drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon       = BOND_LINK_MON_INTERV;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static int all_slaves_active = 0;
 109 static struct bond_params bonding_defaults;
 110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 111
 112 module_param(max_bonds, int, 0);
 113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 114 module_param(tx_queues, int, 0);
 115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 118                                "failover event (alias of num_unsol_na)");
 119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 121                                "failover event (alias of num_grat_arp)");
 122 module_param(miimon, int, 0);
 123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 124 module_param(updelay, int, 0);
 125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 126 module_param(downdelay, int, 0);
 127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 128                             "in milliseconds");
 129 module_param(use_carrier, int, 0);
 130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 131                               "0 for off, 1 for on (default)");
 132 module_param(mode, charp, 0);
 133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 134                        "1 for active-backup, 2 for balance-xor, "
 135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 136                        "6 for balance-alb");
 137 module_param(primary, charp, 0);
 138 MODULE_PARM_DESC(primary, "Primary network device to use");
 139 module_param(primary_reselect, charp, 0);
 140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 141                                    "once it comes up; "
 142                                    "0 for always (default), "
 143                                    "1 for only if speed of primary is "
 144                                    "better, "
 145                                    "2 for only on active slave "
 146                                    "failure");
 147 module_param(lacp_rate, charp, 0);
 148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 149                             "0 for slow, 1 for fast");
 150 module_param(ad_select, charp, 0);
 151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 152                             "0 for stable (default), 1 for bandwidth, "
 153                             "2 for count");
 154 module_param(min_links, int, 0);
 155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 156
 157 module_param(xmit_hash_policy, charp, 0);
 158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 159                                    "0 for layer 2 (default), 1 for layer 3+4, "
 160                                    "2 for layer 2+3");
 161 module_param(arp_interval, int, 0);
 162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 163 module_param_array(arp_ip_target, charp, NULL, 0);
 164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 165 module_param(arp_validate, charp, 0);
 166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 167                                "0 for none (default), 1 for active, "
 168                                "2 for backup, 3 for all");
 169 module_param(fail_over_mac, charp, 0);
 170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 171                                 "the same MAC; 0 for none (default), "
 172                                 "1 for active, 2 for follow");
 173 module_param(all_slaves_active, int, 0);
 174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 175                                      "by setting active flag for all slaves; "
 176                                      "0 for never (default), 1 for always.");
 177 module_param(resend_igmp, int, 0);
 178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 179                               "link failure");
 180
 181 /*----------------------------- Global variables ----------------------------*/
 182
 183 #ifdef CONFIG_NET_POLL_CONTROLLER
 184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 185 #endif
 186
 187 int bond_net_id __read_mostly;
 188
 189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 190 static int arp_ip_count;
 191 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 193 static int lacp_fast;
 194
 195 const struct bond_parm_tbl bond_lacp_tbl[] = {
 196 {       "slow",         AD_LACP_SLOW},
 197 {       "fast",         AD_LACP_FAST},
 198 {       NULL,           -1},
 199 };
 200
 201 const struct bond_parm_tbl bond_mode_tbl[] = {
 202 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 203 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 204 {       "balance-xor",          BOND_MODE_XOR},
 205 {       "broadcast",            BOND_MODE_BROADCAST},
 206 {       "802.3ad",              BOND_MODE_8023AD},
 207 {       "balance-tlb",          BOND_MODE_TLB},
 208 {       "balance-alb",          BOND_MODE_ALB},
 209 {       NULL,                   -1},
 210 };
 211
 212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 213 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 214 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 215 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 216 {       NULL,                   -1},
 217 };
 218
 219 const struct bond_parm_tbl arp_validate_tbl[] = {
 220 {       "none",                 BOND_ARP_VALIDATE_NONE},
 221 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 222 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 223 {       "all",                  BOND_ARP_VALIDATE_ALL},
 224 {       NULL,                   -1},
 225 };
 226
 227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 228 {       "none",                 BOND_FOM_NONE},
 229 {       "active",               BOND_FOM_ACTIVE},
 230 {       "follow",               BOND_FOM_FOLLOW},
 231 {       NULL,                   -1},
 232 };
 233
 234 const struct bond_parm_tbl pri_reselect_tbl[] = {
 235 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 236 {       "better",               BOND_PRI_RESELECT_BETTER},
 237 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 238 {       NULL,                   -1},
 239 };
 240
 241 struct bond_parm_tbl ad_select_tbl[] = {
 242 {       "stable",       BOND_AD_STABLE},
 243 {       "bandwidth",    BOND_AD_BANDWIDTH},
 244 {       "count",        BOND_AD_COUNT},
 245 {       NULL,           -1},
 246 };
 247
 248 /*-------------------------- Forward declarations ---------------------------*/
 249
 250 static int bond_init(struct net_device *bond_dev);
 251 static void bond_uninit(struct net_device *bond_dev);
 252
 253 /*---------------------------- General routines -----------------------------*/
 254
 255 const char *bond_mode_name(int mode)
 256 {
 257         static const char *names[] = {
 258                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 259                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 260                 [BOND_MODE_XOR] = "load balancing (xor)",
 261                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 262                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 263                 [BOND_MODE_TLB] = "transmit load balancing",
 264                 [BOND_MODE_ALB] = "adaptive load balancing",
 265         };
 266
 267         if (mode < 0 || mode > BOND_MODE_ALB)
 268                 return "unknown";
 269
 270         return names[mode];
 271 }
 272
 273 /*---------------------------------- VLAN -----------------------------------*/
 274
 275 /**
 276  * bond_add_vlan - add a new vlan id on bond
 277  * @bond: bond that got the notification
 278  * @vlan_id: the vlan id to add
 279  *
 280  * Returns -ENOMEM if allocation failed.
 281  */
 282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 283 {
 284         struct vlan_entry *vlan;
 285
 286         pr_debug("bond: %s, vlan id %d\n",
 287                  (bond ? bond->dev->name : "None"), vlan_id);
 288
 289         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 290         if (!vlan)
 291                 return -ENOMEM;
 292
 293         INIT_LIST_HEAD(&vlan->vlan_list);
 294         vlan->vlan_id = vlan_id;
 295
 296         write_lock_bh(&bond->lock);
 297
 298         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 299
 300         write_unlock_bh(&bond->lock);
 301
 302         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 303
 304         return 0;
 305 }
 306
 307 /**
 308  * bond_del_vlan - delete a vlan id from bond
 309  * @bond: bond that got the notification
 310  * @vlan_id: the vlan id to delete
 311  *
 312  * returns -ENODEV if @vlan_id was not found in @bond.
 313  */
 314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 315 {
 316         struct vlan_entry *vlan;
 317         int res = -ENODEV;
 318
 319         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 320
 321         block_netpoll_tx();
 322         write_lock_bh(&bond->lock);
 323
 324         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 325                 if (vlan->vlan_id == vlan_id) {
 326                         list_del(&vlan->vlan_list);
 327
 328                         if (bond_is_lb(bond))
 329                                 bond_alb_clear_vlan(bond, vlan_id);
 330
 331                         pr_debug("removed VLAN ID %d from bond %s\n",
 332                                  vlan_id, bond->dev->name);
 333
 334                         kfree(vlan);
 335
 336                         res = 0;
 337                         goto out;
 338                 }
 339         }
 340
 341         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 342                  vlan_id, bond->dev->name);
 343
 344 out:
 345         write_unlock_bh(&bond->lock);
 346         unblock_netpoll_tx();
 347         return res;
 348 }
 349
 350 /**
 351  * bond_next_vlan - safely skip to the next item in the vlans list.
 352  * @bond: the bond we're working on
 353  * @curr: item we're advancing from
 354  *
 355  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 356  * or @curr->next otherwise (even if it is @curr itself again).
 357  *
 358  * Caller must hold bond->lock
 359  */
 360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 361 {
 362         struct vlan_entry *next, *last;
 363
 364         if (list_empty(&bond->vlan_list))
 365                 return NULL;
 366
 367         if (!curr) {
 368                 next = list_entry(bond->vlan_list.next,
 369                                   struct vlan_entry, vlan_list);
 370         } else {
 371                 last = list_entry(bond->vlan_list.prev,
 372                                   struct vlan_entry, vlan_list);
 373                 if (last == curr) {
 374                         next = list_entry(bond->vlan_list.next,
 375                                           struct vlan_entry, vlan_list);
 376                 } else {
 377                         next = list_entry(curr->vlan_list.next,
 378                                           struct vlan_entry, vlan_list);
 379                 }
 380         }
 381
 382         return next;
 383 }
 384
 385 /**
 386  * bond_dev_queue_xmit - Prepare skb for xmit.
 387  *
 388  * @bond: bond device that got this skb for tx.
 389  * @skb: hw accel VLAN tagged skb to transmit
 390  * @slave_dev: slave that is supposed to xmit this skbuff
 391  */
 392 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 393                         struct net_device *slave_dev)
 394 {
 395         skb->dev = slave_dev;
 396
 397         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 398                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 399         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 400
 401         if (unlikely(netpoll_tx_running(bond->dev)))
 402                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 403         else
 404                 dev_queue_xmit(skb);
 405
 406         return 0;
 407 }
 408
 409 /*
 410  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 411  * We don't protect the slave list iteration with a lock because:
 412  * a. This operation is performed in IOCTL context,
 413  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 414  * c. Holding a lock with BH disabled while directly calling a base driver
 415  *    entry point is generally a BAD idea.
 416  *
 417  * The design of synchronization/protection for this operation in the 8021q
 418  * module is good for one or more VLAN devices over a single physical device
 419  * and cannot be extended for a teaming solution like bonding, so there is a
 420  * potential race condition here where a net device from the vlan group might
 421  * be referenced (either by a base driver or the 8021q code) while it is being
 422  * removed from the system. However, it turns out we're not making matters
 423  * worse, and if it works for regular VLAN usage it will work here too.
 424 */
 425
 426 /**
 427  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 428  * @bond_dev: bonding net device that got called
 429  * @vid: vlan id being added
 430  */
 431 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 432 {
 433         struct bonding *bond = netdev_priv(bond_dev);
 434         struct slave *slave, *stop_at;
 435         int i, res;
 436
 437         bond_for_each_slave(bond, slave, i) {
 438                 res = vlan_vid_add(slave->dev, vid);
 439                 if (res)
 440                         goto unwind;
 441         }
 442
 443         res = bond_add_vlan(bond, vid);
 444         if (res) {
 445                 pr_err("%s: Error: Failed to add vlan id %d\n",
 446                        bond_dev->name, vid);
 447                 return res;
 448         }
 449
 450         return 0;
 451
 452 unwind:
 453         /* unwind from head to the slave that failed */
 454         stop_at = slave;
 455         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
 456                 vlan_vid_del(slave->dev, vid);
 457
 458         return res;
 459 }
 460
 461 /**
 462  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 463  * @bond_dev: bonding net device that got called
 464  * @vid: vlan id being removed
 465  */
 466 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 467 {
 468         struct bonding *bond = netdev_priv(bond_dev);
 469         struct slave *slave;
 470         int i, res;
 471
 472         bond_for_each_slave(bond, slave, i)
 473                 vlan_vid_del(slave->dev, vid);
 474
 475         res = bond_del_vlan(bond, vid);
 476         if (res) {
 477                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 478                        bond_dev->name, vid);
 479                 return res;
 480         }
 481
 482         return 0;
 483 }
 484
 485 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 486 {
 487         struct vlan_entry *vlan;
 488         int res;
 489
 490         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 491                 res = vlan_vid_add(slave_dev, vlan->vlan_id);
 492                 if (res)
 493                         pr_warning("%s: Failed to add vlan id %d to device %s\n",
 494                                    bond->dev->name, vlan->vlan_id,
 495                                    slave_dev->name);
 496         }
 497 }
 498
 499 static void bond_del_vlans_from_slave(struct bonding *bond,
 500                                       struct net_device *slave_dev)
 501 {
 502         struct vlan_entry *vlan;
 503
 504         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 505                 if (!vlan->vlan_id)
 506                         continue;
 507                 vlan_vid_del(slave_dev, vlan->vlan_id);
 508         }
 509 }
 510
 511 /*------------------------------- Link status -------------------------------*/
 512
 513 /*
 514  * Set the carrier state for the master according to the state of its
 515  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 516  * do special 802.3ad magic.
 517  *
 518  * Returns zero if carrier state does not change, nonzero if it does.
 519  */
 520 static int bond_set_carrier(struct bonding *bond)
 521 {
 522         struct slave *slave;
 523         int i;
 524
 525         if (bond->slave_cnt == 0)
 526                 goto down;
 527
 528         if (bond->params.mode == BOND_MODE_8023AD)
 529                 return bond_3ad_set_carrier(bond);
 530
 531         bond_for_each_slave(bond, slave, i) {
 532                 if (slave->link == BOND_LINK_UP) {
 533                         if (!netif_carrier_ok(bond->dev)) {
 534                                 netif_carrier_on(bond->dev);
 535                                 return 1;
 536                         }
 537                         return 0;
 538                 }
 539         }
 540
 541 down:
 542         if (netif_carrier_ok(bond->dev)) {
 543                 netif_carrier_off(bond->dev);
 544                 return 1;
 545         }
 546         return 0;
 547 }
 548
 549 /*
 550  * Get link speed and duplex from the slave's base driver
 551  * using ethtool. If for some reason the call fails or the
 552  * values are invalid, set speed and duplex to -1,
 553  * and return.
 554  */
 555 static void bond_update_speed_duplex(struct slave *slave)
 556 {
 557         struct net_device *slave_dev = slave->dev;
 558         struct ethtool_cmd ecmd;
 559         u32 slave_speed;
 560         int res;
 561
 562         slave->speed = SPEED_UNKNOWN;
 563         slave->duplex = DUPLEX_UNKNOWN;
 564
 565         res = __ethtool_get_settings(slave_dev, &ecmd);
 566         if (res < 0)
 567                 return;
 568
 569         slave_speed = ethtool_cmd_speed(&ecmd);
 570         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 571                 return;
 572
 573         switch (ecmd.duplex) {
 574         case DUPLEX_FULL:
 575         case DUPLEX_HALF:
 576                 break;
 577         default:
 578                 return;
 579         }
 580
 581         slave->speed = slave_speed;
 582         slave->duplex = ecmd.duplex;
 583
 584         return;
 585 }
 586
 587 /*
 588  * if <dev> supports MII link status reporting, check its link status.
 589  *
 590  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 591  * depending upon the setting of the use_carrier parameter.
 592  *
 593  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 594  * can't tell and just pretend it is), or 0, meaning that the link is
 595  * down.
 596  *
 597  * If reporting is non-zero, instead of faking link up, return -1 if
 598  * both ETHTOOL and MII ioctls fail (meaning the device does not
 599  * support them).  If use_carrier is set, return whatever it says.
 600  * It'd be nice if there was a good way to tell if a driver supports
 601  * netif_carrier, but there really isn't.
 602  */
 603 static int bond_check_dev_link(struct bonding *bond,
 604                                struct net_device *slave_dev, int reporting)
 605 {
 606         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 607         int (*ioctl)(struct net_device *, struct ifreq *, int);
 608         struct ifreq ifr;
 609         struct mii_ioctl_data *mii;
 610
 611         if (!reporting && !netif_running(slave_dev))
 612                 return 0;
 613
 614         if (bond->params.use_carrier)
 615                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 616
 617         /* Try to get link status using Ethtool first. */
 618         if (slave_dev->ethtool_ops->get_link)
 619                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 620                         BMSR_LSTATUS : 0;
 621
 622         /* Ethtool can't be used, fallback to MII ioctls. */
 623         ioctl = slave_ops->ndo_do_ioctl;
 624         if (ioctl) {
 625                 /* TODO: set pointer to correct ioctl on a per team member */
 626                 /*       bases to make this more efficient. that is, once  */
 627                 /*       we determine the correct ioctl, we will always    */
 628                 /*       call it and not the others for that team          */
 629                 /*       member.                                           */
 630
 631                 /*
 632                  * We cannot assume that SIOCGMIIPHY will also read a
 633                  * register; not all network drivers (e.g., e100)
 634                  * support that.
 635                  */
 636
 637                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 638                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 639                 mii = if_mii(&ifr);
 640                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 641                         mii->reg_num = MII_BMSR;
 642                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 643                                 return mii->val_out & BMSR_LSTATUS;
 644                 }
 645         }
 646
 647         /*
 648          * If reporting, report that either there's no dev->do_ioctl,
 649          * or both SIOCGMIIREG and get_link failed (meaning that we
 650          * cannot report link status).  If not reporting, pretend
 651          * we're ok.
 652          */
 653         return reporting ? -1 : BMSR_LSTATUS;
 654 }
 655
 656 /*----------------------------- Multicast list ------------------------------*/
 657
 658 /*
 659  * Push the promiscuity flag down to appropriate slaves
 660  */
 661 static int bond_set_promiscuity(struct bonding *bond, int inc)
 662 {
 663         int err = 0;
 664         if (USES_PRIMARY(bond->params.mode)) {
 665                 /* write lock already acquired */
 666                 if (bond->curr_active_slave) {
 667                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 668                                                   inc);
 669                 }
 670         } else {
 671                 struct slave *slave;
 672                 int i;
 673                 bond_for_each_slave(bond, slave, i) {
 674                         err = dev_set_promiscuity(slave->dev, inc);
 675                         if (err)
 676                                 return err;
 677                 }
 678         }
 679         return err;
 680 }
 681
 682 /*
 683  * Push the allmulti flag down to all slaves
 684  */
 685 static int bond_set_allmulti(struct bonding *bond, int inc)
 686 {
 687         int err = 0;
 688         if (USES_PRIMARY(bond->params.mode)) {
 689                 /* write lock already acquired */
 690                 if (bond->curr_active_slave) {
 691                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 692                                                inc);
 693                 }
 694         } else {
 695                 struct slave *slave;
 696                 int i;
 697                 bond_for_each_slave(bond, slave, i) {
 698                         err = dev_set_allmulti(slave->dev, inc);
 699                         if (err)
 700                                 return err;
 701                 }
 702         }
 703         return err;
 704 }
 705
 706 /*
 707  * Add a Multicast address to slaves
 708  * according to mode
 709  */
 710 static void bond_mc_add(struct bonding *bond, void *addr)
 711 {
 712         if (USES_PRIMARY(bond->params.mode)) {
 713                 /* write lock already acquired */
 714                 if (bond->curr_active_slave)
 715                         dev_mc_add(bond->curr_active_slave->dev, addr);
 716         } else {
 717                 struct slave *slave;
 718                 int i;
 719
 720                 bond_for_each_slave(bond, slave, i)
 721                         dev_mc_add(slave->dev, addr);
 722         }
 723 }
 724
 725 /*
 726  * Remove a multicast address from slave
 727  * according to mode
 728  */
 729 static void bond_mc_del(struct bonding *bond, void *addr)
 730 {
 731         if (USES_PRIMARY(bond->params.mode)) {
 732                 /* write lock already acquired */
 733                 if (bond->curr_active_slave)
 734                         dev_mc_del(bond->curr_active_slave->dev, addr);
 735         } else {
 736                 struct slave *slave;
 737                 int i;
 738                 bond_for_each_slave(bond, slave, i) {
 739                         dev_mc_del(slave->dev, addr);
 740                 }
 741         }
 742 }
 743
 744
 745 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 746 {
 747         struct in_device *in_dev;
 748
 749         in_dev = __in_dev_get_rcu(dev);
 750         if (in_dev)
 751                 ip_mc_rejoin_groups(in_dev);
 752 }
 753
 754 /*
 755  * Retrieve the list of registered multicast addresses for the bonding
 756  * device and retransmit an IGMP JOIN request to the current active
 757  * slave.
 758  */
 759 static void bond_resend_igmp_join_requests(struct bonding *bond)
 760 {
 761         struct net_device *bond_dev, *vlan_dev, *upper_dev;
 762         struct vlan_entry *vlan;
 763
 764         rcu_read_lock();
 765         read_lock(&bond->lock);
 766
 767         bond_dev = bond->dev;
 768
 769         /* rejoin all groups on bond device */
 770         __bond_resend_igmp_join_requests(bond_dev);
 771
 772         /*
 773          * if bond is enslaved to a bridge,
 774          * then rejoin all groups on its master
 775          */
 776         upper_dev = netdev_master_upper_dev_get_rcu(bond_dev);
 777         if (upper_dev && upper_dev->priv_flags & IFF_EBRIDGE)
 778                 __bond_resend_igmp_join_requests(upper_dev);
 779
 780         /* rejoin all groups on vlan devices */
 781         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 782                 vlan_dev = __vlan_find_dev_deep(bond_dev,
 783                                                 vlan->vlan_id);
 784                 if (vlan_dev)
 785                         __bond_resend_igmp_join_requests(vlan_dev);
 786         }
 787
 788         if (--bond->igmp_retrans > 0)
 789                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 790
 791         read_unlock(&bond->lock);
 792         rcu_read_unlock();
 793 }
 794
 795 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 796 {
 797         struct bonding *bond = container_of(work, struct bonding,
 798                                             mcast_work.work);
 799         rcu_read_lock();
 800         bond_resend_igmp_join_requests(bond);
 801         rcu_read_unlock();
 802 }
 803
 804 /*
 805  * flush all members of flush->mc_list from device dev->mc_list
 806  */
 807 static void bond_mc_list_flush(struct net_device *bond_dev,
 808                                struct net_device *slave_dev)
 809 {
 810         struct bonding *bond = netdev_priv(bond_dev);
 811         struct netdev_hw_addr *ha;
 812
 813         netdev_for_each_mc_addr(ha, bond_dev)
 814                 dev_mc_del(slave_dev, ha->addr);
 815
 816         if (bond->params.mode == BOND_MODE_8023AD) {
 817                 /* del lacpdu mc addr from mc list */
 818                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 819
 820                 dev_mc_del(slave_dev, lacpdu_multicast);
 821         }
 822 }
 823
 824 /*--------------------------- Active slave change ---------------------------*/
 825
 826 /*
 827  * Update the mc list and multicast-related flags for the new and
 828  * old active slaves (if any) according to the multicast mode, and
 829  * promiscuous flags unconditionally.
 830  */
 831 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 832                          struct slave *old_active)
 833 {
 834         struct netdev_hw_addr *ha;
 835
 836         if (!USES_PRIMARY(bond->params.mode))
 837                 /* nothing to do -  mc list is already up-to-date on
 838                  * all slaves
 839                  */
 840                 return;
 841
 842         if (old_active) {
 843                 if (bond->dev->flags & IFF_PROMISC)
 844                         dev_set_promiscuity(old_active->dev, -1);
 845
 846                 if (bond->dev->flags & IFF_ALLMULTI)
 847                         dev_set_allmulti(old_active->dev, -1);
 848
 849                 netdev_for_each_mc_addr(ha, bond->dev)
 850                         dev_mc_del(old_active->dev, ha->addr);
 851         }
 852
 853         if (new_active) {
 854                 /* FIXME: Signal errors upstream. */
 855                 if (bond->dev->flags & IFF_PROMISC)
 856                         dev_set_promiscuity(new_active->dev, 1);
 857
 858                 if (bond->dev->flags & IFF_ALLMULTI)
 859                         dev_set_allmulti(new_active->dev, 1);
 860
 861                 netdev_for_each_mc_addr(ha, bond->dev)
 862                         dev_mc_add(new_active->dev, ha->addr);
 863         }
 864 }
 865
 866 /*
 867  * bond_do_fail_over_mac
 868  *
 869  * Perform special MAC address swapping for fail_over_mac settings
 870  *
 871  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 872  */
 873 static void bond_do_fail_over_mac(struct bonding *bond,
 874                                   struct slave *new_active,
 875                                   struct slave *old_active)
 876         __releases(&bond->curr_slave_lock)
 877         __releases(&bond->lock)
 878         __acquires(&bond->lock)
 879         __acquires(&bond->curr_slave_lock)
 880 {
 881         u8 tmp_mac[ETH_ALEN];
 882         struct sockaddr saddr;
 883         int rv;
 884
 885         switch (bond->params.fail_over_mac) {
 886         case BOND_FOM_ACTIVE:
 887                 if (new_active) {
 888                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 889                                new_active->dev->addr_len);
 890                         write_unlock_bh(&bond->curr_slave_lock);
 891                         read_unlock(&bond->lock);
 892                         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
 893                         read_lock(&bond->lock);
 894                         write_lock_bh(&bond->curr_slave_lock);
 895                 }
 896                 break;
 897         case BOND_FOM_FOLLOW:
 898                 /*
 899                  * if new_active && old_active, swap them
 900                  * if just old_active, do nothing (going to no active slave)
 901                  * if just new_active, set new_active to bond's MAC
 902                  */
 903                 if (!new_active)
 904                         return;
 905
 906                 write_unlock_bh(&bond->curr_slave_lock);
 907                 read_unlock(&bond->lock);
 908
 909                 if (old_active) {
 910                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 911                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 912                                ETH_ALEN);
 913                         saddr.sa_family = new_active->dev->type;
 914                 } else {
 915                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 916                         saddr.sa_family = bond->dev->type;
 917                 }
 918
 919                 rv = dev_set_mac_address(new_active->dev, &saddr);
 920                 if (rv) {
 921                         pr_err("%s: Error %d setting MAC of slave %s\n",
 922                                bond->dev->name, -rv, new_active->dev->name);
 923                         goto out;
 924                 }
 925
 926                 if (!old_active)
 927                         goto out;
 928
 929                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 930                 saddr.sa_family = old_active->dev->type;
 931
 932                 rv = dev_set_mac_address(old_active->dev, &saddr);
 933                 if (rv)
 934                         pr_err("%s: Error %d setting MAC of slave %s\n",
 935                                bond->dev->name, -rv, new_active->dev->name);
 936 out:
 937                 read_lock(&bond->lock);
 938                 write_lock_bh(&bond->curr_slave_lock);
 939                 break;
 940         default:
 941                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 942                        bond->dev->name, bond->params.fail_over_mac);
 943                 break;
 944         }
 945
 946 }
 947
 948 static bool bond_should_change_active(struct bonding *bond)
 949 {
 950         struct slave *prim = bond->primary_slave;
 951         struct slave *curr = bond->curr_active_slave;
 952
 953         if (!prim || !curr || curr->link != BOND_LINK_UP)
 954                 return true;
 955         if (bond->force_primary) {
 956                 bond->force_primary = false;
 957                 return true;
 958         }
 959         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 960             (prim->speed < curr->speed ||
 961              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 962                 return false;
 963         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 964                 return false;
 965         return true;
 966 }
 967
 968 /**
 969  * find_best_interface - select the best available slave to be the active one
 970  * @bond: our bonding struct
 971  *
 972  * Warning: Caller must hold curr_slave_lock for writing.
 973  */
 974 static struct slave *bond_find_best_slave(struct bonding *bond)
 975 {
 976         struct slave *new_active, *old_active;
 977         struct slave *bestslave = NULL;
 978         int mintime = bond->params.updelay;
 979         int i;
 980
 981         new_active = bond->curr_active_slave;
 982
 983         if (!new_active) { /* there were no active slaves left */
 984                 if (bond->slave_cnt > 0)   /* found one slave */
 985                         new_active = bond->first_slave;
 986                 else
 987                         return NULL; /* still no slave, return NULL */
 988         }
 989
 990         if ((bond->primary_slave) &&
 991             bond->primary_slave->link == BOND_LINK_UP &&
 992             bond_should_change_active(bond)) {
 993                 new_active = bond->primary_slave;
 994         }
 995
 996         /* remember where to stop iterating over the slaves */
 997         old_active = new_active;
 998
 999         bond_for_each_slave_from(bond, new_active, i, old_active) {
1000                 if (new_active->link == BOND_LINK_UP) {
1001                         return new_active;
1002                 } else if (new_active->link == BOND_LINK_BACK &&
1003                            IS_UP(new_active->dev)) {
1004                         /* link up, but waiting for stabilization */
1005                         if (new_active->delay < mintime) {
1006                                 mintime = new_active->delay;
1007                                 bestslave = new_active;
1008                         }
1009                 }
1010         }
1011
1012         return bestslave;
1013 }
1014
1015 static bool bond_should_notify_peers(struct bonding *bond)
1016 {
1017         struct slave *slave = bond->curr_active_slave;
1018
1019         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1020                  bond->dev->name, slave ? slave->dev->name : "NULL");
1021
1022         if (!slave || !bond->send_peer_notif ||
1023             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1024                 return false;
1025
1026         bond->send_peer_notif--;
1027         return true;
1028 }
1029
1030 /**
1031  * change_active_interface - change the active slave into the specified one
1032  * @bond: our bonding struct
1033  * @new: the new slave to make the active one
1034  *
1035  * Set the new slave to the bond's settings and unset them on the old
1036  * curr_active_slave.
1037  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1038  *
1039  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1040  * because it is apparently the best available slave we have, even though its
1041  * updelay hasn't timed out yet.
1042  *
1043  * If new_active is not NULL, caller must hold bond->lock for read and
1044  * curr_slave_lock for write_bh.
1045  */
1046 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1047 {
1048         struct slave *old_active = bond->curr_active_slave;
1049
1050         if (old_active == new_active)
1051                 return;
1052
1053         if (new_active) {
1054                 new_active->jiffies = jiffies;
1055
1056                 if (new_active->link == BOND_LINK_BACK) {
1057                         if (USES_PRIMARY(bond->params.mode)) {
1058                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1059                                         bond->dev->name, new_active->dev->name,
1060                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1061                         }
1062
1063                         new_active->delay = 0;
1064                         new_active->link = BOND_LINK_UP;
1065
1066                         if (bond->params.mode == BOND_MODE_8023AD)
1067                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1068
1069                         if (bond_is_lb(bond))
1070                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1071                 } else {
1072                         if (USES_PRIMARY(bond->params.mode)) {
1073                                 pr_info("%s: making interface %s the new active one.\n",
1074                                         bond->dev->name, new_active->dev->name);
1075                         }
1076                 }
1077         }
1078
1079         if (USES_PRIMARY(bond->params.mode))
1080                 bond_mc_swap(bond, new_active, old_active);
1081
1082         if (bond_is_lb(bond)) {
1083                 bond_alb_handle_active_change(bond, new_active);
1084                 if (old_active)
1085                         bond_set_slave_inactive_flags(old_active);
1086                 if (new_active)
1087                         bond_set_slave_active_flags(new_active);
1088         } else {
1089                 bond->curr_active_slave = new_active;
1090         }
1091
1092         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1093                 if (old_active)
1094                         bond_set_slave_inactive_flags(old_active);
1095
1096                 if (new_active) {
1097                         bool should_notify_peers = false;
1098
1099                         bond_set_slave_active_flags(new_active);
1100
1101                         if (bond->params.fail_over_mac)
1102                                 bond_do_fail_over_mac(bond, new_active,
1103                                                       old_active);
1104
1105                         if (netif_running(bond->dev)) {
1106                                 bond->send_peer_notif =
1107                                         bond->params.num_peer_notif;
1108                                 should_notify_peers =
1109                                         bond_should_notify_peers(bond);
1110                         }
1111
1112                         write_unlock_bh(&bond->curr_slave_lock);
1113                         read_unlock(&bond->lock);
1114
1115                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1116                         if (should_notify_peers)
1117                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1118                                                          bond->dev);
1119
1120                         read_lock(&bond->lock);
1121                         write_lock_bh(&bond->curr_slave_lock);
1122                 }
1123         }
1124
1125         /* resend IGMP joins since active slave has changed or
1126          * all were sent on curr_active_slave.
1127          * resend only if bond is brought up with the affected
1128          * bonding modes and the retransmission is enabled */
1129         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1130             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1131              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1132                 bond->igmp_retrans = bond->params.resend_igmp;
1133                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1134         }
1135 }
1136
1137 /**
1138  * bond_select_active_slave - select a new active slave, if needed
1139  * @bond: our bonding struct
1140  *
1141  * This functions should be called when one of the following occurs:
1142  * - The old curr_active_slave has been released or lost its link.
1143  * - The primary_slave has got its link back.
1144  * - A slave has got its link back and there's no old curr_active_slave.
1145  *
1146  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1147  */
1148 void bond_select_active_slave(struct bonding *bond)
1149 {
1150         struct slave *best_slave;
1151         int rv;
1152
1153         best_slave = bond_find_best_slave(bond);
1154         if (best_slave != bond->curr_active_slave) {
1155                 bond_change_active_slave(bond, best_slave);
1156                 rv = bond_set_carrier(bond);
1157                 if (!rv)
1158                         return;
1159
1160                 if (netif_carrier_ok(bond->dev)) {
1161                         pr_info("%s: first active interface up!\n",
1162                                 bond->dev->name);
1163                 } else {
1164                         pr_info("%s: now running without any active interface !\n",
1165                                 bond->dev->name);
1166                 }
1167         }
1168 }
1169
1170 /*--------------------------- slave list handling ---------------------------*/
1171
1172 /*
1173  * This function attaches the slave to the end of list.
1174  *
1175  * bond->lock held for writing by caller.
1176  */
1177 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1178 {
1179         if (bond->first_slave == NULL) { /* attaching the first slave */
1180                 new_slave->next = new_slave;
1181                 new_slave->prev = new_slave;
1182                 bond->first_slave = new_slave;
1183         } else {
1184                 new_slave->next = bond->first_slave;
1185                 new_slave->prev = bond->first_slave->prev;
1186                 new_slave->next->prev = new_slave;
1187                 new_slave->prev->next = new_slave;
1188         }
1189
1190         bond->slave_cnt++;
1191 }
1192
1193 /*
1194  * This function detaches the slave from the list.
1195  * WARNING: no check is made to verify if the slave effectively
1196  * belongs to <bond>.
1197  * Nothing is freed on return, structures are just unchained.
1198  * If any slave pointer in bond was pointing to <slave>,
1199  * it should be changed by the calling function.
1200  *
1201  * bond->lock held for writing by caller.
1202  */
1203 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1204 {
1205         if (slave->next)
1206                 slave->next->prev = slave->prev;
1207
1208         if (slave->prev)
1209                 slave->prev->next = slave->next;
1210
1211         if (bond->first_slave == slave) { /* slave is the first slave */
1212                 if (bond->slave_cnt > 1) { /* there are more slave */
1213                         bond->first_slave = slave->next;
1214                 } else {
1215                         bond->first_slave = NULL; /* slave was the last one */
1216                 }
1217         }
1218
1219         slave->next = NULL;
1220         slave->prev = NULL;
1221         bond->slave_cnt--;
1222 }
1223
1224 #ifdef CONFIG_NET_POLL_CONTROLLER
1225 static inline int slave_enable_netpoll(struct slave *slave)
1226 {
1227         struct netpoll *np;
1228         int err = 0;
1229
1230         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1231         err = -ENOMEM;
1232         if (!np)
1233                 goto out;
1234
1235         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1236         if (err) {
1237                 kfree(np);
1238                 goto out;
1239         }
1240         slave->np = np;
1241 out:
1242         return err;
1243 }
1244 static inline void slave_disable_netpoll(struct slave *slave)
1245 {
1246         struct netpoll *np = slave->np;
1247
1248         if (!np)
1249                 return;
1250
1251         slave->np = NULL;
1252         __netpoll_free_rcu(np);
1253 }
1254 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1255 {
1256         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1257                 return false;
1258         if (!slave_dev->netdev_ops->ndo_poll_controller)
1259                 return false;
1260         return true;
1261 }
1262
1263 static void bond_poll_controller(struct net_device *bond_dev)
1264 {
1265 }
1266
1267 static void __bond_netpoll_cleanup(struct bonding *bond)
1268 {
1269         struct slave *slave;
1270         int i;
1271
1272         bond_for_each_slave(bond, slave, i)
1273                 if (IS_UP(slave->dev))
1274                         slave_disable_netpoll(slave);
1275 }
1276 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1277 {
1278         struct bonding *bond = netdev_priv(bond_dev);
1279
1280         read_lock(&bond->lock);
1281         __bond_netpoll_cleanup(bond);
1282         read_unlock(&bond->lock);
1283 }
1284
1285 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1286 {
1287         struct bonding *bond = netdev_priv(dev);
1288         struct slave *slave;
1289         int i, err = 0;
1290
1291         read_lock(&bond->lock);
1292         bond_for_each_slave(bond, slave, i) {
1293                 err = slave_enable_netpoll(slave);
1294                 if (err) {
1295                         __bond_netpoll_cleanup(bond);
1296                         break;
1297                 }
1298         }
1299         read_unlock(&bond->lock);
1300         return err;
1301 }
1302
1303 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1304 {
1305         return bond->dev->npinfo;
1306 }
1307
1308 #else
1309 static inline int slave_enable_netpoll(struct slave *slave)
1310 {
1311         return 0;
1312 }
1313 static inline void slave_disable_netpoll(struct slave *slave)
1314 {
1315 }
1316 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1317 {
1318 }
1319 #endif
1320
1321 /*---------------------------------- IOCTL ----------------------------------*/
1322
1323 static int bond_sethwaddr(struct net_device *bond_dev,
1324                           struct net_device *slave_dev)
1325 {
1326         pr_debug("bond_dev=%p\n", bond_dev);
1327         pr_debug("slave_dev=%p\n", slave_dev);
1328         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1329         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1330         return 0;
1331 }
1332
1333 static netdev_features_t bond_fix_features(struct net_device *dev,
1334         netdev_features_t features)
1335 {
1336         struct slave *slave;
1337         struct bonding *bond = netdev_priv(dev);
1338         netdev_features_t mask;
1339         int i;
1340
1341         read_lock(&bond->lock);
1342
1343         if (!bond->first_slave) {
1344                 /* Disable adding VLANs to empty bond. But why? --mq */
1345                 features |= NETIF_F_VLAN_CHALLENGED;
1346                 goto out;
1347         }
1348
1349         mask = features;
1350         features &= ~NETIF_F_ONE_FOR_ALL;
1351         features |= NETIF_F_ALL_FOR_ALL;
1352
1353         bond_for_each_slave(bond, slave, i) {
1354                 features = netdev_increment_features(features,
1355                                                      slave->dev->features,
1356                                                      mask);
1357         }
1358
1359 out:
1360         read_unlock(&bond->lock);
1361         return features;
1362 }
1363
1364 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1365                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1366                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1367
1368 static void bond_compute_features(struct bonding *bond)
1369 {
1370         struct slave *slave;
1371         struct net_device *bond_dev = bond->dev;
1372         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1373         unsigned short max_hard_header_len = ETH_HLEN;
1374         unsigned int gso_max_size = GSO_MAX_SIZE;
1375         u16 gso_max_segs = GSO_MAX_SEGS;
1376         int i;
1377         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1378
1379         read_lock(&bond->lock);
1380
1381         if (!bond->first_slave)
1382                 goto done;
1383
1384         bond_for_each_slave(bond, slave, i) {
1385                 vlan_features = netdev_increment_features(vlan_features,
1386                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1387
1388                 dst_release_flag &= slave->dev->priv_flags;
1389                 if (slave->dev->hard_header_len > max_hard_header_len)
1390                         max_hard_header_len = slave->dev->hard_header_len;
1391
1392                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1393                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1394         }
1395
1396 done:
1397         bond_dev->vlan_features = vlan_features;
1398         bond_dev->hard_header_len = max_hard_header_len;
1399         bond_dev->gso_max_segs = gso_max_segs;
1400         netif_set_gso_max_size(bond_dev, gso_max_size);
1401
1402         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1403         bond_dev->priv_flags = flags | dst_release_flag;
1404
1405         read_unlock(&bond->lock);
1406
1407         netdev_change_features(bond_dev);
1408 }
1409
1410 static void bond_setup_by_slave(struct net_device *bond_dev,
1411                                 struct net_device *slave_dev)
1412 {
1413         struct bonding *bond = netdev_priv(bond_dev);
1414
1415         bond_dev->header_ops        = slave_dev->header_ops;
1416
1417         bond_dev->type              = slave_dev->type;
1418         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1419         bond_dev->addr_len          = slave_dev->addr_len;
1420
1421         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1422                 slave_dev->addr_len);
1423         bond->setup_by_slave = 1;
1424 }
1425
1426 /* On bonding slaves other than the currently active slave, suppress
1427  * duplicates except for alb non-mcast/bcast.
1428  */
1429 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1430                                             struct slave *slave,
1431                                             struct bonding *bond)
1432 {
1433         if (bond_is_slave_inactive(slave)) {
1434                 if (bond->params.mode == BOND_MODE_ALB &&
1435                     skb->pkt_type != PACKET_BROADCAST &&
1436                     skb->pkt_type != PACKET_MULTICAST)
1437                         return false;
1438                 return true;
1439         }
1440         return false;
1441 }
1442
1443 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1444 {
1445         struct sk_buff *skb = *pskb;
1446         struct slave *slave;
1447         struct bonding *bond;
1448         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1449                           struct slave *);
1450         int ret = RX_HANDLER_ANOTHER;
1451
1452         skb = skb_share_check(skb, GFP_ATOMIC);
1453         if (unlikely(!skb))
1454                 return RX_HANDLER_CONSUMED;
1455
1456         *pskb = skb;
1457
1458         slave = bond_slave_get_rcu(skb->dev);
1459         bond = slave->bond;
1460
1461         if (bond->params.arp_interval)
1462                 slave->dev->last_rx = jiffies;
1463
1464         recv_probe = ACCESS_ONCE(bond->recv_probe);
1465         if (recv_probe) {
1466                 ret = recv_probe(skb, bond, slave);
1467                 if (ret == RX_HANDLER_CONSUMED) {
1468                         consume_skb(skb);
1469                         return ret;
1470                 }
1471         }
1472
1473         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1474                 return RX_HANDLER_EXACT;
1475         }
1476
1477         skb->dev = bond->dev;
1478
1479         if (bond->params.mode == BOND_MODE_ALB &&
1480             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1481             skb->pkt_type == PACKET_HOST) {
1482
1483                 if (unlikely(skb_cow_head(skb,
1484                                           skb->data - skb_mac_header(skb)))) {
1485                         kfree_skb(skb);
1486                         return RX_HANDLER_CONSUMED;
1487                 }
1488                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1489         }
1490
1491         return ret;
1492 }
1493
1494 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1495                                       struct net_device *slave_dev)
1496 {
1497         int err;
1498
1499         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1500         if (err)
1501                 return err;
1502         slave_dev->flags |= IFF_SLAVE;
1503         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1504         return 0;
1505 }
1506
1507 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1508                                   struct net_device *slave_dev)
1509 {
1510         netdev_upper_dev_unlink(slave_dev, bond_dev);
1511         slave_dev->flags &= ~IFF_SLAVE;
1512         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1513 }
1514
1515 /* enslave device <slave> to bond device <master> */
1516 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1517 {
1518         struct bonding *bond = netdev_priv(bond_dev);
1519         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1520         struct slave *new_slave = NULL;
1521         struct netdev_hw_addr *ha;
1522         struct sockaddr addr;
1523         int link_reporting;
1524         int res = 0;
1525
1526         if (!bond->params.use_carrier &&
1527             slave_dev->ethtool_ops->get_link == NULL &&
1528             slave_ops->ndo_do_ioctl == NULL) {
1529                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1530                            bond_dev->name, slave_dev->name);
1531         }
1532
1533         /* already enslaved */
1534         if (slave_dev->flags & IFF_SLAVE) {
1535                 pr_debug("Error, Device was already enslaved\n");
1536                 return -EBUSY;
1537         }
1538
1539         /* vlan challenged mutual exclusion */
1540         /* no need to lock since we're protected by rtnl_lock */
1541         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1542                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1543                 if (vlan_uses_dev(bond_dev)) {
1544                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1545                                bond_dev->name, slave_dev->name, bond_dev->name);
1546                         return -EPERM;
1547                 } else {
1548                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1549                                    bond_dev->name, slave_dev->name,
1550                                    slave_dev->name, bond_dev->name);
1551                 }
1552         } else {
1553                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1554         }
1555
1556         /*
1557          * Old ifenslave binaries are no longer supported.  These can
1558          * be identified with moderate accuracy by the state of the slave:
1559          * the current ifenslave will set the interface down prior to
1560          * enslaving it; the old ifenslave will not.
1561          */
1562         if ((slave_dev->flags & IFF_UP)) {
1563                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1564                        slave_dev->name);
1565                 res = -EPERM;
1566                 goto err_undo_flags;
1567         }
1568
1569         /* set bonding device ether type by slave - bonding netdevices are
1570          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1571          * there is a need to override some of the type dependent attribs/funcs.
1572          *
1573          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1574          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1575          */
1576         if (bond->slave_cnt == 0) {
1577                 if (bond_dev->type != slave_dev->type) {
1578                         pr_debug("%s: change device type from %d to %d\n",
1579                                  bond_dev->name,
1580                                  bond_dev->type, slave_dev->type);
1581
1582                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1583                                                        bond_dev);
1584                         res = notifier_to_errno(res);
1585                         if (res) {
1586                                 pr_err("%s: refused to change device type\n",
1587                                        bond_dev->name);
1588                                 res = -EBUSY;
1589                                 goto err_undo_flags;
1590                         }
1591
1592                         /* Flush unicast and multicast addresses */
1593                         dev_uc_flush(bond_dev);
1594                         dev_mc_flush(bond_dev);
1595
1596                         if (slave_dev->type != ARPHRD_ETHER)
1597                                 bond_setup_by_slave(bond_dev, slave_dev);
1598                         else {
1599                                 ether_setup(bond_dev);
1600                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1601                         }
1602
1603                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1604                                                  bond_dev);
1605                 }
1606         } else if (bond_dev->type != slave_dev->type) {
1607                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1608                        slave_dev->name,
1609                        slave_dev->type, bond_dev->type);
1610                 res = -EINVAL;
1611                 goto err_undo_flags;
1612         }
1613
1614         if (slave_ops->ndo_set_mac_address == NULL) {
1615                 if (bond->slave_cnt == 0) {
1616                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1617                                    bond_dev->name);
1618                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1619                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1620                         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",
1621                                bond_dev->name);
1622                         res = -EOPNOTSUPP;
1623                         goto err_undo_flags;
1624                 }
1625         }
1626
1627         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1628
1629         /* If this is the first slave, then we need to set the master's hardware
1630          * address to be the same as the slave's. */
1631         if (is_zero_ether_addr(bond->dev->dev_addr))
1632                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1633                        slave_dev->addr_len);
1634
1635
1636         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1637         if (!new_slave) {
1638                 res = -ENOMEM;
1639                 goto err_undo_flags;
1640         }
1641
1642         /*
1643          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1644          * is set via sysfs or module option if desired.
1645          */
1646         new_slave->queue_id = 0;
1647
1648         /* Save slave's original mtu and then set it to match the bond */
1649         new_slave->original_mtu = slave_dev->mtu;
1650         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1651         if (res) {
1652                 pr_debug("Error %d calling dev_set_mtu\n", res);
1653                 goto err_free;
1654         }
1655
1656         /*
1657          * Save slave's original ("permanent") mac address for modes
1658          * that need it, and for restoring it upon release, and then
1659          * set it to the master's address
1660          */
1661         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1662
1663         if (!bond->params.fail_over_mac) {
1664                 /*
1665                  * Set slave to master's mac address.  The application already
1666                  * set the master's mac address to that of the first slave
1667                  */
1668                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1669                 addr.sa_family = slave_dev->type;
1670                 res = dev_set_mac_address(slave_dev, &addr);
1671                 if (res) {
1672                         pr_debug("Error %d calling set_mac_address\n", res);
1673                         goto err_restore_mtu;
1674                 }
1675         }
1676
1677         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1678         if (res) {
1679                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1680                 goto err_restore_mac;
1681         }
1682
1683         /* open the slave since the application closed it */
1684         res = dev_open(slave_dev);
1685         if (res) {
1686                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1687                 goto err_unset_master;
1688         }
1689
1690         new_slave->bond = bond;
1691         new_slave->dev = slave_dev;
1692         slave_dev->priv_flags |= IFF_BONDING;
1693
1694         if (bond_is_lb(bond)) {
1695                 /* bond_alb_init_slave() must be called before all other stages since
1696                  * it might fail and we do not want to have to undo everything
1697                  */
1698                 res = bond_alb_init_slave(bond, new_slave);
1699                 if (res)
1700                         goto err_close;
1701         }
1702
1703         /* If the mode USES_PRIMARY, then the new slave gets the
1704          * master's promisc (and mc) settings only if it becomes the
1705          * curr_active_slave, and that is taken care of later when calling
1706          * bond_change_active()
1707          */
1708         if (!USES_PRIMARY(bond->params.mode)) {
1709                 /* set promiscuity level to new slave */
1710                 if (bond_dev->flags & IFF_PROMISC) {
1711                         res = dev_set_promiscuity(slave_dev, 1);
1712                         if (res)
1713                                 goto err_close;
1714                 }
1715
1716                 /* set allmulti level to new slave */
1717                 if (bond_dev->flags & IFF_ALLMULTI) {
1718                         res = dev_set_allmulti(slave_dev, 1);
1719                         if (res)
1720                                 goto err_close;
1721                 }
1722
1723                 netif_addr_lock_bh(bond_dev);
1724                 /* upload master's mc_list to new slave */
1725                 netdev_for_each_mc_addr(ha, bond_dev)
1726                         dev_mc_add(slave_dev, ha->addr);
1727                 netif_addr_unlock_bh(bond_dev);
1728         }
1729
1730         if (bond->params.mode == BOND_MODE_8023AD) {
1731                 /* add lacpdu mc addr to mc list */
1732                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1733
1734                 dev_mc_add(slave_dev, lacpdu_multicast);
1735         }
1736
1737         bond_add_vlans_on_slave(bond, slave_dev);
1738
1739         write_lock_bh(&bond->lock);
1740
1741         bond_attach_slave(bond, new_slave);
1742
1743         new_slave->delay = 0;
1744         new_slave->link_failure_count = 0;
1745
1746         write_unlock_bh(&bond->lock);
1747
1748         bond_compute_features(bond);
1749
1750         read_lock(&bond->lock);
1751
1752         new_slave->last_arp_rx = jiffies -
1753                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1754
1755         if (bond->params.miimon && !bond->params.use_carrier) {
1756                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1757
1758                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1759                         /*
1760                          * miimon is set but a bonded network driver
1761                          * does not support ETHTOOL/MII and
1762                          * arp_interval is not set.  Note: if
1763                          * use_carrier is enabled, we will never go
1764                          * here (because netif_carrier is always
1765                          * supported); thus, we don't need to change
1766                          * the messages for netif_carrier.
1767                          */
1768                         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",
1769                                bond_dev->name, slave_dev->name);
1770                 } else if (link_reporting == -1) {
1771                         /* unable get link status using mii/ethtool */
1772                         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",
1773                                    bond_dev->name, slave_dev->name);
1774                 }
1775         }
1776
1777         /* check for initial state */
1778         if (bond->params.miimon) {
1779                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1780                         if (bond->params.updelay) {
1781                                 new_slave->link = BOND_LINK_BACK;
1782                                 new_slave->delay = bond->params.updelay;
1783                         } else {
1784                                 new_slave->link = BOND_LINK_UP;
1785                         }
1786                 } else {
1787                         new_slave->link = BOND_LINK_DOWN;
1788                 }
1789         } else if (bond->params.arp_interval) {
1790                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1791                         BOND_LINK_UP : BOND_LINK_DOWN);
1792         } else {
1793                 new_slave->link = BOND_LINK_UP;
1794         }
1795
1796         if (new_slave->link != BOND_LINK_DOWN)
1797                 new_slave->jiffies = jiffies;
1798         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1799                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1800                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1801
1802         bond_update_speed_duplex(new_slave);
1803
1804         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1805                 /* if there is a primary slave, remember it */
1806                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1807                         bond->primary_slave = new_slave;
1808                         bond->force_primary = true;
1809                 }
1810         }
1811
1812         write_lock_bh(&bond->curr_slave_lock);
1813
1814         switch (bond->params.mode) {
1815         case BOND_MODE_ACTIVEBACKUP:
1816                 bond_set_slave_inactive_flags(new_slave);
1817                 bond_select_active_slave(bond);
1818                 break;
1819         case BOND_MODE_8023AD:
1820                 /* in 802.3ad mode, the internal mechanism
1821                  * will activate the slaves in the selected
1822                  * aggregator
1823                  */
1824                 bond_set_slave_inactive_flags(new_slave);
1825                 /* if this is the first slave */
1826                 if (bond->slave_cnt == 1) {
1827                         SLAVE_AD_INFO(new_slave).id = 1;
1828                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1829                          * can be called only after the mac address of the bond is set
1830                          */
1831                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1832                 } else {
1833                         SLAVE_AD_INFO(new_slave).id =
1834                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1835                 }
1836
1837                 bond_3ad_bind_slave(new_slave);
1838                 break;
1839         case BOND_MODE_TLB:
1840         case BOND_MODE_ALB:
1841                 bond_set_active_slave(new_slave);
1842                 bond_set_slave_inactive_flags(new_slave);
1843                 bond_select_active_slave(bond);
1844                 break;
1845         default:
1846                 pr_debug("This slave is always active in trunk mode\n");
1847
1848                 /* always active in trunk mode */
1849                 bond_set_active_slave(new_slave);
1850
1851                 /* In trunking mode there is little meaning to curr_active_slave
1852                  * anyway (it holds no special properties of the bond device),
1853                  * so we can change it without calling change_active_interface()
1854                  */
1855                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1856                         bond->curr_active_slave = new_slave;
1857
1858                 break;
1859         } /* switch(bond_mode) */
1860
1861         write_unlock_bh(&bond->curr_slave_lock);
1862
1863         bond_set_carrier(bond);
1864
1865 #ifdef CONFIG_NET_POLL_CONTROLLER
1866         slave_dev->npinfo = bond_netpoll_info(bond);
1867         if (slave_dev->npinfo) {
1868                 if (slave_enable_netpoll(new_slave)) {
1869                         read_unlock(&bond->lock);
1870                         pr_info("Error, %s: master_dev is using netpoll, "
1871                                  "but new slave device does not support netpoll.\n",
1872                                  bond_dev->name);
1873                         res = -EBUSY;
1874                         goto err_detach;
1875                 }
1876         }
1877 #endif
1878
1879         read_unlock(&bond->lock);
1880
1881         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1882         if (res)
1883                 goto err_detach;
1884
1885         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1886                                          new_slave);
1887         if (res) {
1888                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1889                 goto err_dest_symlinks;
1890         }
1891
1892         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1893                 bond_dev->name, slave_dev->name,
1894                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1895                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1896
1897         /* enslave is successful */
1898         return 0;
1899
1900 /* Undo stages on error */
1901 err_dest_symlinks:
1902         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1903
1904 err_detach:
1905         write_lock_bh(&bond->lock);
1906         bond_detach_slave(bond, new_slave);
1907         write_unlock_bh(&bond->lock);
1908
1909 err_close:
1910         dev_close(slave_dev);
1911
1912 err_unset_master:
1913         bond_upper_dev_unlink(bond_dev, slave_dev);
1914
1915 err_restore_mac:
1916         if (!bond->params.fail_over_mac) {
1917                 /* XXX TODO - fom follow mode needs to change master's
1918                  * MAC if this slave's MAC is in use by the bond, or at
1919                  * least print a warning.
1920                  */
1921                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1922                 addr.sa_family = slave_dev->type;
1923                 dev_set_mac_address(slave_dev, &addr);
1924         }
1925
1926 err_restore_mtu:
1927         dev_set_mtu(slave_dev, new_slave->original_mtu);
1928
1929 err_free:
1930         kfree(new_slave);
1931
1932 err_undo_flags:
1933         bond_compute_features(bond);
1934
1935         return res;
1936 }
1937
1938 /*
1939  * Try to release the slave device <slave> from the bond device <master>
1940  * It is legal to access curr_active_slave without a lock because all the function
1941  * is write-locked.
1942  *
1943  * The rules for slave state should be:
1944  *   for Active/Backup:
1945  *     Active stays on all backups go down
1946  *   for Bonded connections:
1947  *     The first up interface should be left on and all others downed.
1948  */
1949 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1950 {
1951         struct bonding *bond = netdev_priv(bond_dev);
1952         struct slave *slave, *oldcurrent;
1953         struct sockaddr addr;
1954         netdev_features_t old_features = bond_dev->features;
1955
1956         /* slave is not a slave or master is not master of this slave */
1957         if (!(slave_dev->flags & IFF_SLAVE) ||
1958             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1959                 pr_err("%s: Error: cannot release %s.\n",
1960                        bond_dev->name, slave_dev->name);
1961                 return -EINVAL;
1962         }
1963
1964         block_netpoll_tx();
1965         call_netdevice_notifiers(NETDEV_RELEASE, bond_dev);
1966         write_lock_bh(&bond->lock);
1967
1968         slave = bond_get_slave_by_dev(bond, slave_dev);
1969         if (!slave) {
1970                 /* not a slave of this bond */
1971                 pr_info("%s: %s not enslaved\n",
1972                         bond_dev->name, slave_dev->name);
1973                 write_unlock_bh(&bond->lock);
1974                 unblock_netpoll_tx();
1975                 return -EINVAL;
1976         }
1977
1978         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1979          * for this slave anymore.
1980          */
1981         netdev_rx_handler_unregister(slave_dev);
1982         write_unlock_bh(&bond->lock);
1983         synchronize_net();
1984         write_lock_bh(&bond->lock);
1985
1986         if (!bond->params.fail_over_mac) {
1987                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1988                     bond->slave_cnt > 1)
1989                         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",
1990                                    bond_dev->name, slave_dev->name,
1991                                    slave->perm_hwaddr,
1992                                    bond_dev->name, slave_dev->name);
1993         }
1994
1995         /* Inform AD package of unbinding of slave. */
1996         if (bond->params.mode == BOND_MODE_8023AD) {
1997                 /* must be called before the slave is
1998                  * detached from the list
1999                  */
2000                 bond_3ad_unbind_slave(slave);
2001         }
2002
2003         pr_info("%s: releasing %s interface %s\n",
2004                 bond_dev->name,
2005                 bond_is_active_slave(slave) ? "active" : "backup",
2006                 slave_dev->name);
2007
2008         oldcurrent = bond->curr_active_slave;
2009
2010         bond->current_arp_slave = NULL;
2011
2012         /* release the slave from its bond */
2013         bond_detach_slave(bond, slave);
2014
2015         if (bond->primary_slave == slave)
2016                 bond->primary_slave = NULL;
2017
2018         if (oldcurrent == slave)
2019                 bond_change_active_slave(bond, NULL);
2020
2021         if (bond_is_lb(bond)) {
2022                 /* Must be called only after the slave has been
2023                  * detached from the list and the curr_active_slave
2024                  * has been cleared (if our_slave == old_current),
2025                  * but before a new active slave is selected.
2026                  */
2027                 write_unlock_bh(&bond->lock);
2028                 bond_alb_deinit_slave(bond, slave);
2029                 write_lock_bh(&bond->lock);
2030         }
2031
2032         if (oldcurrent == slave) {
2033                 /*
2034                  * Note that we hold RTNL over this sequence, so there
2035                  * is no concern that another slave add/remove event
2036                  * will interfere.
2037                  */
2038                 write_unlock_bh(&bond->lock);
2039                 read_lock(&bond->lock);
2040                 write_lock_bh(&bond->curr_slave_lock);
2041
2042                 bond_select_active_slave(bond);
2043
2044                 write_unlock_bh(&bond->curr_slave_lock);
2045                 read_unlock(&bond->lock);
2046                 write_lock_bh(&bond->lock);
2047         }
2048
2049         if (bond->slave_cnt == 0) {
2050                 bond_set_carrier(bond);
2051
2052                 /* if the last slave was removed, zero the mac address
2053                  * of the master so it will be set by the application
2054                  * to the mac address of the first slave
2055                  */
2056                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2057
2058                 if (bond_vlan_used(bond)) {
2059                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2060                                    bond_dev->name, bond_dev->name);
2061                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2062                                    bond_dev->name);
2063                 }
2064         }
2065
2066         write_unlock_bh(&bond->lock);
2067         unblock_netpoll_tx();
2068
2069         if (bond->slave_cnt == 0)
2070                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2071
2072         bond_compute_features(bond);
2073         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2074             (old_features & NETIF_F_VLAN_CHALLENGED))
2075                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2076                         bond_dev->name, slave_dev->name, bond_dev->name);
2077
2078         /* must do this from outside any spinlocks */
2079         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2080
2081         bond_del_vlans_from_slave(bond, slave_dev);
2082
2083         /* If the mode USES_PRIMARY, then we should only remove its
2084          * promisc and mc settings if it was the curr_active_slave, but that was
2085          * already taken care of above when we detached the slave
2086          */
2087         if (!USES_PRIMARY(bond->params.mode)) {
2088                 /* unset promiscuity level from slave */
2089                 if (bond_dev->flags & IFF_PROMISC)
2090                         dev_set_promiscuity(slave_dev, -1);
2091
2092                 /* unset allmulti level from slave */
2093                 if (bond_dev->flags & IFF_ALLMULTI)
2094                         dev_set_allmulti(slave_dev, -1);
2095
2096                 /* flush master's mc_list from slave */
2097                 netif_addr_lock_bh(bond_dev);
2098                 bond_mc_list_flush(bond_dev, slave_dev);
2099                 netif_addr_unlock_bh(bond_dev);
2100         }
2101
2102         bond_upper_dev_unlink(bond_dev, slave_dev);
2103
2104         slave_disable_netpoll(slave);
2105
2106         /* close slave before restoring its mac address */
2107         dev_close(slave_dev);
2108
2109         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2110                 /* restore original ("permanent") mac address */
2111                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2112                 addr.sa_family = slave_dev->type;
2113                 dev_set_mac_address(slave_dev, &addr);
2114         }
2115
2116         dev_set_mtu(slave_dev, slave->original_mtu);
2117
2118         slave_dev->priv_flags &= ~IFF_BONDING;
2119
2120         kfree(slave);
2121
2122         return 0;  /* deletion OK */
2123 }
2124
2125 /*
2126 * First release a slave and then destroy the bond if no more slaves are left.
2127 * Must be under rtnl_lock when this function is called.
2128 */
2129 static int  bond_release_and_destroy(struct net_device *bond_dev,
2130                                      struct net_device *slave_dev)
2131 {
2132         struct bonding *bond = netdev_priv(bond_dev);
2133         int ret;
2134
2135         ret = bond_release(bond_dev, slave_dev);
2136         if ((ret == 0) && (bond->slave_cnt == 0)) {
2137                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2138                 pr_info("%s: destroying bond %s.\n",
2139                         bond_dev->name, bond_dev->name);
2140                 unregister_netdevice(bond_dev);
2141         }
2142         return ret;
2143 }
2144
2145 /*
2146  * This function releases all slaves.
2147  */
2148 static int bond_release_all(struct net_device *bond_dev)
2149 {
2150         struct bonding *bond = netdev_priv(bond_dev);
2151         struct slave *slave;
2152         struct net_device *slave_dev;
2153         struct sockaddr addr;
2154
2155         write_lock_bh(&bond->lock);
2156
2157         netif_carrier_off(bond_dev);
2158
2159         if (bond->slave_cnt == 0)
2160                 goto out;
2161
2162         bond->current_arp_slave = NULL;
2163         bond->primary_slave = NULL;
2164         bond_change_active_slave(bond, NULL);
2165
2166         while ((slave = bond->first_slave) != NULL) {
2167                 /* Inform AD package of unbinding of slave
2168                  * before slave is detached from the list.
2169                  */
2170                 if (bond->params.mode == BOND_MODE_8023AD)
2171                         bond_3ad_unbind_slave(slave);
2172
2173                 slave_dev = slave->dev;
2174                 bond_detach_slave(bond, slave);
2175
2176                 /* now that the slave is detached, unlock and perform
2177                  * all the undo steps that should not be called from
2178                  * within a lock.
2179                  */
2180                 write_unlock_bh(&bond->lock);
2181
2182                 /* unregister rx_handler early so bond_handle_frame wouldn't
2183                  * be called for this slave anymore.
2184                  */
2185                 netdev_rx_handler_unregister(slave_dev);
2186                 synchronize_net();
2187
2188                 if (bond_is_lb(bond)) {
2189                         /* must be called only after the slave
2190                          * has been detached from the list
2191                          */
2192                         bond_alb_deinit_slave(bond, slave);
2193                 }
2194
2195                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2196                 bond_del_vlans_from_slave(bond, slave_dev);
2197
2198                 /* If the mode USES_PRIMARY, then we should only remove its
2199                  * promisc and mc settings if it was the curr_active_slave, but that was
2200                  * already taken care of above when we detached the slave
2201                  */
2202                 if (!USES_PRIMARY(bond->params.mode)) {
2203                         /* unset promiscuity level from slave */
2204                         if (bond_dev->flags & IFF_PROMISC)
2205                                 dev_set_promiscuity(slave_dev, -1);
2206
2207                         /* unset allmulti level from slave */
2208                         if (bond_dev->flags & IFF_ALLMULTI)
2209                                 dev_set_allmulti(slave_dev, -1);
2210
2211                         /* flush master's mc_list from slave */
2212                         netif_addr_lock_bh(bond_dev);
2213                         bond_mc_list_flush(bond_dev, slave_dev);
2214                         netif_addr_unlock_bh(bond_dev);
2215                 }
2216
2217                 bond_upper_dev_unlink(bond_dev, slave_dev);
2218
2219                 slave_disable_netpoll(slave);
2220
2221                 /* close slave before restoring its mac address */
2222                 dev_close(slave_dev);
2223
2224                 if (!bond->params.fail_over_mac) {
2225                         /* restore original ("permanent") mac address*/
2226                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2227                         addr.sa_family = slave_dev->type;
2228                         dev_set_mac_address(slave_dev, &addr);
2229                 }
2230
2231                 kfree(slave);
2232
2233                 /* re-acquire the lock before getting the next slave */
2234                 write_lock_bh(&bond->lock);
2235         }
2236
2237         /* zero the mac address of the master so it will be
2238          * set by the application to the mac address of the
2239          * first slave
2240          */
2241         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2242
2243         if (bond_vlan_used(bond)) {
2244                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2245                            bond_dev->name, bond_dev->name);
2246                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2247                            bond_dev->name);
2248         }
2249
2250         pr_info("%s: released all slaves\n", bond_dev->name);
2251
2252 out:
2253         write_unlock_bh(&bond->lock);
2254
2255         bond_compute_features(bond);
2256
2257         return 0;
2258 }
2259
2260 /*
2261  * This function changes the active slave to slave <slave_dev>.
2262  * It returns -EINVAL in the following cases.
2263  *  - <slave_dev> is not found in the list.
2264  *  - There is not active slave now.
2265  *  - <slave_dev> is already active.
2266  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2267  *  - <slave_dev> is not running.
2268  * In these cases, this function does nothing.
2269  * In the other cases, current_slave pointer is changed and 0 is returned.
2270  */
2271 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2272 {
2273         struct bonding *bond = netdev_priv(bond_dev);
2274         struct slave *old_active = NULL;
2275         struct slave *new_active = NULL;
2276         int res = 0;
2277
2278         if (!USES_PRIMARY(bond->params.mode))
2279                 return -EINVAL;
2280
2281         /* Verify that bond_dev is indeed the master of slave_dev */
2282         if (!(slave_dev->flags & IFF_SLAVE) ||
2283             !netdev_has_upper_dev(slave_dev, bond_dev))
2284                 return -EINVAL;
2285
2286         read_lock(&bond->lock);
2287
2288         read_lock(&bond->curr_slave_lock);
2289         old_active = bond->curr_active_slave;
2290         read_unlock(&bond->curr_slave_lock);
2291
2292         new_active = bond_get_slave_by_dev(bond, slave_dev);
2293
2294         /*
2295          * Changing to the current active: do nothing; return success.
2296          */
2297         if (new_active && (new_active == old_active)) {
2298                 read_unlock(&bond->lock);
2299                 return 0;
2300         }
2301
2302         if ((new_active) &&
2303             (old_active) &&
2304             (new_active->link == BOND_LINK_UP) &&
2305             IS_UP(new_active->dev)) {
2306                 block_netpoll_tx();
2307                 write_lock_bh(&bond->curr_slave_lock);
2308                 bond_change_active_slave(bond, new_active);
2309                 write_unlock_bh(&bond->curr_slave_lock);
2310                 unblock_netpoll_tx();
2311         } else
2312                 res = -EINVAL;
2313
2314         read_unlock(&bond->lock);
2315
2316         return res;
2317 }
2318
2319 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2320 {
2321         struct bonding *bond = netdev_priv(bond_dev);
2322
2323         info->bond_mode = bond->params.mode;
2324         info->miimon = bond->params.miimon;
2325
2326         read_lock(&bond->lock);
2327         info->num_slaves = bond->slave_cnt;
2328         read_unlock(&bond->lock);
2329
2330         return 0;
2331 }
2332
2333 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2334 {
2335         struct bonding *bond = netdev_priv(bond_dev);
2336         struct slave *slave;
2337         int i, res = -ENODEV;
2338
2339         read_lock(&bond->lock);
2340
2341         bond_for_each_slave(bond, slave, i) {
2342                 if (i == (int)info->slave_id) {
2343                         res = 0;
2344                         strcpy(info->slave_name, slave->dev->name);
2345                         info->link = slave->link;
2346                         info->state = bond_slave_state(slave);
2347                         info->link_failure_count = slave->link_failure_count;
2348                         break;
2349                 }
2350         }
2351
2352         read_unlock(&bond->lock);
2353
2354         return res;
2355 }
2356
2357 /*-------------------------------- Monitoring -------------------------------*/
2358
2359
2360 static int bond_miimon_inspect(struct bonding *bond)
2361 {
2362         struct slave *slave;
2363         int i, link_state, commit = 0;
2364         bool ignore_updelay;
2365
2366         ignore_updelay = !bond->curr_active_slave ? true : false;
2367
2368         bond_for_each_slave(bond, slave, i) {
2369                 slave->new_link = BOND_LINK_NOCHANGE;
2370
2371                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2372
2373                 switch (slave->link) {
2374                 case BOND_LINK_UP:
2375                         if (link_state)
2376                                 continue;
2377
2378                         slave->link = BOND_LINK_FAIL;
2379                         slave->delay = bond->params.downdelay;
2380                         if (slave->delay) {
2381                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2382                                         bond->dev->name,
2383                                         (bond->params.mode ==
2384                                          BOND_MODE_ACTIVEBACKUP) ?
2385                                         (bond_is_active_slave(slave) ?
2386                                          "active " : "backup ") : "",
2387                                         slave->dev->name,
2388                                         bond->params.downdelay * bond->params.miimon);
2389                         }
2390                         /*FALLTHRU*/
2391                 case BOND_LINK_FAIL:
2392                         if (link_state) {
2393                                 /*
2394                                  * recovered before downdelay expired
2395                                  */
2396                                 slave->link = BOND_LINK_UP;
2397                                 slave->jiffies = jiffies;
2398                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2399                                         bond->dev->name,
2400                                         (bond->params.downdelay - slave->delay) *
2401                                         bond->params.miimon,
2402                                         slave->dev->name);
2403                                 continue;
2404                         }
2405
2406                         if (slave->delay <= 0) {
2407                                 slave->new_link = BOND_LINK_DOWN;
2408                                 commit++;
2409                                 continue;
2410                         }
2411
2412                         slave->delay--;
2413                         break;
2414
2415                 case BOND_LINK_DOWN:
2416                         if (!link_state)
2417                                 continue;
2418
2419                         slave->link = BOND_LINK_BACK;
2420                         slave->delay = bond->params.updelay;
2421
2422                         if (slave->delay) {
2423                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2424                                         bond->dev->name, slave->dev->name,
2425                                         ignore_updelay ? 0 :
2426                                         bond->params.updelay *
2427                                         bond->params.miimon);
2428                         }
2429                         /*FALLTHRU*/
2430                 case BOND_LINK_BACK:
2431                         if (!link_state) {
2432                                 slave->link = BOND_LINK_DOWN;
2433                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2434                                         bond->dev->name,
2435                                         (bond->params.updelay - slave->delay) *
2436                                         bond->params.miimon,
2437                                         slave->dev->name);
2438
2439                                 continue;
2440                         }
2441
2442                         if (ignore_updelay)
2443                                 slave->delay = 0;
2444
2445                         if (slave->delay <= 0) {
2446                                 slave->new_link = BOND_LINK_UP;
2447                                 commit++;
2448                                 ignore_updelay = false;
2449                                 continue;
2450                         }
2451
2452                         slave->delay--;
2453                         break;
2454                 }
2455         }
2456
2457         return commit;
2458 }
2459
2460 static void bond_miimon_commit(struct bonding *bond)
2461 {
2462         struct slave *slave;
2463         int i;
2464
2465         bond_for_each_slave(bond, slave, i) {
2466                 switch (slave->new_link) {
2467                 case BOND_LINK_NOCHANGE:
2468                         continue;
2469
2470                 case BOND_LINK_UP:
2471                         slave->link = BOND_LINK_UP;
2472                         slave->jiffies = jiffies;
2473
2474                         if (bond->params.mode == BOND_MODE_8023AD) {
2475                                 /* prevent it from being the active one */
2476                                 bond_set_backup_slave(slave);
2477                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2478                                 /* make it immediately active */
2479                                 bond_set_active_slave(slave);
2480                         } else if (slave != bond->primary_slave) {
2481                                 /* prevent it from being the active one */
2482                                 bond_set_backup_slave(slave);
2483                         }
2484
2485                         bond_update_speed_duplex(slave);
2486
2487                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2488                                 bond->dev->name, slave->dev->name,
2489                                 slave->speed, slave->duplex ? "full" : "half");
2490
2491                         /* notify ad that the link status has changed */
2492                         if (bond->params.mode == BOND_MODE_8023AD)
2493                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2494
2495                         if (bond_is_lb(bond))
2496                                 bond_alb_handle_link_change(bond, slave,
2497                                                             BOND_LINK_UP);
2498
2499                         if (!bond->curr_active_slave ||
2500                             (slave == bond->primary_slave))
2501                                 goto do_failover;
2502
2503                         continue;
2504
2505                 case BOND_LINK_DOWN:
2506                         if (slave->link_failure_count < UINT_MAX)
2507                                 slave->link_failure_count++;
2508
2509                         slave->link = BOND_LINK_DOWN;
2510
2511                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2512                             bond->params.mode == BOND_MODE_8023AD)
2513                                 bond_set_slave_inactive_flags(slave);
2514
2515                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2516                                 bond->dev->name, slave->dev->name);
2517
2518                         if (bond->params.mode == BOND_MODE_8023AD)
2519                                 bond_3ad_handle_link_change(slave,
2520                                                             BOND_LINK_DOWN);
2521
2522                         if (bond_is_lb(bond))
2523                                 bond_alb_handle_link_change(bond, slave,
2524                                                             BOND_LINK_DOWN);
2525
2526                         if (slave == bond->curr_active_slave)
2527                                 goto do_failover;
2528
2529                         continue;
2530
2531                 default:
2532                         pr_err("%s: invalid new link %d on slave %s\n",
2533                                bond->dev->name, slave->new_link,
2534                                slave->dev->name);
2535                         slave->new_link = BOND_LINK_NOCHANGE;
2536
2537                         continue;
2538                 }
2539
2540 do_failover:
2541                 ASSERT_RTNL();
2542                 block_netpoll_tx();
2543                 write_lock_bh(&bond->curr_slave_lock);
2544                 bond_select_active_slave(bond);
2545                 write_unlock_bh(&bond->curr_slave_lock);
2546                 unblock_netpoll_tx();
2547         }
2548
2549         bond_set_carrier(bond);
2550 }
2551
2552 /*
2553  * bond_mii_monitor
2554  *
2555  * Really a wrapper that splits the mii monitor into two phases: an
2556  * inspection, then (if inspection indicates something needs to be done)
2557  * an acquisition of appropriate locks followed by a commit phase to
2558  * implement whatever link state changes are indicated.
2559  */
2560 void bond_mii_monitor(struct work_struct *work)
2561 {
2562         struct bonding *bond = container_of(work, struct bonding,
2563                                             mii_work.work);
2564         bool should_notify_peers = false;
2565         unsigned long delay;
2566
2567         read_lock(&bond->lock);
2568
2569         delay = msecs_to_jiffies(bond->params.miimon);
2570
2571         if (bond->slave_cnt == 0)
2572                 goto re_arm;
2573
2574         should_notify_peers = bond_should_notify_peers(bond);
2575
2576         if (bond_miimon_inspect(bond)) {
2577                 read_unlock(&bond->lock);
2578
2579                 /* Race avoidance with bond_close cancel of workqueue */
2580                 if (!rtnl_trylock()) {
2581                         read_lock(&bond->lock);
2582                         delay = 1;
2583                         should_notify_peers = false;
2584                         goto re_arm;
2585                 }
2586
2587                 read_lock(&bond->lock);
2588
2589                 bond_miimon_commit(bond);
2590
2591                 read_unlock(&bond->lock);
2592                 rtnl_unlock();  /* might sleep, hold no other locks */
2593                 read_lock(&bond->lock);
2594         }
2595
2596 re_arm:
2597         if (bond->params.miimon)
2598                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2599
2600         read_unlock(&bond->lock);
2601
2602         if (should_notify_peers) {
2603                 if (!rtnl_trylock()) {
2604                         read_lock(&bond->lock);
2605                         bond->send_peer_notif++;
2606                         read_unlock(&bond->lock);
2607                         return;
2608                 }
2609                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2610                 rtnl_unlock();
2611         }
2612 }
2613
2614 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2615 {
2616         struct vlan_entry *vlan;
2617         struct net_device *vlan_dev;
2618
2619         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2620                 return 1;
2621
2622         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2623                 rcu_read_lock();
2624                 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2625                 rcu_read_unlock();
2626                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2627                         return 1;
2628         }
2629
2630         return 0;
2631 }
2632
2633 /*
2634  * We go to the (large) trouble of VLAN tagging ARP frames because
2635  * switches in VLAN mode (especially if ports are configured as
2636  * "native" to a VLAN) might not pass non-tagged frames.
2637  */
2638 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2639 {
2640         struct sk_buff *skb;
2641
2642         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2643                  slave_dev->name, dest_ip, src_ip, vlan_id);
2644
2645         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2646                          NULL, slave_dev->dev_addr, NULL);
2647
2648         if (!skb) {
2649                 pr_err("ARP packet allocation failed\n");
2650                 return;
2651         }
2652         if (vlan_id) {
2653                 skb = vlan_put_tag(skb, vlan_id);
2654                 if (!skb) {
2655                         pr_err("failed to insert VLAN tag\n");
2656                         return;
2657                 }
2658         }
2659         arp_xmit(skb);
2660 }
2661
2662
2663 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2664 {
2665         int i, vlan_id;
2666         __be32 *targets = bond->params.arp_targets;
2667         struct vlan_entry *vlan;
2668         struct net_device *vlan_dev = NULL;
2669         struct rtable *rt;
2670
2671         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2672                 __be32 addr;
2673                 if (!targets[i])
2674                         break;
2675                 pr_debug("basa: target %x\n", targets[i]);
2676                 if (!bond_vlan_used(bond)) {
2677                         pr_debug("basa: empty vlan: arp_send\n");
2678                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2679                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2680                                       addr, 0);
2681                         continue;
2682                 }
2683
2684                 /*
2685                  * If VLANs are configured, we do a route lookup to
2686                  * determine which VLAN interface would be used, so we
2687                  * can tag the ARP with the proper VLAN tag.
2688                  */
2689                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2690                                      RTO_ONLINK, 0);
2691                 if (IS_ERR(rt)) {
2692                         if (net_ratelimit()) {
2693                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2694                                            bond->dev->name, &targets[i]);
2695                         }
2696                         continue;
2697                 }
2698
2699                 /*
2700                  * This target is not on a VLAN
2701                  */
2702                 if (rt->dst.dev == bond->dev) {
2703                         ip_rt_put(rt);
2704                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2705                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2706                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2707                                       addr, 0);
2708                         continue;
2709                 }
2710
2711                 vlan_id = 0;
2712                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2713                         rcu_read_lock();
2714                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2715                                                         vlan->vlan_id);
2716                         rcu_read_unlock();
2717                         if (vlan_dev == rt->dst.dev) {
2718                                 vlan_id = vlan->vlan_id;
2719                                 pr_debug("basa: vlan match on %s %d\n",
2720                                        vlan_dev->name, vlan_id);
2721                                 break;
2722                         }
2723                 }
2724
2725                 if (vlan_id && vlan_dev) {
2726                         ip_rt_put(rt);
2727                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2728                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2729                                       addr, vlan_id);
2730                         continue;
2731                 }
2732
2733                 if (net_ratelimit()) {
2734                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2735                                    bond->dev->name, &targets[i],
2736                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2737                 }
2738                 ip_rt_put(rt);
2739         }
2740 }
2741
2742 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2743 {
2744         int i;
2745         __be32 *targets = bond->params.arp_targets;
2746
2747         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2748                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2749                          &sip, &tip, i, &targets[i],
2750                          bond_has_this_ip(bond, tip));
2751                 if (sip == targets[i]) {
2752                         if (bond_has_this_ip(bond, tip))
2753                                 slave->last_arp_rx = jiffies;
2754                         return;
2755                 }
2756         }
2757 }
2758
2759 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2760                         struct slave *slave)
2761 {
2762         struct arphdr *arp = (struct arphdr *)skb->data;
2763         unsigned char *arp_ptr;
2764         __be32 sip, tip;
2765         int alen;
2766
2767         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2768                 return RX_HANDLER_ANOTHER;
2769
2770         read_lock(&bond->lock);
2771         alen = arp_hdr_len(bond->dev);
2772
2773         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2774                  bond->dev->name, skb->dev->name);
2775
2776         if (alen > skb_headlen(skb)) {
2777                 arp = kmalloc(alen, GFP_ATOMIC);
2778                 if (!arp)
2779                         goto out_unlock;
2780                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2781                         goto out_unlock;
2782         }
2783
2784         if (arp->ar_hln != bond->dev->addr_len ||
2785             skb->pkt_type == PACKET_OTHERHOST ||
2786             skb->pkt_type == PACKET_LOOPBACK ||
2787             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2788             arp->ar_pro != htons(ETH_P_IP) ||
2789             arp->ar_pln != 4)
2790                 goto out_unlock;
2791
2792         arp_ptr = (unsigned char *)(arp + 1);
2793         arp_ptr += bond->dev->addr_len;
2794         memcpy(&sip, arp_ptr, 4);
2795         arp_ptr += 4 + bond->dev->addr_len;
2796         memcpy(&tip, arp_ptr, 4);
2797
2798         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2799                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2800                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2801                  &sip, &tip);
2802
2803         /*
2804          * Backup slaves won't see the ARP reply, but do come through
2805          * here for each ARP probe (so we swap the sip/tip to validate
2806          * the probe).  In a "redundant switch, common router" type of
2807          * configuration, the ARP probe will (hopefully) travel from
2808          * the active, through one switch, the router, then the other
2809          * switch before reaching the backup.
2810          */
2811         if (bond_is_active_slave(slave))
2812                 bond_validate_arp(bond, slave, sip, tip);
2813         else
2814                 bond_validate_arp(bond, slave, tip, sip);
2815
2816 out_unlock:
2817         read_unlock(&bond->lock);
2818         if (arp != (struct arphdr *)skb->data)
2819                 kfree(arp);
2820         return RX_HANDLER_ANOTHER;
2821 }
2822
2823 /*
2824  * this function is called regularly to monitor each slave's link
2825  * ensuring that traffic is being sent and received when arp monitoring
2826  * is used in load-balancing mode. if the adapter has been dormant, then an
2827  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2828  * arp monitoring in active backup mode.
2829  */
2830 void bond_loadbalance_arp_mon(struct work_struct *work)
2831 {
2832         struct bonding *bond = container_of(work, struct bonding,
2833                                             arp_work.work);
2834         struct slave *slave, *oldcurrent;
2835         int do_failover = 0;
2836         int delta_in_ticks, extra_ticks;
2837         int i;
2838
2839         read_lock(&bond->lock);
2840
2841         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2842         extra_ticks = delta_in_ticks / 2;
2843
2844         if (bond->slave_cnt == 0)
2845                 goto re_arm;
2846
2847         read_lock(&bond->curr_slave_lock);
2848         oldcurrent = bond->curr_active_slave;
2849         read_unlock(&bond->curr_slave_lock);
2850
2851         /* see if any of the previous devices are up now (i.e. they have
2852          * xmt and rcv traffic). the curr_active_slave does not come into
2853          * the picture unless it is null. also, slave->jiffies is not needed
2854          * here because we send an arp on each slave and give a slave as
2855          * long as it needs to get the tx/rx within the delta.
2856          * TODO: what about up/down delay in arp mode? it wasn't here before
2857          *       so it can wait
2858          */
2859         bond_for_each_slave(bond, slave, i) {
2860                 unsigned long trans_start = dev_trans_start(slave->dev);
2861
2862                 if (slave->link != BOND_LINK_UP) {
2863                         if (time_in_range(jiffies,
2864                                 trans_start - delta_in_ticks,
2865                                 trans_start + delta_in_ticks + extra_ticks) &&
2866                             time_in_range(jiffies,
2867                                 slave->dev->last_rx - delta_in_ticks,
2868                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2869
2870                                 slave->link  = BOND_LINK_UP;
2871                                 bond_set_active_slave(slave);
2872
2873                                 /* primary_slave has no meaning in round-robin
2874                                  * mode. the window of a slave being up and
2875                                  * curr_active_slave being null after enslaving
2876                                  * is closed.
2877                                  */
2878                                 if (!oldcurrent) {
2879                                         pr_info("%s: link status definitely up for interface %s, ",
2880                                                 bond->dev->name,
2881                                                 slave->dev->name);
2882                                         do_failover = 1;
2883                                 } else {
2884                                         pr_info("%s: interface %s is now up\n",
2885                                                 bond->dev->name,
2886                                                 slave->dev->name);
2887                                 }
2888                         }
2889                 } else {
2890                         /* slave->link == BOND_LINK_UP */
2891
2892                         /* not all switches will respond to an arp request
2893                          * when the source ip is 0, so don't take the link down
2894                          * if we don't know our ip yet
2895                          */
2896                         if (!time_in_range(jiffies,
2897                                 trans_start - delta_in_ticks,
2898                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2899                             !time_in_range(jiffies,
2900                                 slave->dev->last_rx - delta_in_ticks,
2901                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2902
2903                                 slave->link  = BOND_LINK_DOWN;
2904                                 bond_set_backup_slave(slave);
2905
2906                                 if (slave->link_failure_count < UINT_MAX)
2907                                         slave->link_failure_count++;
2908
2909                                 pr_info("%s: interface %s is now down.\n",
2910                                         bond->dev->name,
2911                                         slave->dev->name);
2912
2913                                 if (slave == oldcurrent)
2914                                         do_failover = 1;
2915                         }
2916                 }
2917
2918                 /* note: if switch is in round-robin mode, all links
2919                  * must tx arp to ensure all links rx an arp - otherwise
2920                  * links may oscillate or not come up at all; if switch is
2921                  * in something like xor mode, there is nothing we can
2922                  * do - all replies will be rx'ed on same link causing slaves
2923                  * to be unstable during low/no traffic periods
2924                  */
2925                 if (IS_UP(slave->dev))
2926                         bond_arp_send_all(bond, slave);
2927         }
2928
2929         if (do_failover) {
2930                 block_netpoll_tx();
2931                 write_lock_bh(&bond->curr_slave_lock);
2932
2933                 bond_select_active_slave(bond);
2934
2935                 write_unlock_bh(&bond->curr_slave_lock);
2936                 unblock_netpoll_tx();
2937         }
2938
2939 re_arm:
2940         if (bond->params.arp_interval)
2941                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2942
2943         read_unlock(&bond->lock);
2944 }
2945
2946 /*
2947  * Called to inspect slaves for active-backup mode ARP monitor link state
2948  * changes.  Sets new_link in slaves to specify what action should take
2949  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2950  * to link states must be committed.
2951  *
2952  * Called with bond->lock held for read.
2953  */
2954 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2955 {
2956         struct slave *slave;
2957         int i, commit = 0;
2958         unsigned long trans_start;
2959         int extra_ticks;
2960
2961         /* All the time comparisons below need some extra time. Otherwise, on
2962          * fast networks the ARP probe/reply may arrive within the same jiffy
2963          * as it was sent.  Then, the next time the ARP monitor is run, one
2964          * arp_interval will already have passed in the comparisons.
2965          */
2966         extra_ticks = delta_in_ticks / 2;
2967
2968         bond_for_each_slave(bond, slave, i) {
2969                 slave->new_link = BOND_LINK_NOCHANGE;
2970
2971                 if (slave->link != BOND_LINK_UP) {
2972                         if (time_in_range(jiffies,
2973                                 slave_last_rx(bond, slave) - delta_in_ticks,
2974                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2975
2976                                 slave->new_link = BOND_LINK_UP;
2977                                 commit++;
2978                         }
2979
2980                         continue;
2981                 }
2982
2983                 /*
2984                  * Give slaves 2*delta after being enslaved or made
2985                  * active.  This avoids bouncing, as the last receive
2986                  * times need a full ARP monitor cycle to be updated.
2987                  */
2988                 if (time_in_range(jiffies,
2989                                   slave->jiffies - delta_in_ticks,
2990                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2991                         continue;
2992
2993                 /*
2994                  * Backup slave is down if:
2995                  * - No current_arp_slave AND
2996                  * - more than 3*delta since last receive AND
2997                  * - the bond has an IP address
2998                  *
2999                  * Note: a non-null current_arp_slave indicates
3000                  * the curr_active_slave went down and we are
3001                  * searching for a new one; under this condition
3002                  * we only take the curr_active_slave down - this
3003                  * gives each slave a chance to tx/rx traffic
3004                  * before being taken out
3005                  */
3006                 if (!bond_is_active_slave(slave) &&
3007                     !bond->current_arp_slave &&
3008                     !time_in_range(jiffies,
3009                         slave_last_rx(bond, slave) - delta_in_ticks,
3010                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
3011
3012                         slave->new_link = BOND_LINK_DOWN;
3013                         commit++;
3014                 }
3015
3016                 /*
3017                  * Active slave is down if:
3018                  * - more than 2*delta since transmitting OR
3019                  * - (more than 2*delta since receive AND
3020                  *    the bond has an IP address)
3021                  */
3022                 trans_start = dev_trans_start(slave->dev);
3023                 if (bond_is_active_slave(slave) &&
3024                     (!time_in_range(jiffies,
3025                         trans_start - delta_in_ticks,
3026                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
3027                      !time_in_range(jiffies,
3028                         slave_last_rx(bond, slave) - delta_in_ticks,
3029                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
3030
3031                         slave->new_link = BOND_LINK_DOWN;
3032                         commit++;
3033                 }
3034         }
3035
3036         return commit;
3037 }
3038
3039 /*
3040  * Called to commit link state changes noted by inspection step of
3041  * active-backup mode ARP monitor.
3042  *
3043  * Called with RTNL and bond->lock for read.
3044  */
3045 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3046 {
3047         struct slave *slave;
3048         int i;
3049         unsigned long trans_start;
3050
3051         bond_for_each_slave(bond, slave, i) {
3052                 switch (slave->new_link) {
3053                 case BOND_LINK_NOCHANGE:
3054                         continue;
3055
3056                 case BOND_LINK_UP:
3057                         trans_start = dev_trans_start(slave->dev);
3058                         if ((!bond->curr_active_slave &&
3059                              time_in_range(jiffies,
3060                                            trans_start - delta_in_ticks,
3061                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
3062                             bond->curr_active_slave != slave) {
3063                                 slave->link = BOND_LINK_UP;
3064                                 if (bond->current_arp_slave) {
3065                                         bond_set_slave_inactive_flags(
3066                                                 bond->current_arp_slave);
3067                                         bond->current_arp_slave = NULL;
3068                                 }
3069
3070                                 pr_info("%s: link status definitely up for interface %s.\n",
3071                                         bond->dev->name, slave->dev->name);
3072
3073                                 if (!bond->curr_active_slave ||
3074                                     (slave == bond->primary_slave))
3075                                         goto do_failover;
3076
3077                         }
3078
3079                         continue;
3080
3081                 case BOND_LINK_DOWN:
3082                         if (slave->link_failure_count < UINT_MAX)
3083                                 slave->link_failure_count++;
3084
3085                         slave->link = BOND_LINK_DOWN;
3086                         bond_set_slave_inactive_flags(slave);
3087
3088                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3089                                 bond->dev->name, slave->dev->name);
3090
3091                         if (slave == bond->curr_active_slave) {
3092                                 bond->current_arp_slave = NULL;
3093                                 goto do_failover;
3094                         }
3095
3096                         continue;
3097
3098                 default:
3099                         pr_err("%s: impossible: new_link %d on slave %s\n",
3100                                bond->dev->name, slave->new_link,
3101                                slave->dev->name);
3102                         continue;
3103                 }
3104
3105 do_failover:
3106                 ASSERT_RTNL();
3107                 block_netpoll_tx();
3108                 write_lock_bh(&bond->curr_slave_lock);
3109                 bond_select_active_slave(bond);
3110                 write_unlock_bh(&bond->curr_slave_lock);
3111                 unblock_netpoll_tx();
3112         }
3113
3114         bond_set_carrier(bond);
3115 }
3116
3117 /*
3118  * Send ARP probes for active-backup mode ARP monitor.
3119  *
3120  * Called with bond->lock held for read.
3121  */
3122 static void bond_ab_arp_probe(struct bonding *bond)
3123 {
3124         struct slave *slave;
3125         int i;
3126
3127         read_lock(&bond->curr_slave_lock);
3128
3129         if (bond->current_arp_slave && bond->curr_active_slave)
3130                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3131                         bond->current_arp_slave->dev->name,
3132                         bond->curr_active_slave->dev->name);
3133
3134         if (bond->curr_active_slave) {
3135                 bond_arp_send_all(bond, bond->curr_active_slave);
3136                 read_unlock(&bond->curr_slave_lock);
3137                 return;
3138         }
3139
3140         read_unlock(&bond->curr_slave_lock);
3141
3142         /* if we don't have a curr_active_slave, search for the next available
3143          * backup slave from the current_arp_slave and make it the candidate
3144          * for becoming the curr_active_slave
3145          */
3146
3147         if (!bond->current_arp_slave) {
3148                 bond->current_arp_slave = bond->first_slave;
3149                 if (!bond->current_arp_slave)
3150                         return;
3151         }
3152
3153         bond_set_slave_inactive_flags(bond->current_arp_slave);
3154
3155         /* search for next candidate */
3156         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3157                 if (IS_UP(slave->dev)) {
3158                         slave->link = BOND_LINK_BACK;
3159                         bond_set_slave_active_flags(slave);
3160                         bond_arp_send_all(bond, slave);
3161                         slave->jiffies = jiffies;
3162                         bond->current_arp_slave = slave;
3163                         break;
3164                 }
3165
3166                 /* if the link state is up at this point, we
3167                  * mark it down - this can happen if we have
3168                  * simultaneous link failures and
3169                  * reselect_active_interface doesn't make this
3170                  * one the current slave so it is still marked
3171                  * up when it is actually down
3172                  */
3173                 if (slave->link == BOND_LINK_UP) {
3174                         slave->link = BOND_LINK_DOWN;
3175                         if (slave->link_failure_count < UINT_MAX)
3176                                 slave->link_failure_count++;
3177
3178                         bond_set_slave_inactive_flags(slave);
3179
3180                         pr_info("%s: backup interface %s is now down.\n",
3181                                 bond->dev->name, slave->dev->name);
3182                 }
3183         }
3184 }
3185
3186 void bond_activebackup_arp_mon(struct work_struct *work)
3187 {
3188         struct bonding *bond = container_of(work, struct bonding,
3189                                             arp_work.work);
3190         bool should_notify_peers = false;
3191         int delta_in_ticks;
3192
3193         read_lock(&bond->lock);
3194
3195         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3196
3197         if (bond->slave_cnt == 0)
3198                 goto re_arm;
3199
3200         should_notify_peers = bond_should_notify_peers(bond);
3201
3202         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3203                 read_unlock(&bond->lock);
3204
3205                 /* Race avoidance with bond_close flush of workqueue */
3206                 if (!rtnl_trylock()) {
3207                         read_lock(&bond->lock);
3208                         delta_in_ticks = 1;
3209                         should_notify_peers = false;
3210                         goto re_arm;
3211                 }
3212
3213                 read_lock(&bond->lock);
3214
3215                 bond_ab_arp_commit(bond, delta_in_ticks);
3216
3217                 read_unlock(&bond->lock);
3218                 rtnl_unlock();
3219                 read_lock(&bond->lock);
3220         }
3221
3222         bond_ab_arp_probe(bond);
3223
3224 re_arm:
3225         if (bond->params.arp_interval)
3226                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3227
3228         read_unlock(&bond->lock);
3229
3230         if (should_notify_peers) {
3231                 if (!rtnl_trylock()) {
3232                         read_lock(&bond->lock);
3233                         bond->send_peer_notif++;
3234                         read_unlock(&bond->lock);
3235                         return;
3236                 }
3237                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3238                 rtnl_unlock();
3239         }
3240 }
3241
3242 /*-------------------------- netdev event handling --------------------------*/
3243
3244 /*
3245  * Change device name
3246  */
3247 static int bond_event_changename(struct bonding *bond)
3248 {
3249         bond_remove_proc_entry(bond);
3250         bond_create_proc_entry(bond);
3251
3252         bond_debug_reregister(bond);
3253
3254         return NOTIFY_DONE;
3255 }
3256
3257 static int bond_master_netdev_event(unsigned long event,
3258                                     struct net_device *bond_dev)
3259 {
3260         struct bonding *event_bond = netdev_priv(bond_dev);
3261
3262         switch (event) {
3263         case NETDEV_CHANGENAME:
3264                 return bond_event_changename(event_bond);
3265         case NETDEV_UNREGISTER:
3266                 bond_remove_proc_entry(event_bond);
3267                 break;
3268         case NETDEV_REGISTER:
3269                 bond_create_proc_entry(event_bond);
3270                 break;
3271         default:
3272                 break;
3273         }
3274
3275         return NOTIFY_DONE;
3276 }
3277
3278 static int bond_slave_netdev_event(unsigned long event,
3279                                    struct net_device *slave_dev)
3280 {
3281         struct slave *slave = bond_slave_get_rtnl(slave_dev);
3282         struct bonding *bond = slave->bond;
3283         struct net_device *bond_dev = slave->bond->dev;
3284         u32 old_speed;
3285         u8 old_duplex;
3286
3287         switch (event) {
3288         case NETDEV_UNREGISTER:
3289                 if (bond->setup_by_slave)
3290                         bond_release_and_destroy(bond_dev, slave_dev);
3291                 else
3292                         bond_release(bond_dev, slave_dev);
3293                 break;
3294         case NETDEV_UP:
3295         case NETDEV_CHANGE:
3296                 old_speed = slave->speed;
3297                 old_duplex = slave->duplex;
3298
3299                 bond_update_speed_duplex(slave);
3300
3301                 if (bond->params.mode == BOND_MODE_8023AD) {
3302                         if (old_speed != slave->speed)
3303                                 bond_3ad_adapter_speed_changed(slave);
3304                         if (old_duplex != slave->duplex)
3305                                 bond_3ad_adapter_duplex_changed(slave);
3306                 }
3307                 break;
3308         case NETDEV_DOWN:
3309                 /*
3310                  * ... Or is it this?
3311                  */
3312                 break;
3313         case NETDEV_CHANGEMTU:
3314                 /*
3315                  * TODO: Should slaves be allowed to
3316                  * independently alter their MTU?  For
3317                  * an active-backup bond, slaves need
3318                  * not be the same type of device, so
3319                  * MTUs may vary.  For other modes,
3320                  * slaves arguably should have the
3321                  * same MTUs. To do this, we'd need to
3322                  * take over the slave's change_mtu
3323                  * function for the duration of their
3324                  * servitude.
3325                  */
3326                 break;
3327         case NETDEV_CHANGENAME:
3328                 /*
3329                  * TODO: handle changing the primary's name
3330                  */
3331                 break;
3332         case NETDEV_FEAT_CHANGE:
3333                 bond_compute_features(bond);
3334                 break;
3335         default:
3336                 break;
3337         }
3338
3339         return NOTIFY_DONE;
3340 }
3341
3342 /*
3343  * bond_netdev_event: handle netdev notifier chain events.
3344  *
3345  * This function receives events for the netdev chain.  The caller (an
3346  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3347  * locks for us to safely manipulate the slave devices (RTNL lock,
3348  * dev_probe_lock).
3349  */
3350 static int bond_netdev_event(struct notifier_block *this,
3351                              unsigned long event, void *ptr)
3352 {
3353         struct net_device *event_dev = (struct net_device *)ptr;
3354
3355         pr_debug("event_dev: %s, event: %lx\n",
3356                  event_dev ? event_dev->name : "None",
3357                  event);
3358
3359         if (!(event_dev->priv_flags & IFF_BONDING))
3360                 return NOTIFY_DONE;
3361
3362         if (event_dev->flags & IFF_MASTER) {
3363                 pr_debug("IFF_MASTER\n");
3364                 return bond_master_netdev_event(event, event_dev);
3365         }
3366
3367         if (event_dev->flags & IFF_SLAVE) {
3368                 pr_debug("IFF_SLAVE\n");
3369                 return bond_slave_netdev_event(event, event_dev);
3370         }
3371
3372         return NOTIFY_DONE;
3373 }
3374
3375 static struct notifier_block bond_netdev_notifier = {
3376         .notifier_call = bond_netdev_event,
3377 };
3378
3379 /*---------------------------- Hashing Policies -----------------------------*/
3380
3381 /*
3382  * Hash for the output device based upon layer 2 data
3383  */
3384 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3385 {
3386         struct ethhdr *data = (struct ethhdr *)skb->data;
3387
3388         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3389                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3390
3391         return 0;
3392 }
3393
3394 /*
3395  * Hash for the output device based upon layer 2 and layer 3 data. If
3396  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3397  */
3398 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3399 {
3400         struct ethhdr *data = (struct ethhdr *)skb->data;
3401         struct iphdr *iph;
3402         struct ipv6hdr *ipv6h;
3403         u32 v6hash;
3404         __be32 *s, *d;
3405
3406         if (skb->protocol == htons(ETH_P_IP) &&
3407             skb_network_header_len(skb) >= sizeof(*iph)) {
3408                 iph = ip_hdr(skb);
3409                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3410                         (data->h_dest[5] ^ data->h_source[5])) % count;
3411         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3412                    skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3413                 ipv6h = ipv6_hdr(skb);
3414                 s = &ipv6h->saddr.s6_addr32[0];
3415                 d = &ipv6h->daddr.s6_addr32[0];
3416                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3417                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3418                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3419         }
3420
3421         return bond_xmit_hash_policy_l2(skb, count);
3422 }
3423
3424 /*
3425  * Hash for the output device based upon layer 3 and layer 4 data. If
3426  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3427  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3428  */
3429 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3430 {
3431         u32 layer4_xor = 0;
3432         struct iphdr *iph;
3433         struct ipv6hdr *ipv6h;
3434         __be32 *s, *d;
3435         __be16 *layer4hdr;
3436
3437         if (skb->protocol == htons(ETH_P_IP) &&
3438             skb_network_header_len(skb) >= sizeof(*iph)) {
3439                 iph = ip_hdr(skb);
3440                 if (!ip_is_fragment(iph) &&
3441                     (iph->protocol == IPPROTO_TCP ||
3442                      iph->protocol == IPPROTO_UDP) &&
3443                     (skb_headlen(skb) - skb_network_offset(skb) >=
3444                      iph->ihl * sizeof(u32) + sizeof(*layer4hdr) * 2)) {
3445                         layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3446                         layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3447                 }
3448                 return (layer4_xor ^
3449                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3450         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3451                    skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3452                 ipv6h = ipv6_hdr(skb);
3453                 if ((ipv6h->nexthdr == IPPROTO_TCP ||
3454                      ipv6h->nexthdr == IPPROTO_UDP) &&
3455                     (skb_headlen(skb) - skb_network_offset(skb) >=
3456                      sizeof(*ipv6h) + sizeof(*layer4hdr) * 2)) {
3457                         layer4hdr = (__be16 *)(ipv6h + 1);
3458                         layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3459                 }
3460                 s = &ipv6h->saddr.s6_addr32[0];
3461                 d = &ipv6h->daddr.s6_addr32[0];
3462                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3463                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3464                                (layer4_xor >> 8);
3465                 return layer4_xor % count;
3466         }
3467
3468         return bond_xmit_hash_policy_l2(skb, count);
3469 }
3470
3471 /*-------------------------- Device entry points ----------------------------*/
3472
3473 static void bond_work_init_all(struct bonding *bond)
3474 {
3475         INIT_DELAYED_WORK(&bond->mcast_work,
3476                           bond_resend_igmp_join_requests_delayed);
3477         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3478         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3479         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3480                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3481         else
3482                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3483         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3484 }
3485
3486 static void bond_work_cancel_all(struct bonding *bond)
3487 {
3488         cancel_delayed_work_sync(&bond->mii_work);
3489         cancel_delayed_work_sync(&bond->arp_work);
3490         cancel_delayed_work_sync(&bond->alb_work);
3491         cancel_delayed_work_sync(&bond->ad_work);
3492         cancel_delayed_work_sync(&bond->mcast_work);
3493 }
3494
3495 static int bond_open(struct net_device *bond_dev)
3496 {
3497         struct bonding *bond = netdev_priv(bond_dev);
3498         struct slave *slave;
3499         int i;
3500
3501         /* reset slave->backup and slave->inactive */
3502         read_lock(&bond->lock);
3503         if (bond->slave_cnt > 0) {
3504                 read_lock(&bond->curr_slave_lock);
3505                 bond_for_each_slave(bond, slave, i) {
3506                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3507                                 && (slave != bond->curr_active_slave)) {
3508                                 bond_set_slave_inactive_flags(slave);
3509                         } else {
3510                                 bond_set_slave_active_flags(slave);
3511                         }
3512                 }
3513                 read_unlock(&bond->curr_slave_lock);
3514         }
3515         read_unlock(&bond->lock);
3516
3517         bond_work_init_all(bond);
3518
3519         if (bond_is_lb(bond)) {
3520                 /* bond_alb_initialize must be called before the timer
3521                  * is started.
3522                  */
3523                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3524                         return -ENOMEM;
3525                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3526         }
3527
3528         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3529                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3530
3531         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3532                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3533                 if (bond->params.arp_validate)
3534                         bond->recv_probe = bond_arp_rcv;
3535         }
3536
3537         if (bond->params.mode == BOND_MODE_8023AD) {
3538                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3539                 /* register to receive LACPDUs */
3540                 bond->recv_probe = bond_3ad_lacpdu_recv;
3541                 bond_3ad_initiate_agg_selection(bond, 1);
3542         }
3543
3544         return 0;
3545 }
3546
3547 static int bond_close(struct net_device *bond_dev)
3548 {
3549         struct bonding *bond = netdev_priv(bond_dev);
3550
3551         write_lock_bh(&bond->lock);
3552         bond->send_peer_notif = 0;
3553         write_unlock_bh(&bond->lock);
3554
3555         bond_work_cancel_all(bond);
3556         if (bond_is_lb(bond)) {
3557                 /* Must be called only after all
3558                  * slaves have been released
3559                  */
3560                 bond_alb_deinitialize(bond);
3561         }
3562         bond->recv_probe = NULL;
3563
3564         return 0;
3565 }
3566
3567 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3568                                                 struct rtnl_link_stats64 *stats)
3569 {
3570         struct bonding *bond = netdev_priv(bond_dev);
3571         struct rtnl_link_stats64 temp;
3572         struct slave *slave;
3573         int i;
3574
3575         memset(stats, 0, sizeof(*stats));
3576
3577         read_lock_bh(&bond->lock);
3578
3579         bond_for_each_slave(bond, slave, i) {
3580                 const struct rtnl_link_stats64 *sstats =
3581                         dev_get_stats(slave->dev, &temp);
3582
3583                 stats->rx_packets += sstats->rx_packets;
3584                 stats->rx_bytes += sstats->rx_bytes;
3585                 stats->rx_errors += sstats->rx_errors;
3586                 stats->rx_dropped += sstats->rx_dropped;
3587
3588                 stats->tx_packets += sstats->tx_packets;
3589                 stats->tx_bytes += sstats->tx_bytes;
3590                 stats->tx_errors += sstats->tx_errors;
3591                 stats->tx_dropped += sstats->tx_dropped;
3592
3593                 stats->multicast += sstats->multicast;
3594                 stats->collisions += sstats->collisions;
3595
3596                 stats->rx_length_errors += sstats->rx_length_errors;
3597                 stats->rx_over_errors += sstats->rx_over_errors;
3598                 stats->rx_crc_errors += sstats->rx_crc_errors;
3599                 stats->rx_frame_errors += sstats->rx_frame_errors;
3600                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3601                 stats->rx_missed_errors += sstats->rx_missed_errors;
3602
3603                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3604                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3605                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3606                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3607                 stats->tx_window_errors += sstats->tx_window_errors;
3608         }
3609
3610         read_unlock_bh(&bond->lock);
3611
3612         return stats;
3613 }
3614
3615 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3616 {
3617         struct net_device *slave_dev = NULL;
3618         struct ifbond k_binfo;
3619         struct ifbond __user *u_binfo = NULL;
3620         struct ifslave k_sinfo;
3621         struct ifslave __user *u_sinfo = NULL;
3622         struct mii_ioctl_data *mii = NULL;
3623         int res = 0;
3624
3625         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3626
3627         switch (cmd) {
3628         case SIOCGMIIPHY:
3629                 mii = if_mii(ifr);
3630                 if (!mii)
3631                         return -EINVAL;
3632
3633                 mii->phy_id = 0;
3634                 /* Fall Through */
3635         case SIOCGMIIREG:
3636                 /*
3637                  * We do this again just in case we were called by SIOCGMIIREG
3638                  * instead of SIOCGMIIPHY.
3639                  */
3640                 mii = if_mii(ifr);
3641                 if (!mii)
3642                         return -EINVAL;
3643
3644
3645                 if (mii->reg_num == 1) {
3646                         struct bonding *bond = netdev_priv(bond_dev);
3647                         mii->val_out = 0;
3648                         read_lock(&bond->lock);
3649                         read_lock(&bond->curr_slave_lock);
3650                         if (netif_carrier_ok(bond->dev))
3651                                 mii->val_out = BMSR_LSTATUS;
3652
3653                         read_unlock(&bond->curr_slave_lock);
3654                         read_unlock(&bond->lock);
3655                 }
3656
3657                 return 0;
3658         case BOND_INFO_QUERY_OLD:
3659         case SIOCBONDINFOQUERY:
3660                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3661
3662                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3663                         return -EFAULT;
3664
3665                 res = bond_info_query(bond_dev, &k_binfo);
3666                 if (res == 0 &&
3667                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3668                         return -EFAULT;
3669
3670                 return res;
3671         case BOND_SLAVE_INFO_QUERY_OLD:
3672         case SIOCBONDSLAVEINFOQUERY:
3673                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3674
3675                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3676                         return -EFAULT;
3677
3678                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3679                 if (res == 0 &&
3680                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3681                         return -EFAULT;
3682
3683                 return res;
3684         default:
3685                 /* Go on */
3686                 break;
3687         }
3688
3689         if (!capable(CAP_NET_ADMIN))
3690                 return -EPERM;
3691
3692         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3693
3694         pr_debug("slave_dev=%p:\n", slave_dev);
3695
3696         if (!slave_dev)
3697                 res = -ENODEV;
3698         else {
3699                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3700                 switch (cmd) {
3701                 case BOND_ENSLAVE_OLD:
3702                 case SIOCBONDENSLAVE:
3703                         res = bond_enslave(bond_dev, slave_dev);
3704                         break;
3705                 case BOND_RELEASE_OLD:
3706                 case SIOCBONDRELEASE:
3707                         res = bond_release(bond_dev, slave_dev);
3708                         break;
3709                 case BOND_SETHWADDR_OLD:
3710                 case SIOCBONDSETHWADDR:
3711                         res = bond_sethwaddr(bond_dev, slave_dev);
3712                         break;
3713                 case BOND_CHANGE_ACTIVE_OLD:
3714                 case SIOCBONDCHANGEACTIVE:
3715                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3716                         break;
3717                 default:
3718                         res = -EOPNOTSUPP;
3719                 }
3720
3721                 dev_put(slave_dev);
3722         }
3723
3724         return res;
3725 }
3726
3727 static bool bond_addr_in_mc_list(unsigned char *addr,
3728                                  struct netdev_hw_addr_list *list,
3729                                  int addrlen)
3730 {
3731         struct netdev_hw_addr *ha;
3732
3733         netdev_hw_addr_list_for_each(ha, list)
3734                 if (!memcmp(ha->addr, addr, addrlen))
3735                         return true;
3736
3737         return false;
3738 }
3739
3740 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3741 {
3742         struct bonding *bond = netdev_priv(bond_dev);
3743
3744         if (change & IFF_PROMISC)
3745                 bond_set_promiscuity(bond,
3746                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3747
3748         if (change & IFF_ALLMULTI)
3749                 bond_set_allmulti(bond,
3750                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3751 }
3752
3753 static void bond_set_multicast_list(struct net_device *bond_dev)
3754 {
3755         struct bonding *bond = netdev_priv(bond_dev);
3756         struct netdev_hw_addr *ha;
3757         bool found;
3758
3759         read_lock(&bond->lock);
3760
3761         /* looking for addresses to add to slaves' mc list */
3762         netdev_for_each_mc_addr(ha, bond_dev) {
3763                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3764                                              bond_dev->addr_len);
3765                 if (!found)
3766                         bond_mc_add(bond, ha->addr);
3767         }
3768
3769         /* looking for addresses to delete from slaves' list */
3770         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3771                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3772                                              bond_dev->addr_len);
3773                 if (!found)
3774                         bond_mc_del(bond, ha->addr);
3775         }
3776
3777         /* save master's multicast list */
3778         __hw_addr_flush(&bond->mc_list);
3779         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3780                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3781
3782         read_unlock(&bond->lock);
3783 }
3784
3785 static int bond_neigh_init(struct neighbour *n)
3786 {
3787         struct bonding *bond = netdev_priv(n->dev);
3788         struct slave *slave = bond->first_slave;
3789         const struct net_device_ops *slave_ops;
3790         struct neigh_parms parms;
3791         int ret;
3792
3793         if (!slave)
3794                 return 0;
3795
3796         slave_ops = slave->dev->netdev_ops;
3797
3798         if (!slave_ops->ndo_neigh_setup)
3799                 return 0;
3800
3801         parms.neigh_setup = NULL;
3802         parms.neigh_cleanup = NULL;
3803         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3804         if (ret)
3805                 return ret;
3806
3807         /*
3808          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3809          * after the last slave has been detached.  Assumes that all slaves
3810          * utilize the same neigh_cleanup (true at this writing as only user
3811          * is ipoib).
3812          */
3813         n->parms->neigh_cleanup = parms.neigh_cleanup;
3814
3815         if (!parms.neigh_setup)
3816                 return 0;
3817
3818         return parms.neigh_setup(n);
3819 }
3820
3821 /*
3822  * The bonding ndo_neigh_setup is called at init time beofre any
3823  * slave exists. So we must declare proxy setup function which will
3824  * be used at run time to resolve the actual slave neigh param setup.
3825  */
3826 static int bond_neigh_setup(struct net_device *dev,
3827                             struct neigh_parms *parms)
3828 {
3829         parms->neigh_setup   = bond_neigh_init;
3830
3831         return 0;
3832 }
3833
3834 /*
3835  * Change the MTU of all of a master's slaves to match the master
3836  */
3837 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3838 {
3839         struct bonding *bond = netdev_priv(bond_dev);
3840         struct slave *slave, *stop_at;
3841         int res = 0;
3842         int i;
3843
3844         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3845                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3846
3847         /* Can't hold bond->lock with bh disabled here since
3848          * some base drivers panic. On the other hand we can't
3849          * hold bond->lock without bh disabled because we'll
3850          * deadlock. The only solution is to rely on the fact
3851          * that we're under rtnl_lock here, and the slaves
3852          * list won't change. This doesn't solve the problem
3853          * of setting the slave's MTU while it is
3854          * transmitting, but the assumption is that the base
3855          * driver can handle that.
3856          *
3857          * TODO: figure out a way to safely iterate the slaves
3858          * list, but without holding a lock around the actual
3859          * call to the base driver.
3860          */
3861
3862         bond_for_each_slave(bond, slave, i) {
3863                 pr_debug("s %p s->p %p c_m %p\n",
3864                          slave,
3865                          slave->prev,
3866                          slave->dev->netdev_ops->ndo_change_mtu);
3867
3868                 res = dev_set_mtu(slave->dev, new_mtu);
3869
3870                 if (res) {
3871                         /* If we failed to set the slave's mtu to the new value
3872                          * we must abort the operation even in ACTIVE_BACKUP
3873                          * mode, because if we allow the backup slaves to have
3874                          * different mtu values than the active slave we'll
3875                          * need to change their mtu when doing a failover. That
3876                          * means changing their mtu from timer context, which
3877                          * is probably not a good idea.
3878                          */
3879                         pr_debug("err %d %s\n", res, slave->dev->name);
3880                         goto unwind;
3881                 }
3882         }
3883
3884         bond_dev->mtu = new_mtu;
3885
3886         return 0;
3887
3888 unwind:
3889         /* unwind from head to the slave that failed */
3890         stop_at = slave;
3891         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3892                 int tmp_res;
3893
3894                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3895                 if (tmp_res) {
3896                         pr_debug("unwind err %d dev %s\n",
3897                                  tmp_res, slave->dev->name);
3898                 }
3899         }
3900
3901         return res;
3902 }
3903
3904 /*
3905  * Change HW address
3906  *
3907  * Note that many devices must be down to change the HW address, and
3908  * downing the master releases all slaves.  We can make bonds full of
3909  * bonding devices to test this, however.
3910  */
3911 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3912 {
3913         struct bonding *bond = netdev_priv(bond_dev);
3914         struct sockaddr *sa = addr, tmp_sa;
3915         struct slave *slave, *stop_at;
3916         int res = 0;
3917         int i;
3918
3919         if (bond->params.mode == BOND_MODE_ALB)
3920                 return bond_alb_set_mac_address(bond_dev, addr);
3921
3922
3923         pr_debug("bond=%p, name=%s\n",
3924                  bond, bond_dev ? bond_dev->name : "None");
3925
3926         /*
3927          * If fail_over_mac is set to active, do nothing and return
3928          * success.  Returning an error causes ifenslave to fail.
3929          */
3930         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3931                 return 0;
3932
3933         if (!is_valid_ether_addr(sa->sa_data))
3934                 return -EADDRNOTAVAIL;
3935
3936         /* Can't hold bond->lock with bh disabled here since
3937          * some base drivers panic. On the other hand we can't
3938          * hold bond->lock without bh disabled because we'll
3939          * deadlock. The only solution is to rely on the fact
3940          * that we're under rtnl_lock here, and the slaves
3941          * list won't change. This doesn't solve the problem
3942          * of setting the slave's hw address while it is
3943          * transmitting, but the assumption is that the base
3944          * driver can handle that.
3945          *
3946          * TODO: figure out a way to safely iterate the slaves
3947          * list, but without holding a lock around the actual
3948          * call to the base driver.
3949          */
3950
3951         bond_for_each_slave(bond, slave, i) {
3952                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3953                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3954
3955                 if (slave_ops->ndo_set_mac_address == NULL) {
3956                         res = -EOPNOTSUPP;
3957                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3958                         goto unwind;
3959                 }
3960
3961                 res = dev_set_mac_address(slave->dev, addr);
3962                 if (res) {
3963                         /* TODO: consider downing the slave
3964                          * and retry ?
3965                          * User should expect communications
3966                          * breakage anyway until ARP finish
3967                          * updating, so...
3968                          */
3969                         pr_debug("err %d %s\n", res, slave->dev->name);
3970                         goto unwind;
3971                 }
3972         }
3973
3974         /* success */
3975         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3976         return 0;
3977
3978 unwind:
3979         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3980         tmp_sa.sa_family = bond_dev->type;
3981
3982         /* unwind from head to the slave that failed */
3983         stop_at = slave;
3984         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3985                 int tmp_res;
3986
3987                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3988                 if (tmp_res) {
3989                         pr_debug("unwind err %d dev %s\n",
3990                                  tmp_res, slave->dev->name);
3991                 }
3992         }
3993
3994         return res;
3995 }
3996
3997 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3998 {
3999         struct bonding *bond = netdev_priv(bond_dev);
4000         struct slave *slave, *start_at;
4001         int i, slave_no, res = 1;
4002         struct iphdr *iph = ip_hdr(skb);
4003
4004         /*
4005          * Start with the curr_active_slave that joined the bond as the
4006          * default for sending IGMP traffic.  For failover purposes one
4007          * needs to maintain some consistency for the interface that will
4008          * send the join/membership reports.  The curr_active_slave found
4009          * will send all of this type of traffic.
4010          */
4011         if ((iph->protocol == IPPROTO_IGMP) &&
4012             (skb->protocol == htons(ETH_P_IP))) {
4013
4014                 read_lock(&bond->curr_slave_lock);
4015                 slave = bond->curr_active_slave;
4016                 read_unlock(&bond->curr_slave_lock);
4017
4018                 if (!slave)
4019                         goto out;
4020         } else {
4021                 /*
4022                  * Concurrent TX may collide on rr_tx_counter; we accept
4023                  * that as being rare enough not to justify using an
4024                  * atomic op here.
4025                  */
4026                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4027
4028                 bond_for_each_slave(bond, slave, i) {
4029                         slave_no--;
4030                         if (slave_no < 0)
4031                                 break;
4032                 }
4033         }
4034
4035         start_at = slave;
4036         bond_for_each_slave_from(bond, slave, i, start_at) {
4037                 if (IS_UP(slave->dev) &&
4038                     (slave->link == BOND_LINK_UP) &&
4039                     bond_is_active_slave(slave)) {
4040                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4041                         break;
4042                 }
4043         }
4044
4045 out:
4046         if (res) {
4047                 /* no suitable interface, frame not sent */
4048                 kfree_skb(skb);
4049         }
4050
4051         return NETDEV_TX_OK;
4052 }
4053
4054
4055 /*
4056  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4057  * the bond has a usable interface.
4058  */
4059 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4060 {
4061         struct bonding *bond = netdev_priv(bond_dev);
4062         int res = 1;
4063
4064         read_lock(&bond->curr_slave_lock);
4065
4066         if (bond->curr_active_slave)
4067                 res = bond_dev_queue_xmit(bond, skb,
4068                         bond->curr_active_slave->dev);
4069
4070         read_unlock(&bond->curr_slave_lock);
4071
4072         if (res)
4073                 /* no suitable interface, frame not sent */
4074                 kfree_skb(skb);
4075
4076         return NETDEV_TX_OK;
4077 }
4078
4079 /*
4080  * In bond_xmit_xor() , we determine the output device by using a pre-
4081  * determined xmit_hash_policy(), If the selected device is not enabled,
4082  * find the next active slave.
4083  */
4084 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4085 {
4086         struct bonding *bond = netdev_priv(bond_dev);
4087         struct slave *slave, *start_at;
4088         int slave_no;
4089         int i;
4090         int res = 1;
4091
4092         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4093
4094         bond_for_each_slave(bond, slave, i) {
4095                 slave_no--;
4096                 if (slave_no < 0)
4097                         break;
4098         }
4099
4100         start_at = slave;
4101
4102         bond_for_each_slave_from(bond, slave, i, start_at) {
4103                 if (IS_UP(slave->dev) &&
4104                     (slave->link == BOND_LINK_UP) &&
4105                     bond_is_active_slave(slave)) {
4106                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4107                         break;
4108                 }
4109         }
4110
4111         if (res) {
4112                 /* no suitable interface, frame not sent */
4113                 kfree_skb(skb);
4114         }
4115
4116         return NETDEV_TX_OK;
4117 }
4118
4119 /*
4120  * in broadcast mode, we send everything to all usable interfaces.
4121  */
4122 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4123 {
4124         struct bonding *bond = netdev_priv(bond_dev);
4125         struct slave *slave, *start_at;
4126         struct net_device *tx_dev = NULL;
4127         int i;
4128         int res = 1;
4129
4130         read_lock(&bond->curr_slave_lock);
4131         start_at = bond->curr_active_slave;
4132         read_unlock(&bond->curr_slave_lock);
4133
4134         if (!start_at)
4135                 goto out;
4136
4137         bond_for_each_slave_from(bond, slave, i, start_at) {
4138                 if (IS_UP(slave->dev) &&
4139                     (slave->link == BOND_LINK_UP) &&
4140                     bond_is_active_slave(slave)) {
4141                         if (tx_dev) {
4142                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4143                                 if (!skb2) {
4144                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4145                                                bond_dev->name);
4146                                         continue;
4147                                 }
4148
4149                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4150                                 if (res) {
4151                                         kfree_skb(skb2);
4152                                         continue;
4153                                 }
4154                         }
4155                         tx_dev = slave->dev;
4156                 }
4157         }
4158
4159         if (tx_dev)
4160                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4161
4162 out:
4163         if (res)
4164                 /* no suitable interface, frame not sent */
4165                 kfree_skb(skb);
4166
4167         /* frame sent to all suitable interfaces */
4168         return NETDEV_TX_OK;
4169 }
4170
4171 /*------------------------- Device initialization ---------------------------*/
4172
4173 static void bond_set_xmit_hash_policy(struct bonding *bond)
4174 {
4175         switch (bond->params.xmit_policy) {
4176         case BOND_XMIT_POLICY_LAYER23:
4177                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4178                 break;
4179         case BOND_XMIT_POLICY_LAYER34:
4180                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4181                 break;
4182         case BOND_XMIT_POLICY_LAYER2:
4183         default:
4184                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4185                 break;
4186         }
4187 }
4188
4189 /*
4190  * Lookup the slave that corresponds to a qid
4191  */
4192 static inline int bond_slave_override(struct bonding *bond,
4193                                       struct sk_buff *skb)
4194 {
4195         int i, res = 1;
4196         struct slave *slave = NULL;
4197         struct slave *check_slave;
4198
4199         if (!skb->queue_mapping)
4200                 return 1;
4201
4202         /* Find out if any slaves have the same mapping as this skb. */
4203         bond_for_each_slave(bond, check_slave, i) {
4204                 if (check_slave->queue_id == skb->queue_mapping) {
4205                         slave = check_slave;
4206                         break;
4207                 }
4208         }
4209
4210         /* If the slave isn't UP, use default transmit policy. */
4211         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4212             (slave->link == BOND_LINK_UP)) {
4213                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4214         }
4215
4216         return res;
4217 }
4218
4219
4220 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4221 {
4222         /*
4223          * This helper function exists to help dev_pick_tx get the correct
4224          * destination queue.  Using a helper function skips a call to
4225          * skb_tx_hash and will put the skbs in the queue we expect on their
4226          * way down to the bonding driver.
4227          */
4228         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4229
4230         /*
4231          * Save the original txq to restore before passing to the driver
4232          */
4233         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4234
4235         if (unlikely(txq >= dev->real_num_tx_queues)) {
4236                 do {
4237                         txq -= dev->real_num_tx_queues;
4238                 } while (txq >= dev->real_num_tx_queues);
4239         }
4240         return txq;
4241 }
4242
4243 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4244 {
4245         struct bonding *bond = netdev_priv(dev);
4246
4247         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4248                 if (!bond_slave_override(bond, skb))
4249                         return NETDEV_TX_OK;
4250         }
4251
4252         switch (bond->params.mode) {
4253         case BOND_MODE_ROUNDROBIN:
4254                 return bond_xmit_roundrobin(skb, dev);
4255         case BOND_MODE_ACTIVEBACKUP:
4256                 return bond_xmit_activebackup(skb, dev);
4257         case BOND_MODE_XOR:
4258                 return bond_xmit_xor(skb, dev);
4259         case BOND_MODE_BROADCAST:
4260                 return bond_xmit_broadcast(skb, dev);
4261         case BOND_MODE_8023AD:
4262                 return bond_3ad_xmit_xor(skb, dev);
4263         case BOND_MODE_ALB:
4264         case BOND_MODE_TLB:
4265                 return bond_alb_xmit(skb, dev);
4266         default:
4267                 /* Should never happen, mode already checked */
4268                 pr_err("%s: Error: Unknown bonding mode %d\n",
4269                        dev->name, bond->params.mode);
4270                 WARN_ON_ONCE(1);
4271                 kfree_skb(skb);
4272                 return NETDEV_TX_OK;
4273         }
4274 }
4275
4276 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4277 {
4278         struct bonding *bond = netdev_priv(dev);
4279         netdev_tx_t ret = NETDEV_TX_OK;
4280
4281         /*
4282          * If we risk deadlock from transmitting this in the
4283          * netpoll path, tell netpoll to queue the frame for later tx
4284          */
4285         if (is_netpoll_tx_blocked(dev))
4286                 return NETDEV_TX_BUSY;
4287
4288         read_lock(&bond->lock);
4289
4290         if (bond->slave_cnt)
4291                 ret = __bond_start_xmit(skb, dev);
4292         else
4293                 kfree_skb(skb);
4294
4295         read_unlock(&bond->lock);
4296
4297         return ret;
4298 }
4299
4300 /*
4301  * set bond mode specific net device operations
4302  */
4303 void bond_set_mode_ops(struct bonding *bond, int mode)
4304 {
4305         struct net_device *bond_dev = bond->dev;
4306
4307         switch (mode) {
4308         case BOND_MODE_ROUNDROBIN:
4309                 break;
4310         case BOND_MODE_ACTIVEBACKUP:
4311                 break;
4312         case BOND_MODE_XOR:
4313                 bond_set_xmit_hash_policy(bond);
4314                 break;
4315         case BOND_MODE_BROADCAST:
4316                 break;
4317         case BOND_MODE_8023AD:
4318                 bond_set_xmit_hash_policy(bond);
4319                 break;
4320         case BOND_MODE_ALB:
4321                 /* FALLTHRU */
4322         case BOND_MODE_TLB:
4323                 break;
4324         default:
4325                 /* Should never happen, mode already checked */
4326                 pr_err("%s: Error: Unknown bonding mode %d\n",
4327                        bond_dev->name, mode);
4328                 break;
4329         }
4330 }
4331
4332 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4333                                      struct ethtool_drvinfo *drvinfo)
4334 {
4335         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4336         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4337         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4338                  BOND_ABI_VERSION);
4339 }
4340
4341 static const struct ethtool_ops bond_ethtool_ops = {
4342         .get_drvinfo            = bond_ethtool_get_drvinfo,
4343         .get_link               = ethtool_op_get_link,
4344 };
4345
4346 static const struct net_device_ops bond_netdev_ops = {
4347         .ndo_init               = bond_init,
4348         .ndo_uninit             = bond_uninit,
4349         .ndo_open               = bond_open,
4350         .ndo_stop               = bond_close,
4351         .ndo_start_xmit         = bond_start_xmit,
4352         .ndo_select_queue       = bond_select_queue,
4353         .ndo_get_stats64        = bond_get_stats,
4354         .ndo_do_ioctl           = bond_do_ioctl,
4355         .ndo_change_rx_flags    = bond_change_rx_flags,
4356         .ndo_set_rx_mode        = bond_set_multicast_list,
4357         .ndo_change_mtu         = bond_change_mtu,
4358         .ndo_set_mac_address    = bond_set_mac_address,
4359         .ndo_neigh_setup        = bond_neigh_setup,
4360         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4361         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4362 #ifdef CONFIG_NET_POLL_CONTROLLER
4363         .ndo_netpoll_setup      = bond_netpoll_setup,
4364         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4365         .ndo_poll_controller    = bond_poll_controller,
4366 #endif
4367         .ndo_add_slave          = bond_enslave,
4368         .ndo_del_slave          = bond_release,
4369         .ndo_fix_features       = bond_fix_features,
4370 };
4371
4372 static void bond_destructor(struct net_device *bond_dev)
4373 {
4374         struct bonding *bond = netdev_priv(bond_dev);
4375         if (bond->wq)
4376                 destroy_workqueue(bond->wq);
4377         free_netdev(bond_dev);
4378 }
4379
4380 static void bond_setup(struct net_device *bond_dev)
4381 {
4382         struct bonding *bond = netdev_priv(bond_dev);
4383
4384         /* initialize rwlocks */
4385         rwlock_init(&bond->lock);
4386         rwlock_init(&bond->curr_slave_lock);
4387
4388         bond->params = bonding_defaults;
4389
4390         /* Initialize pointers */
4391         bond->dev = bond_dev;
4392         INIT_LIST_HEAD(&bond->vlan_list);
4393
4394         /* Initialize the device entry points */
4395         ether_setup(bond_dev);
4396         bond_dev->netdev_ops = &bond_netdev_ops;
4397         bond_dev->ethtool_ops = &bond_ethtool_ops;
4398         bond_set_mode_ops(bond, bond->params.mode);
4399
4400         bond_dev->destructor = bond_destructor;
4401
4402         /* Initialize the device options */
4403         bond_dev->tx_queue_len = 0;
4404         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4405         bond_dev->priv_flags |= IFF_BONDING;
4406         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4407
4408         /* At first, we block adding VLANs. That's the only way to
4409          * prevent problems that occur when adding VLANs over an
4410          * empty bond. The block will be removed once non-challenged
4411          * slaves are enslaved.
4412          */
4413         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4414
4415         /* don't acquire bond device's netif_tx_lock when
4416          * transmitting */
4417         bond_dev->features |= NETIF_F_LLTX;
4418
4419         /* By default, we declare the bond to be fully
4420          * VLAN hardware accelerated capable. Special
4421          * care is taken in the various xmit functions
4422          * when there are slaves that are not hw accel
4423          * capable
4424          */
4425
4426         bond_dev->hw_features = BOND_VLAN_FEATURES |
4427                                 NETIF_F_HW_VLAN_TX |
4428                                 NETIF_F_HW_VLAN_RX |
4429                                 NETIF_F_HW_VLAN_FILTER;
4430
4431         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4432         bond_dev->features |= bond_dev->hw_features;
4433 }
4434
4435 /*
4436 * Destroy a bonding device.
4437 * Must be under rtnl_lock when this function is called.
4438 */
4439 static void bond_uninit(struct net_device *bond_dev)
4440 {
4441         struct bonding *bond = netdev_priv(bond_dev);
4442         struct vlan_entry *vlan, *tmp;
4443
4444         bond_netpoll_cleanup(bond_dev);
4445
4446         /* Release the bonded slaves */
4447         bond_release_all(bond_dev);
4448
4449         list_del(&bond->bond_list);
4450
4451         bond_debug_unregister(bond);
4452
4453         __hw_addr_flush(&bond->mc_list);
4454
4455         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4456                 list_del(&vlan->vlan_list);
4457                 kfree(vlan);
4458         }
4459 }
4460
4461 /*------------------------- Module initialization ---------------------------*/
4462
4463 /*
4464  * Convert string input module parms.  Accept either the
4465  * number of the mode or its string name.  A bit complicated because
4466  * some mode names are substrings of other names, and calls from sysfs
4467  * may have whitespace in the name (trailing newlines, for example).
4468  */
4469 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4470 {
4471         int modeint = -1, i, rv;
4472         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4473
4474         for (p = (char *)buf; *p; p++)
4475                 if (!(isdigit(*p) || isspace(*p)))
4476                         break;
4477
4478         if (*p)
4479                 rv = sscanf(buf, "%20s", modestr);
4480         else
4481                 rv = sscanf(buf, "%d", &modeint);
4482
4483         if (!rv)
4484                 return -1;
4485
4486         for (i = 0; tbl[i].modename; i++) {
4487                 if (modeint == tbl[i].mode)
4488                         return tbl[i].mode;
4489                 if (strcmp(modestr, tbl[i].modename) == 0)
4490                         return tbl[i].mode;
4491         }
4492
4493         return -1;
4494 }
4495
4496 static int bond_check_params(struct bond_params *params)
4497 {
4498         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4499
4500         /*
4501          * Convert string parameters.
4502          */
4503         if (mode) {
4504                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4505                 if (bond_mode == -1) {
4506                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4507                                mode == NULL ? "NULL" : mode);
4508                         return -EINVAL;
4509                 }
4510         }
4511
4512         if (xmit_hash_policy) {
4513                 if ((bond_mode != BOND_MODE_XOR) &&
4514                     (bond_mode != BOND_MODE_8023AD)) {
4515                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4516                                bond_mode_name(bond_mode));
4517                 } else {
4518                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4519                                                         xmit_hashtype_tbl);
4520                         if (xmit_hashtype == -1) {
4521                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4522                                        xmit_hash_policy == NULL ? "NULL" :
4523                                        xmit_hash_policy);
4524                                 return -EINVAL;
4525                         }
4526                 }
4527         }
4528
4529         if (lacp_rate) {
4530                 if (bond_mode != BOND_MODE_8023AD) {
4531                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4532                                 bond_mode_name(bond_mode));
4533                 } else {
4534                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4535                         if (lacp_fast == -1) {
4536                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4537                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4538                                 return -EINVAL;
4539                         }
4540                 }
4541         }
4542
4543         if (ad_select) {
4544                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4545                 if (params->ad_select == -1) {
4546                         pr_err("Error: Invalid ad_select \"%s\"\n",
4547                                ad_select == NULL ? "NULL" : ad_select);
4548                         return -EINVAL;
4549                 }
4550
4551                 if (bond_mode != BOND_MODE_8023AD) {
4552                         pr_warning("ad_select param only affects 802.3ad mode\n");
4553                 }
4554         } else {
4555                 params->ad_select = BOND_AD_STABLE;
4556         }
4557
4558         if (max_bonds < 0) {
4559                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4560                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4561                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4562         }
4563
4564         if (miimon < 0) {
4565                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4566                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4567                 miimon = BOND_LINK_MON_INTERV;
4568         }
4569
4570         if (updelay < 0) {
4571                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4572                            updelay, INT_MAX);
4573                 updelay = 0;
4574         }
4575
4576         if (downdelay < 0) {
4577                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4578                            downdelay, INT_MAX);
4579                 downdelay = 0;
4580         }
4581
4582         if ((use_carrier != 0) && (use_carrier != 1)) {
4583                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4584                            use_carrier);
4585                 use_carrier = 1;
4586         }
4587
4588         if (num_peer_notif < 0 || num_peer_notif > 255) {
4589                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4590                            num_peer_notif);
4591                 num_peer_notif = 1;
4592         }
4593
4594         /* reset values for 802.3ad */
4595         if (bond_mode == BOND_MODE_8023AD) {
4596                 if (!miimon) {
4597                         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");
4598                         pr_warning("Forcing miimon to 100msec\n");
4599                         miimon = 100;
4600                 }
4601         }
4602
4603         if (tx_queues < 1 || tx_queues > 255) {
4604                 pr_warning("Warning: tx_queues (%d) should be between "
4605                            "1 and 255, resetting to %d\n",
4606                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4607                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4608         }
4609
4610         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4611                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4612                            "not of valid value (0/1), so it was set to "
4613                            "0\n", all_slaves_active);
4614                 all_slaves_active = 0;
4615         }
4616
4617         if (resend_igmp < 0 || resend_igmp > 255) {
4618                 pr_warning("Warning: resend_igmp (%d) should be between "
4619                            "0 and 255, resetting to %d\n",
4620                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4621                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4622         }
4623
4624         /* reset values for TLB/ALB */
4625         if ((bond_mode == BOND_MODE_TLB) ||
4626             (bond_mode == BOND_MODE_ALB)) {
4627                 if (!miimon) {
4628                         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");
4629                         pr_warning("Forcing miimon to 100msec\n");
4630                         miimon = 100;
4631                 }
4632         }
4633
4634         if (bond_mode == BOND_MODE_ALB) {
4635                 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",
4636                           updelay);
4637         }
4638
4639         if (!miimon) {
4640                 if (updelay || downdelay) {
4641                         /* just warn the user the up/down delay will have
4642                          * no effect since miimon is zero...
4643                          */
4644                         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",
4645                                    updelay, downdelay);
4646                 }
4647         } else {
4648                 /* don't allow arp monitoring */
4649                 if (arp_interval) {
4650                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4651                                    miimon, arp_interval);
4652                         arp_interval = 0;
4653                 }
4654
4655                 if ((updelay % miimon) != 0) {
4656                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4657                                    updelay, miimon,
4658                                    (updelay / miimon) * miimon);
4659                 }
4660
4661                 updelay /= miimon;
4662
4663                 if ((downdelay % miimon) != 0) {
4664                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4665                                    downdelay, miimon,
4666                                    (downdelay / miimon) * miimon);
4667                 }
4668
4669                 downdelay /= miimon;
4670         }
4671
4672         if (arp_interval < 0) {
4673                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4674                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4675                 arp_interval = BOND_LINK_ARP_INTERV;
4676         }
4677
4678         for (arp_ip_count = 0;
4679              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4680              arp_ip_count++) {
4681                 /* not complete check, but should be good enough to
4682                    catch mistakes */
4683                 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4684                 if (!isdigit(arp_ip_target[arp_ip_count][0]) ||
4685                     ip == 0 || ip == htonl(INADDR_BROADCAST)) {
4686                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4687                                    arp_ip_target[arp_ip_count]);
4688                         arp_interval = 0;
4689                 } else {
4690                         arp_target[arp_ip_count] = ip;
4691                 }
4692         }
4693
4694         if (arp_interval && !arp_ip_count) {
4695                 /* don't allow arping if no arp_ip_target given... */
4696                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4697                            arp_interval);
4698                 arp_interval = 0;
4699         }
4700
4701         if (arp_validate) {
4702                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4703                         pr_err("arp_validate only supported in active-backup mode\n");
4704                         return -EINVAL;
4705                 }
4706                 if (!arp_interval) {
4707                         pr_err("arp_validate requires arp_interval\n");
4708                         return -EINVAL;
4709                 }
4710
4711                 arp_validate_value = bond_parse_parm(arp_validate,
4712                                                      arp_validate_tbl);
4713                 if (arp_validate_value == -1) {
4714                         pr_err("Error: invalid arp_validate \"%s\"\n",
4715                                arp_validate == NULL ? "NULL" : arp_validate);
4716                         return -EINVAL;
4717                 }
4718         } else
4719                 arp_validate_value = 0;
4720
4721         if (miimon) {
4722                 pr_info("MII link monitoring set to %d ms\n", miimon);
4723         } else if (arp_interval) {
4724                 int i;
4725
4726                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4727                         arp_interval,
4728                         arp_validate_tbl[arp_validate_value].modename,
4729                         arp_ip_count);
4730
4731                 for (i = 0; i < arp_ip_count; i++)
4732                         pr_info(" %s", arp_ip_target[i]);
4733
4734                 pr_info("\n");
4735
4736         } else if (max_bonds) {
4737                 /* miimon and arp_interval not set, we need one so things
4738                  * work as expected, see bonding.txt for details
4739                  */
4740                 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");
4741         }
4742
4743         if (primary && !USES_PRIMARY(bond_mode)) {
4744                 /* currently, using a primary only makes sense
4745                  * in active backup, TLB or ALB modes
4746                  */
4747                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4748                            primary, bond_mode_name(bond_mode));
4749                 primary = NULL;
4750         }
4751
4752         if (primary && primary_reselect) {
4753                 primary_reselect_value = bond_parse_parm(primary_reselect,
4754                                                          pri_reselect_tbl);
4755                 if (primary_reselect_value == -1) {
4756                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4757                                primary_reselect ==
4758                                         NULL ? "NULL" : primary_reselect);
4759                         return -EINVAL;
4760                 }
4761         } else {
4762                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4763         }
4764
4765         if (fail_over_mac) {
4766                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4767                                                       fail_over_mac_tbl);
4768                 if (fail_over_mac_value == -1) {
4769                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4770                                arp_validate == NULL ? "NULL" : arp_validate);
4771                         return -EINVAL;
4772                 }
4773
4774                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4775                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4776         } else {
4777                 fail_over_mac_value = BOND_FOM_NONE;
4778         }
4779
4780         /* fill params struct with the proper values */
4781         params->mode = bond_mode;
4782         params->xmit_policy = xmit_hashtype;
4783         params->miimon = miimon;
4784         params->num_peer_notif = num_peer_notif;
4785         params->arp_interval = arp_interval;
4786         params->arp_validate = arp_validate_value;
4787         params->updelay = updelay;
4788         params->downdelay = downdelay;
4789         params->use_carrier = use_carrier;
4790         params->lacp_fast = lacp_fast;
4791         params->primary[0] = 0;
4792         params->primary_reselect = primary_reselect_value;
4793         params->fail_over_mac = fail_over_mac_value;
4794         params->tx_queues = tx_queues;
4795         params->all_slaves_active = all_slaves_active;
4796         params->resend_igmp = resend_igmp;
4797         params->min_links = min_links;
4798
4799         if (primary) {
4800                 strncpy(params->primary, primary, IFNAMSIZ);
4801                 params->primary[IFNAMSIZ - 1] = 0;
4802         }
4803
4804         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4805
4806         return 0;
4807 }
4808
4809 static struct lock_class_key bonding_netdev_xmit_lock_key;
4810 static struct lock_class_key bonding_netdev_addr_lock_key;
4811 static struct lock_class_key bonding_tx_busylock_key;
4812
4813 static void bond_set_lockdep_class_one(struct net_device *dev,
4814                                        struct netdev_queue *txq,
4815                                        void *_unused)
4816 {
4817         lockdep_set_class(&txq->_xmit_lock,
4818                           &bonding_netdev_xmit_lock_key);
4819 }
4820
4821 static void bond_set_lockdep_class(struct net_device *dev)
4822 {
4823         lockdep_set_class(&dev->addr_list_lock,
4824                           &bonding_netdev_addr_lock_key);
4825         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4826         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4827 }
4828
4829 /*
4830  * Called from registration process
4831  */
4832 static int bond_init(struct net_device *bond_dev)
4833 {
4834         struct bonding *bond = netdev_priv(bond_dev);
4835         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4836         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4837
4838         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4839
4840         /*
4841          * Initialize locks that may be required during
4842          * en/deslave operations.  All of the bond_open work
4843          * (of which this is part) should really be moved to
4844          * a phase prior to dev_open
4845          */
4846         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4847         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4848
4849         bond->wq = create_singlethread_workqueue(bond_dev->name);
4850         if (!bond->wq)
4851                 return -ENOMEM;
4852
4853         bond_set_lockdep_class(bond_dev);
4854
4855         list_add_tail(&bond->bond_list, &bn->dev_list);
4856
4857         bond_prepare_sysfs_group(bond);
4858
4859         bond_debug_register(bond);
4860
4861         __hw_addr_init(&bond->mc_list);
4862         return 0;
4863 }
4864
4865 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4866 {
4867         if (tb[IFLA_ADDRESS]) {
4868                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4869                         return -EINVAL;
4870                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4871                         return -EADDRNOTAVAIL;
4872         }
4873         return 0;
4874 }
4875
4876 static unsigned int bond_get_num_tx_queues(void)
4877 {
4878         return tx_queues;
4879 }
4880
4881 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4882         .kind                   = "bond",
4883         .priv_size              = sizeof(struct bonding),
4884         .setup                  = bond_setup,
4885         .validate               = bond_validate,
4886         .get_num_tx_queues      = bond_get_num_tx_queues,
4887         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4888                                                              as for TX queues */
4889 };
4890
4891 /* Create a new bond based on the specified name and bonding parameters.
4892  * If name is NULL, obtain a suitable "bond%d" name for us.
4893  * Caller must NOT hold rtnl_lock; we need to release it here before we
4894  * set up our sysfs entries.
4895  */
4896 int bond_create(struct net *net, const char *name)
4897 {
4898         struct net_device *bond_dev;
4899         int res;
4900
4901         rtnl_lock();
4902
4903         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4904                                    name ? name : "bond%d",
4905                                    bond_setup, tx_queues);
4906         if (!bond_dev) {
4907                 pr_err("%s: eek! can't alloc netdev!\n", name);
4908                 rtnl_unlock();
4909                 return -ENOMEM;
4910         }
4911
4912         dev_net_set(bond_dev, net);
4913         bond_dev->rtnl_link_ops = &bond_link_ops;
4914
4915         res = register_netdevice(bond_dev);
4916
4917         netif_carrier_off(bond_dev);
4918
4919         rtnl_unlock();
4920         if (res < 0)
4921                 bond_destructor(bond_dev);
4922         return res;
4923 }
4924
4925 static int __net_init bond_net_init(struct net *net)
4926 {
4927         struct bond_net *bn = net_generic(net, bond_net_id);
4928
4929         bn->net = net;
4930         INIT_LIST_HEAD(&bn->dev_list);
4931
4932         bond_create_proc_dir(bn);
4933         bond_create_sysfs(bn);
4934
4935         return 0;
4936 }
4937
4938 static void __net_exit bond_net_exit(struct net *net)
4939 {
4940         struct bond_net *bn = net_generic(net, bond_net_id);
4941
4942         bond_destroy_sysfs(bn);
4943         bond_destroy_proc_dir(bn);
4944 }
4945
4946 static struct pernet_operations bond_net_ops = {
4947         .init = bond_net_init,
4948         .exit = bond_net_exit,
4949         .id   = &bond_net_id,
4950         .size = sizeof(struct bond_net),
4951 };
4952
4953 static int __init bonding_init(void)
4954 {
4955         int i;
4956         int res;
4957
4958         pr_info("%s", bond_version);
4959
4960         res = bond_check_params(&bonding_defaults);
4961         if (res)
4962                 goto out;
4963
4964         res = register_pernet_subsys(&bond_net_ops);
4965         if (res)
4966                 goto out;
4967
4968         res = rtnl_link_register(&bond_link_ops);
4969         if (res)
4970                 goto err_link;
4971
4972         bond_create_debugfs();
4973
4974         for (i = 0; i < max_bonds; i++) {
4975                 res = bond_create(&init_net, NULL);
4976                 if (res)
4977                         goto err;
4978         }
4979
4980         register_netdevice_notifier(&bond_netdev_notifier);
4981 out:
4982         return res;
4983 err:
4984         rtnl_link_unregister(&bond_link_ops);
4985 err_link:
4986         unregister_pernet_subsys(&bond_net_ops);
4987         goto out;
4988
4989 }
4990
4991 static void __exit bonding_exit(void)
4992 {
4993         unregister_netdevice_notifier(&bond_netdev_notifier);
4994
4995         bond_destroy_debugfs();
4996
4997         rtnl_link_unregister(&bond_link_ops);
4998         unregister_pernet_subsys(&bond_net_ops);
4999
5000 #ifdef CONFIG_NET_POLL_CONTROLLER
5001         /*
5002          * Make sure we don't have an imbalance on our netpoll blocking
5003          */
5004         WARN_ON(atomic_read(&netpoll_block_tx));
5005 #endif
5006 }
5007
5008 module_init(bonding_init);
5009 module_exit(bonding_exit);
5010 MODULE_LICENSE("GPL");
5011 MODULE_VERSION(DRV_VERSION);
5012 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5013 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5014 MODULE_ALIAS_RTNL_LINK("bond");