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