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                 netif_addr_lock_bh(bond->dev);
 850                 netdev_for_each_mc_addr(ha, bond->dev)
 851                         dev_mc_del(old_active->dev, ha->addr);
 852                 netif_addr_unlock_bh(bond->dev);
 853         }
 854
 855         if (new_active) {
 856                 /* FIXME: Signal errors upstream. */
 857                 if (bond->dev->flags & IFF_PROMISC)
 858                         dev_set_promiscuity(new_active->dev, 1);
 859
 860                 if (bond->dev->flags & IFF_ALLMULTI)
 861                         dev_set_allmulti(new_active->dev, 1);
 862
 863                 netif_addr_lock_bh(bond->dev);
 864                 netdev_for_each_mc_addr(ha, bond->dev)
 865                         dev_mc_add(new_active->dev, ha->addr);
 866                 netif_addr_unlock_bh(bond->dev);
 867         }
 868 }
 869
 870 /*
 871  * bond_do_fail_over_mac
 872  *
 873  * Perform special MAC address swapping for fail_over_mac settings
 874  *
 875  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 876  */
 877 static void bond_do_fail_over_mac(struct bonding *bond,
 878                                   struct slave *new_active,
 879                                   struct slave *old_active)
 880         __releases(&bond->curr_slave_lock)
 881         __releases(&bond->lock)
 882         __acquires(&bond->lock)
 883         __acquires(&bond->curr_slave_lock)
 884 {
 885         u8 tmp_mac[ETH_ALEN];
 886         struct sockaddr saddr;
 887         int rv;
 888
 889         switch (bond->params.fail_over_mac) {
 890         case BOND_FOM_ACTIVE:
 891                 if (new_active) {
 892                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 893                                new_active->dev->addr_len);
 894                         write_unlock_bh(&bond->curr_slave_lock);
 895                         read_unlock(&bond->lock);
 896                         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
 897                         read_lock(&bond->lock);
 898                         write_lock_bh(&bond->curr_slave_lock);
 899                 }
 900                 break;
 901         case BOND_FOM_FOLLOW:
 902                 /*
 903                  * if new_active && old_active, swap them
 904                  * if just old_active, do nothing (going to no active slave)
 905                  * if just new_active, set new_active to bond's MAC
 906                  */
 907                 if (!new_active)
 908                         return;
 909
 910                 write_unlock_bh(&bond->curr_slave_lock);
 911                 read_unlock(&bond->lock);
 912
 913                 if (old_active) {
 914                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 915                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 916                                ETH_ALEN);
 917                         saddr.sa_family = new_active->dev->type;
 918                 } else {
 919                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 920                         saddr.sa_family = bond->dev->type;
 921                 }
 922
 923                 rv = dev_set_mac_address(new_active->dev, &saddr);
 924                 if (rv) {
 925                         pr_err("%s: Error %d setting MAC of slave %s\n",
 926                                bond->dev->name, -rv, new_active->dev->name);
 927                         goto out;
 928                 }
 929
 930                 if (!old_active)
 931                         goto out;
 932
 933                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 934                 saddr.sa_family = old_active->dev->type;
 935
 936                 rv = dev_set_mac_address(old_active->dev, &saddr);
 937                 if (rv)
 938                         pr_err("%s: Error %d setting MAC of slave %s\n",
 939                                bond->dev->name, -rv, new_active->dev->name);
 940 out:
 941                 read_lock(&bond->lock);
 942                 write_lock_bh(&bond->curr_slave_lock);
 943                 break;
 944         default:
 945                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 946                        bond->dev->name, bond->params.fail_over_mac);
 947                 break;
 948         }
 949
 950 }
 951
 952 static bool bond_should_change_active(struct bonding *bond)
 953 {
 954         struct slave *prim = bond->primary_slave;
 955         struct slave *curr = bond->curr_active_slave;
 956
 957         if (!prim || !curr || curr->link != BOND_LINK_UP)
 958                 return true;
 959         if (bond->force_primary) {
 960                 bond->force_primary = false;
 961                 return true;
 962         }
 963         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 964             (prim->speed < curr->speed ||
 965              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 966                 return false;
 967         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 968                 return false;
 969         return true;
 970 }
 971
 972 /**
 973  * find_best_interface - select the best available slave to be the active one
 974  * @bond: our bonding struct
 975  *
 976  * Warning: Caller must hold curr_slave_lock for writing.
 977  */
 978 static struct slave *bond_find_best_slave(struct bonding *bond)
 979 {
 980         struct slave *new_active, *old_active;
 981         struct slave *bestslave = NULL;
 982         int mintime = bond->params.updelay;
 983         int i;
 984
 985         new_active = bond->curr_active_slave;
 986
 987         if (!new_active) { /* there were no active slaves left */
 988                 if (bond->slave_cnt > 0)   /* found one slave */
 989                         new_active = bond->first_slave;
 990                 else
 991                         return NULL; /* still no slave, return NULL */
 992         }
 993
 994         if ((bond->primary_slave) &&
 995             bond->primary_slave->link == BOND_LINK_UP &&
 996             bond_should_change_active(bond)) {
 997                 new_active = bond->primary_slave;
 998         }
 999
1000         /* remember where to stop iterating over the slaves */
1001         old_active = new_active;
1002
1003         bond_for_each_slave_from(bond, new_active, i, old_active) {
1004                 if (new_active->link == BOND_LINK_UP) {
1005                         return new_active;
1006                 } else if (new_active->link == BOND_LINK_BACK &&
1007                            IS_UP(new_active->dev)) {
1008                         /* link up, but waiting for stabilization */
1009                         if (new_active->delay < mintime) {
1010                                 mintime = new_active->delay;
1011                                 bestslave = new_active;
1012                         }
1013                 }
1014         }
1015
1016         return bestslave;
1017 }
1018
1019 static bool bond_should_notify_peers(struct bonding *bond)
1020 {
1021         struct slave *slave = bond->curr_active_slave;
1022
1023         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1024                  bond->dev->name, slave ? slave->dev->name : "NULL");
1025
1026         if (!slave || !bond->send_peer_notif ||
1027             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1028                 return false;
1029
1030         bond->send_peer_notif--;
1031         return true;
1032 }
1033
1034 /**
1035  * change_active_interface - change the active slave into the specified one
1036  * @bond: our bonding struct
1037  * @new: the new slave to make the active one
1038  *
1039  * Set the new slave to the bond's settings and unset them on the old
1040  * curr_active_slave.
1041  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1042  *
1043  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1044  * because it is apparently the best available slave we have, even though its
1045  * updelay hasn't timed out yet.
1046  *
1047  * If new_active is not NULL, caller must hold bond->lock for read and
1048  * curr_slave_lock for write_bh.
1049  */
1050 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1051 {
1052         struct slave *old_active = bond->curr_active_slave;
1053
1054         if (old_active == new_active)
1055                 return;
1056
1057         if (new_active) {
1058                 new_active->jiffies = jiffies;
1059
1060                 if (new_active->link == BOND_LINK_BACK) {
1061                         if (USES_PRIMARY(bond->params.mode)) {
1062                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1063                                         bond->dev->name, new_active->dev->name,
1064                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1065                         }
1066
1067                         new_active->delay = 0;
1068                         new_active->link = BOND_LINK_UP;
1069
1070                         if (bond->params.mode == BOND_MODE_8023AD)
1071                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1072
1073                         if (bond_is_lb(bond))
1074                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1075                 } else {
1076                         if (USES_PRIMARY(bond->params.mode)) {
1077                                 pr_info("%s: making interface %s the new active one.\n",
1078                                         bond->dev->name, new_active->dev->name);
1079                         }
1080                 }
1081         }
1082
1083         if (USES_PRIMARY(bond->params.mode))
1084                 bond_mc_swap(bond, new_active, old_active);
1085
1086         if (bond_is_lb(bond)) {
1087                 bond_alb_handle_active_change(bond, new_active);
1088                 if (old_active)
1089                         bond_set_slave_inactive_flags(old_active);
1090                 if (new_active)
1091                         bond_set_slave_active_flags(new_active);
1092         } else {
1093                 bond->curr_active_slave = new_active;
1094         }
1095
1096         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1097                 if (old_active)
1098                         bond_set_slave_inactive_flags(old_active);
1099
1100                 if (new_active) {
1101                         bool should_notify_peers = false;
1102
1103                         bond_set_slave_active_flags(new_active);
1104
1105                         if (bond->params.fail_over_mac)
1106                                 bond_do_fail_over_mac(bond, new_active,
1107                                                       old_active);
1108
1109                         if (netif_running(bond->dev)) {
1110                                 bond->send_peer_notif =
1111                                         bond->params.num_peer_notif;
1112                                 should_notify_peers =
1113                                         bond_should_notify_peers(bond);
1114                         }
1115
1116                         write_unlock_bh(&bond->curr_slave_lock);
1117                         read_unlock(&bond->lock);
1118
1119                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1120                         if (should_notify_peers)
1121                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1122                                                          bond->dev);
1123
1124                         read_lock(&bond->lock);
1125                         write_lock_bh(&bond->curr_slave_lock);
1126                 }
1127         }
1128
1129         /* resend IGMP joins since active slave has changed or
1130          * all were sent on curr_active_slave.
1131          * resend only if bond is brought up with the affected
1132          * bonding modes and the retransmission is enabled */
1133         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1134             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1135              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1136                 bond->igmp_retrans = bond->params.resend_igmp;
1137                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1138         }
1139 }
1140
1141 /**
1142  * bond_select_active_slave - select a new active slave, if needed
1143  * @bond: our bonding struct
1144  *
1145  * This functions should be called when one of the following occurs:
1146  * - The old curr_active_slave has been released or lost its link.
1147  * - The primary_slave has got its link back.
1148  * - A slave has got its link back and there's no old curr_active_slave.
1149  *
1150  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1151  */
1152 void bond_select_active_slave(struct bonding *bond)
1153 {
1154         struct slave *best_slave;
1155         int rv;
1156
1157         best_slave = bond_find_best_slave(bond);
1158         if (best_slave != bond->curr_active_slave) {
1159                 bond_change_active_slave(bond, best_slave);
1160                 rv = bond_set_carrier(bond);
1161                 if (!rv)
1162                         return;
1163
1164                 if (netif_carrier_ok(bond->dev)) {
1165                         pr_info("%s: first active interface up!\n",
1166                                 bond->dev->name);
1167                 } else {
1168                         pr_info("%s: now running without any active interface !\n",
1169                                 bond->dev->name);
1170                 }
1171         }
1172 }
1173
1174 /*--------------------------- slave list handling ---------------------------*/
1175
1176 /*
1177  * This function attaches the slave to the end of list.
1178  *
1179  * bond->lock held for writing by caller.
1180  */
1181 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1182 {
1183         if (bond->first_slave == NULL) { /* attaching the first slave */
1184                 new_slave->next = new_slave;
1185                 new_slave->prev = new_slave;
1186                 bond->first_slave = new_slave;
1187         } else {
1188                 new_slave->next = bond->first_slave;
1189                 new_slave->prev = bond->first_slave->prev;
1190                 new_slave->next->prev = new_slave;
1191                 new_slave->prev->next = new_slave;
1192         }
1193
1194         bond->slave_cnt++;
1195 }
1196
1197 /*
1198  * This function detaches the slave from the list.
1199  * WARNING: no check is made to verify if the slave effectively
1200  * belongs to <bond>.
1201  * Nothing is freed on return, structures are just unchained.
1202  * If any slave pointer in bond was pointing to <slave>,
1203  * it should be changed by the calling function.
1204  *
1205  * bond->lock held for writing by caller.
1206  */
1207 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1208 {
1209         if (slave->next)
1210                 slave->next->prev = slave->prev;
1211
1212         if (slave->prev)
1213                 slave->prev->next = slave->next;
1214
1215         if (bond->first_slave == slave) { /* slave is the first slave */
1216                 if (bond->slave_cnt > 1) { /* there are more slave */
1217                         bond->first_slave = slave->next;
1218                 } else {
1219                         bond->first_slave = NULL; /* slave was the last one */
1220                 }
1221         }
1222
1223         slave->next = NULL;
1224         slave->prev = NULL;
1225         bond->slave_cnt--;
1226 }
1227
1228 #ifdef CONFIG_NET_POLL_CONTROLLER
1229 static inline int slave_enable_netpoll(struct slave *slave)
1230 {
1231         struct netpoll *np;
1232         int err = 0;
1233
1234         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1235         err = -ENOMEM;
1236         if (!np)
1237                 goto out;
1238
1239         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1240         if (err) {
1241                 kfree(np);
1242                 goto out;
1243         }
1244         slave->np = np;
1245 out:
1246         return err;
1247 }
1248 static inline void slave_disable_netpoll(struct slave *slave)
1249 {
1250         struct netpoll *np = slave->np;
1251
1252         if (!np)
1253                 return;
1254
1255         slave->np = NULL;
1256         __netpoll_free_async(np);
1257 }
1258 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1259 {
1260         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1261                 return false;
1262         if (!slave_dev->netdev_ops->ndo_poll_controller)
1263                 return false;
1264         return true;
1265 }
1266
1267 static void bond_poll_controller(struct net_device *bond_dev)
1268 {
1269 }
1270
1271 static void __bond_netpoll_cleanup(struct bonding *bond)
1272 {
1273         struct slave *slave;
1274         int i;
1275
1276         bond_for_each_slave(bond, slave, i)
1277                 if (IS_UP(slave->dev))
1278                         slave_disable_netpoll(slave);
1279 }
1280 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1281 {
1282         struct bonding *bond = netdev_priv(bond_dev);
1283
1284         read_lock(&bond->lock);
1285         __bond_netpoll_cleanup(bond);
1286         read_unlock(&bond->lock);
1287 }
1288
1289 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1290 {
1291         struct bonding *bond = netdev_priv(dev);
1292         struct slave *slave;
1293         int i, err = 0;
1294
1295         read_lock(&bond->lock);
1296         bond_for_each_slave(bond, slave, i) {
1297                 err = slave_enable_netpoll(slave);
1298                 if (err) {
1299                         __bond_netpoll_cleanup(bond);
1300                         break;
1301                 }
1302         }
1303         read_unlock(&bond->lock);
1304         return err;
1305 }
1306
1307 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1308 {
1309         return bond->dev->npinfo;
1310 }
1311
1312 #else
1313 static inline int slave_enable_netpoll(struct slave *slave)
1314 {
1315         return 0;
1316 }
1317 static inline void slave_disable_netpoll(struct slave *slave)
1318 {
1319 }
1320 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1321 {
1322 }
1323 #endif
1324
1325 /*---------------------------------- IOCTL ----------------------------------*/
1326
1327 static void bond_set_dev_addr(struct net_device *bond_dev,
1328                               struct net_device *slave_dev)
1329 {
1330         pr_debug("bond_dev=%p\n", bond_dev);
1331         pr_debug("slave_dev=%p\n", slave_dev);
1332         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1333         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1334         bond_dev->addr_assign_type = NET_ADDR_SET;
1335         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
1336 }
1337
1338 static netdev_features_t bond_fix_features(struct net_device *dev,
1339         netdev_features_t features)
1340 {
1341         struct slave *slave;
1342         struct bonding *bond = netdev_priv(dev);
1343         netdev_features_t mask;
1344         int i;
1345
1346         read_lock(&bond->lock);
1347
1348         if (!bond->first_slave) {
1349                 /* Disable adding VLANs to empty bond. But why? --mq */
1350                 features |= NETIF_F_VLAN_CHALLENGED;
1351                 goto out;
1352         }
1353
1354         mask = features;
1355         features &= ~NETIF_F_ONE_FOR_ALL;
1356         features |= NETIF_F_ALL_FOR_ALL;
1357
1358         bond_for_each_slave(bond, slave, i) {
1359                 features = netdev_increment_features(features,
1360                                                      slave->dev->features,
1361                                                      mask);
1362         }
1363
1364 out:
1365         read_unlock(&bond->lock);
1366         return features;
1367 }
1368
1369 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1370                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1371                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1372
1373 static void bond_compute_features(struct bonding *bond)
1374 {
1375         struct slave *slave;
1376         struct net_device *bond_dev = bond->dev;
1377         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1378         unsigned short max_hard_header_len = ETH_HLEN;
1379         unsigned int gso_max_size = GSO_MAX_SIZE;
1380         u16 gso_max_segs = GSO_MAX_SEGS;
1381         int i;
1382         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1383
1384         read_lock(&bond->lock);
1385
1386         if (!bond->first_slave)
1387                 goto done;
1388
1389         bond_for_each_slave(bond, slave, i) {
1390                 vlan_features = netdev_increment_features(vlan_features,
1391                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1392
1393                 dst_release_flag &= slave->dev->priv_flags;
1394                 if (slave->dev->hard_header_len > max_hard_header_len)
1395                         max_hard_header_len = slave->dev->hard_header_len;
1396
1397                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1398                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1399         }
1400
1401 done:
1402         bond_dev->vlan_features = vlan_features;
1403         bond_dev->hard_header_len = max_hard_header_len;
1404         bond_dev->gso_max_segs = gso_max_segs;
1405         netif_set_gso_max_size(bond_dev, gso_max_size);
1406
1407         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1408         bond_dev->priv_flags = flags | dst_release_flag;
1409
1410         read_unlock(&bond->lock);
1411
1412         netdev_change_features(bond_dev);
1413 }
1414
1415 static void bond_setup_by_slave(struct net_device *bond_dev,
1416                                 struct net_device *slave_dev)
1417 {
1418         struct bonding *bond = netdev_priv(bond_dev);
1419
1420         bond_dev->header_ops        = slave_dev->header_ops;
1421
1422         bond_dev->type              = slave_dev->type;
1423         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1424         bond_dev->addr_len          = slave_dev->addr_len;
1425
1426         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1427                 slave_dev->addr_len);
1428         bond->setup_by_slave = 1;
1429 }
1430
1431 /* On bonding slaves other than the currently active slave, suppress
1432  * duplicates except for alb non-mcast/bcast.
1433  */
1434 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1435                                             struct slave *slave,
1436                                             struct bonding *bond)
1437 {
1438         if (bond_is_slave_inactive(slave)) {
1439                 if (bond->params.mode == BOND_MODE_ALB &&
1440                     skb->pkt_type != PACKET_BROADCAST &&
1441                     skb->pkt_type != PACKET_MULTICAST)
1442                         return false;
1443                 return true;
1444         }
1445         return false;
1446 }
1447
1448 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1449 {
1450         struct sk_buff *skb = *pskb;
1451         struct slave *slave;
1452         struct bonding *bond;
1453         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1454                           struct slave *);
1455         int ret = RX_HANDLER_ANOTHER;
1456
1457         skb = skb_share_check(skb, GFP_ATOMIC);
1458         if (unlikely(!skb))
1459                 return RX_HANDLER_CONSUMED;
1460
1461         *pskb = skb;
1462
1463         slave = bond_slave_get_rcu(skb->dev);
1464         bond = slave->bond;
1465
1466         if (bond->params.arp_interval)
1467                 slave->dev->last_rx = jiffies;
1468
1469         recv_probe = ACCESS_ONCE(bond->recv_probe);
1470         if (recv_probe) {
1471                 ret = recv_probe(skb, bond, slave);
1472                 if (ret == RX_HANDLER_CONSUMED) {
1473                         consume_skb(skb);
1474                         return ret;
1475                 }
1476         }
1477
1478         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1479                 return RX_HANDLER_EXACT;
1480         }
1481
1482         skb->dev = bond->dev;
1483
1484         if (bond->params.mode == BOND_MODE_ALB &&
1485             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1486             skb->pkt_type == PACKET_HOST) {
1487
1488                 if (unlikely(skb_cow_head(skb,
1489                                           skb->data - skb_mac_header(skb)))) {
1490                         kfree_skb(skb);
1491                         return RX_HANDLER_CONSUMED;
1492                 }
1493                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1494         }
1495
1496         return ret;
1497 }
1498
1499 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1500                                       struct net_device *slave_dev)
1501 {
1502         int err;
1503
1504         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1505         if (err)
1506                 return err;
1507         slave_dev->flags |= IFF_SLAVE;
1508         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1509         return 0;
1510 }
1511
1512 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1513                                   struct net_device *slave_dev)
1514 {
1515         netdev_upper_dev_unlink(slave_dev, bond_dev);
1516         slave_dev->flags &= ~IFF_SLAVE;
1517         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1518 }
1519
1520 /* enslave device <slave> to bond device <master> */
1521 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1522 {
1523         struct bonding *bond = netdev_priv(bond_dev);
1524         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1525         struct slave *new_slave = NULL;
1526         struct netdev_hw_addr *ha;
1527         struct sockaddr addr;
1528         int link_reporting;
1529         int res = 0;
1530
1531         if (!bond->params.use_carrier &&
1532             slave_dev->ethtool_ops->get_link == NULL &&
1533             slave_ops->ndo_do_ioctl == NULL) {
1534                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1535                            bond_dev->name, slave_dev->name);
1536         }
1537
1538         /* already enslaved */
1539         if (slave_dev->flags & IFF_SLAVE) {
1540                 pr_debug("Error, Device was already enslaved\n");
1541                 return -EBUSY;
1542         }
1543
1544         /* vlan challenged mutual exclusion */
1545         /* no need to lock since we're protected by rtnl_lock */
1546         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1547                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1548                 if (vlan_uses_dev(bond_dev)) {
1549                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1550                                bond_dev->name, slave_dev->name, bond_dev->name);
1551                         return -EPERM;
1552                 } else {
1553                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1554                                    bond_dev->name, slave_dev->name,
1555                                    slave_dev->name, bond_dev->name);
1556                 }
1557         } else {
1558                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1559         }
1560
1561         /*
1562          * Old ifenslave binaries are no longer supported.  These can
1563          * be identified with moderate accuracy by the state of the slave:
1564          * the current ifenslave will set the interface down prior to
1565          * enslaving it; the old ifenslave will not.
1566          */
1567         if ((slave_dev->flags & IFF_UP)) {
1568                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1569                        slave_dev->name);
1570                 res = -EPERM;
1571                 goto err_undo_flags;
1572         }
1573
1574         /* set bonding device ether type by slave - bonding netdevices are
1575          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1576          * there is a need to override some of the type dependent attribs/funcs.
1577          *
1578          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1579          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1580          */
1581         if (bond->slave_cnt == 0) {
1582                 if (bond_dev->type != slave_dev->type) {
1583                         pr_debug("%s: change device type from %d to %d\n",
1584                                  bond_dev->name,
1585                                  bond_dev->type, slave_dev->type);
1586
1587                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1588                                                        bond_dev);
1589                         res = notifier_to_errno(res);
1590                         if (res) {
1591                                 pr_err("%s: refused to change device type\n",
1592                                        bond_dev->name);
1593                                 res = -EBUSY;
1594                                 goto err_undo_flags;
1595                         }
1596
1597                         /* Flush unicast and multicast addresses */
1598                         dev_uc_flush(bond_dev);
1599                         dev_mc_flush(bond_dev);
1600
1601                         if (slave_dev->type != ARPHRD_ETHER)
1602                                 bond_setup_by_slave(bond_dev, slave_dev);
1603                         else {
1604                                 ether_setup(bond_dev);
1605                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1606                         }
1607
1608                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1609                                                  bond_dev);
1610                 }
1611         } else if (bond_dev->type != slave_dev->type) {
1612                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1613                        slave_dev->name,
1614                        slave_dev->type, bond_dev->type);
1615                 res = -EINVAL;
1616                 goto err_undo_flags;
1617         }
1618
1619         if (slave_ops->ndo_set_mac_address == NULL) {
1620                 if (bond->slave_cnt == 0) {
1621                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1622                                    bond_dev->name);
1623                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1624                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1625                         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",
1626                                bond_dev->name);
1627                         res = -EOPNOTSUPP;
1628                         goto err_undo_flags;
1629                 }
1630         }
1631
1632         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1633
1634         /* If this is the first slave, then we need to set the master's hardware
1635          * address to be the same as the slave's. */
1636         if (bond->slave_cnt == 0 && bond->dev_addr_from_first)
1637                 bond_set_dev_addr(bond->dev, slave_dev);
1638
1639         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1640         if (!new_slave) {
1641                 res = -ENOMEM;
1642                 goto err_undo_flags;
1643         }
1644
1645         /*
1646          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1647          * is set via sysfs or module option if desired.
1648          */
1649         new_slave->queue_id = 0;
1650
1651         /* Save slave's original mtu and then set it to match the bond */
1652         new_slave->original_mtu = slave_dev->mtu;
1653         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1654         if (res) {
1655                 pr_debug("Error %d calling dev_set_mtu\n", res);
1656                 goto err_free;
1657         }
1658
1659         /*
1660          * Save slave's original ("permanent") mac address for modes
1661          * that need it, and for restoring it upon release, and then
1662          * set it to the master's address
1663          */
1664         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1665
1666         if (!bond->params.fail_over_mac) {
1667                 /*
1668                  * Set slave to master's mac address.  The application already
1669                  * set the master's mac address to that of the first slave
1670                  */
1671                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1672                 addr.sa_family = slave_dev->type;
1673                 res = dev_set_mac_address(slave_dev, &addr);
1674                 if (res) {
1675                         pr_debug("Error %d calling set_mac_address\n", res);
1676                         goto err_restore_mtu;
1677                 }
1678         }
1679
1680         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1681         if (res) {
1682                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1683                 goto err_restore_mac;
1684         }
1685
1686         /* open the slave since the application closed it */
1687         res = dev_open(slave_dev);
1688         if (res) {
1689                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1690                 goto err_unset_master;
1691         }
1692
1693         new_slave->bond = bond;
1694         new_slave->dev = slave_dev;
1695         slave_dev->priv_flags |= IFF_BONDING;
1696
1697         if (bond_is_lb(bond)) {
1698                 /* bond_alb_init_slave() must be called before all other stages since
1699                  * it might fail and we do not want to have to undo everything
1700                  */
1701                 res = bond_alb_init_slave(bond, new_slave);
1702                 if (res)
1703                         goto err_close;
1704         }
1705
1706         /* If the mode USES_PRIMARY, then the new slave gets the
1707          * master's promisc (and mc) settings only if it becomes the
1708          * curr_active_slave, and that is taken care of later when calling
1709          * bond_change_active()
1710          */
1711         if (!USES_PRIMARY(bond->params.mode)) {
1712                 /* set promiscuity level to new slave */
1713                 if (bond_dev->flags & IFF_PROMISC) {
1714                         res = dev_set_promiscuity(slave_dev, 1);
1715                         if (res)
1716                                 goto err_close;
1717                 }
1718
1719                 /* set allmulti level to new slave */
1720                 if (bond_dev->flags & IFF_ALLMULTI) {
1721                         res = dev_set_allmulti(slave_dev, 1);
1722                         if (res)
1723                                 goto err_close;
1724                 }
1725
1726                 netif_addr_lock_bh(bond_dev);
1727                 /* upload master's mc_list to new slave */
1728                 netdev_for_each_mc_addr(ha, bond_dev)
1729                         dev_mc_add(slave_dev, ha->addr);
1730                 netif_addr_unlock_bh(bond_dev);
1731         }
1732
1733         if (bond->params.mode == BOND_MODE_8023AD) {
1734                 /* add lacpdu mc addr to mc list */
1735                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1736
1737                 dev_mc_add(slave_dev, lacpdu_multicast);
1738         }
1739
1740         bond_add_vlans_on_slave(bond, slave_dev);
1741
1742         write_lock_bh(&bond->lock);
1743
1744         bond_attach_slave(bond, new_slave);
1745
1746         new_slave->delay = 0;
1747         new_slave->link_failure_count = 0;
1748
1749         write_unlock_bh(&bond->lock);
1750
1751         bond_compute_features(bond);
1752
1753         bond_update_speed_duplex(new_slave);
1754
1755         read_lock(&bond->lock);
1756
1757         new_slave->last_arp_rx = jiffies -
1758                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1759
1760         if (bond->params.miimon && !bond->params.use_carrier) {
1761                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1762
1763                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1764                         /*
1765                          * miimon is set but a bonded network driver
1766                          * does not support ETHTOOL/MII and
1767                          * arp_interval is not set.  Note: if
1768                          * use_carrier is enabled, we will never go
1769                          * here (because netif_carrier is always
1770                          * supported); thus, we don't need to change
1771                          * the messages for netif_carrier.
1772                          */
1773                         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",
1774                                bond_dev->name, slave_dev->name);
1775                 } else if (link_reporting == -1) {
1776                         /* unable get link status using mii/ethtool */
1777                         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",
1778                                    bond_dev->name, slave_dev->name);
1779                 }
1780         }
1781
1782         /* check for initial state */
1783         if (bond->params.miimon) {
1784                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1785                         if (bond->params.updelay) {
1786                                 new_slave->link = BOND_LINK_BACK;
1787                                 new_slave->delay = bond->params.updelay;
1788                         } else {
1789                                 new_slave->link = BOND_LINK_UP;
1790                         }
1791                 } else {
1792                         new_slave->link = BOND_LINK_DOWN;
1793                 }
1794         } else if (bond->params.arp_interval) {
1795                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1796                         BOND_LINK_UP : BOND_LINK_DOWN);
1797         } else {
1798                 new_slave->link = BOND_LINK_UP;
1799         }
1800
1801         if (new_slave->link != BOND_LINK_DOWN)
1802                 new_slave->jiffies = jiffies;
1803         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1804                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1805                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1806
1807         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1808                 /* if there is a primary slave, remember it */
1809                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1810                         bond->primary_slave = new_slave;
1811                         bond->force_primary = true;
1812                 }
1813         }
1814
1815         write_lock_bh(&bond->curr_slave_lock);
1816
1817         switch (bond->params.mode) {
1818         case BOND_MODE_ACTIVEBACKUP:
1819                 bond_set_slave_inactive_flags(new_slave);
1820                 bond_select_active_slave(bond);
1821                 break;
1822         case BOND_MODE_8023AD:
1823                 /* in 802.3ad mode, the internal mechanism
1824                  * will activate the slaves in the selected
1825                  * aggregator
1826                  */
1827                 bond_set_slave_inactive_flags(new_slave);
1828                 /* if this is the first slave */
1829                 if (bond->slave_cnt == 1) {
1830                         SLAVE_AD_INFO(new_slave).id = 1;
1831                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1832                          * can be called only after the mac address of the bond is set
1833                          */
1834                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1835                 } else {
1836                         SLAVE_AD_INFO(new_slave).id =
1837                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1838                 }
1839
1840                 bond_3ad_bind_slave(new_slave);
1841                 break;
1842         case BOND_MODE_TLB:
1843         case BOND_MODE_ALB:
1844                 bond_set_active_slave(new_slave);
1845                 bond_set_slave_inactive_flags(new_slave);
1846                 bond_select_active_slave(bond);
1847                 break;
1848         default:
1849                 pr_debug("This slave is always active in trunk mode\n");
1850
1851                 /* always active in trunk mode */
1852                 bond_set_active_slave(new_slave);
1853
1854                 /* In trunking mode there is little meaning to curr_active_slave
1855                  * anyway (it holds no special properties of the bond device),
1856                  * so we can change it without calling change_active_interface()
1857                  */
1858                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1859                         bond->curr_active_slave = new_slave;
1860
1861                 break;
1862         } /* switch(bond_mode) */
1863
1864         write_unlock_bh(&bond->curr_slave_lock);
1865
1866         bond_set_carrier(bond);
1867
1868 #ifdef CONFIG_NET_POLL_CONTROLLER
1869         slave_dev->npinfo = bond_netpoll_info(bond);
1870         if (slave_dev->npinfo) {
1871                 if (slave_enable_netpoll(new_slave)) {
1872                         read_unlock(&bond->lock);
1873                         pr_info("Error, %s: master_dev is using netpoll, "
1874                                  "but new slave device does not support netpoll.\n",
1875                                  bond_dev->name);
1876                         res = -EBUSY;
1877                         goto err_detach;
1878                 }
1879         }
1880 #endif
1881
1882         read_unlock(&bond->lock);
1883
1884         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1885         if (res)
1886                 goto err_detach;
1887
1888         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1889                                          new_slave);
1890         if (res) {
1891                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1892                 goto err_dest_symlinks;
1893         }
1894
1895         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1896                 bond_dev->name, slave_dev->name,
1897                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1898                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1899
1900         /* enslave is successful */
1901         return 0;
1902
1903 /* Undo stages on error */
1904 err_dest_symlinks:
1905         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1906
1907 err_detach:
1908         if (!USES_PRIMARY(bond->params.mode)) {
1909                 netif_addr_lock_bh(bond_dev);
1910                 bond_mc_list_flush(bond_dev, slave_dev);
1911                 netif_addr_unlock_bh(bond_dev);
1912         }
1913         bond_del_vlans_from_slave(bond, slave_dev);
1914         write_lock_bh(&bond->lock);
1915         bond_detach_slave(bond, new_slave);
1916         if (bond->primary_slave == new_slave)
1917                 bond->primary_slave = NULL;
1918         if (bond->curr_active_slave == new_slave) {
1919                 bond_change_active_slave(bond, NULL);
1920                 write_unlock_bh(&bond->lock);
1921                 read_lock(&bond->lock);
1922                 write_lock_bh(&bond->curr_slave_lock);
1923                 bond_select_active_slave(bond);
1924                 write_unlock_bh(&bond->curr_slave_lock);
1925                 read_unlock(&bond->lock);
1926         } else {
1927                 write_unlock_bh(&bond->lock);
1928         }
1929         slave_disable_netpoll(new_slave);
1930
1931 err_close:
1932         slave_dev->priv_flags &= ~IFF_BONDING;
1933         dev_close(slave_dev);
1934
1935 err_unset_master:
1936         bond_upper_dev_unlink(bond_dev, slave_dev);
1937
1938 err_restore_mac:
1939         if (!bond->params.fail_over_mac) {
1940                 /* XXX TODO - fom follow mode needs to change master's
1941                  * MAC if this slave's MAC is in use by the bond, or at
1942                  * least print a warning.
1943                  */
1944                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1945                 addr.sa_family = slave_dev->type;
1946                 dev_set_mac_address(slave_dev, &addr);
1947         }
1948
1949 err_restore_mtu:
1950         dev_set_mtu(slave_dev, new_slave->original_mtu);
1951
1952 err_free:
1953         kfree(new_slave);
1954
1955 err_undo_flags:
1956         bond_compute_features(bond);
1957
1958         return res;
1959 }
1960
1961 /*
1962  * Try to release the slave device <slave> from the bond device <master>
1963  * It is legal to access curr_active_slave without a lock because all the function
1964  * is write-locked. If "all" is true it means that the function is being called
1965  * while destroying a bond interface and all slaves are being released.
1966  *
1967  * The rules for slave state should be:
1968  *   for Active/Backup:
1969  *     Active stays on all backups go down
1970  *   for Bonded connections:
1971  *     The first up interface should be left on and all others downed.
1972  */
1973 static int __bond_release_one(struct net_device *bond_dev,
1974                               struct net_device *slave_dev,
1975                               bool all)
1976 {
1977         struct bonding *bond = netdev_priv(bond_dev);
1978         struct slave *slave, *oldcurrent;
1979         struct sockaddr addr;
1980         netdev_features_t old_features = bond_dev->features;
1981
1982         /* slave is not a slave or master is not master of this slave */
1983         if (!(slave_dev->flags & IFF_SLAVE) ||
1984             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1985                 pr_err("%s: Error: cannot release %s.\n",
1986                        bond_dev->name, slave_dev->name);
1987                 return -EINVAL;
1988         }
1989
1990         block_netpoll_tx();
1991         write_lock_bh(&bond->lock);
1992
1993         slave = bond_get_slave_by_dev(bond, slave_dev);
1994         if (!slave) {
1995                 /* not a slave of this bond */
1996                 pr_info("%s: %s not enslaved\n",
1997                         bond_dev->name, slave_dev->name);
1998                 write_unlock_bh(&bond->lock);
1999                 unblock_netpoll_tx();
2000                 return -EINVAL;
2001         }
2002
2003         write_unlock_bh(&bond->lock);
2004         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2005          * for this slave anymore.
2006          */
2007         netdev_rx_handler_unregister(slave_dev);
2008         write_lock_bh(&bond->lock);
2009
2010         if (!all && !bond->params.fail_over_mac) {
2011                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
2012                     bond->slave_cnt > 1)
2013                         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",
2014                                    bond_dev->name, slave_dev->name,
2015                                    slave->perm_hwaddr,
2016                                    bond_dev->name, slave_dev->name);
2017         }
2018
2019         /* Inform AD package of unbinding of slave. */
2020         if (bond->params.mode == BOND_MODE_8023AD) {
2021                 /* must be called before the slave is
2022                  * detached from the list
2023                  */
2024                 bond_3ad_unbind_slave(slave);
2025         }
2026
2027         pr_info("%s: releasing %s interface %s\n",
2028                 bond_dev->name,
2029                 bond_is_active_slave(slave) ? "active" : "backup",
2030                 slave_dev->name);
2031
2032         oldcurrent = bond->curr_active_slave;
2033
2034         bond->current_arp_slave = NULL;
2035
2036         /* release the slave from its bond */
2037         bond_detach_slave(bond, slave);
2038
2039         if (bond->primary_slave == slave)
2040                 bond->primary_slave = NULL;
2041
2042         if (oldcurrent == slave)
2043                 bond_change_active_slave(bond, NULL);
2044
2045         if (bond_is_lb(bond)) {
2046                 /* Must be called only after the slave has been
2047                  * detached from the list and the curr_active_slave
2048                  * has been cleared (if our_slave == old_current),
2049                  * but before a new active slave is selected.
2050                  */
2051                 write_unlock_bh(&bond->lock);
2052                 bond_alb_deinit_slave(bond, slave);
2053                 write_lock_bh(&bond->lock);
2054         }
2055
2056         if (all) {
2057                 bond->curr_active_slave = NULL;
2058         } else if (oldcurrent == slave) {
2059                 /*
2060                  * Note that we hold RTNL over this sequence, so there
2061                  * is no concern that another slave add/remove event
2062                  * will interfere.
2063                  */
2064                 write_unlock_bh(&bond->lock);
2065                 read_lock(&bond->lock);
2066                 write_lock_bh(&bond->curr_slave_lock);
2067
2068                 bond_select_active_slave(bond);
2069
2070                 write_unlock_bh(&bond->curr_slave_lock);
2071                 read_unlock(&bond->lock);
2072                 write_lock_bh(&bond->lock);
2073         }
2074
2075         if (bond->slave_cnt == 0) {
2076                 bond_set_carrier(bond);
2077                 eth_hw_addr_random(bond_dev);
2078                 bond->dev_addr_from_first = true;
2079
2080                 if (bond_vlan_used(bond)) {
2081                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2082                                    bond_dev->name, bond_dev->name);
2083                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2084                                    bond_dev->name);
2085                 }
2086         }
2087
2088         write_unlock_bh(&bond->lock);
2089         unblock_netpoll_tx();
2090
2091         if (bond->slave_cnt == 0) {
2092                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2093                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2094         }
2095
2096         bond_compute_features(bond);
2097         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2098             (old_features & NETIF_F_VLAN_CHALLENGED))
2099                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2100                         bond_dev->name, slave_dev->name, bond_dev->name);
2101
2102         /* must do this from outside any spinlocks */
2103         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2104
2105         bond_del_vlans_from_slave(bond, slave_dev);
2106
2107         /* If the mode USES_PRIMARY, then we should only remove its
2108          * promisc and mc settings if it was the curr_active_slave, but that was
2109          * already taken care of above when we detached the slave
2110          */
2111         if (!USES_PRIMARY(bond->params.mode)) {
2112                 /* unset promiscuity level from slave */
2113                 if (bond_dev->flags & IFF_PROMISC)
2114                         dev_set_promiscuity(slave_dev, -1);
2115
2116                 /* unset allmulti level from slave */
2117                 if (bond_dev->flags & IFF_ALLMULTI)
2118                         dev_set_allmulti(slave_dev, -1);
2119
2120                 /* flush master's mc_list from slave */
2121                 netif_addr_lock_bh(bond_dev);
2122                 bond_mc_list_flush(bond_dev, slave_dev);
2123                 netif_addr_unlock_bh(bond_dev);
2124         }
2125
2126         bond_upper_dev_unlink(bond_dev, slave_dev);
2127
2128         slave_disable_netpoll(slave);
2129
2130         /* close slave before restoring its mac address */
2131         dev_close(slave_dev);
2132
2133         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2134                 /* restore original ("permanent") mac address */
2135                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2136                 addr.sa_family = slave_dev->type;
2137                 dev_set_mac_address(slave_dev, &addr);
2138         }
2139
2140         dev_set_mtu(slave_dev, slave->original_mtu);
2141
2142         slave_dev->priv_flags &= ~IFF_BONDING;
2143
2144         kfree(slave);
2145
2146         return 0;  /* deletion OK */
2147 }
2148
2149 /* A wrapper used because of ndo_del_link */
2150 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2151 {
2152         return __bond_release_one(bond_dev, slave_dev, false);
2153 }
2154
2155 /*
2156 * First release a slave and then destroy the bond if no more slaves are left.
2157 * Must be under rtnl_lock when this function is called.
2158 */
2159 static int  bond_release_and_destroy(struct net_device *bond_dev,
2160                                      struct net_device *slave_dev)
2161 {
2162         struct bonding *bond = netdev_priv(bond_dev);
2163         int ret;
2164
2165         ret = bond_release(bond_dev, slave_dev);
2166         if ((ret == 0) && (bond->slave_cnt == 0)) {
2167                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2168                 pr_info("%s: destroying bond %s.\n",
2169                         bond_dev->name, bond_dev->name);
2170                 unregister_netdevice(bond_dev);
2171         }
2172         return ret;
2173 }
2174
2175 /*
2176  * This function changes the active slave to slave <slave_dev>.
2177  * It returns -EINVAL in the following cases.
2178  *  - <slave_dev> is not found in the list.
2179  *  - There is not active slave now.
2180  *  - <slave_dev> is already active.
2181  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2182  *  - <slave_dev> is not running.
2183  * In these cases, this function does nothing.
2184  * In the other cases, current_slave pointer is changed and 0 is returned.
2185  */
2186 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2187 {
2188         struct bonding *bond = netdev_priv(bond_dev);
2189         struct slave *old_active = NULL;
2190         struct slave *new_active = NULL;
2191         int res = 0;
2192
2193         if (!USES_PRIMARY(bond->params.mode))
2194                 return -EINVAL;
2195
2196         /* Verify that bond_dev is indeed the master of slave_dev */
2197         if (!(slave_dev->flags & IFF_SLAVE) ||
2198             !netdev_has_upper_dev(slave_dev, bond_dev))
2199                 return -EINVAL;
2200
2201         read_lock(&bond->lock);
2202
2203         read_lock(&bond->curr_slave_lock);
2204         old_active = bond->curr_active_slave;
2205         read_unlock(&bond->curr_slave_lock);
2206
2207         new_active = bond_get_slave_by_dev(bond, slave_dev);
2208
2209         /*
2210          * Changing to the current active: do nothing; return success.
2211          */
2212         if (new_active && (new_active == old_active)) {
2213                 read_unlock(&bond->lock);
2214                 return 0;
2215         }
2216
2217         if ((new_active) &&
2218             (old_active) &&
2219             (new_active->link == BOND_LINK_UP) &&
2220             IS_UP(new_active->dev)) {
2221                 block_netpoll_tx();
2222                 write_lock_bh(&bond->curr_slave_lock);
2223                 bond_change_active_slave(bond, new_active);
2224                 write_unlock_bh(&bond->curr_slave_lock);
2225                 unblock_netpoll_tx();
2226         } else
2227                 res = -EINVAL;
2228
2229         read_unlock(&bond->lock);
2230
2231         return res;
2232 }
2233
2234 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2235 {
2236         struct bonding *bond = netdev_priv(bond_dev);
2237
2238         info->bond_mode = bond->params.mode;
2239         info->miimon = bond->params.miimon;
2240
2241         read_lock(&bond->lock);
2242         info->num_slaves = bond->slave_cnt;
2243         read_unlock(&bond->lock);
2244
2245         return 0;
2246 }
2247
2248 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2249 {
2250         struct bonding *bond = netdev_priv(bond_dev);
2251         struct slave *slave;
2252         int i, res = -ENODEV;
2253
2254         read_lock(&bond->lock);
2255
2256         bond_for_each_slave(bond, slave, i) {
2257                 if (i == (int)info->slave_id) {
2258                         res = 0;
2259                         strcpy(info->slave_name, slave->dev->name);
2260                         info->link = slave->link;
2261                         info->state = bond_slave_state(slave);
2262                         info->link_failure_count = slave->link_failure_count;
2263                         break;
2264                 }
2265         }
2266
2267         read_unlock(&bond->lock);
2268
2269         return res;
2270 }
2271
2272 /*-------------------------------- Monitoring -------------------------------*/
2273
2274
2275 static int bond_miimon_inspect(struct bonding *bond)
2276 {
2277         struct slave *slave;
2278         int i, link_state, commit = 0;
2279         bool ignore_updelay;
2280
2281         ignore_updelay = !bond->curr_active_slave ? true : false;
2282
2283         bond_for_each_slave(bond, slave, i) {
2284                 slave->new_link = BOND_LINK_NOCHANGE;
2285
2286                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2287
2288                 switch (slave->link) {
2289                 case BOND_LINK_UP:
2290                         if (link_state)
2291                                 continue;
2292
2293                         slave->link = BOND_LINK_FAIL;
2294                         slave->delay = bond->params.downdelay;
2295                         if (slave->delay) {
2296                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2297                                         bond->dev->name,
2298                                         (bond->params.mode ==
2299                                          BOND_MODE_ACTIVEBACKUP) ?
2300                                         (bond_is_active_slave(slave) ?
2301                                          "active " : "backup ") : "",
2302                                         slave->dev->name,
2303                                         bond->params.downdelay * bond->params.miimon);
2304                         }
2305                         /*FALLTHRU*/
2306                 case BOND_LINK_FAIL:
2307                         if (link_state) {
2308                                 /*
2309                                  * recovered before downdelay expired
2310                                  */
2311                                 slave->link = BOND_LINK_UP;
2312                                 slave->jiffies = jiffies;
2313                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2314                                         bond->dev->name,
2315                                         (bond->params.downdelay - slave->delay) *
2316                                         bond->params.miimon,
2317                                         slave->dev->name);
2318                                 continue;
2319                         }
2320
2321                         if (slave->delay <= 0) {
2322                                 slave->new_link = BOND_LINK_DOWN;
2323                                 commit++;
2324                                 continue;
2325                         }
2326
2327                         slave->delay--;
2328                         break;
2329
2330                 case BOND_LINK_DOWN:
2331                         if (!link_state)
2332                                 continue;
2333
2334                         slave->link = BOND_LINK_BACK;
2335                         slave->delay = bond->params.updelay;
2336
2337                         if (slave->delay) {
2338                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2339                                         bond->dev->name, slave->dev->name,
2340                                         ignore_updelay ? 0 :
2341                                         bond->params.updelay *
2342                                         bond->params.miimon);
2343                         }
2344                         /*FALLTHRU*/
2345                 case BOND_LINK_BACK:
2346                         if (!link_state) {
2347                                 slave->link = BOND_LINK_DOWN;
2348                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2349                                         bond->dev->name,
2350                                         (bond->params.updelay - slave->delay) *
2351                                         bond->params.miimon,
2352                                         slave->dev->name);
2353
2354                                 continue;
2355                         }
2356
2357                         if (ignore_updelay)
2358                                 slave->delay = 0;
2359
2360                         if (slave->delay <= 0) {
2361                                 slave->new_link = BOND_LINK_UP;
2362                                 commit++;
2363                                 ignore_updelay = false;
2364                                 continue;
2365                         }
2366
2367                         slave->delay--;
2368                         break;
2369                 }
2370         }
2371
2372         return commit;
2373 }
2374
2375 static void bond_miimon_commit(struct bonding *bond)
2376 {
2377         struct slave *slave;
2378         int i;
2379
2380         bond_for_each_slave(bond, slave, i) {
2381                 switch (slave->new_link) {
2382                 case BOND_LINK_NOCHANGE:
2383                         continue;
2384
2385                 case BOND_LINK_UP:
2386                         slave->link = BOND_LINK_UP;
2387                         slave->jiffies = jiffies;
2388
2389                         if (bond->params.mode == BOND_MODE_8023AD) {
2390                                 /* prevent it from being the active one */
2391                                 bond_set_backup_slave(slave);
2392                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2393                                 /* make it immediately active */
2394                                 bond_set_active_slave(slave);
2395                         } else if (slave != bond->primary_slave) {
2396                                 /* prevent it from being the active one */
2397                                 bond_set_backup_slave(slave);
2398                         }
2399
2400                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2401                                 bond->dev->name, slave->dev->name,
2402                                 slave->speed, slave->duplex ? "full" : "half");
2403
2404                         /* notify ad that the link status has changed */
2405                         if (bond->params.mode == BOND_MODE_8023AD)
2406                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2407
2408                         if (bond_is_lb(bond))
2409                                 bond_alb_handle_link_change(bond, slave,
2410                                                             BOND_LINK_UP);
2411
2412                         if (!bond->curr_active_slave ||
2413                             (slave == bond->primary_slave))
2414                                 goto do_failover;
2415
2416                         continue;
2417
2418                 case BOND_LINK_DOWN:
2419                         if (slave->link_failure_count < UINT_MAX)
2420                                 slave->link_failure_count++;
2421
2422                         slave->link = BOND_LINK_DOWN;
2423
2424                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2425                             bond->params.mode == BOND_MODE_8023AD)
2426                                 bond_set_slave_inactive_flags(slave);
2427
2428                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2429                                 bond->dev->name, slave->dev->name);
2430
2431                         if (bond->params.mode == BOND_MODE_8023AD)
2432                                 bond_3ad_handle_link_change(slave,
2433                                                             BOND_LINK_DOWN);
2434
2435                         if (bond_is_lb(bond))
2436                                 bond_alb_handle_link_change(bond, slave,
2437                                                             BOND_LINK_DOWN);
2438
2439                         if (slave == bond->curr_active_slave)
2440                                 goto do_failover;
2441
2442                         continue;
2443
2444                 default:
2445                         pr_err("%s: invalid new link %d on slave %s\n",
2446                                bond->dev->name, slave->new_link,
2447                                slave->dev->name);
2448                         slave->new_link = BOND_LINK_NOCHANGE;
2449
2450                         continue;
2451                 }
2452
2453 do_failover:
2454                 ASSERT_RTNL();
2455                 block_netpoll_tx();
2456                 write_lock_bh(&bond->curr_slave_lock);
2457                 bond_select_active_slave(bond);
2458                 write_unlock_bh(&bond->curr_slave_lock);
2459                 unblock_netpoll_tx();
2460         }
2461
2462         bond_set_carrier(bond);
2463 }
2464
2465 /*
2466  * bond_mii_monitor
2467  *
2468  * Really a wrapper that splits the mii monitor into two phases: an
2469  * inspection, then (if inspection indicates something needs to be done)
2470  * an acquisition of appropriate locks followed by a commit phase to
2471  * implement whatever link state changes are indicated.
2472  */
2473 void bond_mii_monitor(struct work_struct *work)
2474 {
2475         struct bonding *bond = container_of(work, struct bonding,
2476                                             mii_work.work);
2477         bool should_notify_peers = false;
2478         unsigned long delay;
2479
2480         read_lock(&bond->lock);
2481
2482         delay = msecs_to_jiffies(bond->params.miimon);
2483
2484         if (bond->slave_cnt == 0)
2485                 goto re_arm;
2486
2487         should_notify_peers = bond_should_notify_peers(bond);
2488
2489         if (bond_miimon_inspect(bond)) {
2490                 read_unlock(&bond->lock);
2491
2492                 /* Race avoidance with bond_close cancel of workqueue */
2493                 if (!rtnl_trylock()) {
2494                         read_lock(&bond->lock);
2495                         delay = 1;
2496                         should_notify_peers = false;
2497                         goto re_arm;
2498                 }
2499
2500                 read_lock(&bond->lock);
2501
2502                 bond_miimon_commit(bond);
2503
2504                 read_unlock(&bond->lock);
2505                 rtnl_unlock();  /* might sleep, hold no other locks */
2506                 read_lock(&bond->lock);
2507         }
2508
2509 re_arm:
2510         if (bond->params.miimon)
2511                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2512
2513         read_unlock(&bond->lock);
2514
2515         if (should_notify_peers) {
2516                 if (!rtnl_trylock()) {
2517                         read_lock(&bond->lock);
2518                         bond->send_peer_notif++;
2519                         read_unlock(&bond->lock);
2520                         return;
2521                 }
2522                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2523                 rtnl_unlock();
2524         }
2525 }
2526
2527 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2528 {
2529         struct vlan_entry *vlan;
2530         struct net_device *vlan_dev;
2531
2532         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2533                 return 1;
2534
2535         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2536                 rcu_read_lock();
2537                 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2538                 rcu_read_unlock();
2539                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2540                         return 1;
2541         }
2542
2543         return 0;
2544 }
2545
2546 /*
2547  * We go to the (large) trouble of VLAN tagging ARP frames because
2548  * switches in VLAN mode (especially if ports are configured as
2549  * "native" to a VLAN) might not pass non-tagged frames.
2550  */
2551 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2552 {
2553         struct sk_buff *skb;
2554
2555         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2556                  slave_dev->name, dest_ip, src_ip, vlan_id);
2557
2558         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2559                          NULL, slave_dev->dev_addr, NULL);
2560
2561         if (!skb) {
2562                 pr_err("ARP packet allocation failed\n");
2563                 return;
2564         }
2565         if (vlan_id) {
2566                 skb = vlan_put_tag(skb, vlan_id);
2567                 if (!skb) {
2568                         pr_err("failed to insert VLAN tag\n");
2569                         return;
2570                 }
2571         }
2572         arp_xmit(skb);
2573 }
2574
2575
2576 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2577 {
2578         int i, vlan_id;
2579         __be32 *targets = bond->params.arp_targets;
2580         struct vlan_entry *vlan;
2581         struct net_device *vlan_dev = NULL;
2582         struct rtable *rt;
2583
2584         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2585                 __be32 addr;
2586                 if (!targets[i])
2587                         break;
2588                 pr_debug("basa: target %x\n", targets[i]);
2589                 if (!bond_vlan_used(bond)) {
2590                         pr_debug("basa: empty vlan: arp_send\n");
2591                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2592                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2593                                       addr, 0);
2594                         continue;
2595                 }
2596
2597                 /*
2598                  * If VLANs are configured, we do a route lookup to
2599                  * determine which VLAN interface would be used, so we
2600                  * can tag the ARP with the proper VLAN tag.
2601                  */
2602                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2603                                      RTO_ONLINK, 0);
2604                 if (IS_ERR(rt)) {
2605                         if (net_ratelimit()) {
2606                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2607                                            bond->dev->name, &targets[i]);
2608                         }
2609                         continue;
2610                 }
2611
2612                 /*
2613                  * This target is not on a VLAN
2614                  */
2615                 if (rt->dst.dev == bond->dev) {
2616                         ip_rt_put(rt);
2617                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2618                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2619                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2620                                       addr, 0);
2621                         continue;
2622                 }
2623
2624                 vlan_id = 0;
2625                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2626                         rcu_read_lock();
2627                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2628                                                         vlan->vlan_id);
2629                         rcu_read_unlock();
2630                         if (vlan_dev == rt->dst.dev) {
2631                                 vlan_id = vlan->vlan_id;
2632                                 pr_debug("basa: vlan match on %s %d\n",
2633                                        vlan_dev->name, vlan_id);
2634                                 break;
2635                         }
2636                 }
2637
2638                 if (vlan_id && vlan_dev) {
2639                         ip_rt_put(rt);
2640                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2641                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2642                                       addr, vlan_id);
2643                         continue;
2644                 }
2645
2646                 if (net_ratelimit()) {
2647                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2648                                    bond->dev->name, &targets[i],
2649                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2650                 }
2651                 ip_rt_put(rt);
2652         }
2653 }
2654
2655 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2656 {
2657         int i;
2658         __be32 *targets = bond->params.arp_targets;
2659
2660         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2661                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2662                          &sip, &tip, i, &targets[i],
2663                          bond_has_this_ip(bond, tip));
2664                 if (sip == targets[i]) {
2665                         if (bond_has_this_ip(bond, tip))
2666                                 slave->last_arp_rx = jiffies;
2667                         return;
2668                 }
2669         }
2670 }
2671
2672 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2673                         struct slave *slave)
2674 {
2675         struct arphdr *arp = (struct arphdr *)skb->data;
2676         unsigned char *arp_ptr;
2677         __be32 sip, tip;
2678         int alen;
2679
2680         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2681                 return RX_HANDLER_ANOTHER;
2682
2683         read_lock(&bond->lock);
2684         alen = arp_hdr_len(bond->dev);
2685
2686         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2687                  bond->dev->name, skb->dev->name);
2688
2689         if (alen > skb_headlen(skb)) {
2690                 arp = kmalloc(alen, GFP_ATOMIC);
2691                 if (!arp)
2692                         goto out_unlock;
2693                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2694                         goto out_unlock;
2695         }
2696
2697         if (arp->ar_hln != bond->dev->addr_len ||
2698             skb->pkt_type == PACKET_OTHERHOST ||
2699             skb->pkt_type == PACKET_LOOPBACK ||
2700             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2701             arp->ar_pro != htons(ETH_P_IP) ||
2702             arp->ar_pln != 4)
2703                 goto out_unlock;
2704
2705         arp_ptr = (unsigned char *)(arp + 1);
2706         arp_ptr += bond->dev->addr_len;
2707         memcpy(&sip, arp_ptr, 4);
2708         arp_ptr += 4 + bond->dev->addr_len;
2709         memcpy(&tip, arp_ptr, 4);
2710
2711         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2712                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2713                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2714                  &sip, &tip);
2715
2716         /*
2717          * Backup slaves won't see the ARP reply, but do come through
2718          * here for each ARP probe (so we swap the sip/tip to validate
2719          * the probe).  In a "redundant switch, common router" type of
2720          * configuration, the ARP probe will (hopefully) travel from
2721          * the active, through one switch, the router, then the other
2722          * switch before reaching the backup.
2723          */
2724         if (bond_is_active_slave(slave))
2725                 bond_validate_arp(bond, slave, sip, tip);
2726         else
2727                 bond_validate_arp(bond, slave, tip, sip);
2728
2729 out_unlock:
2730         read_unlock(&bond->lock);
2731         if (arp != (struct arphdr *)skb->data)
2732                 kfree(arp);
2733         return RX_HANDLER_ANOTHER;
2734 }
2735
2736 /*
2737  * this function is called regularly to monitor each slave's link
2738  * ensuring that traffic is being sent and received when arp monitoring
2739  * is used in load-balancing mode. if the adapter has been dormant, then an
2740  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2741  * arp monitoring in active backup mode.
2742  */
2743 void bond_loadbalance_arp_mon(struct work_struct *work)
2744 {
2745         struct bonding *bond = container_of(work, struct bonding,
2746                                             arp_work.work);
2747         struct slave *slave, *oldcurrent;
2748         int do_failover = 0;
2749         int delta_in_ticks, extra_ticks;
2750         int i;
2751
2752         read_lock(&bond->lock);
2753
2754         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2755         extra_ticks = delta_in_ticks / 2;
2756
2757         if (bond->slave_cnt == 0)
2758                 goto re_arm;
2759
2760         read_lock(&bond->curr_slave_lock);
2761         oldcurrent = bond->curr_active_slave;
2762         read_unlock(&bond->curr_slave_lock);
2763
2764         /* see if any of the previous devices are up now (i.e. they have
2765          * xmt and rcv traffic). the curr_active_slave does not come into
2766          * the picture unless it is null. also, slave->jiffies is not needed
2767          * here because we send an arp on each slave and give a slave as
2768          * long as it needs to get the tx/rx within the delta.
2769          * TODO: what about up/down delay in arp mode? it wasn't here before
2770          *       so it can wait
2771          */
2772         bond_for_each_slave(bond, slave, i) {
2773                 unsigned long trans_start = dev_trans_start(slave->dev);
2774
2775                 if (slave->link != BOND_LINK_UP) {
2776                         if (time_in_range(jiffies,
2777                                 trans_start - delta_in_ticks,
2778                                 trans_start + delta_in_ticks + extra_ticks) &&
2779                             time_in_range(jiffies,
2780                                 slave->dev->last_rx - delta_in_ticks,
2781                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2782
2783                                 slave->link  = BOND_LINK_UP;
2784                                 bond_set_active_slave(slave);
2785
2786                                 /* primary_slave has no meaning in round-robin
2787                                  * mode. the window of a slave being up and
2788                                  * curr_active_slave being null after enslaving
2789                                  * is closed.
2790                                  */
2791                                 if (!oldcurrent) {
2792                                         pr_info("%s: link status definitely up for interface %s, ",
2793                                                 bond->dev->name,
2794                                                 slave->dev->name);
2795                                         do_failover = 1;
2796                                 } else {
2797                                         pr_info("%s: interface %s is now up\n",
2798                                                 bond->dev->name,
2799                                                 slave->dev->name);
2800                                 }
2801                         }
2802                 } else {
2803                         /* slave->link == BOND_LINK_UP */
2804
2805                         /* not all switches will respond to an arp request
2806                          * when the source ip is 0, so don't take the link down
2807                          * if we don't know our ip yet
2808                          */
2809                         if (!time_in_range(jiffies,
2810                                 trans_start - delta_in_ticks,
2811                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2812                             !time_in_range(jiffies,
2813                                 slave->dev->last_rx - delta_in_ticks,
2814                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2815
2816                                 slave->link  = BOND_LINK_DOWN;
2817                                 bond_set_backup_slave(slave);
2818
2819                                 if (slave->link_failure_count < UINT_MAX)
2820                                         slave->link_failure_count++;
2821
2822                                 pr_info("%s: interface %s is now down.\n",
2823                                         bond->dev->name,
2824                                         slave->dev->name);
2825
2826                                 if (slave == oldcurrent)
2827                                         do_failover = 1;
2828                         }
2829                 }
2830
2831                 /* note: if switch is in round-robin mode, all links
2832                  * must tx arp to ensure all links rx an arp - otherwise
2833                  * links may oscillate or not come up at all; if switch is
2834                  * in something like xor mode, there is nothing we can
2835                  * do - all replies will be rx'ed on same link causing slaves
2836                  * to be unstable during low/no traffic periods
2837                  */
2838                 if (IS_UP(slave->dev))
2839                         bond_arp_send_all(bond, slave);
2840         }
2841
2842         if (do_failover) {
2843                 block_netpoll_tx();
2844                 write_lock_bh(&bond->curr_slave_lock);
2845
2846                 bond_select_active_slave(bond);
2847
2848                 write_unlock_bh(&bond->curr_slave_lock);
2849                 unblock_netpoll_tx();
2850         }
2851
2852 re_arm:
2853         if (bond->params.arp_interval)
2854                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2855
2856         read_unlock(&bond->lock);
2857 }
2858
2859 /*
2860  * Called to inspect slaves for active-backup mode ARP monitor link state
2861  * changes.  Sets new_link in slaves to specify what action should take
2862  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2863  * to link states must be committed.
2864  *
2865  * Called with bond->lock held for read.
2866  */
2867 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2868 {
2869         struct slave *slave;
2870         int i, commit = 0;
2871         unsigned long trans_start;
2872         int extra_ticks;
2873
2874         /* All the time comparisons below need some extra time. Otherwise, on
2875          * fast networks the ARP probe/reply may arrive within the same jiffy
2876          * as it was sent.  Then, the next time the ARP monitor is run, one
2877          * arp_interval will already have passed in the comparisons.
2878          */
2879         extra_ticks = delta_in_ticks / 2;
2880
2881         bond_for_each_slave(bond, slave, i) {
2882                 slave->new_link = BOND_LINK_NOCHANGE;
2883
2884                 if (slave->link != BOND_LINK_UP) {
2885                         if (time_in_range(jiffies,
2886                                 slave_last_rx(bond, slave) - delta_in_ticks,
2887                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2888
2889                                 slave->new_link = BOND_LINK_UP;
2890                                 commit++;
2891                         }
2892
2893                         continue;
2894                 }
2895
2896                 /*
2897                  * Give slaves 2*delta after being enslaved or made
2898                  * active.  This avoids bouncing, as the last receive
2899                  * times need a full ARP monitor cycle to be updated.
2900                  */
2901                 if (time_in_range(jiffies,
2902                                   slave->jiffies - delta_in_ticks,
2903                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2904                         continue;
2905
2906                 /*
2907                  * Backup slave is down if:
2908                  * - No current_arp_slave AND
2909                  * - more than 3*delta since last receive AND
2910                  * - the bond has an IP address
2911                  *
2912                  * Note: a non-null current_arp_slave indicates
2913                  * the curr_active_slave went down and we are
2914                  * searching for a new one; under this condition
2915                  * we only take the curr_active_slave down - this
2916                  * gives each slave a chance to tx/rx traffic
2917                  * before being taken out
2918                  */
2919                 if (!bond_is_active_slave(slave) &&
2920                     !bond->current_arp_slave &&
2921                     !time_in_range(jiffies,
2922                         slave_last_rx(bond, slave) - delta_in_ticks,
2923                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2924
2925                         slave->new_link = BOND_LINK_DOWN;
2926                         commit++;
2927                 }
2928
2929                 /*
2930                  * Active slave is down if:
2931                  * - more than 2*delta since transmitting OR
2932                  * - (more than 2*delta since receive AND
2933                  *    the bond has an IP address)
2934                  */
2935                 trans_start = dev_trans_start(slave->dev);
2936                 if (bond_is_active_slave(slave) &&
2937                     (!time_in_range(jiffies,
2938                         trans_start - delta_in_ticks,
2939                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
2940                      !time_in_range(jiffies,
2941                         slave_last_rx(bond, slave) - delta_in_ticks,
2942                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2943
2944                         slave->new_link = BOND_LINK_DOWN;
2945                         commit++;
2946                 }
2947         }
2948
2949         return commit;
2950 }
2951
2952 /*
2953  * Called to commit link state changes noted by inspection step of
2954  * active-backup mode ARP monitor.
2955  *
2956  * Called with RTNL and bond->lock for read.
2957  */
2958 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2959 {
2960         struct slave *slave;
2961         int i;
2962         unsigned long trans_start;
2963
2964         bond_for_each_slave(bond, slave, i) {
2965                 switch (slave->new_link) {
2966                 case BOND_LINK_NOCHANGE:
2967                         continue;
2968
2969                 case BOND_LINK_UP:
2970                         trans_start = dev_trans_start(slave->dev);
2971                         if ((!bond->curr_active_slave &&
2972                              time_in_range(jiffies,
2973                                            trans_start - delta_in_ticks,
2974                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2975                             bond->curr_active_slave != slave) {
2976                                 slave->link = BOND_LINK_UP;
2977                                 if (bond->current_arp_slave) {
2978                                         bond_set_slave_inactive_flags(
2979                                                 bond->current_arp_slave);
2980                                         bond->current_arp_slave = NULL;
2981                                 }
2982
2983                                 pr_info("%s: link status definitely up for interface %s.\n",
2984                                         bond->dev->name, slave->dev->name);
2985
2986                                 if (!bond->curr_active_slave ||
2987                                     (slave == bond->primary_slave))
2988                                         goto do_failover;
2989
2990                         }
2991
2992                         continue;
2993
2994                 case BOND_LINK_DOWN:
2995                         if (slave->link_failure_count < UINT_MAX)
2996                                 slave->link_failure_count++;
2997
2998                         slave->link = BOND_LINK_DOWN;
2999                         bond_set_slave_inactive_flags(slave);
3000
3001                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3002                                 bond->dev->name, slave->dev->name);
3003
3004                         if (slave == bond->curr_active_slave) {
3005                                 bond->current_arp_slave = NULL;
3006                                 goto do_failover;
3007                         }
3008
3009                         continue;
3010
3011                 default:
3012                         pr_err("%s: impossible: new_link %d on slave %s\n",
3013                                bond->dev->name, slave->new_link,
3014                                slave->dev->name);
3015                         continue;
3016                 }
3017
3018 do_failover:
3019                 ASSERT_RTNL();
3020                 block_netpoll_tx();
3021                 write_lock_bh(&bond->curr_slave_lock);
3022                 bond_select_active_slave(bond);
3023                 write_unlock_bh(&bond->curr_slave_lock);
3024                 unblock_netpoll_tx();
3025         }
3026
3027         bond_set_carrier(bond);
3028 }
3029
3030 /*
3031  * Send ARP probes for active-backup mode ARP monitor.
3032  *
3033  * Called with bond->lock held for read.
3034  */
3035 static void bond_ab_arp_probe(struct bonding *bond)
3036 {
3037         struct slave *slave;
3038         int i;
3039
3040         read_lock(&bond->curr_slave_lock);
3041
3042         if (bond->current_arp_slave && bond->curr_active_slave)
3043                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3044                         bond->current_arp_slave->dev->name,
3045                         bond->curr_active_slave->dev->name);
3046
3047         if (bond->curr_active_slave) {
3048                 bond_arp_send_all(bond, bond->curr_active_slave);
3049                 read_unlock(&bond->curr_slave_lock);
3050                 return;
3051         }
3052
3053         read_unlock(&bond->curr_slave_lock);
3054
3055         /* if we don't have a curr_active_slave, search for the next available
3056          * backup slave from the current_arp_slave and make it the candidate
3057          * for becoming the curr_active_slave
3058          */
3059
3060         if (!bond->current_arp_slave) {
3061                 bond->current_arp_slave = bond->first_slave;
3062                 if (!bond->current_arp_slave)
3063                         return;
3064         }
3065
3066         bond_set_slave_inactive_flags(bond->current_arp_slave);
3067
3068         /* search for next candidate */
3069         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3070                 if (IS_UP(slave->dev)) {
3071                         slave->link = BOND_LINK_BACK;
3072                         bond_set_slave_active_flags(slave);
3073                         bond_arp_send_all(bond, slave);
3074                         slave->jiffies = jiffies;
3075                         bond->current_arp_slave = slave;
3076                         break;
3077                 }
3078
3079                 /* if the link state is up at this point, we
3080                  * mark it down - this can happen if we have
3081                  * simultaneous link failures and
3082                  * reselect_active_interface doesn't make this
3083                  * one the current slave so it is still marked
3084                  * up when it is actually down
3085                  */
3086                 if (slave->link == BOND_LINK_UP) {
3087                         slave->link = BOND_LINK_DOWN;
3088                         if (slave->link_failure_count < UINT_MAX)
3089                                 slave->link_failure_count++;
3090
3091                         bond_set_slave_inactive_flags(slave);
3092
3093                         pr_info("%s: backup interface %s is now down.\n",
3094                                 bond->dev->name, slave->dev->name);
3095                 }
3096         }
3097 }
3098
3099 void bond_activebackup_arp_mon(struct work_struct *work)
3100 {
3101         struct bonding *bond = container_of(work, struct bonding,
3102                                             arp_work.work);
3103         bool should_notify_peers = false;
3104         int delta_in_ticks;
3105
3106         read_lock(&bond->lock);
3107
3108         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3109
3110         if (bond->slave_cnt == 0)
3111                 goto re_arm;
3112
3113         should_notify_peers = bond_should_notify_peers(bond);
3114
3115         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3116                 read_unlock(&bond->lock);
3117
3118                 /* Race avoidance with bond_close flush of workqueue */
3119                 if (!rtnl_trylock()) {
3120                         read_lock(&bond->lock);
3121                         delta_in_ticks = 1;
3122                         should_notify_peers = false;
3123                         goto re_arm;
3124                 }
3125
3126                 read_lock(&bond->lock);
3127
3128                 bond_ab_arp_commit(bond, delta_in_ticks);
3129
3130                 read_unlock(&bond->lock);
3131                 rtnl_unlock();
3132                 read_lock(&bond->lock);
3133         }
3134
3135         bond_ab_arp_probe(bond);
3136
3137 re_arm:
3138         if (bond->params.arp_interval)
3139                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3140
3141         read_unlock(&bond->lock);
3142
3143         if (should_notify_peers) {
3144                 if (!rtnl_trylock()) {
3145                         read_lock(&bond->lock);
3146                         bond->send_peer_notif++;
3147                         read_unlock(&bond->lock);
3148                         return;
3149                 }
3150                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3151                 rtnl_unlock();
3152         }
3153 }
3154
3155 /*-------------------------- netdev event handling --------------------------*/
3156
3157 /*
3158  * Change device name
3159  */
3160 static int bond_event_changename(struct bonding *bond)
3161 {
3162         bond_remove_proc_entry(bond);
3163         bond_create_proc_entry(bond);
3164
3165         bond_debug_reregister(bond);
3166
3167         return NOTIFY_DONE;
3168 }
3169
3170 static int bond_master_netdev_event(unsigned long event,
3171                                     struct net_device *bond_dev)
3172 {
3173         struct bonding *event_bond = netdev_priv(bond_dev);
3174
3175         switch (event) {
3176         case NETDEV_CHANGENAME:
3177                 return bond_event_changename(event_bond);
3178         case NETDEV_UNREGISTER:
3179                 bond_remove_proc_entry(event_bond);
3180                 break;
3181         case NETDEV_REGISTER:
3182                 bond_create_proc_entry(event_bond);
3183                 break;
3184         default:
3185                 break;
3186         }
3187
3188         return NOTIFY_DONE;
3189 }
3190
3191 static int bond_slave_netdev_event(unsigned long event,
3192                                    struct net_device *slave_dev)
3193 {
3194         struct slave *slave = bond_slave_get_rtnl(slave_dev);
3195         struct bonding *bond;
3196         struct net_device *bond_dev;
3197         u32 old_speed;
3198         u8 old_duplex;
3199
3200         /* A netdev event can be generated while enslaving a device
3201          * before netdev_rx_handler_register is called in which case
3202          * slave will be NULL
3203          */
3204         if (!slave)
3205                 return NOTIFY_DONE;
3206         bond_dev = slave->bond->dev;
3207         bond = slave->bond;
3208
3209         switch (event) {
3210         case NETDEV_UNREGISTER:
3211                 if (bond->setup_by_slave)
3212                         bond_release_and_destroy(bond_dev, slave_dev);
3213                 else
3214                         bond_release(bond_dev, slave_dev);
3215                 break;
3216         case NETDEV_UP:
3217         case NETDEV_CHANGE:
3218                 old_speed = slave->speed;
3219                 old_duplex = slave->duplex;
3220
3221                 bond_update_speed_duplex(slave);
3222
3223                 if (bond->params.mode == BOND_MODE_8023AD) {
3224                         if (old_speed != slave->speed)
3225                                 bond_3ad_adapter_speed_changed(slave);
3226                         if (old_duplex != slave->duplex)
3227                                 bond_3ad_adapter_duplex_changed(slave);
3228                 }
3229                 break;
3230         case NETDEV_DOWN:
3231                 /*
3232                  * ... Or is it this?
3233                  */
3234                 break;
3235         case NETDEV_CHANGEMTU:
3236                 /*
3237                  * TODO: Should slaves be allowed to
3238                  * independently alter their MTU?  For
3239                  * an active-backup bond, slaves need
3240                  * not be the same type of device, so
3241                  * MTUs may vary.  For other modes,
3242                  * slaves arguably should have the
3243                  * same MTUs. To do this, we'd need to
3244                  * take over the slave's change_mtu
3245                  * function for the duration of their
3246                  * servitude.
3247                  */
3248                 break;
3249         case NETDEV_CHANGENAME:
3250                 /*
3251                  * TODO: handle changing the primary's name
3252                  */
3253                 break;
3254         case NETDEV_FEAT_CHANGE:
3255                 bond_compute_features(bond);
3256                 break;
3257         default:
3258                 break;
3259         }
3260
3261         return NOTIFY_DONE;
3262 }
3263
3264 /*
3265  * bond_netdev_event: handle netdev notifier chain events.
3266  *
3267  * This function receives events for the netdev chain.  The caller (an
3268  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3269  * locks for us to safely manipulate the slave devices (RTNL lock,
3270  * dev_probe_lock).
3271  */
3272 static int bond_netdev_event(struct notifier_block *this,
3273                              unsigned long event, void *ptr)
3274 {
3275         struct net_device *event_dev = (struct net_device *)ptr;
3276
3277         pr_debug("event_dev: %s, event: %lx\n",
3278                  event_dev ? event_dev->name : "None",
3279                  event);
3280
3281         if (!(event_dev->priv_flags & IFF_BONDING))
3282                 return NOTIFY_DONE;
3283
3284         if (event_dev->flags & IFF_MASTER) {
3285                 pr_debug("IFF_MASTER\n");
3286                 return bond_master_netdev_event(event, event_dev);
3287         }
3288
3289         if (event_dev->flags & IFF_SLAVE) {
3290                 pr_debug("IFF_SLAVE\n");
3291                 return bond_slave_netdev_event(event, event_dev);
3292         }
3293
3294         return NOTIFY_DONE;
3295 }
3296
3297 static struct notifier_block bond_netdev_notifier = {
3298         .notifier_call = bond_netdev_event,
3299 };
3300
3301 /*---------------------------- Hashing Policies -----------------------------*/
3302
3303 /*
3304  * Hash for the output device based upon layer 2 data
3305  */
3306 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3307 {
3308         struct ethhdr *data = (struct ethhdr *)skb->data;
3309
3310         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3311                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3312
3313         return 0;
3314 }
3315
3316 /*
3317  * Hash for the output device based upon layer 2 and layer 3 data. If
3318  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3319  */
3320 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3321 {
3322         const struct ethhdr *data;
3323         const struct iphdr *iph;
3324         const struct ipv6hdr *ipv6h;
3325         u32 v6hash;
3326         const __be32 *s, *d;
3327
3328         if (skb->protocol == htons(ETH_P_IP) &&
3329             pskb_network_may_pull(skb, sizeof(*iph))) {
3330                 iph = ip_hdr(skb);
3331                 data = (struct ethhdr *)skb->data;
3332                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3333                         (data->h_dest[5] ^ data->h_source[5])) % count;
3334         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3335                    pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3336                 ipv6h = ipv6_hdr(skb);
3337                 data = (struct ethhdr *)skb->data;
3338                 s = &ipv6h->saddr.s6_addr32[0];
3339                 d = &ipv6h->daddr.s6_addr32[0];
3340                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3341                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3342                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3343         }
3344
3345         return bond_xmit_hash_policy_l2(skb, count);
3346 }
3347
3348 /*
3349  * Hash for the output device based upon layer 3 and layer 4 data. If
3350  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3351  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3352  */
3353 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3354 {
3355         u32 layer4_xor = 0;
3356         const struct iphdr *iph;
3357         const struct ipv6hdr *ipv6h;
3358         const __be32 *s, *d;
3359         const __be16 *l4 = NULL;
3360         __be16 _l4[2];
3361         int noff = skb_network_offset(skb);
3362         int poff;
3363
3364         if (skb->protocol == htons(ETH_P_IP) &&
3365             pskb_may_pull(skb, noff + sizeof(*iph))) {
3366                 iph = ip_hdr(skb);
3367                 poff = proto_ports_offset(iph->protocol);
3368
3369                 if (!ip_is_fragment(iph) && poff >= 0) {
3370                         l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3371                                                 sizeof(_l4), &_l4);
3372                         if (l4)
3373                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3374                 }
3375                 return (layer4_xor ^
3376                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3377         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3378                    pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3379                 ipv6h = ipv6_hdr(skb);
3380                 poff = proto_ports_offset(ipv6h->nexthdr);
3381                 if (poff >= 0) {
3382                         l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3383                                                 sizeof(_l4), &_l4);
3384                         if (l4)
3385                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3386                 }
3387                 s = &ipv6h->saddr.s6_addr32[0];
3388                 d = &ipv6h->daddr.s6_addr32[0];
3389                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3390                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3391                                (layer4_xor >> 8);
3392                 return layer4_xor % count;
3393         }
3394
3395         return bond_xmit_hash_policy_l2(skb, count);
3396 }
3397
3398 /*-------------------------- Device entry points ----------------------------*/
3399
3400 static void bond_work_init_all(struct bonding *bond)
3401 {
3402         INIT_DELAYED_WORK(&bond->mcast_work,
3403                           bond_resend_igmp_join_requests_delayed);
3404         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3405         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3406         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3407                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3408         else
3409                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3410         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3411 }
3412
3413 static void bond_work_cancel_all(struct bonding *bond)
3414 {
3415         cancel_delayed_work_sync(&bond->mii_work);
3416         cancel_delayed_work_sync(&bond->arp_work);
3417         cancel_delayed_work_sync(&bond->alb_work);
3418         cancel_delayed_work_sync(&bond->ad_work);
3419         cancel_delayed_work_sync(&bond->mcast_work);
3420 }
3421
3422 static int bond_open(struct net_device *bond_dev)
3423 {
3424         struct bonding *bond = netdev_priv(bond_dev);
3425         struct slave *slave;
3426         int i;
3427
3428         /* reset slave->backup and slave->inactive */
3429         read_lock(&bond->lock);
3430         if (bond->slave_cnt > 0) {
3431                 read_lock(&bond->curr_slave_lock);
3432                 bond_for_each_slave(bond, slave, i) {
3433                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3434                                 && (slave != bond->curr_active_slave)) {
3435                                 bond_set_slave_inactive_flags(slave);
3436                         } else {
3437                                 bond_set_slave_active_flags(slave);
3438                         }
3439                 }
3440                 read_unlock(&bond->curr_slave_lock);
3441         }
3442         read_unlock(&bond->lock);
3443
3444         bond_work_init_all(bond);
3445
3446         if (bond_is_lb(bond)) {
3447                 /* bond_alb_initialize must be called before the timer
3448                  * is started.
3449                  */
3450                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3451                         return -ENOMEM;
3452                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3453         }
3454
3455         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3456                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3457
3458         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3459                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3460                 if (bond->params.arp_validate)
3461                         bond->recv_probe = bond_arp_rcv;
3462         }
3463
3464         if (bond->params.mode == BOND_MODE_8023AD) {
3465                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3466                 /* register to receive LACPDUs */
3467                 bond->recv_probe = bond_3ad_lacpdu_recv;
3468                 bond_3ad_initiate_agg_selection(bond, 1);
3469         }
3470
3471         return 0;
3472 }
3473
3474 static int bond_close(struct net_device *bond_dev)
3475 {
3476         struct bonding *bond = netdev_priv(bond_dev);
3477
3478         write_lock_bh(&bond->lock);
3479         bond->send_peer_notif = 0;
3480         write_unlock_bh(&bond->lock);
3481
3482         bond_work_cancel_all(bond);
3483         if (bond_is_lb(bond)) {
3484                 /* Must be called only after all
3485                  * slaves have been released
3486                  */
3487                 bond_alb_deinitialize(bond);
3488         }
3489         bond->recv_probe = NULL;
3490
3491         return 0;
3492 }
3493
3494 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3495                                                 struct rtnl_link_stats64 *stats)
3496 {
3497         struct bonding *bond = netdev_priv(bond_dev);
3498         struct rtnl_link_stats64 temp;
3499         struct slave *slave;
3500         int i;
3501
3502         memset(stats, 0, sizeof(*stats));
3503
3504         read_lock_bh(&bond->lock);
3505
3506         bond_for_each_slave(bond, slave, i) {
3507                 const struct rtnl_link_stats64 *sstats =
3508                         dev_get_stats(slave->dev, &temp);
3509
3510                 stats->rx_packets += sstats->rx_packets;
3511                 stats->rx_bytes += sstats->rx_bytes;
3512                 stats->rx_errors += sstats->rx_errors;
3513                 stats->rx_dropped += sstats->rx_dropped;
3514
3515                 stats->tx_packets += sstats->tx_packets;
3516                 stats->tx_bytes += sstats->tx_bytes;
3517                 stats->tx_errors += sstats->tx_errors;
3518                 stats->tx_dropped += sstats->tx_dropped;
3519
3520                 stats->multicast += sstats->multicast;
3521                 stats->collisions += sstats->collisions;
3522
3523                 stats->rx_length_errors += sstats->rx_length_errors;
3524                 stats->rx_over_errors += sstats->rx_over_errors;
3525                 stats->rx_crc_errors += sstats->rx_crc_errors;
3526                 stats->rx_frame_errors += sstats->rx_frame_errors;
3527                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3528                 stats->rx_missed_errors += sstats->rx_missed_errors;
3529
3530                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3531                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3532                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3533                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3534                 stats->tx_window_errors += sstats->tx_window_errors;
3535         }
3536
3537         read_unlock_bh(&bond->lock);
3538
3539         return stats;
3540 }
3541
3542 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3543 {
3544         struct net_device *slave_dev = NULL;
3545         struct ifbond k_binfo;
3546         struct ifbond __user *u_binfo = NULL;
3547         struct ifslave k_sinfo;
3548         struct ifslave __user *u_sinfo = NULL;
3549         struct mii_ioctl_data *mii = NULL;
3550         struct net *net;
3551         int res = 0;
3552
3553         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3554
3555         switch (cmd) {
3556         case SIOCGMIIPHY:
3557                 mii = if_mii(ifr);
3558                 if (!mii)
3559                         return -EINVAL;
3560
3561                 mii->phy_id = 0;
3562                 /* Fall Through */
3563         case SIOCGMIIREG:
3564                 /*
3565                  * We do this again just in case we were called by SIOCGMIIREG
3566                  * instead of SIOCGMIIPHY.
3567                  */
3568                 mii = if_mii(ifr);
3569                 if (!mii)
3570                         return -EINVAL;
3571
3572
3573                 if (mii->reg_num == 1) {
3574                         struct bonding *bond = netdev_priv(bond_dev);
3575                         mii->val_out = 0;
3576                         read_lock(&bond->lock);
3577                         read_lock(&bond->curr_slave_lock);
3578                         if (netif_carrier_ok(bond->dev))
3579                                 mii->val_out = BMSR_LSTATUS;
3580
3581                         read_unlock(&bond->curr_slave_lock);
3582                         read_unlock(&bond->lock);
3583                 }
3584
3585                 return 0;
3586         case BOND_INFO_QUERY_OLD:
3587         case SIOCBONDINFOQUERY:
3588                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3589
3590                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3591                         return -EFAULT;
3592
3593                 res = bond_info_query(bond_dev, &k_binfo);
3594                 if (res == 0 &&
3595                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3596                         return -EFAULT;
3597
3598                 return res;
3599         case BOND_SLAVE_INFO_QUERY_OLD:
3600         case SIOCBONDSLAVEINFOQUERY:
3601                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3602
3603                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3604                         return -EFAULT;
3605
3606                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3607                 if (res == 0 &&
3608                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3609                         return -EFAULT;
3610
3611                 return res;
3612         default:
3613                 /* Go on */
3614                 break;
3615         }
3616
3617         net = dev_net(bond_dev);
3618
3619         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3620                 return -EPERM;
3621
3622         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3623
3624         pr_debug("slave_dev=%p:\n", slave_dev);
3625
3626         if (!slave_dev)
3627                 res = -ENODEV;
3628         else {
3629                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3630                 switch (cmd) {
3631                 case BOND_ENSLAVE_OLD:
3632                 case SIOCBONDENSLAVE:
3633                         res = bond_enslave(bond_dev, slave_dev);
3634                         break;
3635                 case BOND_RELEASE_OLD:
3636                 case SIOCBONDRELEASE:
3637                         res = bond_release(bond_dev, slave_dev);
3638                         break;
3639                 case BOND_SETHWADDR_OLD:
3640                 case SIOCBONDSETHWADDR:
3641                         bond_set_dev_addr(bond_dev, slave_dev);
3642                         res = 0;
3643                         break;
3644                 case BOND_CHANGE_ACTIVE_OLD:
3645                 case SIOCBONDCHANGEACTIVE:
3646                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3647                         break;
3648                 default:
3649                         res = -EOPNOTSUPP;
3650                 }
3651
3652                 dev_put(slave_dev);
3653         }
3654
3655         return res;
3656 }
3657
3658 static bool bond_addr_in_mc_list(unsigned char *addr,
3659                                  struct netdev_hw_addr_list *list,
3660                                  int addrlen)
3661 {
3662         struct netdev_hw_addr *ha;
3663
3664         netdev_hw_addr_list_for_each(ha, list)
3665                 if (!memcmp(ha->addr, addr, addrlen))
3666                         return true;
3667
3668         return false;
3669 }
3670
3671 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3672 {
3673         struct bonding *bond = netdev_priv(bond_dev);
3674
3675         if (change & IFF_PROMISC)
3676                 bond_set_promiscuity(bond,
3677                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3678
3679         if (change & IFF_ALLMULTI)
3680                 bond_set_allmulti(bond,
3681                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3682 }
3683
3684 static void bond_set_multicast_list(struct net_device *bond_dev)
3685 {
3686         struct bonding *bond = netdev_priv(bond_dev);
3687         struct netdev_hw_addr *ha;
3688         bool found;
3689
3690         read_lock(&bond->lock);
3691
3692         /* looking for addresses to add to slaves' mc list */
3693         netdev_for_each_mc_addr(ha, bond_dev) {
3694                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3695                                              bond_dev->addr_len);
3696                 if (!found)
3697                         bond_mc_add(bond, ha->addr);
3698         }
3699
3700         /* looking for addresses to delete from slaves' list */
3701         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3702                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3703                                              bond_dev->addr_len);
3704                 if (!found)
3705                         bond_mc_del(bond, ha->addr);
3706         }
3707
3708         /* save master's multicast list */
3709         __hw_addr_flush(&bond->mc_list);
3710         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3711                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3712
3713         read_unlock(&bond->lock);
3714 }
3715
3716 static int bond_neigh_init(struct neighbour *n)
3717 {
3718         struct bonding *bond = netdev_priv(n->dev);
3719         struct slave *slave = bond->first_slave;
3720         const struct net_device_ops *slave_ops;
3721         struct neigh_parms parms;
3722         int ret;
3723
3724         if (!slave)
3725                 return 0;
3726
3727         slave_ops = slave->dev->netdev_ops;
3728
3729         if (!slave_ops->ndo_neigh_setup)
3730                 return 0;
3731
3732         parms.neigh_setup = NULL;
3733         parms.neigh_cleanup = NULL;
3734         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3735         if (ret)
3736                 return ret;
3737
3738         /*
3739          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3740          * after the last slave has been detached.  Assumes that all slaves
3741          * utilize the same neigh_cleanup (true at this writing as only user
3742          * is ipoib).
3743          */
3744         n->parms->neigh_cleanup = parms.neigh_cleanup;
3745
3746         if (!parms.neigh_setup)
3747                 return 0;
3748
3749         return parms.neigh_setup(n);
3750 }
3751
3752 /*
3753  * The bonding ndo_neigh_setup is called at init time beofre any
3754  * slave exists. So we must declare proxy setup function which will
3755  * be used at run time to resolve the actual slave neigh param setup.
3756  */
3757 static int bond_neigh_setup(struct net_device *dev,
3758                             struct neigh_parms *parms)
3759 {
3760         parms->neigh_setup   = bond_neigh_init;
3761
3762         return 0;
3763 }
3764
3765 /*
3766  * Change the MTU of all of a master's slaves to match the master
3767  */
3768 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3769 {
3770         struct bonding *bond = netdev_priv(bond_dev);
3771         struct slave *slave, *stop_at;
3772         int res = 0;
3773         int i;
3774
3775         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3776                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3777
3778         /* Can't hold bond->lock with bh disabled here since
3779          * some base drivers panic. On the other hand we can't
3780          * hold bond->lock without bh disabled because we'll
3781          * deadlock. The only solution is to rely on the fact
3782          * that we're under rtnl_lock here, and the slaves
3783          * list won't change. This doesn't solve the problem
3784          * of setting the slave's MTU while it is
3785          * transmitting, but the assumption is that the base
3786          * driver can handle that.
3787          *
3788          * TODO: figure out a way to safely iterate the slaves
3789          * list, but without holding a lock around the actual
3790          * call to the base driver.
3791          */
3792
3793         bond_for_each_slave(bond, slave, i) {
3794                 pr_debug("s %p s->p %p c_m %p\n",
3795                          slave,
3796                          slave->prev,
3797                          slave->dev->netdev_ops->ndo_change_mtu);
3798
3799                 res = dev_set_mtu(slave->dev, new_mtu);
3800
3801                 if (res) {
3802                         /* If we failed to set the slave's mtu to the new value
3803                          * we must abort the operation even in ACTIVE_BACKUP
3804                          * mode, because if we allow the backup slaves to have
3805                          * different mtu values than the active slave we'll
3806                          * need to change their mtu when doing a failover. That
3807                          * means changing their mtu from timer context, which
3808                          * is probably not a good idea.
3809                          */
3810                         pr_debug("err %d %s\n", res, slave->dev->name);
3811                         goto unwind;
3812                 }
3813         }
3814
3815         bond_dev->mtu = new_mtu;
3816
3817         return 0;
3818
3819 unwind:
3820         /* unwind from head to the slave that failed */
3821         stop_at = slave;
3822         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3823                 int tmp_res;
3824
3825                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3826                 if (tmp_res) {
3827                         pr_debug("unwind err %d dev %s\n",
3828                                  tmp_res, slave->dev->name);
3829                 }
3830         }
3831
3832         return res;
3833 }
3834
3835 /*
3836  * Change HW address
3837  *
3838  * Note that many devices must be down to change the HW address, and
3839  * downing the master releases all slaves.  We can make bonds full of
3840  * bonding devices to test this, however.
3841  */
3842 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3843 {
3844         struct bonding *bond = netdev_priv(bond_dev);
3845         struct sockaddr *sa = addr, tmp_sa;
3846         struct slave *slave, *stop_at;
3847         int res = 0;
3848         int i;
3849
3850         if (bond->params.mode == BOND_MODE_ALB)
3851                 return bond_alb_set_mac_address(bond_dev, addr);
3852
3853
3854         pr_debug("bond=%p, name=%s\n",
3855                  bond, bond_dev ? bond_dev->name : "None");
3856
3857         /*
3858          * If fail_over_mac is set to active, do nothing and return
3859          * success.  Returning an error causes ifenslave to fail.
3860          */
3861         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3862                 return 0;
3863
3864         if (!is_valid_ether_addr(sa->sa_data))
3865                 return -EADDRNOTAVAIL;
3866
3867         /* Can't hold bond->lock with bh disabled here since
3868          * some base drivers panic. On the other hand we can't
3869          * hold bond->lock without bh disabled because we'll
3870          * deadlock. The only solution is to rely on the fact
3871          * that we're under rtnl_lock here, and the slaves
3872          * list won't change. This doesn't solve the problem
3873          * of setting the slave's hw address while it is
3874          * transmitting, but the assumption is that the base
3875          * driver can handle that.
3876          *
3877          * TODO: figure out a way to safely iterate the slaves
3878          * list, but without holding a lock around the actual
3879          * call to the base driver.
3880          */
3881
3882         bond_for_each_slave(bond, slave, i) {
3883                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3884                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3885
3886                 if (slave_ops->ndo_set_mac_address == NULL) {
3887                         res = -EOPNOTSUPP;
3888                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3889                         goto unwind;
3890                 }
3891
3892                 res = dev_set_mac_address(slave->dev, addr);
3893                 if (res) {
3894                         /* TODO: consider downing the slave
3895                          * and retry ?
3896                          * User should expect communications
3897                          * breakage anyway until ARP finish
3898                          * updating, so...
3899                          */
3900                         pr_debug("err %d %s\n", res, slave->dev->name);
3901                         goto unwind;
3902                 }
3903         }
3904
3905         /* success */
3906         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3907         return 0;
3908
3909 unwind:
3910         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3911         tmp_sa.sa_family = bond_dev->type;
3912
3913         /* unwind from head to the slave that failed */
3914         stop_at = slave;
3915         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3916                 int tmp_res;
3917
3918                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3919                 if (tmp_res) {
3920                         pr_debug("unwind err %d dev %s\n",
3921                                  tmp_res, slave->dev->name);
3922                 }
3923         }
3924
3925         return res;
3926 }
3927
3928 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3929 {
3930         struct bonding *bond = netdev_priv(bond_dev);
3931         struct slave *slave, *start_at;
3932         int i, slave_no, res = 1;
3933         struct iphdr *iph = ip_hdr(skb);
3934
3935         /*
3936          * Start with the curr_active_slave that joined the bond as the
3937          * default for sending IGMP traffic.  For failover purposes one
3938          * needs to maintain some consistency for the interface that will
3939          * send the join/membership reports.  The curr_active_slave found
3940          * will send all of this type of traffic.
3941          */
3942         if ((iph->protocol == IPPROTO_IGMP) &&
3943             (skb->protocol == htons(ETH_P_IP))) {
3944
3945                 read_lock(&bond->curr_slave_lock);
3946                 slave = bond->curr_active_slave;
3947                 read_unlock(&bond->curr_slave_lock);
3948
3949                 if (!slave)
3950                         goto out;
3951         } else {
3952                 /*
3953                  * Concurrent TX may collide on rr_tx_counter; we accept
3954                  * that as being rare enough not to justify using an
3955                  * atomic op here.
3956                  */
3957                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3958
3959                 bond_for_each_slave(bond, slave, i) {
3960                         slave_no--;
3961                         if (slave_no < 0)
3962                                 break;
3963                 }
3964         }
3965
3966         start_at = slave;
3967         bond_for_each_slave_from(bond, slave, i, start_at) {
3968                 if (IS_UP(slave->dev) &&
3969                     (slave->link == BOND_LINK_UP) &&
3970                     bond_is_active_slave(slave)) {
3971                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3972                         break;
3973                 }
3974         }
3975
3976 out:
3977         if (res) {
3978                 /* no suitable interface, frame not sent */
3979                 kfree_skb(skb);
3980         }
3981
3982         return NETDEV_TX_OK;
3983 }
3984
3985
3986 /*
3987  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3988  * the bond has a usable interface.
3989  */
3990 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3991 {
3992         struct bonding *bond = netdev_priv(bond_dev);
3993         int res = 1;
3994
3995         read_lock(&bond->curr_slave_lock);
3996
3997         if (bond->curr_active_slave)
3998                 res = bond_dev_queue_xmit(bond, skb,
3999                         bond->curr_active_slave->dev);
4000
4001         read_unlock(&bond->curr_slave_lock);
4002
4003         if (res)
4004                 /* no suitable interface, frame not sent */
4005                 kfree_skb(skb);
4006
4007         return NETDEV_TX_OK;
4008 }
4009
4010 /*
4011  * In bond_xmit_xor() , we determine the output device by using a pre-
4012  * determined xmit_hash_policy(), If the selected device is not enabled,
4013  * find the next active slave.
4014  */
4015 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4016 {
4017         struct bonding *bond = netdev_priv(bond_dev);
4018         struct slave *slave, *start_at;
4019         int slave_no;
4020         int i;
4021         int res = 1;
4022
4023         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4024
4025         bond_for_each_slave(bond, slave, i) {
4026                 slave_no--;
4027                 if (slave_no < 0)
4028                         break;
4029         }
4030
4031         start_at = slave;
4032
4033         bond_for_each_slave_from(bond, slave, i, start_at) {
4034                 if (IS_UP(slave->dev) &&
4035                     (slave->link == BOND_LINK_UP) &&
4036                     bond_is_active_slave(slave)) {
4037                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4038                         break;
4039                 }
4040         }
4041
4042         if (res) {
4043                 /* no suitable interface, frame not sent */
4044                 kfree_skb(skb);
4045         }
4046
4047         return NETDEV_TX_OK;
4048 }
4049
4050 /*
4051  * in broadcast mode, we send everything to all usable interfaces.
4052  */
4053 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4054 {
4055         struct bonding *bond = netdev_priv(bond_dev);
4056         struct slave *slave, *start_at;
4057         struct net_device *tx_dev = NULL;
4058         int i;
4059         int res = 1;
4060
4061         read_lock(&bond->curr_slave_lock);
4062         start_at = bond->curr_active_slave;
4063         read_unlock(&bond->curr_slave_lock);
4064
4065         if (!start_at)
4066                 goto out;
4067
4068         bond_for_each_slave_from(bond, slave, i, start_at) {
4069                 if (IS_UP(slave->dev) &&
4070                     (slave->link == BOND_LINK_UP) &&
4071                     bond_is_active_slave(slave)) {
4072                         if (tx_dev) {
4073                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4074                                 if (!skb2) {
4075                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4076                                                bond_dev->name);
4077                                         continue;
4078                                 }
4079
4080                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4081                                 if (res) {
4082                                         kfree_skb(skb2);
4083                                         continue;
4084                                 }
4085                         }
4086                         tx_dev = slave->dev;
4087                 }
4088         }
4089
4090         if (tx_dev)
4091                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4092
4093 out:
4094         if (res)
4095                 /* no suitable interface, frame not sent */
4096                 kfree_skb(skb);
4097
4098         /* frame sent to all suitable interfaces */
4099         return NETDEV_TX_OK;
4100 }
4101
4102 /*------------------------- Device initialization ---------------------------*/
4103
4104 static void bond_set_xmit_hash_policy(struct bonding *bond)
4105 {
4106         switch (bond->params.xmit_policy) {
4107         case BOND_XMIT_POLICY_LAYER23:
4108                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4109                 break;
4110         case BOND_XMIT_POLICY_LAYER34:
4111                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4112                 break;
4113         case BOND_XMIT_POLICY_LAYER2:
4114         default:
4115                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4116                 break;
4117         }
4118 }
4119
4120 /*
4121  * Lookup the slave that corresponds to a qid
4122  */
4123 static inline int bond_slave_override(struct bonding *bond,
4124                                       struct sk_buff *skb)
4125 {
4126         int i, res = 1;
4127         struct slave *slave = NULL;
4128         struct slave *check_slave;
4129
4130         if (!skb->queue_mapping)
4131                 return 1;
4132
4133         /* Find out if any slaves have the same mapping as this skb. */
4134         bond_for_each_slave(bond, check_slave, i) {
4135                 if (check_slave->queue_id == skb->queue_mapping) {
4136                         slave = check_slave;
4137                         break;
4138                 }
4139         }
4140
4141         /* If the slave isn't UP, use default transmit policy. */
4142         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4143             (slave->link == BOND_LINK_UP)) {
4144                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4145         }
4146
4147         return res;
4148 }
4149
4150
4151 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4152 {
4153         /*
4154          * This helper function exists to help dev_pick_tx get the correct
4155          * destination queue.  Using a helper function skips a call to
4156          * skb_tx_hash and will put the skbs in the queue we expect on their
4157          * way down to the bonding driver.
4158          */
4159         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4160
4161         /*
4162          * Save the original txq to restore before passing to the driver
4163          */
4164         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4165
4166         if (unlikely(txq >= dev->real_num_tx_queues)) {
4167                 do {
4168                         txq -= dev->real_num_tx_queues;
4169                 } while (txq >= dev->real_num_tx_queues);
4170         }
4171         return txq;
4172 }
4173
4174 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4175 {
4176         struct bonding *bond = netdev_priv(dev);
4177
4178         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4179                 if (!bond_slave_override(bond, skb))
4180                         return NETDEV_TX_OK;
4181         }
4182
4183         switch (bond->params.mode) {
4184         case BOND_MODE_ROUNDROBIN:
4185                 return bond_xmit_roundrobin(skb, dev);
4186         case BOND_MODE_ACTIVEBACKUP:
4187                 return bond_xmit_activebackup(skb, dev);
4188         case BOND_MODE_XOR:
4189                 return bond_xmit_xor(skb, dev);
4190         case BOND_MODE_BROADCAST:
4191                 return bond_xmit_broadcast(skb, dev);
4192         case BOND_MODE_8023AD:
4193                 return bond_3ad_xmit_xor(skb, dev);
4194         case BOND_MODE_ALB:
4195         case BOND_MODE_TLB:
4196                 return bond_alb_xmit(skb, dev);
4197         default:
4198                 /* Should never happen, mode already checked */
4199                 pr_err("%s: Error: Unknown bonding mode %d\n",
4200                        dev->name, bond->params.mode);
4201                 WARN_ON_ONCE(1);
4202                 kfree_skb(skb);
4203                 return NETDEV_TX_OK;
4204         }
4205 }
4206
4207 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4208 {
4209         struct bonding *bond = netdev_priv(dev);
4210         netdev_tx_t ret = NETDEV_TX_OK;
4211
4212         /*
4213          * If we risk deadlock from transmitting this in the
4214          * netpoll path, tell netpoll to queue the frame for later tx
4215          */
4216         if (is_netpoll_tx_blocked(dev))
4217                 return NETDEV_TX_BUSY;
4218
4219         read_lock(&bond->lock);
4220
4221         if (bond->slave_cnt)
4222                 ret = __bond_start_xmit(skb, dev);
4223         else
4224                 kfree_skb(skb);
4225
4226         read_unlock(&bond->lock);
4227
4228         return ret;
4229 }
4230
4231 /*
4232  * set bond mode specific net device operations
4233  */
4234 void bond_set_mode_ops(struct bonding *bond, int mode)
4235 {
4236         struct net_device *bond_dev = bond->dev;
4237
4238         switch (mode) {
4239         case BOND_MODE_ROUNDROBIN:
4240                 break;
4241         case BOND_MODE_ACTIVEBACKUP:
4242                 break;
4243         case BOND_MODE_XOR:
4244                 bond_set_xmit_hash_policy(bond);
4245                 break;
4246         case BOND_MODE_BROADCAST:
4247                 break;
4248         case BOND_MODE_8023AD:
4249                 bond_set_xmit_hash_policy(bond);
4250                 break;
4251         case BOND_MODE_ALB:
4252                 /* FALLTHRU */
4253         case BOND_MODE_TLB:
4254                 break;
4255         default:
4256                 /* Should never happen, mode already checked */
4257                 pr_err("%s: Error: Unknown bonding mode %d\n",
4258                        bond_dev->name, mode);
4259                 break;
4260         }
4261 }
4262
4263 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4264                                      struct ethtool_drvinfo *drvinfo)
4265 {
4266         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4267         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4268         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4269                  BOND_ABI_VERSION);
4270 }
4271
4272 static const struct ethtool_ops bond_ethtool_ops = {
4273         .get_drvinfo            = bond_ethtool_get_drvinfo,
4274         .get_link               = ethtool_op_get_link,
4275 };
4276
4277 static const struct net_device_ops bond_netdev_ops = {
4278         .ndo_init               = bond_init,
4279         .ndo_uninit             = bond_uninit,
4280         .ndo_open               = bond_open,
4281         .ndo_stop               = bond_close,
4282         .ndo_start_xmit         = bond_start_xmit,
4283         .ndo_select_queue       = bond_select_queue,
4284         .ndo_get_stats64        = bond_get_stats,
4285         .ndo_do_ioctl           = bond_do_ioctl,
4286         .ndo_change_rx_flags    = bond_change_rx_flags,
4287         .ndo_set_rx_mode        = bond_set_multicast_list,
4288         .ndo_change_mtu         = bond_change_mtu,
4289         .ndo_set_mac_address    = bond_set_mac_address,
4290         .ndo_neigh_setup        = bond_neigh_setup,
4291         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4292         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4293 #ifdef CONFIG_NET_POLL_CONTROLLER
4294         .ndo_netpoll_setup      = bond_netpoll_setup,
4295         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4296         .ndo_poll_controller    = bond_poll_controller,
4297 #endif
4298         .ndo_add_slave          = bond_enslave,
4299         .ndo_del_slave          = bond_release,
4300         .ndo_fix_features       = bond_fix_features,
4301 };
4302
4303 static const struct device_type bond_type = {
4304         .name = "bond",
4305 };
4306
4307 static void bond_destructor(struct net_device *bond_dev)
4308 {
4309         struct bonding *bond = netdev_priv(bond_dev);
4310         if (bond->wq)
4311                 destroy_workqueue(bond->wq);
4312         free_netdev(bond_dev);
4313 }
4314
4315 static void bond_setup(struct net_device *bond_dev)
4316 {
4317         struct bonding *bond = netdev_priv(bond_dev);
4318
4319         /* initialize rwlocks */
4320         rwlock_init(&bond->lock);
4321         rwlock_init(&bond->curr_slave_lock);
4322
4323         bond->params = bonding_defaults;
4324
4325         /* Initialize pointers */
4326         bond->dev = bond_dev;
4327         INIT_LIST_HEAD(&bond->vlan_list);
4328
4329         /* Initialize the device entry points */
4330         ether_setup(bond_dev);
4331         bond_dev->netdev_ops = &bond_netdev_ops;
4332         bond_dev->ethtool_ops = &bond_ethtool_ops;
4333         bond_set_mode_ops(bond, bond->params.mode);
4334
4335         bond_dev->destructor = bond_destructor;
4336
4337         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4338
4339         /* Initialize the device options */
4340         bond_dev->tx_queue_len = 0;
4341         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4342         bond_dev->priv_flags |= IFF_BONDING;
4343         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4344
4345         /* At first, we block adding VLANs. That's the only way to
4346          * prevent problems that occur when adding VLANs over an
4347          * empty bond. The block will be removed once non-challenged
4348          * slaves are enslaved.
4349          */
4350         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4351
4352         /* don't acquire bond device's netif_tx_lock when
4353          * transmitting */
4354         bond_dev->features |= NETIF_F_LLTX;
4355
4356         /* By default, we declare the bond to be fully
4357          * VLAN hardware accelerated capable. Special
4358          * care is taken in the various xmit functions
4359          * when there are slaves that are not hw accel
4360          * capable
4361          */
4362
4363         bond_dev->hw_features = BOND_VLAN_FEATURES |
4364                                 NETIF_F_HW_VLAN_TX |
4365                                 NETIF_F_HW_VLAN_RX |
4366                                 NETIF_F_HW_VLAN_FILTER;
4367
4368         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4369         bond_dev->features |= bond_dev->hw_features;
4370 }
4371
4372 /*
4373 * Destroy a bonding device.
4374 * Must be under rtnl_lock when this function is called.
4375 */
4376 static void bond_uninit(struct net_device *bond_dev)
4377 {
4378         struct bonding *bond = netdev_priv(bond_dev);
4379         struct vlan_entry *vlan, *tmp;
4380
4381         bond_netpoll_cleanup(bond_dev);
4382
4383         /* Release the bonded slaves */
4384         while (bond->first_slave != NULL)
4385                 __bond_release_one(bond_dev, bond->first_slave->dev, true);
4386         pr_info("%s: released all slaves\n", bond_dev->name);
4387
4388         list_del(&bond->bond_list);
4389
4390         bond_debug_unregister(bond);
4391
4392         __hw_addr_flush(&bond->mc_list);
4393
4394         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4395                 list_del(&vlan->vlan_list);
4396                 kfree(vlan);
4397         }
4398 }
4399
4400 /*------------------------- Module initialization ---------------------------*/
4401
4402 /*
4403  * Convert string input module parms.  Accept either the
4404  * number of the mode or its string name.  A bit complicated because
4405  * some mode names are substrings of other names, and calls from sysfs
4406  * may have whitespace in the name (trailing newlines, for example).
4407  */
4408 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4409 {
4410         int modeint = -1, i, rv;
4411         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4412
4413         for (p = (char *)buf; *p; p++)
4414                 if (!(isdigit(*p) || isspace(*p)))
4415                         break;
4416
4417         if (*p)
4418                 rv = sscanf(buf, "%20s", modestr);
4419         else
4420                 rv = sscanf(buf, "%d", &modeint);
4421
4422         if (!rv)
4423                 return -1;
4424
4425         for (i = 0; tbl[i].modename; i++) {
4426                 if (modeint == tbl[i].mode)
4427                         return tbl[i].mode;
4428                 if (strcmp(modestr, tbl[i].modename) == 0)
4429                         return tbl[i].mode;
4430         }
4431
4432         return -1;
4433 }
4434
4435 static int bond_check_params(struct bond_params *params)
4436 {
4437         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4438
4439         /*
4440          * Convert string parameters.
4441          */
4442         if (mode) {
4443                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4444                 if (bond_mode == -1) {
4445                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4446                                mode == NULL ? "NULL" : mode);
4447                         return -EINVAL;
4448                 }
4449         }
4450
4451         if (xmit_hash_policy) {
4452                 if ((bond_mode != BOND_MODE_XOR) &&
4453                     (bond_mode != BOND_MODE_8023AD)) {
4454                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4455                                bond_mode_name(bond_mode));
4456                 } else {
4457                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4458                                                         xmit_hashtype_tbl);
4459                         if (xmit_hashtype == -1) {
4460                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4461                                        xmit_hash_policy == NULL ? "NULL" :
4462                                        xmit_hash_policy);
4463                                 return -EINVAL;
4464                         }
4465                 }
4466         }
4467
4468         if (lacp_rate) {
4469                 if (bond_mode != BOND_MODE_8023AD) {
4470                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4471                                 bond_mode_name(bond_mode));
4472                 } else {
4473                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4474                         if (lacp_fast == -1) {
4475                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4476                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4477                                 return -EINVAL;
4478                         }
4479                 }
4480         }
4481
4482         if (ad_select) {
4483                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4484                 if (params->ad_select == -1) {
4485                         pr_err("Error: Invalid ad_select \"%s\"\n",
4486                                ad_select == NULL ? "NULL" : ad_select);
4487                         return -EINVAL;
4488                 }
4489
4490                 if (bond_mode != BOND_MODE_8023AD) {
4491                         pr_warning("ad_select param only affects 802.3ad mode\n");
4492                 }
4493         } else {
4494                 params->ad_select = BOND_AD_STABLE;
4495         }
4496
4497         if (max_bonds < 0) {
4498                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4499                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4500                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4501         }
4502
4503         if (miimon < 0) {
4504                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4505                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4506                 miimon = BOND_LINK_MON_INTERV;
4507         }
4508
4509         if (updelay < 0) {
4510                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4511                            updelay, INT_MAX);
4512                 updelay = 0;
4513         }
4514
4515         if (downdelay < 0) {
4516                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4517                            downdelay, INT_MAX);
4518                 downdelay = 0;
4519         }
4520
4521         if ((use_carrier != 0) && (use_carrier != 1)) {
4522                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4523                            use_carrier);
4524                 use_carrier = 1;
4525         }
4526
4527         if (num_peer_notif < 0 || num_peer_notif > 255) {
4528                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4529                            num_peer_notif);
4530                 num_peer_notif = 1;
4531         }
4532
4533         /* reset values for 802.3ad */
4534         if (bond_mode == BOND_MODE_8023AD) {
4535                 if (!miimon) {
4536                         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");
4537                         pr_warning("Forcing miimon to 100msec\n");
4538                         miimon = 100;
4539                 }
4540         }
4541
4542         if (tx_queues < 1 || tx_queues > 255) {
4543                 pr_warning("Warning: tx_queues (%d) should be between "
4544                            "1 and 255, resetting to %d\n",
4545                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4546                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4547         }
4548
4549         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4550                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4551                            "not of valid value (0/1), so it was set to "
4552                            "0\n", all_slaves_active);
4553                 all_slaves_active = 0;
4554         }
4555
4556         if (resend_igmp < 0 || resend_igmp > 255) {
4557                 pr_warning("Warning: resend_igmp (%d) should be between "
4558                            "0 and 255, resetting to %d\n",
4559                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4560                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4561         }
4562
4563         /* reset values for TLB/ALB */
4564         if ((bond_mode == BOND_MODE_TLB) ||
4565             (bond_mode == BOND_MODE_ALB)) {
4566                 if (!miimon) {
4567                         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");
4568                         pr_warning("Forcing miimon to 100msec\n");
4569                         miimon = 100;
4570                 }
4571         }
4572
4573         if (bond_mode == BOND_MODE_ALB) {
4574                 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",
4575                           updelay);
4576         }
4577
4578         if (!miimon) {
4579                 if (updelay || downdelay) {
4580                         /* just warn the user the up/down delay will have
4581                          * no effect since miimon is zero...
4582                          */
4583                         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",
4584                                    updelay, downdelay);
4585                 }
4586         } else {
4587                 /* don't allow arp monitoring */
4588                 if (arp_interval) {
4589                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4590                                    miimon, arp_interval);
4591                         arp_interval = 0;
4592                 }
4593
4594                 if ((updelay % miimon) != 0) {
4595                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4596                                    updelay, miimon,
4597                                    (updelay / miimon) * miimon);
4598                 }
4599
4600                 updelay /= miimon;
4601
4602                 if ((downdelay % miimon) != 0) {
4603                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4604                                    downdelay, miimon,
4605                                    (downdelay / miimon) * miimon);
4606                 }
4607
4608                 downdelay /= miimon;
4609         }
4610
4611         if (arp_interval < 0) {
4612                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4613                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4614                 arp_interval = BOND_LINK_ARP_INTERV;
4615         }
4616
4617         for (arp_ip_count = 0;
4618              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4619              arp_ip_count++) {
4620                 /* not complete check, but should be good enough to
4621                    catch mistakes */
4622                 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4623                 if (!isdigit(arp_ip_target[arp_ip_count][0]) ||
4624                     ip == 0 || ip == htonl(INADDR_BROADCAST)) {
4625                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4626                                    arp_ip_target[arp_ip_count]);
4627                         arp_interval = 0;
4628                 } else {
4629                         arp_target[arp_ip_count] = ip;
4630                 }
4631         }
4632
4633         if (arp_interval && !arp_ip_count) {
4634                 /* don't allow arping if no arp_ip_target given... */
4635                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4636                            arp_interval);
4637                 arp_interval = 0;
4638         }
4639
4640         if (arp_validate) {
4641                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4642                         pr_err("arp_validate only supported in active-backup mode\n");
4643                         return -EINVAL;
4644                 }
4645                 if (!arp_interval) {
4646                         pr_err("arp_validate requires arp_interval\n");
4647                         return -EINVAL;
4648                 }
4649
4650                 arp_validate_value = bond_parse_parm(arp_validate,
4651                                                      arp_validate_tbl);
4652                 if (arp_validate_value == -1) {
4653                         pr_err("Error: invalid arp_validate \"%s\"\n",
4654                                arp_validate == NULL ? "NULL" : arp_validate);
4655                         return -EINVAL;
4656                 }
4657         } else
4658                 arp_validate_value = 0;
4659
4660         if (miimon) {
4661                 pr_info("MII link monitoring set to %d ms\n", miimon);
4662         } else if (arp_interval) {
4663                 int i;
4664
4665                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4666                         arp_interval,
4667                         arp_validate_tbl[arp_validate_value].modename,
4668                         arp_ip_count);
4669
4670                 for (i = 0; i < arp_ip_count; i++)
4671                         pr_info(" %s", arp_ip_target[i]);
4672
4673                 pr_info("\n");
4674
4675         } else if (max_bonds) {
4676                 /* miimon and arp_interval not set, we need one so things
4677                  * work as expected, see bonding.txt for details
4678                  */
4679                 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");
4680         }
4681
4682         if (primary && !USES_PRIMARY(bond_mode)) {
4683                 /* currently, using a primary only makes sense
4684                  * in active backup, TLB or ALB modes
4685                  */
4686                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4687                            primary, bond_mode_name(bond_mode));
4688                 primary = NULL;
4689         }
4690
4691         if (primary && primary_reselect) {
4692                 primary_reselect_value = bond_parse_parm(primary_reselect,
4693                                                          pri_reselect_tbl);
4694                 if (primary_reselect_value == -1) {
4695                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4696                                primary_reselect ==
4697                                         NULL ? "NULL" : primary_reselect);
4698                         return -EINVAL;
4699                 }
4700         } else {
4701                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4702         }
4703
4704         if (fail_over_mac) {
4705                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4706                                                       fail_over_mac_tbl);
4707                 if (fail_over_mac_value == -1) {
4708                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4709                                arp_validate == NULL ? "NULL" : arp_validate);
4710                         return -EINVAL;
4711                 }
4712
4713                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4714                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4715         } else {
4716                 fail_over_mac_value = BOND_FOM_NONE;
4717         }
4718
4719         /* fill params struct with the proper values */
4720         params->mode = bond_mode;
4721         params->xmit_policy = xmit_hashtype;
4722         params->miimon = miimon;
4723         params->num_peer_notif = num_peer_notif;
4724         params->arp_interval = arp_interval;
4725         params->arp_validate = arp_validate_value;
4726         params->updelay = updelay;
4727         params->downdelay = downdelay;
4728         params->use_carrier = use_carrier;
4729         params->lacp_fast = lacp_fast;
4730         params->primary[0] = 0;
4731         params->primary_reselect = primary_reselect_value;
4732         params->fail_over_mac = fail_over_mac_value;
4733         params->tx_queues = tx_queues;
4734         params->all_slaves_active = all_slaves_active;
4735         params->resend_igmp = resend_igmp;
4736         params->min_links = min_links;
4737
4738         if (primary) {
4739                 strncpy(params->primary, primary, IFNAMSIZ);
4740                 params->primary[IFNAMSIZ - 1] = 0;
4741         }
4742
4743         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4744
4745         return 0;
4746 }
4747
4748 static struct lock_class_key bonding_netdev_xmit_lock_key;
4749 static struct lock_class_key bonding_netdev_addr_lock_key;
4750 static struct lock_class_key bonding_tx_busylock_key;
4751
4752 static void bond_set_lockdep_class_one(struct net_device *dev,
4753                                        struct netdev_queue *txq,
4754                                        void *_unused)
4755 {
4756         lockdep_set_class(&txq->_xmit_lock,
4757                           &bonding_netdev_xmit_lock_key);
4758 }
4759
4760 static void bond_set_lockdep_class(struct net_device *dev)
4761 {
4762         lockdep_set_class(&dev->addr_list_lock,
4763                           &bonding_netdev_addr_lock_key);
4764         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4765         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4766 }
4767
4768 /*
4769  * Called from registration process
4770  */
4771 static int bond_init(struct net_device *bond_dev)
4772 {
4773         struct bonding *bond = netdev_priv(bond_dev);
4774         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4775         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4776
4777         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4778
4779         /*
4780          * Initialize locks that may be required during
4781          * en/deslave operations.  All of the bond_open work
4782          * (of which this is part) should really be moved to
4783          * a phase prior to dev_open
4784          */
4785         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4786         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4787
4788         bond->wq = create_singlethread_workqueue(bond_dev->name);
4789         if (!bond->wq)
4790                 return -ENOMEM;
4791
4792         bond_set_lockdep_class(bond_dev);
4793
4794         list_add_tail(&bond->bond_list, &bn->dev_list);
4795
4796         bond_prepare_sysfs_group(bond);
4797
4798         bond_debug_register(bond);
4799
4800         /* Ensure valid dev_addr */
4801         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4802             bond_dev->addr_assign_type == NET_ADDR_PERM) {
4803                 eth_hw_addr_random(bond_dev);
4804                 bond->dev_addr_from_first = true;
4805         }
4806
4807         __hw_addr_init(&bond->mc_list);
4808         return 0;
4809 }
4810
4811 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4812 {
4813         if (tb[IFLA_ADDRESS]) {
4814                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4815                         return -EINVAL;
4816                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4817                         return -EADDRNOTAVAIL;
4818         }
4819         return 0;
4820 }
4821
4822 static unsigned int bond_get_num_tx_queues(void)
4823 {
4824         return tx_queues;
4825 }
4826
4827 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4828         .kind                   = "bond",
4829         .priv_size              = sizeof(struct bonding),
4830         .setup                  = bond_setup,
4831         .validate               = bond_validate,
4832         .get_num_tx_queues      = bond_get_num_tx_queues,
4833         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4834                                                              as for TX queues */
4835 };
4836
4837 /* Create a new bond based on the specified name and bonding parameters.
4838  * If name is NULL, obtain a suitable "bond%d" name for us.
4839  * Caller must NOT hold rtnl_lock; we need to release it here before we
4840  * set up our sysfs entries.
4841  */
4842 int bond_create(struct net *net, const char *name)
4843 {
4844         struct net_device *bond_dev;
4845         int res;
4846
4847         rtnl_lock();
4848
4849         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4850                                    name ? name : "bond%d",
4851                                    bond_setup, tx_queues);
4852         if (!bond_dev) {
4853                 pr_err("%s: eek! can't alloc netdev!\n", name);
4854                 rtnl_unlock();
4855                 return -ENOMEM;
4856         }
4857
4858         dev_net_set(bond_dev, net);
4859         bond_dev->rtnl_link_ops = &bond_link_ops;
4860
4861         res = register_netdevice(bond_dev);
4862
4863         netif_carrier_off(bond_dev);
4864
4865         rtnl_unlock();
4866         if (res < 0)
4867                 bond_destructor(bond_dev);
4868         return res;
4869 }
4870
4871 static int __net_init bond_net_init(struct net *net)
4872 {
4873         struct bond_net *bn = net_generic(net, bond_net_id);
4874
4875         bn->net = net;
4876         INIT_LIST_HEAD(&bn->dev_list);
4877
4878         bond_create_proc_dir(bn);
4879         bond_create_sysfs(bn);
4880
4881         return 0;
4882 }
4883
4884 static void __net_exit bond_net_exit(struct net *net)
4885 {
4886         struct bond_net *bn = net_generic(net, bond_net_id);
4887         struct bonding *bond, *tmp_bond;
4888         LIST_HEAD(list);
4889
4890         bond_destroy_sysfs(bn);
4891         bond_destroy_proc_dir(bn);
4892
4893         /* Kill off any bonds created after unregistering bond rtnl ops */
4894         rtnl_lock();
4895         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4896                 unregister_netdevice_queue(bond->dev, &list);
4897         unregister_netdevice_many(&list);
4898         rtnl_unlock();
4899 }
4900
4901 static struct pernet_operations bond_net_ops = {
4902         .init = bond_net_init,
4903         .exit = bond_net_exit,
4904         .id   = &bond_net_id,
4905         .size = sizeof(struct bond_net),
4906 };
4907
4908 static int __init bonding_init(void)
4909 {
4910         int i;
4911         int res;
4912
4913         pr_info("%s", bond_version);
4914
4915         res = bond_check_params(&bonding_defaults);
4916         if (res)
4917                 goto out;
4918
4919         res = register_pernet_subsys(&bond_net_ops);
4920         if (res)
4921                 goto out;
4922
4923         res = rtnl_link_register(&bond_link_ops);
4924         if (res)
4925                 goto err_link;
4926
4927         bond_create_debugfs();
4928
4929         for (i = 0; i < max_bonds; i++) {
4930                 res = bond_create(&init_net, NULL);
4931                 if (res)
4932                         goto err;
4933         }
4934
4935         register_netdevice_notifier(&bond_netdev_notifier);
4936 out:
4937         return res;
4938 err:
4939         rtnl_link_unregister(&bond_link_ops);
4940 err_link:
4941         unregister_pernet_subsys(&bond_net_ops);
4942         goto out;
4943
4944 }
4945
4946 static void __exit bonding_exit(void)
4947 {
4948         unregister_netdevice_notifier(&bond_netdev_notifier);
4949
4950         bond_destroy_debugfs();
4951
4952         rtnl_link_unregister(&bond_link_ops);
4953         unregister_pernet_subsys(&bond_net_ops);
4954
4955 #ifdef CONFIG_NET_POLL_CONTROLLER
4956         /*
4957          * Make sure we don't have an imbalance on our netpoll blocking
4958          */
4959         WARN_ON(atomic_read(&netpoll_block_tx));
4960 #endif
4961 }
4962
4963 module_init(bonding_init);
4964 module_exit(bonding_exit);
4965 MODULE_LICENSE("GPL");
4966 MODULE_VERSION(DRV_VERSION);
4967 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4968 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4969 MODULE_ALIAS_RTNL_LINK("bond");