drivers/net/bonding/bond_main.c

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