) {
4614                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4615                                    miimon, arp_interval);
4616                         arp_interval = 0;
4617                 }
4618
4619                 if ((updelay % miimon) != 0) {
4620                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4621                                    updelay, miimon,
4622                                    (updelay / miimon) * miimon);
4623                 }
4624
4625                 updelay /= miimon;
4626
4627                 if ((downdelay % miimon) != 0) {
4628                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4629                                    downdelay, miimon,
4630                                    (downdelay / miimon) * miimon);
4631                 }
4632
4633                 downdelay /= miimon;
4634         }
4635
4636         if (arp_interval < 0) {
4637                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4638                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4639                 arp_interval = BOND_LINK_ARP_INTERV;
4640         }
4641
4642         for (arp_ip_count = 0;
4643              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4644              arp_ip_count++) {
4645                 /* not complete check, but should be good enough to
4646                    catch mistakes */
4647                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4648                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4649                                    arp_ip_target[arp_ip_count]);
4650                         arp_interval = 0;
4651                 } else {
4652                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4653                         arp_target[arp_ip_count] = ip;
4654                 }
4655         }
4656
4657         if (arp_interval && !arp_ip_count) {
4658                 /* don't allow arping if no arp_ip_target given... */
4659                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4660                            arp_interval);
4661                 arp_interval = 0;
4662         }
4663
4664         if (arp_validate) {
4665                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4666                         pr_err("arp_validate only supported in active-backup mode\n");
4667                         return -EINVAL;
4668                 }
4669                 if (!arp_interval) {
4670                         pr_err("arp_validate requires arp_interval\n");
4671                         return -EINVAL;
4672                 }
4673
4674                 arp_validate_value = bond_parse_parm(arp_validate,
4675                                                      arp_validate_tbl);
4676                 if (arp_validate_value == -1) {
4677                         pr_err("Error: invalid arp_validate \"%s\"\n",
4678                                arp_validate == NULL ? "NULL" : arp_validate);
4679                         return -EINVAL;
4680                 }
4681         } else
4682                 arp_validate_value = 0;
4683
4684         if (miimon) {
4685                 pr_info("MII link monitoring set to %d ms\n", miimon);
4686         } else if (arp_interval) {
4687                 int i;
4688
4689                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4690                         arp_interval,
4691                         arp_validate_tbl[arp_validate_value].modename,
4692                         arp_ip_count);
4693
4694                 for (i = 0; i < arp_ip_count; i++)
4695                         pr_info(" %s", arp_ip_target[i]);
4696
4697                 pr_info("\n");
4698
4699         } else if (max_bonds) {
4700                 /* miimon and arp_interval not set, we need one so things
4701                  * work as expected, see bonding.txt for details
4702                  */
4703                 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");
4704         }
4705
4706         if (primary && !USES_PRIMARY(bond_mode)) {
4707                 /* currently, using a primary only makes sense
4708                  * in active backup, TLB or ALB modes
4709                  */
4710                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4711                            primary, bond_mode_name(bond_mode));
4712                 primary = NULL;
4713         }
4714
4715         if (primary && primary_reselect) {
4716                 primary_reselect_value = bond_parse_parm(primary_reselect,
4717                                                          pri_reselect_tbl);
4718                 if (primary_reselect_value == -1) {
4719                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4720                                primary_reselect ==
4721                                         NULL ? "NULL" : primary_reselect);
4722                         return -EINVAL;
4723                 }
4724         } else {
4725                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4726         }
4727
4728         if (fail_over_mac) {
4729                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4730                                                       fail_over_mac_tbl);
4731                 if (fail_over_mac_value == -1) {
4732                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4733                                arp_validate == NULL ? "NULL" : arp_validate);
4734                         return -EINVAL;
4735                 }
4736
4737                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4738                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4739         } else {
4740                 fail_over_mac_value = BOND_FOM_NONE;
4741         }
4742
4743         /* fill params struct with the proper values */
4744         params->mode = bond_mode;
4745         params->xmit_policy = xmit_hashtype;
4746         params->miimon = miimon;
4747         params->num_peer_notif = num_peer_notif;
4748         params->arp_interval = arp_interval;
4749         params->arp_validate = arp_validate_value;
4750         params->updelay = updelay;
4751         params->downdelay = downdelay;
4752         params->use_carrier = use_carrier;
4753         params->lacp_fast = lacp_fast;
4754         params->primary[0] = 0;
4755         params->primary_reselect = primary_reselect_value;
4756         params->fail_over_mac = fail_over_mac_value;
4757         params->tx_queues = tx_queues;
4758         params->all_slaves_active = all_slaves_active;
4759         params->resend_igmp = resend_igmp;
4760         params->min_links = min_links;
4761
4762         if (primary) {
4763                 strncpy(params->primary, primary, IFNAMSIZ);
4764                 params->primary[IFNAMSIZ - 1] = 0;
4765         }
4766
4767         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4768
4769         return 0;
4770 }
4771
4772 static struct lock_class_key bonding_netdev_xmit_lock_key;
4773 static struct lock_class_key bonding_netdev_addr_lock_key;
4774
4775 static void bond_set_lockdep_class_one(struct net_device *dev,
4776                                        struct netdev_queue *txq,
4777                                        void *_unused)
4778 {
4779         lockdep_set_class(&txq->_xmit_lock,
4780                           &bonding_netdev_xmit_lock_key);
4781 }
4782
4783 static void bond_set_lockdep_class(struct net_device *dev)
4784 {
4785         lockdep_set_class(&dev->addr_list_lock,
4786                           &bonding_netdev_addr_lock_key);
4787         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4788 }
4789
4790 /*
4791  * Called from registration process
4792  */
4793 static int bond_init(struct net_device *bond_dev)
4794 {
4795         struct bonding *bond = netdev_priv(bond_dev);
4796         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4797         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4798
4799         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4800
4801         /*
4802          * Initialize locks that may be required during
4803          * en/deslave operations.  All of the bond_open work
4804          * (of which this is part) should really be moved to
4805          * a phase prior to dev_open
4806          */
4807         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4808         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4809
4810         bond->wq = create_singlethread_workqueue(bond_dev->name);
4811         if (!bond->wq)
4812                 return -ENOMEM;
4813
4814         bond_set_lockdep_class(bond_dev);
4815
4816         bond_create_proc_entry(bond);
4817         list_add_tail(&bond->bond_list, &bn->dev_list);
4818
4819         bond_prepare_sysfs_group(bond);
4820
4821         bond_debug_register(bond);
4822
4823         __hw_addr_init(&bond->mc_list);
4824         return 0;
4825 }
4826
4827 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4828 {
4829         if (tb[IFLA_ADDRESS]) {
4830                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4831                         return -EINVAL;
4832                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4833                         return -EADDRNOTAVAIL;
4834         }
4835         return 0;
4836 }
4837
4838 static int bond_get_tx_queues(struct net *net, struct nlattr *tb[])
4839 {
4840         return tx_queues;
4841 }
4842
4843 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4844         .kind           = "bond",
4845         .priv_size      = sizeof(struct bonding),
4846         .setup          = bond_setup,
4847         .validate       = bond_validate,
4848         .get_tx_queues  = bond_get_tx_queues,
4849 };
4850
4851 /* Create a new bond based on the specified name and bonding parameters.
4852  * If name is NULL, obtain a suitable "bond%d" name for us.
4853  * Caller must NOT hold rtnl_lock; we need to release it here before we
4854  * set up our sysfs entries.
4855  */
4856 int bond_create(struct net *net, const char *name)
4857 {
4858         struct net_device *bond_dev;
4859         int res;
4860
4861         rtnl_lock();
4862
4863         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4864                                    name ? name : "bond%d",
4865                                    bond_setup, tx_queues);
4866         if (!bond_dev) {
4867                 pr_err("%s: eek! can't alloc netdev!\n", name);
4868                 rtnl_unlock();
4869                 return -ENOMEM;
4870         }
4871
4872         dev_net_set(bond_dev, net);
4873         bond_dev->rtnl_link_ops = &bond_link_ops;
4874
4875         res = register_netdevice(bond_dev);
4876
4877         netif_carrier_off(bond_dev);
4878
4879         rtnl_unlock();
4880         if (res < 0)
4881                 bond_destructor(bond_dev);
4882         return res;
4883 }
4884
4885 static int __net_init bond_net_init(struct net *net)
4886 {
4887         struct bond_net *bn = net_generic(net, bond_net_id);
4888
4889         bn->net = net;
4890         INIT_LIST_HEAD(&bn->dev_list);
4891
4892         bond_create_proc_dir(bn);
4893         bond_create_sysfs(bn);
4894
4895         return 0;
4896 }
4897
4898 static void __net_exit bond_net_exit(struct net *net)
4899 {
4900         struct bond_net *bn = net_generic(net, bond_net_id);
4901
4902         bond_destroy_sysfs(bn);
4903         bond_destroy_proc_dir(bn);
4904 }
4905
4906 static struct pernet_operations bond_net_ops = {
4907         .init = bond_net_init,
4908         .exit = bond_net_exit,
4909         .id   = &bond_net_id,
4910         .size = sizeof(struct bond_net),
4911 };
4912
4913 static int __init bonding_init(void)
4914 {
4915         int i;
4916         int res;
4917
4918         pr_info("%s", bond_version);
4919
4920         res = bond_check_params(&bonding_defaults);
4921         if (res)
4922                 goto out;
4923
4924         res = register_pernet_subsys(&bond_net_ops);
4925         if (res)
4926                 goto out;
4927
4928         res = rtnl_link_register(&bond_link_ops);
4929         if (res)
4930                 goto err_link;
4931
4932         bond_create_debugfs();
4933
4934         for (i = 0; i < max_bonds; i++) {
4935                 res = bond_create(&init_net, NULL);
4936                 if (res)
4937                         goto err;
4938         }
4939
4940         register_netdevice_notifier(&bond_netdev_notifier);
4941 out:
4942         return res;
4943 err:
4944         rtnl_link_unregister(&bond_link_ops);
4945 err_link:
4946         unregister_pernet_subsys(&bond_net_ops);
4947         goto out;
4948
4949 }
4950
4951 static void __exit bonding_exit(void)
4952 {
4953         unregister_netdevice_notifier(&bond_netdev_notifier);
4954
4955         bond_destroy_debugfs();
4956
4957         rtnl_link_unregister(&bond_link_ops);
4958         unregister_pernet_subsys(&bond_net_ops);
4959
4960 #ifdef CONFIG_NET_POLL_CONTROLLER
4961         /*
4962          * Make sure we don't have an imbalance on our netpoll blocking
4963          */
4964         WARN_ON(atomic_read(&netpoll_block_tx));
4965 #endif
4966 }
4967
4968 module_init(bonding_init);
4969 module_exit(bonding_exit);
4970 MODULE_LICENSE("GPL");
4971 MODULE_VERSION(DRV_VERSION);
4972 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4973 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4974 MODULE_ALIAS_RTNL_LINK("bond");