drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon       = BOND_LINK_MON_INTERV;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *arp_all_targets;
 108 static char *fail_over_mac;
 109 static int all_slaves_active;
 110 static struct bond_params bonding_defaults;
 111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 112
 113 module_param(max_bonds, int, 0);
 114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 115 module_param(tx_queues, int, 0);
 116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 117 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 118 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 119                                "failover event (alias of num_unsol_na)");
 120 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 122                                "failover event (alias of num_grat_arp)");
 123 module_param(miimon, int, 0);
 124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 125 module_param(updelay, int, 0);
 126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 127 module_param(downdelay, int, 0);
 128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 129                             "in milliseconds");
 130 module_param(use_carrier, int, 0);
 131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 132                               "0 for off, 1 for on (default)");
 133 module_param(mode, charp, 0);
 134 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 135                        "1 for active-backup, 2 for balance-xor, "
 136                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 137                        "6 for balance-alb");
 138 module_param(primary, charp, 0);
 139 MODULE_PARM_DESC(primary, "Primary network device to use");
 140 module_param(primary_reselect, charp, 0);
 141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 142                                    "once it comes up; "
 143                                    "0 for always (default), "
 144                                    "1 for only if speed of primary is "
 145                                    "better, "
 146                                    "2 for only on active slave "
 147                                    "failure");
 148 module_param(lacp_rate, charp, 0);
 149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 150                             "0 for slow, 1 for fast");
 151 module_param(ad_select, charp, 0);
 152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 153                             "0 for stable (default), 1 for bandwidth, "
 154                             "2 for count");
 155 module_param(min_links, int, 0);
 156 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 157
 158 module_param(xmit_hash_policy, charp, 0);
 159 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 160                                    "0 for layer 2 (default), 1 for layer 3+4, "
 161                                    "2 for layer 2+3");
 162 module_param(arp_interval, int, 0);
 163 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 164 module_param_array(arp_ip_target, charp, NULL, 0);
 165 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 166 module_param(arp_validate, charp, 0);
 167 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 168                                "0 for none (default), 1 for active, "
 169                                "2 for backup, 3 for all");
 170 module_param(arp_all_targets, charp, 0);
 171 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 172 module_param(fail_over_mac, charp, 0);
 173 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 174                                 "the same MAC; 0 for none (default), "
 175                                 "1 for active, 2 for follow");
 176 module_param(all_slaves_active, int, 0);
 177 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 178                                      "by setting active flag for all slaves; "
 179                                      "0 for never (default), 1 for always.");
 180 module_param(resend_igmp, int, 0);
 181 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 182                               "link failure");
 183
 184 /*----------------------------- Global variables ----------------------------*/
 185
 186 #ifdef CONFIG_NET_POLL_CONTROLLER
 187 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 188 #endif
 189
 190 int bond_net_id __read_mostly;
 191
 192 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 193 static int arp_ip_count;
 194 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 195 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 196 static int lacp_fast;
 197
 198 const struct bond_parm_tbl bond_lacp_tbl[] = {
 199 {       "slow",         AD_LACP_SLOW},
 200 {       "fast",         AD_LACP_FAST},
 201 {       NULL,           -1},
 202 };
 203
 204 const struct bond_parm_tbl bond_mode_tbl[] = {
 205 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 206 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 207 {       "balance-xor",          BOND_MODE_XOR},
 208 {       "broadcast",            BOND_MODE_BROADCAST},
 209 {       "802.3ad",              BOND_MODE_8023AD},
 210 {       "balance-tlb",          BOND_MODE_TLB},
 211 {       "balance-alb",          BOND_MODE_ALB},
 212 {       NULL,                   -1},
 213 };
 214
 215 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 216 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 217 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 218 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 219 {       NULL,                   -1},
 220 };
 221
 222 const struct bond_parm_tbl arp_all_targets_tbl[] = {
 223 {       "any",                  BOND_ARP_TARGETS_ANY},
 224 {       "all",                  BOND_ARP_TARGETS_ALL},
 225 {       NULL,                   -1},
 226 };
 227
 228 const struct bond_parm_tbl arp_validate_tbl[] = {
 229 {       "none",                 BOND_ARP_VALIDATE_NONE},
 230 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 231 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 232 {       "all",                  BOND_ARP_VALIDATE_ALL},
 233 {       NULL,                   -1},
 234 };
 235
 236 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 237 {       "none",                 BOND_FOM_NONE},
 238 {       "active",               BOND_FOM_ACTIVE},
 239 {       "follow",               BOND_FOM_FOLLOW},
 240 {       NULL,                   -1},
 241 };
 242
 243 const struct bond_parm_tbl pri_reselect_tbl[] = {
 244 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 245 {       "better",               BOND_PRI_RESELECT_BETTER},
 246 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 247 {       NULL,                   -1},
 248 };
 249
 250 struct bond_parm_tbl ad_select_tbl[] = {
 251 {       "stable",       BOND_AD_STABLE},
 252 {       "bandwidth",    BOND_AD_BANDWIDTH},
 253 {       "count",        BOND_AD_COUNT},
 254 {       NULL,           -1},
 255 };
 256
 257 /*-------------------------- Forward declarations ---------------------------*/
 258
 259 static int bond_init(struct net_device *bond_dev);
 260 static void bond_uninit(struct net_device *bond_dev);
 261
 262 /*---------------------------- General routines -----------------------------*/
 263
 264 const char *bond_mode_name(int mode)
 265 {
 266         static const char *names[] = {
 267                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 268                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 269                 [BOND_MODE_XOR] = "load balancing (xor)",
 270                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 271                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 272                 [BOND_MODE_TLB] = "transmit load balancing",
 273                 [BOND_MODE_ALB] = "adaptive load balancing",
 274         };
 275
 276         if (mode < 0 || mode > BOND_MODE_ALB)
 277                 return "unknown";
 278
 279         return names[mode];
 280 }
 281
 282 /*---------------------------------- VLAN -----------------------------------*/
 283
 284 /**
 285  * bond_add_vlan - add a new vlan id on bond
 286  * @bond: bond that got the notification
 287  * @vlan_id: the vlan id to add
 288  *
 289  * Returns -ENOMEM if allocation failed.
 290  */
 291 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 292 {
 293         struct vlan_entry *vlan;
 294
 295         pr_debug("bond: %s, vlan id %d\n",
 296                  (bond ? bond->dev->name : "None"), vlan_id);
 297
 298         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 299         if (!vlan)
 300                 return -ENOMEM;
 301
 302         INIT_LIST_HEAD(&vlan->vlan_list);
 303         vlan->vlan_id = vlan_id;
 304
 305         write_lock_bh(&bond->lock);
 306
 307         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 308
 309         write_unlock_bh(&bond->lock);
 310
 311         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 312
 313         return 0;
 314 }
 315
 316 /**
 317  * bond_del_vlan - delete a vlan id from bond
 318  * @bond: bond that got the notification
 319  * @vlan_id: the vlan id to delete
 320  *
 321  * returns -ENODEV if @vlan_id was not found in @bond.
 322  */
 323 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 324 {
 325         struct vlan_entry *vlan;
 326         int res = -ENODEV;
 327
 328         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 329
 330         block_netpoll_tx();
 331         write_lock_bh(&bond->lock);
 332
 333         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 334                 if (vlan->vlan_id == vlan_id) {
 335                         list_del(&vlan->vlan_list);
 336
 337                         if (bond_is_lb(bond))
 338                                 bond_alb_clear_vlan(bond, vlan_id);
 339
 340                         pr_debug("removed VLAN ID %d from bond %s\n",
 341                                  vlan_id, bond->dev->name);
 342
 343                         kfree(vlan);
 344
 345                         res = 0;
 346                         goto out;
 347                 }
 348         }
 349
 350         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 351                  vlan_id, bond->dev->name);
 352
 353 out:
 354         write_unlock_bh(&bond->lock);
 355         unblock_netpoll_tx();
 356         return res;
 357 }
 358
 359 /**
 360  * bond_next_vlan - safely skip to the next item in the vlans list.
 361  * @bond: the bond we're working on
 362  * @curr: item we're advancing from
 363  *
 364  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 365  * or @curr->next otherwise (even if it is @curr itself again).
 366  *
 367  * Caller must hold bond->lock
 368  */
 369 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 370 {
 371         struct vlan_entry *next, *last;
 372
 373         if (list_empty(&bond->vlan_list))
 374                 return NULL;
 375
 376         if (!curr) {
 377                 next = list_entry(bond->vlan_list.next,
 378                                   struct vlan_entry, vlan_list);
 379         } else {
 380                 last = list_entry(bond->vlan_list.prev,
 381                                   struct vlan_entry, vlan_list);
 382                 if (last == curr) {
 383                         next = list_entry(bond->vlan_list.next,
 384                                           struct vlan_entry, vlan_list);
 385                 } else {
 386                         next = list_entry(curr->vlan_list.next,
 387                                           struct vlan_entry, vlan_list);
 388                 }
 389         }
 390
 391         return next;
 392 }
 393
 394 /**
 395  * bond_dev_queue_xmit - Prepare skb for xmit.
 396  *
 397  * @bond: bond device that got this skb for tx.
 398  * @skb: hw accel VLAN tagged skb to transmit
 399  * @slave_dev: slave that is supposed to xmit this skbuff
 400  */
 401 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 402                         struct net_device *slave_dev)
 403 {
 404         skb->dev = slave_dev;
 405
 406         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 407                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 408         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 409
 410         if (unlikely(netpoll_tx_running(bond->dev)))
 411                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 412         else
 413                 dev_queue_xmit(skb);
 414
 415         return 0;
 416 }
 417
 418 /*
 419  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 420  * We don't protect the slave list iteration with a lock because:
 421  * a. This operation is performed in IOCTL context,
 422  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 423  * c. Holding a lock with BH disabled while directly calling a base driver
 424  *    entry point is generally a BAD idea.
 425  *
 426  * The design of synchronization/protection for this operation in the 8021q
 427  * module is good for one or more VLAN devices over a single physical device
 428  * and cannot be extended for a teaming solution like bonding, so there is a
 429  * potential race condition here where a net device from the vlan group might
 430  * be referenced (either by a base driver or the 8021q code) while it is being
 431  * removed from the system. However, it turns out we're not making matters
 432  * worse, and if it works for regular VLAN usage it will work here too.
 433 */
 434
 435 /**
 436  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 437  * @bond_dev: bonding net device that got called
 438  * @vid: vlan id being added
 439  */
 440 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 441                                 __be16 proto, u16 vid)
 442 {
 443         struct bonding *bond = netdev_priv(bond_dev);
 444         struct slave *slave, *stop_at;
 445         int i, res;
 446
 447         bond_for_each_slave(bond, slave, i) {
 448                 res = vlan_vid_add(slave->dev, proto, vid);
 449                 if (res)
 450                         goto unwind;
 451         }
 452
 453         res = bond_add_vlan(bond, vid);
 454         if (res) {
 455                 pr_err("%s: Error: Failed to add vlan id %d\n",
 456                        bond_dev->name, vid);
 457                 return res;
 458         }
 459
 460         return 0;
 461
 462 unwind:
 463         /* unwind from head to the slave that failed */
 464         stop_at = slave;
 465         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
 466                 vlan_vid_del(slave->dev, proto, vid);
 467
 468         return res;
 469 }
 470
 471 /**
 472  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 473  * @bond_dev: bonding net device that got called
 474  * @vid: vlan id being removed
 475  */
 476 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 477                                  __be16 proto, u16 vid)
 478 {
 479         struct bonding *bond = netdev_priv(bond_dev);
 480         struct slave *slave;
 481         int i, res;
 482
 483         bond_for_each_slave(bond, slave, i)
 484                 vlan_vid_del(slave->dev, proto, vid);
 485
 486         res = bond_del_vlan(bond, vid);
 487         if (res) {
 488                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 489                        bond_dev->name, vid);
 490                 return res;
 491         }
 492
 493         return 0;
 494 }
 495
 496 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 497 {
 498         struct vlan_entry *vlan;
 499         int res;
 500
 501         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 502                 res = vlan_vid_add(slave_dev, htons(ETH_P_8021Q),
 503                                    vlan->vlan_id);
 504                 if (res)
 505                         pr_warning("%s: Failed to add vlan id %d to device %s\n",
 506                                    bond->dev->name, vlan->vlan_id,
 507                                    slave_dev->name);
 508         }
 509 }
 510
 511 static void bond_del_vlans_from_slave(struct bonding *bond,
 512                                       struct net_device *slave_dev)
 513 {
 514         struct vlan_entry *vlan;
 515
 516         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 517                 if (!vlan->vlan_id)
 518                         continue;
 519                 vlan_vid_del(slave_dev, htons(ETH_P_8021Q), vlan->vlan_id);
 520         }
 521 }
 522
 523 /*------------------------------- Link status -------------------------------*/
 524
 525 /*
 526  * Set the carrier state for the master according to the state of its
 527  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 528  * do special 802.3ad magic.
 529  *
 530  * Returns zero if carrier state does not change, nonzero if it does.
 531  */
 532 static int bond_set_carrier(struct bonding *bond)
 533 {
 534         struct slave *slave;
 535         int i;
 536
 537         if (bond->slave_cnt == 0)
 538                 goto down;
 539
 540         if (bond->params.mode == BOND_MODE_8023AD)
 541                 return bond_3ad_set_carrier(bond);
 542
 543         bond_for_each_slave(bond, slave, i) {
 544                 if (slave->link == BOND_LINK_UP) {
 545                         if (!netif_carrier_ok(bond->dev)) {
 546                                 netif_carrier_on(bond->dev);
 547                                 return 1;
 548                         }
 549                         return 0;
 550                 }
 551         }
 552
 553 down:
 554         if (netif_carrier_ok(bond->dev)) {
 555                 netif_carrier_off(bond->dev);
 556                 return 1;
 557         }
 558         return 0;
 559 }
 560
 561 /*
 562  * Get link speed and duplex from the slave's base driver
 563  * using ethtool. If for some reason the call fails or the
 564  * values are invalid, set speed and duplex to -1,
 565  * and return.
 566  */
 567 static void bond_update_speed_duplex(struct slave *slave)
 568 {
 569         struct net_device *slave_dev = slave->dev;
 570         struct ethtool_cmd ecmd;
 571         u32 slave_speed;
 572         int res;
 573
 574         slave->speed = SPEED_UNKNOWN;
 575         slave->duplex = DUPLEX_UNKNOWN;
 576
 577         res = __ethtool_get_settings(slave_dev, &ecmd);
 578         if (res < 0)
 579                 return;
 580
 581         slave_speed = ethtool_cmd_speed(&ecmd);
 582         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 583                 return;
 584
 585         switch (ecmd.duplex) {
 586         case DUPLEX_FULL:
 587         case DUPLEX_HALF:
 588                 break;
 589         default:
 590                 return;
 591         }
 592
 593         slave->speed = slave_speed;
 594         slave->duplex = ecmd.duplex;
 595
 596         return;
 597 }
 598
 599 /*
 600  * if <dev> supports MII link status reporting, check its link status.
 601  *
 602  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 603  * depending upon the setting of the use_carrier parameter.
 604  *
 605  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 606  * can't tell and just pretend it is), or 0, meaning that the link is
 607  * down.
 608  *
 609  * If reporting is non-zero, instead of faking link up, return -1 if
 610  * both ETHTOOL and MII ioctls fail (meaning the device does not
 611  * support them).  If use_carrier is set, return whatever it says.
 612  * It'd be nice if there was a good way to tell if a driver supports
 613  * netif_carrier, but there really isn't.
 614  */
 615 static int bond_check_dev_link(struct bonding *bond,
 616                                struct net_device *slave_dev, int reporting)
 617 {
 618         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 619         int (*ioctl)(struct net_device *, struct ifreq *, int);
 620         struct ifreq ifr;
 621         struct mii_ioctl_data *mii;
 622
 623         if (!reporting && !netif_running(slave_dev))
 624                 return 0;
 625
 626         if (bond->params.use_carrier)
 627                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 628
 629         /* Try to get link status using Ethtool first. */
 630         if (slave_dev->ethtool_ops->get_link)
 631                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 632                         BMSR_LSTATUS : 0;
 633
 634         /* Ethtool can't be used, fallback to MII ioctls. */
 635         ioctl = slave_ops->ndo_do_ioctl;
 636         if (ioctl) {
 637                 /* TODO: set pointer to correct ioctl on a per team member */
 638                 /*       bases to make this more efficient. that is, once  */
 639                 /*       we determine the correct ioctl, we will always    */
 640                 /*       call it and not the others for that team          */
 641                 /*       member.                                           */
 642
 643                 /*
 644                  * We cannot assume that SIOCGMIIPHY will also read a
 645                  * register; not all network drivers (e.g., e100)
 646                  * support that.
 647                  */
 648
 649                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 650                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 651                 mii = if_mii(&ifr);
 652                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 653                         mii->reg_num = MII_BMSR;
 654                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 655                                 return mii->val_out & BMSR_LSTATUS;
 656                 }
 657         }
 658
 659         /*
 660          * If reporting, report that either there's no dev->do_ioctl,
 661          * or both SIOCGMIIREG and get_link failed (meaning that we
 662          * cannot report link status).  If not reporting, pretend
 663          * we're ok.
 664          */
 665         return reporting ? -1 : BMSR_LSTATUS;
 666 }
 667
 668 /*----------------------------- Multicast list ------------------------------*/
 669
 670 /*
 671  * Push the promiscuity flag down to appropriate slaves
 672  */
 673 static int bond_set_promiscuity(struct bonding *bond, int inc)
 674 {
 675         int err = 0;
 676         if (USES_PRIMARY(bond->params.mode)) {
 677                 /* write lock already acquired */
 678                 if (bond->curr_active_slave) {
 679                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 680                                                   inc);
 681                 }
 682         } else {
 683                 struct slave *slave;
 684                 int i;
 685                 bond_for_each_slave(bond, slave, i) {
 686                         err = dev_set_promiscuity(slave->dev, inc);
 687                         if (err)
 688                                 return err;
 689                 }
 690         }
 691         return err;
 692 }
 693
 694 /*
 695  * Push the allmulti flag down to all slaves
 696  */
 697 static int bond_set_allmulti(struct bonding *bond, int inc)
 698 {
 699         int err = 0;
 700         if (USES_PRIMARY(bond->params.mode)) {
 701                 /* write lock already acquired */
 702                 if (bond->curr_active_slave) {
 703                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 704                                                inc);
 705                 }
 706         } else {
 707                 struct slave *slave;
 708                 int i;
 709                 bond_for_each_slave(bond, slave, i) {
 710                         err = dev_set_allmulti(slave->dev, inc);
 711                         if (err)
 712                                 return err;
 713                 }
 714         }
 715         return err;
 716 }
 717
 718 /*
 719  * Retrieve the list of registered multicast addresses for the bonding
 720  * device and retransmit an IGMP JOIN request to the current active
 721  * slave.
 722  */
 723 static void bond_resend_igmp_join_requests(struct bonding *bond)
 724 {
 725         if (!rtnl_trylock()) {
 726                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
 727                 return;
 728         }
 729         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 730         rtnl_unlock();
 731
 732         /* We use curr_slave_lock to protect against concurrent access to
 733          * igmp_retrans from multiple running instances of this function and
 734          * bond_change_active_slave
 735          */
 736         write_lock_bh(&bond->curr_slave_lock);
 737         if (bond->igmp_retrans > 1) {
 738                 bond->igmp_retrans--;
 739                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 740         }
 741         write_unlock_bh(&bond->curr_slave_lock);
 742         read_unlock(&bond->lock);
 743 }
 744
 745 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 746 {
 747         struct bonding *bond = container_of(work, struct bonding,
 748                                             mcast_work.work);
 749
 750         bond_resend_igmp_join_requests(bond);
 751 }
 752
 753 /* Flush bond's hardware addresses from slave
 754  */
 755 static void bond_hw_addr_flush(struct net_device *bond_dev,
 756                                struct net_device *slave_dev)
 757 {
 758         struct bonding *bond = netdev_priv(bond_dev);
 759
 760         dev_uc_unsync(slave_dev, bond_dev);
 761         dev_mc_unsync(slave_dev, bond_dev);
 762
 763         if (bond->params.mode == BOND_MODE_8023AD) {
 764                 /* del lacpdu mc addr from mc list */
 765                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 766
 767                 dev_mc_del(slave_dev, lacpdu_multicast);
 768         }
 769 }
 770
 771 /*--------------------------- Active slave change ---------------------------*/
 772
 773 /* Update the hardware address list and promisc/allmulti for the new and
 774  * old active slaves (if any).  Modes that are !USES_PRIMARY keep all
 775  * slaves up date at all times; only the USES_PRIMARY modes need to call
 776  * this function to swap these settings during a failover.
 777  */
 778 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 779                               struct slave *old_active)
 780 {
 781         if (old_active) {
 782                 if (bond->dev->flags & IFF_PROMISC)
 783                         dev_set_promiscuity(old_active->dev, -1);
 784
 785                 if (bond->dev->flags & IFF_ALLMULTI)
 786                         dev_set_allmulti(old_active->dev, -1);
 787
 788                 bond_hw_addr_flush(bond->dev, old_active->dev);
 789         }
 790
 791         if (new_active) {
 792                 /* FIXME: Signal errors upstream. */
 793                 if (bond->dev->flags & IFF_PROMISC)
 794                         dev_set_promiscuity(new_active->dev, 1);
 795
 796                 if (bond->dev->flags & IFF_ALLMULTI)
 797                         dev_set_allmulti(new_active->dev, 1);
 798
 799                 netif_addr_lock_bh(bond->dev);
 800                 dev_uc_sync(new_active->dev, bond->dev);
 801                 dev_mc_sync(new_active->dev, bond->dev);
 802                 netif_addr_unlock_bh(bond->dev);
 803         }
 804 }
 805
 806 /**
 807  * bond_set_dev_addr - clone slave's address to bond
 808  * @bond_dev: bond net device
 809  * @slave_dev: slave net device
 810  *
 811  * Should be called with RTNL held.
 812  */
 813 static void bond_set_dev_addr(struct net_device *bond_dev,
 814                               struct net_device *slave_dev)
 815 {
 816         pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 817                  bond_dev, slave_dev, slave_dev->addr_len);
 818         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 819         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 820         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 821 }
 822
 823 /*
 824  * bond_do_fail_over_mac
 825  *
 826  * Perform special MAC address swapping for fail_over_mac settings
 827  *
 828  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 829  */
 830 static void bond_do_fail_over_mac(struct bonding *bond,
 831                                   struct slave *new_active,
 832                                   struct slave *old_active)
 833         __releases(&bond->curr_slave_lock)
 834         __releases(&bond->lock)
 835         __acquires(&bond->lock)
 836         __acquires(&bond->curr_slave_lock)
 837 {
 838         u8 tmp_mac[ETH_ALEN];
 839         struct sockaddr saddr;
 840         int rv;
 841
 842         switch (bond->params.fail_over_mac) {
 843         case BOND_FOM_ACTIVE:
 844                 if (new_active) {
 845                         write_unlock_bh(&bond->curr_slave_lock);
 846                         read_unlock(&bond->lock);
 847                         bond_set_dev_addr(bond->dev, new_active->dev);
 848                         read_lock(&bond->lock);
 849                         write_lock_bh(&bond->curr_slave_lock);
 850                 }
 851                 break;
 852         case BOND_FOM_FOLLOW:
 853                 /*
 854                  * if new_active && old_active, swap them
 855                  * if just old_active, do nothing (going to no active slave)
 856                  * if just new_active, set new_active to bond's MAC
 857                  */
 858                 if (!new_active)
 859                         return;
 860
 861                 write_unlock_bh(&bond->curr_slave_lock);
 862                 read_unlock(&bond->lock);
 863
 864                 if (old_active) {
 865                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 866                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 867                                ETH_ALEN);
 868                         saddr.sa_family = new_active->dev->type;
 869                 } else {
 870                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 871                         saddr.sa_family = bond->dev->type;
 872                 }
 873
 874                 rv = dev_set_mac_address(new_active->dev, &saddr);
 875                 if (rv) {
 876                         pr_err("%s: Error %d setting MAC of slave %s\n",
 877                                bond->dev->name, -rv, new_active->dev->name);
 878                         goto out;
 879                 }
 880
 881                 if (!old_active)
 882                         goto out;
 883
 884                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 885                 saddr.sa_family = old_active->dev->type;
 886
 887                 rv = dev_set_mac_address(old_active->dev, &saddr);
 888                 if (rv)
 889                         pr_err("%s: Error %d setting MAC of slave %s\n",
 890                                bond->dev->name, -rv, new_active->dev->name);
 891 out:
 892                 read_lock(&bond->lock);
 893                 write_lock_bh(&bond->curr_slave_lock);
 894                 break;
 895         default:
 896                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 897                        bond->dev->name, bond->params.fail_over_mac);
 898                 break;
 899         }
 900
 901 }
 902
 903 static bool bond_should_change_active(struct bonding *bond)
 904 {
 905         struct slave *prim = bond->primary_slave;
 906         struct slave *curr = bond->curr_active_slave;
 907
 908         if (!prim || !curr || curr->link != BOND_LINK_UP)
 909                 return true;
 910         if (bond->force_primary) {
 911                 bond->force_primary = false;
 912                 return true;
 913         }
 914         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 915             (prim->speed < curr->speed ||
 916              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 917                 return false;
 918         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 919                 return false;
 920         return true;
 921 }
 922
 923 /**
 924  * find_best_interface - select the best available slave to be the active one
 925  * @bond: our bonding struct
 926  *
 927  * Warning: Caller must hold curr_slave_lock for writing.
 928  */
 929 static struct slave *bond_find_best_slave(struct bonding *bond)
 930 {
 931         struct slave *new_active, *old_active;
 932         struct slave *bestslave = NULL;
 933         int mintime = bond->params.updelay;
 934         int i;
 935
 936         new_active = bond->curr_active_slave;
 937
 938         if (!new_active) { /* there were no active slaves left */
 939                 if (bond->slave_cnt > 0)   /* found one slave */
 940                         new_active = bond->first_slave;
 941                 else
 942                         return NULL; /* still no slave, return NULL */
 943         }
 944
 945         if ((bond->primary_slave) &&
 946             bond->primary_slave->link == BOND_LINK_UP &&
 947             bond_should_change_active(bond)) {
 948                 new_active = bond->primary_slave;
 949         }
 950
 951         /* remember where to stop iterating over the slaves */
 952         old_active = new_active;
 953
 954         bond_for_each_slave_from(bond, new_active, i, old_active) {
 955                 if (new_active->link == BOND_LINK_UP) {
 956                         return new_active;
 957                 } else if (new_active->link == BOND_LINK_BACK &&
 958                            IS_UP(new_active->dev)) {
 959                         /* link up, but waiting for stabilization */
 960                         if (new_active->delay < mintime) {
 961                                 mintime = new_active->delay;
 962                                 bestslave = new_active;
 963                         }
 964                 }
 965         }
 966
 967         return bestslave;
 968 }
 969
 970 static bool bond_should_notify_peers(struct bonding *bond)
 971 {
 972         struct slave *slave = bond->curr_active_slave;
 973
 974         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
 975                  bond->dev->name, slave ? slave->dev->name : "NULL");
 976
 977         if (!slave || !bond->send_peer_notif ||
 978             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 979                 return false;
 980
 981         bond->send_peer_notif--;
 982         return true;
 983 }
 984
 985 /**
 986  * change_active_interface - change the active slave into the specified one
 987  * @bond: our bonding struct
 988  * @new: the new slave to make the active one
 989  *
 990  * Set the new slave to the bond's settings and unset them on the old
 991  * curr_active_slave.
 992  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 993  *
 994  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 995  * because it is apparently the best available slave we have, even though its
 996  * updelay hasn't timed out yet.
 997  *
 998  * If new_active is not NULL, caller must hold bond->lock for read and
 999  * curr_slave_lock for write_bh.
1000  */
1001 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1002 {
1003         struct slave *old_active = bond->curr_active_slave;
1004
1005         if (old_active == new_active)
1006                 return;
1007
1008         if (new_active) {
1009                 new_active->jiffies = jiffies;
1010
1011                 if (new_active->link == BOND_LINK_BACK) {
1012                         if (USES_PRIMARY(bond->params.mode)) {
1013                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1014                                         bond->dev->name, new_active->dev->name,
1015                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1016                         }
1017
1018                         new_active->delay = 0;
1019                         new_active->link = BOND_LINK_UP;
1020
1021                         if (bond->params.mode == BOND_MODE_8023AD)
1022                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1023
1024                         if (bond_is_lb(bond))
1025                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1026                 } else {
1027                         if (USES_PRIMARY(bond->params.mode)) {
1028                                 pr_info("%s: making interface %s the new active one.\n",
1029                                         bond->dev->name, new_active->dev->name);
1030                         }
1031                 }
1032         }
1033
1034         if (USES_PRIMARY(bond->params.mode))
1035                 bond_hw_addr_swap(bond, new_active, old_active);
1036
1037         if (bond_is_lb(bond)) {
1038                 bond_alb_handle_active_change(bond, new_active);
1039                 if (old_active)
1040                         bond_set_slave_inactive_flags(old_active);
1041                 if (new_active)
1042                         bond_set_slave_active_flags(new_active);
1043         } else {
1044                 bond->curr_active_slave = new_active;
1045         }
1046
1047         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1048                 if (old_active)
1049                         bond_set_slave_inactive_flags(old_active);
1050
1051                 if (new_active) {
1052                         bool should_notify_peers = false;
1053
1054                         bond_set_slave_active_flags(new_active);
1055
1056                         if (bond->params.fail_over_mac)
1057                                 bond_do_fail_over_mac(bond, new_active,
1058                                                       old_active);
1059
1060                         if (netif_running(bond->dev)) {
1061                                 bond->send_peer_notif =
1062                                         bond->params.num_peer_notif;
1063                                 should_notify_peers =
1064                                         bond_should_notify_peers(bond);
1065                         }
1066
1067                         write_unlock_bh(&bond->curr_slave_lock);
1068                         read_unlock(&bond->lock);
1069
1070                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1071                         if (should_notify_peers)
1072                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1073                                                          bond->dev);
1074
1075                         read_lock(&bond->lock);
1076                         write_lock_bh(&bond->curr_slave_lock);
1077                 }
1078         }
1079
1080         /* resend IGMP joins since active slave has changed or
1081          * all were sent on curr_active_slave.
1082          * resend only if bond is brought up with the affected
1083          * bonding modes and the retransmission is enabled */
1084         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1085             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1086              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1087                 bond->igmp_retrans = bond->params.resend_igmp;
1088                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1089         }
1090 }
1091
1092 /**
1093  * bond_select_active_slave - select a new active slave, if needed
1094  * @bond: our bonding struct
1095  *
1096  * This functions should be called when one of the following occurs:
1097  * - The old curr_active_slave has been released or lost its link.
1098  * - The primary_slave has got its link back.
1099  * - A slave has got its link back and there's no old curr_active_slave.
1100  *
1101  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1102  */
1103 void bond_select_active_slave(struct bonding *bond)
1104 {
1105         struct slave *best_slave;
1106         int rv;
1107
1108         best_slave = bond_find_best_slave(bond);
1109         if (best_slave != bond->curr_active_slave) {
1110                 bond_change_active_slave(bond, best_slave);
1111                 rv = bond_set_carrier(bond);
1112                 if (!rv)
1113                         return;
1114
1115                 if (netif_carrier_ok(bond->dev)) {
1116                         pr_info("%s: first active interface up!\n",
1117                                 bond->dev->name);
1118                 } else {
1119                         pr_info("%s: now running without any active interface !\n",
1120                                 bond->dev->name);
1121                 }
1122         }
1123 }
1124
1125 /*--------------------------- slave list handling ---------------------------*/
1126
1127 /*
1128  * This function attaches the slave to the end of list.
1129  *
1130  * bond->lock held for writing by caller.
1131  */
1132 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1133 {
1134         if (bond->first_slave == NULL) { /* attaching the first slave */
1135                 new_slave->next = new_slave;
1136                 new_slave->prev = new_slave;
1137                 bond->first_slave = new_slave;
1138         } else {
1139                 new_slave->next = bond->first_slave;
1140                 new_slave->prev = bond->first_slave->prev;
1141                 new_slave->next->prev = new_slave;
1142                 new_slave->prev->next = new_slave;
1143         }
1144
1145         bond->slave_cnt++;
1146 }
1147
1148 /*
1149  * This function detaches the slave from the list.
1150  * WARNING: no check is made to verify if the slave effectively
1151  * belongs to <bond>.
1152  * Nothing is freed on return, structures are just unchained.
1153  * If any slave pointer in bond was pointing to <slave>,
1154  * it should be changed by the calling function.
1155  *
1156  * bond->lock held for writing by caller.
1157  */
1158 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1159 {
1160         if (slave->next)
1161                 slave->next->prev = slave->prev;
1162
1163         if (slave->prev)
1164                 slave->prev->next = slave->next;
1165
1166         if (bond->first_slave == slave) { /* slave is the first slave */
1167                 if (bond->slave_cnt > 1) { /* there are more slave */
1168                         bond->first_slave = slave->next;
1169                 } else {
1170                         bond->first_slave = NULL; /* slave was the last one */
1171                 }
1172         }
1173
1174         slave->next = NULL;
1175         slave->prev = NULL;
1176         bond->slave_cnt--;
1177 }
1178
1179 #ifdef CONFIG_NET_POLL_CONTROLLER
1180 static inline int slave_enable_netpoll(struct slave *slave)
1181 {
1182         struct netpoll *np;
1183         int err = 0;
1184
1185         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1186         err = -ENOMEM;
1187         if (!np)
1188                 goto out;
1189
1190         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1191         if (err) {
1192                 kfree(np);
1193                 goto out;
1194         }
1195         slave->np = np;
1196 out:
1197         return err;
1198 }
1199 static inline void slave_disable_netpoll(struct slave *slave)
1200 {
1201         struct netpoll *np = slave->np;
1202
1203         if (!np)
1204                 return;
1205
1206         slave->np = NULL;
1207         __netpoll_free_async(np);
1208 }
1209 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1210 {
1211         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1212                 return false;
1213         if (!slave_dev->netdev_ops->ndo_poll_controller)
1214                 return false;
1215         return true;
1216 }
1217
1218 static void bond_poll_controller(struct net_device *bond_dev)
1219 {
1220 }
1221
1222 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1223 {
1224         struct bonding *bond = netdev_priv(bond_dev);
1225         struct slave *slave;
1226         int i;
1227
1228         bond_for_each_slave(bond, slave, i)
1229                 if (IS_UP(slave->dev))
1230                         slave_disable_netpoll(slave);
1231 }
1232
1233 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1234 {
1235         struct bonding *bond = netdev_priv(dev);
1236         struct slave *slave;
1237         int i, err = 0;
1238
1239         bond_for_each_slave(bond, slave, i) {
1240                 err = slave_enable_netpoll(slave);
1241                 if (err) {
1242                         bond_netpoll_cleanup(dev);
1243                         break;
1244                 }
1245         }
1246         return err;
1247 }
1248
1249 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1250 {
1251         return bond->dev->npinfo;
1252 }
1253
1254 #else
1255 static inline int slave_enable_netpoll(struct slave *slave)
1256 {
1257         return 0;
1258 }
1259 static inline void slave_disable_netpoll(struct slave *slave)
1260 {
1261 }
1262 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1263 {
1264 }
1265 #endif
1266
1267 /*---------------------------------- IOCTL ----------------------------------*/
1268
1269 static netdev_features_t bond_fix_features(struct net_device *dev,
1270         netdev_features_t features)
1271 {
1272         struct slave *slave;
1273         struct bonding *bond = netdev_priv(dev);
1274         netdev_features_t mask;
1275         int i;
1276
1277         read_lock(&bond->lock);
1278
1279         if (!bond->first_slave) {
1280                 /* Disable adding VLANs to empty bond. But why? --mq */
1281                 features |= NETIF_F_VLAN_CHALLENGED;
1282                 goto out;
1283         }
1284
1285         mask = features;
1286         features &= ~NETIF_F_ONE_FOR_ALL;
1287         features |= NETIF_F_ALL_FOR_ALL;
1288
1289         bond_for_each_slave(bond, slave, i) {
1290                 features = netdev_increment_features(features,
1291                                                      slave->dev->features,
1292                                                      mask);
1293         }
1294         features = netdev_add_tso_features(features, mask);
1295
1296 out:
1297         read_unlock(&bond->lock);
1298         return features;
1299 }
1300
1301 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1302                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1303                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1304
1305 static void bond_compute_features(struct bonding *bond)
1306 {
1307         struct slave *slave;
1308         struct net_device *bond_dev = bond->dev;
1309         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1310         unsigned short max_hard_header_len = ETH_HLEN;
1311         unsigned int gso_max_size = GSO_MAX_SIZE;
1312         u16 gso_max_segs = GSO_MAX_SEGS;
1313         int i;
1314         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1315
1316         read_lock(&bond->lock);
1317
1318         if (!bond->first_slave)
1319                 goto done;
1320
1321         bond_for_each_slave(bond, slave, i) {
1322                 vlan_features = netdev_increment_features(vlan_features,
1323                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1324
1325                 dst_release_flag &= slave->dev->priv_flags;
1326                 if (slave->dev->hard_header_len > max_hard_header_len)
1327                         max_hard_header_len = slave->dev->hard_header_len;
1328
1329                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1330                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1331         }
1332
1333 done:
1334         bond_dev->vlan_features = vlan_features;
1335         bond_dev->hard_header_len = max_hard_header_len;
1336         bond_dev->gso_max_segs = gso_max_segs;
1337         netif_set_gso_max_size(bond_dev, gso_max_size);
1338
1339         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1340         bond_dev->priv_flags = flags | dst_release_flag;
1341
1342         read_unlock(&bond->lock);
1343
1344         netdev_change_features(bond_dev);
1345 }
1346
1347 static void bond_setup_by_slave(struct net_device *bond_dev,
1348                                 struct net_device *slave_dev)
1349 {
1350         bond_dev->header_ops        = slave_dev->header_ops;
1351
1352         bond_dev->type              = slave_dev->type;
1353         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1354         bond_dev->addr_len          = slave_dev->addr_len;
1355
1356         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1357                 slave_dev->addr_len);
1358 }
1359
1360 /* On bonding slaves other than the currently active slave, suppress
1361  * duplicates except for alb non-mcast/bcast.
1362  */
1363 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1364                                             struct slave *slave,
1365                                             struct bonding *bond)
1366 {
1367         if (bond_is_slave_inactive(slave)) {
1368                 if (bond->params.mode == BOND_MODE_ALB &&
1369                     skb->pkt_type != PACKET_BROADCAST &&
1370                     skb->pkt_type != PACKET_MULTICAST)
1371                         return false;
1372                 return true;
1373         }
1374         return false;
1375 }
1376
1377 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1378 {
1379         struct sk_buff *skb = *pskb;
1380         struct slave *slave;
1381         struct bonding *bond;
1382         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1383                           struct slave *);
1384         int ret = RX_HANDLER_ANOTHER;
1385
1386         skb = skb_share_check(skb, GFP_ATOMIC);
1387         if (unlikely(!skb))
1388                 return RX_HANDLER_CONSUMED;
1389
1390         *pskb = skb;
1391
1392         slave = bond_slave_get_rcu(skb->dev);
1393         bond = slave->bond;
1394
1395         if (bond->params.arp_interval)
1396                 slave->dev->last_rx = jiffies;
1397
1398         recv_probe = ACCESS_ONCE(bond->recv_probe);
1399         if (recv_probe) {
1400                 ret = recv_probe(skb, bond, slave);
1401                 if (ret == RX_HANDLER_CONSUMED) {
1402                         consume_skb(skb);
1403                         return ret;
1404                 }
1405         }
1406
1407         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1408                 return RX_HANDLER_EXACT;
1409         }
1410
1411         skb->dev = bond->dev;
1412
1413         if (bond->params.mode == BOND_MODE_ALB &&
1414             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1415             skb->pkt_type == PACKET_HOST) {
1416
1417                 if (unlikely(skb_cow_head(skb,
1418                                           skb->data - skb_mac_header(skb)))) {
1419                         kfree_skb(skb);
1420                         return RX_HANDLER_CONSUMED;
1421                 }
1422                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1423         }
1424
1425         return ret;
1426 }
1427
1428 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1429                                       struct net_device *slave_dev)
1430 {
1431         int err;
1432
1433         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1434         if (err)
1435                 return err;
1436         slave_dev->flags |= IFF_SLAVE;
1437         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1438         return 0;
1439 }
1440
1441 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1442                                   struct net_device *slave_dev)
1443 {
1444         netdev_upper_dev_unlink(slave_dev, bond_dev);
1445         slave_dev->flags &= ~IFF_SLAVE;
1446         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1447 }
1448
1449 /* enslave device <slave> to bond device <master> */
1450 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1451 {
1452         struct bonding *bond = netdev_priv(bond_dev);
1453         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1454         struct slave *new_slave = NULL;
1455         struct sockaddr addr;
1456         int link_reporting;
1457         int res = 0, i;
1458
1459         if (!bond->params.use_carrier &&
1460             slave_dev->ethtool_ops->get_link == NULL &&
1461             slave_ops->ndo_do_ioctl == NULL) {
1462                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1463                            bond_dev->name, slave_dev->name);
1464         }
1465
1466         /* already enslaved */
1467         if (slave_dev->flags & IFF_SLAVE) {
1468                 pr_debug("Error, Device was already enslaved\n");
1469                 return -EBUSY;
1470         }
1471
1472         /* vlan challenged mutual exclusion */
1473         /* no need to lock since we're protected by rtnl_lock */
1474         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1475                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1476                 if (vlan_uses_dev(bond_dev)) {
1477                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1478                                bond_dev->name, slave_dev->name, bond_dev->name);
1479                         return -EPERM;
1480                 } else {
1481                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1482                                    bond_dev->name, slave_dev->name,
1483                                    slave_dev->name, bond_dev->name);
1484                 }
1485         } else {
1486                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1487         }
1488
1489         /*
1490          * Old ifenslave binaries are no longer supported.  These can
1491          * be identified with moderate accuracy by the state of the slave:
1492          * the current ifenslave will set the interface down prior to
1493          * enslaving it; the old ifenslave will not.
1494          */
1495         if ((slave_dev->flags & IFF_UP)) {
1496                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1497                        slave_dev->name);
1498                 res = -EPERM;
1499                 goto err_undo_flags;
1500         }
1501
1502         /* set bonding device ether type by slave - bonding netdevices are
1503          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1504          * there is a need to override some of the type dependent attribs/funcs.
1505          *
1506          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1507          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1508          */
1509         if (bond->slave_cnt == 0) {
1510                 if (bond_dev->type != slave_dev->type) {
1511                         pr_debug("%s: change device type from %d to %d\n",
1512                                  bond_dev->name,
1513                                  bond_dev->type, slave_dev->type);
1514
1515                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1516                                                        bond_dev);
1517                         res = notifier_to_errno(res);
1518                         if (res) {
1519                                 pr_err("%s: refused to change device type\n",
1520                                        bond_dev->name);
1521                                 res = -EBUSY;
1522                                 goto err_undo_flags;
1523                         }
1524
1525                         /* Flush unicast and multicast addresses */
1526                         dev_uc_flush(bond_dev);
1527                         dev_mc_flush(bond_dev);
1528
1529                         if (slave_dev->type != ARPHRD_ETHER)
1530                                 bond_setup_by_slave(bond_dev, slave_dev);
1531                         else {
1532                                 ether_setup(bond_dev);
1533                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1534                         }
1535
1536                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1537                                                  bond_dev);
1538                 }
1539         } else if (bond_dev->type != slave_dev->type) {
1540                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1541                        slave_dev->name,
1542                        slave_dev->type, bond_dev->type);
1543                 res = -EINVAL;
1544                 goto err_undo_flags;
1545         }
1546
1547         if (slave_ops->ndo_set_mac_address == NULL) {
1548                 if (bond->slave_cnt == 0) {
1549                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1550                                    bond_dev->name);
1551                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1552                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1553                         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",
1554                                bond_dev->name);
1555                         res = -EOPNOTSUPP;
1556                         goto err_undo_flags;
1557                 }
1558         }
1559
1560         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1561
1562         /* If this is the first slave, then we need to set the master's hardware
1563          * address to be the same as the slave's. */
1564         if (!bond->slave_cnt && bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1565                 bond_set_dev_addr(bond->dev, slave_dev);
1566
1567         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1568         if (!new_slave) {
1569                 res = -ENOMEM;
1570                 goto err_undo_flags;
1571         }
1572
1573         /*
1574          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1575          * is set via sysfs or module option if desired.
1576          */
1577         new_slave->queue_id = 0;
1578
1579         /* Save slave's original mtu and then set it to match the bond */
1580         new_slave->original_mtu = slave_dev->mtu;
1581         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1582         if (res) {
1583                 pr_debug("Error %d calling dev_set_mtu\n", res);
1584                 goto err_free;
1585         }
1586
1587         /*
1588          * Save slave's original ("permanent") mac address for modes
1589          * that need it, and for restoring it upon release, and then
1590          * set it to the master's address
1591          */
1592         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1593
1594         if (!bond->params.fail_over_mac) {
1595                 /*
1596                  * Set slave to master's mac address.  The application already
1597                  * set the master's mac address to that of the first slave
1598                  */
1599                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1600                 addr.sa_family = slave_dev->type;
1601                 res = dev_set_mac_address(slave_dev, &addr);
1602                 if (res) {
1603                         pr_debug("Error %d calling set_mac_address\n", res);
1604                         goto err_restore_mtu;
1605                 }
1606         }
1607
1608         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1609         if (res) {
1610                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1611                 goto err_restore_mac;
1612         }
1613
1614         /* open the slave since the application closed it */
1615         res = dev_open(slave_dev);
1616         if (res) {
1617                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1618                 goto err_unset_master;
1619         }
1620
1621         new_slave->bond = bond;
1622         new_slave->dev = slave_dev;
1623         slave_dev->priv_flags |= IFF_BONDING;
1624
1625         if (bond_is_lb(bond)) {
1626                 /* bond_alb_init_slave() must be called before all other stages since
1627                  * it might fail and we do not want to have to undo everything
1628                  */
1629                 res = bond_alb_init_slave(bond, new_slave);
1630                 if (res)
1631                         goto err_close;
1632         }
1633
1634         /* If the mode USES_PRIMARY, then the following is handled by
1635          * bond_change_active_slave().
1636          */
1637         if (!USES_PRIMARY(bond->params.mode)) {
1638                 /* set promiscuity level to new slave */
1639                 if (bond_dev->flags & IFF_PROMISC) {
1640                         res = dev_set_promiscuity(slave_dev, 1);
1641                         if (res)
1642                                 goto err_close;
1643                 }
1644
1645                 /* set allmulti level to new slave */
1646                 if (bond_dev->flags & IFF_ALLMULTI) {
1647                         res = dev_set_allmulti(slave_dev, 1);
1648                         if (res)
1649                                 goto err_close;
1650                 }
1651
1652                 netif_addr_lock_bh(bond_dev);
1653
1654                 dev_mc_sync_multiple(slave_dev, bond_dev);
1655                 dev_uc_sync_multiple(slave_dev, bond_dev);
1656
1657                 netif_addr_unlock_bh(bond_dev);
1658         }
1659
1660         if (bond->params.mode == BOND_MODE_8023AD) {
1661                 /* add lacpdu mc addr to mc list */
1662                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1663
1664                 dev_mc_add(slave_dev, lacpdu_multicast);
1665         }
1666
1667         bond_add_vlans_on_slave(bond, slave_dev);
1668
1669         write_lock_bh(&bond->lock);
1670
1671         bond_attach_slave(bond, new_slave);
1672
1673         new_slave->delay = 0;
1674         new_slave->link_failure_count = 0;
1675
1676         write_unlock_bh(&bond->lock);
1677
1678         bond_compute_features(bond);
1679
1680         bond_update_speed_duplex(new_slave);
1681
1682         read_lock(&bond->lock);
1683
1684         new_slave->last_arp_rx = jiffies -
1685                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1686         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1687                 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1688
1689         if (bond->params.miimon && !bond->params.use_carrier) {
1690                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1691
1692                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1693                         /*
1694                          * miimon is set but a bonded network driver
1695                          * does not support ETHTOOL/MII and
1696                          * arp_interval is not set.  Note: if
1697                          * use_carrier is enabled, we will never go
1698                          * here (because netif_carrier is always
1699                          * supported); thus, we don't need to change
1700                          * the messages for netif_carrier.
1701                          */
1702                         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",
1703                                bond_dev->name, slave_dev->name);
1704                 } else if (link_reporting == -1) {
1705                         /* unable get link status using mii/ethtool */
1706                         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",
1707                                    bond_dev->name, slave_dev->name);
1708                 }
1709         }
1710
1711         /* check for initial state */
1712         if (bond->params.miimon) {
1713                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1714                         if (bond->params.updelay) {
1715                                 new_slave->link = BOND_LINK_BACK;
1716                                 new_slave->delay = bond->params.updelay;
1717                         } else {
1718                                 new_slave->link = BOND_LINK_UP;
1719                         }
1720                 } else {
1721                         new_slave->link = BOND_LINK_DOWN;
1722                 }
1723         } else if (bond->params.arp_interval) {
1724                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1725                         BOND_LINK_UP : BOND_LINK_DOWN);
1726         } else {
1727                 new_slave->link = BOND_LINK_UP;
1728         }
1729
1730         if (new_slave->link != BOND_LINK_DOWN)
1731                 new_slave->jiffies = jiffies;
1732         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1733                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1734                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1735
1736         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1737                 /* if there is a primary slave, remember it */
1738                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1739                         bond->primary_slave = new_slave;
1740                         bond->force_primary = true;
1741                 }
1742         }
1743
1744         write_lock_bh(&bond->curr_slave_lock);
1745
1746         switch (bond->params.mode) {
1747         case BOND_MODE_ACTIVEBACKUP:
1748                 bond_set_slave_inactive_flags(new_slave);
1749                 bond_select_active_slave(bond);
1750                 break;
1751         case BOND_MODE_8023AD:
1752                 /* in 802.3ad mode, the internal mechanism
1753                  * will activate the slaves in the selected
1754                  * aggregator
1755                  */
1756                 bond_set_slave_inactive_flags(new_slave);
1757                 /* if this is the first slave */
1758                 if (bond->slave_cnt == 1) {
1759                         SLAVE_AD_INFO(new_slave).id = 1;
1760                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1761                          * can be called only after the mac address of the bond is set
1762                          */
1763                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1764                 } else {
1765                         SLAVE_AD_INFO(new_slave).id =
1766                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1767                 }
1768
1769                 bond_3ad_bind_slave(new_slave);
1770                 break;
1771         case BOND_MODE_TLB:
1772         case BOND_MODE_ALB:
1773                 bond_set_active_slave(new_slave);
1774                 bond_set_slave_inactive_flags(new_slave);
1775                 bond_select_active_slave(bond);
1776                 break;
1777         default:
1778                 pr_debug("This slave is always active in trunk mode\n");
1779
1780                 /* always active in trunk mode */
1781                 bond_set_active_slave(new_slave);
1782
1783                 /* In trunking mode there is little meaning to curr_active_slave
1784                  * anyway (it holds no special properties of the bond device),
1785                  * so we can change it without calling change_active_interface()
1786                  */
1787                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1788                         bond->curr_active_slave = new_slave;
1789
1790                 break;
1791         } /* switch(bond_mode) */
1792
1793         write_unlock_bh(&bond->curr_slave_lock);
1794
1795         bond_set_carrier(bond);
1796
1797 #ifdef CONFIG_NET_POLL_CONTROLLER
1798         slave_dev->npinfo = bond_netpoll_info(bond);
1799         if (slave_dev->npinfo) {
1800                 if (slave_enable_netpoll(new_slave)) {
1801                         read_unlock(&bond->lock);
1802                         pr_info("Error, %s: master_dev is using netpoll, "
1803                                  "but new slave device does not support netpoll.\n",
1804                                  bond_dev->name);
1805                         res = -EBUSY;
1806                         goto err_detach;
1807                 }
1808         }
1809 #endif
1810
1811         read_unlock(&bond->lock);
1812
1813         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1814         if (res)
1815                 goto err_detach;
1816
1817         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1818                                          new_slave);
1819         if (res) {
1820                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1821                 goto err_dest_symlinks;
1822         }
1823
1824         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1825                 bond_dev->name, slave_dev->name,
1826                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1827                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1828
1829         /* enslave is successful */
1830         return 0;
1831
1832 /* Undo stages on error */
1833 err_dest_symlinks:
1834         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1835
1836 err_detach:
1837         if (!USES_PRIMARY(bond->params.mode))
1838                 bond_hw_addr_flush(bond_dev, slave_dev);
1839
1840         bond_del_vlans_from_slave(bond, slave_dev);
1841         write_lock_bh(&bond->lock);
1842         bond_detach_slave(bond, new_slave);
1843         if (bond->primary_slave == new_slave)
1844                 bond->primary_slave = NULL;
1845         if (bond->curr_active_slave == new_slave) {
1846                 bond_change_active_slave(bond, NULL);
1847                 write_unlock_bh(&bond->lock);
1848                 read_lock(&bond->lock);
1849                 write_lock_bh(&bond->curr_slave_lock);
1850                 bond_select_active_slave(bond);
1851                 write_unlock_bh(&bond->curr_slave_lock);
1852                 read_unlock(&bond->lock);
1853         } else {
1854                 write_unlock_bh(&bond->lock);
1855         }
1856         slave_disable_netpoll(new_slave);
1857
1858 err_close:
1859         slave_dev->priv_flags &= ~IFF_BONDING;
1860         dev_close(slave_dev);
1861
1862 err_unset_master:
1863         bond_upper_dev_unlink(bond_dev, slave_dev);
1864
1865 err_restore_mac:
1866         if (!bond->params.fail_over_mac) {
1867                 /* XXX TODO - fom follow mode needs to change master's
1868                  * MAC if this slave's MAC is in use by the bond, or at
1869                  * least print a warning.
1870                  */
1871                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1872                 addr.sa_family = slave_dev->type;
1873                 dev_set_mac_address(slave_dev, &addr);
1874         }
1875
1876 err_restore_mtu:
1877         dev_set_mtu(slave_dev, new_slave->original_mtu);
1878
1879 err_free:
1880         kfree(new_slave);
1881
1882 err_undo_flags:
1883         bond_compute_features(bond);
1884         /* Enslave of first slave has failed and we need to fix master's mac */
1885         if (bond->slave_cnt == 0 &&
1886             ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1887                 eth_hw_addr_random(bond_dev);
1888
1889         return res;
1890 }
1891
1892 /*
1893  * Try to release the slave device <slave> from the bond device <master>
1894  * It is legal to access curr_active_slave without a lock because all the function
1895  * is write-locked. If "all" is true it means that the function is being called
1896  * while destroying a bond interface and all slaves are being released.
1897  *
1898  * The rules for slave state should be:
1899  *   for Active/Backup:
1900  *     Active stays on all backups go down
1901  *   for Bonded connections:
1902  *     The first up interface should be left on and all others downed.
1903  */
1904 static int __bond_release_one(struct net_device *bond_dev,
1905                               struct net_device *slave_dev,
1906                               bool all)
1907 {
1908         struct bonding *bond = netdev_priv(bond_dev);
1909         struct slave *slave, *oldcurrent;
1910         struct sockaddr addr;
1911         netdev_features_t old_features = bond_dev->features;
1912
1913         /* slave is not a slave or master is not master of this slave */
1914         if (!(slave_dev->flags & IFF_SLAVE) ||
1915             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1916                 pr_err("%s: Error: cannot release %s.\n",
1917                        bond_dev->name, slave_dev->name);
1918                 return -EINVAL;
1919         }
1920
1921         block_netpoll_tx();
1922         write_lock_bh(&bond->lock);
1923
1924         slave = bond_get_slave_by_dev(bond, slave_dev);
1925         if (!slave) {
1926                 /* not a slave of this bond */
1927                 pr_info("%s: %s not enslaved\n",
1928                         bond_dev->name, slave_dev->name);
1929                 write_unlock_bh(&bond->lock);
1930                 unblock_netpoll_tx();
1931                 return -EINVAL;
1932         }
1933
1934         write_unlock_bh(&bond->lock);
1935         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1936          * for this slave anymore.
1937          */
1938         netdev_rx_handler_unregister(slave_dev);
1939         write_lock_bh(&bond->lock);
1940
1941         if (!all && !bond->params.fail_over_mac) {
1942                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1943                     bond->slave_cnt > 1)
1944                         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",
1945                                    bond_dev->name, slave_dev->name,
1946                                    slave->perm_hwaddr,
1947                                    bond_dev->name, slave_dev->name);
1948         }
1949
1950         /* Inform AD package of unbinding of slave. */
1951         if (bond->params.mode == BOND_MODE_8023AD) {
1952                 /* must be called before the slave is
1953                  * detached from the list
1954                  */
1955                 bond_3ad_unbind_slave(slave);
1956         }
1957
1958         pr_info("%s: releasing %s interface %s\n",
1959                 bond_dev->name,
1960                 bond_is_active_slave(slave) ? "active" : "backup",
1961                 slave_dev->name);
1962
1963         oldcurrent = bond->curr_active_slave;
1964
1965         bond->current_arp_slave = NULL;
1966
1967         /* release the slave from its bond */
1968         bond_detach_slave(bond, slave);
1969
1970         if (bond->primary_slave == slave)
1971                 bond->primary_slave = NULL;
1972
1973         if (oldcurrent == slave)
1974                 bond_change_active_slave(bond, NULL);
1975
1976         if (bond_is_lb(bond)) {
1977                 /* Must be called only after the slave has been
1978                  * detached from the list and the curr_active_slave
1979                  * has been cleared (if our_slave == old_current),
1980                  * but before a new active slave is selected.
1981                  */
1982                 write_unlock_bh(&bond->lock);
1983                 bond_alb_deinit_slave(bond, slave);
1984                 write_lock_bh(&bond->lock);
1985         }
1986
1987         if (all) {
1988                 bond->curr_active_slave = NULL;
1989         } else if (oldcurrent == slave) {
1990                 /*
1991                  * Note that we hold RTNL over this sequence, so there
1992                  * is no concern that another slave add/remove event
1993                  * will interfere.
1994                  */
1995                 write_unlock_bh(&bond->lock);
1996                 read_lock(&bond->lock);
1997                 write_lock_bh(&bond->curr_slave_lock);
1998
1999                 bond_select_active_slave(bond);
2000
2001                 write_unlock_bh(&bond->curr_slave_lock);
2002                 read_unlock(&bond->lock);
2003                 write_lock_bh(&bond->lock);
2004         }
2005
2006         if (bond->slave_cnt == 0) {
2007                 bond_set_carrier(bond);
2008                 eth_hw_addr_random(bond_dev);
2009
2010                 if (bond_vlan_used(bond)) {
2011                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2012                                    bond_dev->name, bond_dev->name);
2013                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2014                                    bond_dev->name);
2015                 }
2016         }
2017
2018         write_unlock_bh(&bond->lock);
2019         unblock_netpoll_tx();
2020
2021         if (bond->slave_cnt == 0) {
2022                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2023                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2024         }
2025
2026         bond_compute_features(bond);
2027         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2028             (old_features & NETIF_F_VLAN_CHALLENGED))
2029                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2030                         bond_dev->name, slave_dev->name, bond_dev->name);
2031
2032         /* must do this from outside any spinlocks */
2033         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2034
2035         bond_del_vlans_from_slave(bond, slave_dev);
2036
2037         /* If the mode USES_PRIMARY, then this cases was handled above by
2038          * bond_change_active_slave(..., NULL)
2039          */
2040         if (!USES_PRIMARY(bond->params.mode)) {
2041                 /* unset promiscuity level from slave */
2042                 if (bond_dev->flags & IFF_PROMISC)
2043                         dev_set_promiscuity(slave_dev, -1);
2044
2045                 /* unset allmulti level from slave */
2046                 if (bond_dev->flags & IFF_ALLMULTI)
2047                         dev_set_allmulti(slave_dev, -1);
2048
2049                 bond_hw_addr_flush(bond_dev, slave_dev);
2050         }
2051
2052         bond_upper_dev_unlink(bond_dev, slave_dev);
2053
2054         slave_disable_netpoll(slave);
2055
2056         /* close slave before restoring its mac address */
2057         dev_close(slave_dev);
2058
2059         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2060                 /* restore original ("permanent") mac address */
2061                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2062                 addr.sa_family = slave_dev->type;
2063                 dev_set_mac_address(slave_dev, &addr);
2064         }
2065
2066         dev_set_mtu(slave_dev, slave->original_mtu);
2067
2068         slave_dev->priv_flags &= ~IFF_BONDING;
2069
2070         kfree(slave);
2071
2072         return 0;  /* deletion OK */
2073 }
2074
2075 /* A wrapper used because of ndo_del_link */
2076 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2077 {
2078         return __bond_release_one(bond_dev, slave_dev, false);
2079 }
2080
2081 /*
2082 * First release a slave and then destroy the bond if no more slaves are left.
2083 * Must be under rtnl_lock when this function is called.
2084 */
2085 static int  bond_release_and_destroy(struct net_device *bond_dev,
2086                                      struct net_device *slave_dev)
2087 {
2088         struct bonding *bond = netdev_priv(bond_dev);
2089         int ret;
2090
2091         ret = bond_release(bond_dev, slave_dev);
2092         if ((ret == 0) && (bond->slave_cnt == 0)) {
2093                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2094                 pr_info("%s: destroying bond %s.\n",
2095                         bond_dev->name, bond_dev->name);
2096                 unregister_netdevice(bond_dev);
2097         }
2098         return ret;
2099 }
2100
2101 /*
2102  * This function changes the active slave to slave <slave_dev>.
2103  * It returns -EINVAL in the following cases.
2104  *  - <slave_dev> is not found in the list.
2105  *  - There is not active slave now.
2106  *  - <slave_dev> is already active.
2107  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2108  *  - <slave_dev> is not running.
2109  * In these cases, this function does nothing.
2110  * In the other cases, current_slave pointer is changed and 0 is returned.
2111  */
2112 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2113 {
2114         struct bonding *bond = netdev_priv(bond_dev);
2115         struct slave *old_active = NULL;
2116         struct slave *new_active = NULL;
2117         int res = 0;
2118
2119         if (!USES_PRIMARY(bond->params.mode))
2120                 return -EINVAL;
2121
2122         /* Verify that bond_dev is indeed the master of slave_dev */
2123         if (!(slave_dev->flags & IFF_SLAVE) ||
2124             !netdev_has_upper_dev(slave_dev, bond_dev))
2125                 return -EINVAL;
2126
2127         read_lock(&bond->lock);
2128
2129         read_lock(&bond->curr_slave_lock);
2130         old_active = bond->curr_active_slave;
2131         read_unlock(&bond->curr_slave_lock);
2132
2133         new_active = bond_get_slave_by_dev(bond, slave_dev);
2134
2135         /*
2136          * Changing to the current active: do nothing; return success.
2137          */
2138         if (new_active && (new_active == old_active)) {
2139                 read_unlock(&bond->lock);
2140                 return 0;
2141         }
2142
2143         if ((new_active) &&
2144             (old_active) &&
2145             (new_active->link == BOND_LINK_UP) &&
2146             IS_UP(new_active->dev)) {
2147                 block_netpoll_tx();
2148                 write_lock_bh(&bond->curr_slave_lock);
2149                 bond_change_active_slave(bond, new_active);
2150                 write_unlock_bh(&bond->curr_slave_lock);
2151                 unblock_netpoll_tx();
2152         } else
2153                 res = -EINVAL;
2154
2155         read_unlock(&bond->lock);
2156
2157         return res;
2158 }
2159
2160 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2161 {
2162         struct bonding *bond = netdev_priv(bond_dev);
2163
2164         info->bond_mode = bond->params.mode;
2165         info->miimon = bond->params.miimon;
2166
2167         read_lock(&bond->lock);
2168         info->num_slaves = bond->slave_cnt;
2169         read_unlock(&bond->lock);
2170
2171         return 0;
2172 }
2173
2174 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2175 {
2176         struct bonding *bond = netdev_priv(bond_dev);
2177         struct slave *slave;
2178         int i, res = -ENODEV;
2179
2180         read_lock(&bond->lock);
2181
2182         bond_for_each_slave(bond, slave, i) {
2183                 if (i == (int)info->slave_id) {
2184                         res = 0;
2185                         strcpy(info->slave_name, slave->dev->name);
2186                         info->link = slave->link;
2187                         info->state = bond_slave_state(slave);
2188                         info->link_failure_count = slave->link_failure_count;
2189                         break;
2190                 }
2191         }
2192
2193         read_unlock(&bond->lock);
2194
2195         return res;
2196 }
2197
2198 /*-------------------------------- Monitoring -------------------------------*/
2199
2200
2201 static int bond_miimon_inspect(struct bonding *bond)
2202 {
2203         struct slave *slave;
2204         int i, link_state, commit = 0;
2205         bool ignore_updelay;
2206
2207         ignore_updelay = !bond->curr_active_slave ? true : false;
2208
2209         bond_for_each_slave(bond, slave, i) {
2210                 slave->new_link = BOND_LINK_NOCHANGE;
2211
2212                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2213
2214                 switch (slave->link) {
2215                 case BOND_LINK_UP:
2216                         if (link_state)
2217                                 continue;
2218
2219                         slave->link = BOND_LINK_FAIL;
2220                         slave->delay = bond->params.downdelay;
2221                         if (slave->delay) {
2222                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2223                                         bond->dev->name,
2224                                         (bond->params.mode ==
2225                                          BOND_MODE_ACTIVEBACKUP) ?
2226                                         (bond_is_active_slave(slave) ?
2227                                          "active " : "backup ") : "",
2228                                         slave->dev->name,
2229                                         bond->params.downdelay * bond->params.miimon);
2230                         }
2231                         /*FALLTHRU*/
2232                 case BOND_LINK_FAIL:
2233                         if (link_state) {
2234                                 /*
2235                                  * recovered before downdelay expired
2236                                  */
2237                                 slave->link = BOND_LINK_UP;
2238                                 slave->jiffies = jiffies;
2239                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2240                                         bond->dev->name,
2241                                         (bond->params.downdelay - slave->delay) *
2242                                         bond->params.miimon,
2243                                         slave->dev->name);
2244                                 continue;
2245                         }
2246
2247                         if (slave->delay <= 0) {
2248                                 slave->new_link = BOND_LINK_DOWN;
2249                                 commit++;
2250                                 continue;
2251                         }
2252
2253                         slave->delay--;
2254                         break;
2255
2256                 case BOND_LINK_DOWN:
2257                         if (!link_state)
2258                                 continue;
2259
2260                         slave->link = BOND_LINK_BACK;
2261                         slave->delay = bond->params.updelay;
2262
2263                         if (slave->delay) {
2264                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2265                                         bond->dev->name, slave->dev->name,
2266                                         ignore_updelay ? 0 :
2267                                         bond->params.updelay *
2268                                         bond->params.miimon);
2269                         }
2270                         /*FALLTHRU*/
2271                 case BOND_LINK_BACK:
2272                         if (!link_state) {
2273                                 slave->link = BOND_LINK_DOWN;
2274                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2275                                         bond->dev->name,
2276                                         (bond->params.updelay - slave->delay) *
2277                                         bond->params.miimon,
2278                                         slave->dev->name);
2279
2280                                 continue;
2281                         }
2282
2283                         if (ignore_updelay)
2284                                 slave->delay = 0;
2285
2286                         if (slave->delay <= 0) {
2287                                 slave->new_link = BOND_LINK_UP;
2288                                 commit++;
2289                                 ignore_updelay = false;
2290                                 continue;
2291                         }
2292
2293                         slave->delay--;
2294                         break;
2295                 }
2296         }
2297
2298         return commit;
2299 }
2300
2301 static void bond_miimon_commit(struct bonding *bond)
2302 {
2303         struct slave *slave;
2304         int i;
2305
2306         bond_for_each_slave(bond, slave, i) {
2307                 switch (slave->new_link) {
2308                 case BOND_LINK_NOCHANGE:
2309                         continue;
2310
2311                 case BOND_LINK_UP:
2312                         slave->link = BOND_LINK_UP;
2313                         slave->jiffies = jiffies;
2314
2315                         if (bond->params.mode == BOND_MODE_8023AD) {
2316                                 /* prevent it from being the active one */
2317                                 bond_set_backup_slave(slave);
2318                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2319                                 /* make it immediately active */
2320                                 bond_set_active_slave(slave);
2321                         } else if (slave != bond->primary_slave) {
2322                                 /* prevent it from being the active one */
2323                                 bond_set_backup_slave(slave);
2324                         }
2325
2326                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2327                                 bond->dev->name, slave->dev->name,
2328                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2329                                 slave->duplex ? "full" : "half");
2330
2331                         /* notify ad that the link status has changed */
2332                         if (bond->params.mode == BOND_MODE_8023AD)
2333                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2334
2335                         if (bond_is_lb(bond))
2336                                 bond_alb_handle_link_change(bond, slave,
2337                                                             BOND_LINK_UP);
2338
2339                         if (!bond->curr_active_slave ||
2340                             (slave == bond->primary_slave))
2341                                 goto do_failover;
2342
2343                         continue;
2344
2345                 case BOND_LINK_DOWN:
2346                         if (slave->link_failure_count < UINT_MAX)
2347                                 slave->link_failure_count++;
2348
2349                         slave->link = BOND_LINK_DOWN;
2350
2351                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2352                             bond->params.mode == BOND_MODE_8023AD)
2353                                 bond_set_slave_inactive_flags(slave);
2354
2355                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2356                                 bond->dev->name, slave->dev->name);
2357
2358                         if (bond->params.mode == BOND_MODE_8023AD)
2359                                 bond_3ad_handle_link_change(slave,
2360                                                             BOND_LINK_DOWN);
2361
2362                         if (bond_is_lb(bond))
2363                                 bond_alb_handle_link_change(bond, slave,
2364                                                             BOND_LINK_DOWN);
2365
2366                         if (slave == bond->curr_active_slave)
2367                                 goto do_failover;
2368
2369                         continue;
2370
2371                 default:
2372                         pr_err("%s: invalid new link %d on slave %s\n",
2373                                bond->dev->name, slave->new_link,
2374                                slave->dev->name);
2375                         slave->new_link = BOND_LINK_NOCHANGE;
2376
2377                         continue;
2378                 }
2379
2380 do_failover:
2381                 ASSERT_RTNL();
2382                 block_netpoll_tx();
2383                 write_lock_bh(&bond->curr_slave_lock);
2384                 bond_select_active_slave(bond);
2385                 write_unlock_bh(&bond->curr_slave_lock);
2386                 unblock_netpoll_tx();
2387         }
2388
2389         bond_set_carrier(bond);
2390 }
2391
2392 /*
2393  * bond_mii_monitor
2394  *
2395  * Really a wrapper that splits the mii monitor into two phases: an
2396  * inspection, then (if inspection indicates something needs to be done)
2397  * an acquisition of appropriate locks followed by a commit phase to
2398  * implement whatever link state changes are indicated.
2399  */
2400 void bond_mii_monitor(struct work_struct *work)
2401 {
2402         struct bonding *bond = container_of(work, struct bonding,
2403                                             mii_work.work);
2404         bool should_notify_peers = false;
2405         unsigned long delay;
2406
2407         read_lock(&bond->lock);
2408
2409         delay = msecs_to_jiffies(bond->params.miimon);
2410
2411         if (bond->slave_cnt == 0)
2412                 goto re_arm;
2413
2414         should_notify_peers = bond_should_notify_peers(bond);
2415
2416         if (bond_miimon_inspect(bond)) {
2417                 read_unlock(&bond->lock);
2418
2419                 /* Race avoidance with bond_close cancel of workqueue */
2420                 if (!rtnl_trylock()) {
2421                         read_lock(&bond->lock);
2422                         delay = 1;
2423                         should_notify_peers = false;
2424                         goto re_arm;
2425                 }
2426
2427                 read_lock(&bond->lock);
2428
2429                 bond_miimon_commit(bond);
2430
2431                 read_unlock(&bond->lock);
2432                 rtnl_unlock();  /* might sleep, hold no other locks */
2433                 read_lock(&bond->lock);
2434         }
2435
2436 re_arm:
2437         if (bond->params.miimon)
2438                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2439
2440         read_unlock(&bond->lock);
2441
2442         if (should_notify_peers) {
2443                 if (!rtnl_trylock()) {
2444                         read_lock(&bond->lock);
2445                         bond->send_peer_notif++;
2446                         read_unlock(&bond->lock);
2447                         return;
2448                 }
2449                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2450                 rtnl_unlock();
2451         }
2452 }
2453
2454 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2455 {
2456         struct vlan_entry *vlan;
2457         struct net_device *vlan_dev;
2458
2459         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2460                 return 1;
2461
2462         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2463                 rcu_read_lock();
2464                 vlan_dev = __vlan_find_dev_deep(bond->dev, htons(ETH_P_8021Q),
2465                                                 vlan->vlan_id);
2466                 rcu_read_unlock();
2467                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2468                         return 1;
2469         }
2470
2471         return 0;
2472 }
2473
2474 /*
2475  * We go to the (large) trouble of VLAN tagging ARP frames because
2476  * switches in VLAN mode (especially if ports are configured as
2477  * "native" to a VLAN) might not pass non-tagged frames.
2478  */
2479 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2480 {
2481         struct sk_buff *skb;
2482
2483         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2484                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2485
2486         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2487                          NULL, slave_dev->dev_addr, NULL);
2488
2489         if (!skb) {
2490                 pr_err("ARP packet allocation failed\n");
2491                 return;
2492         }
2493         if (vlan_id) {
2494                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2495                 if (!skb) {
2496                         pr_err("failed to insert VLAN tag\n");
2497                         return;
2498                 }
2499         }
2500         arp_xmit(skb);
2501 }
2502
2503
2504 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2505 {
2506         int i, vlan_id;
2507         __be32 *targets = bond->params.arp_targets;
2508         struct vlan_entry *vlan;
2509         struct net_device *vlan_dev = NULL;
2510         struct rtable *rt;
2511
2512         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2513                 __be32 addr;
2514                 if (!targets[i])
2515                         break;
2516                 pr_debug("basa: target %pI4\n", &targets[i]);
2517                 if (!bond_vlan_used(bond)) {
2518                         pr_debug("basa: empty vlan: arp_send\n");
2519                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2520                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2521                                       addr, 0);
2522                         continue;
2523                 }
2524
2525                 /*
2526                  * If VLANs are configured, we do a route lookup to
2527                  * determine which VLAN interface would be used, so we
2528                  * can tag the ARP with the proper VLAN tag.
2529                  */
2530                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2531                                      RTO_ONLINK, 0);
2532                 if (IS_ERR(rt)) {
2533                         if (net_ratelimit()) {
2534                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2535                                            bond->dev->name, &targets[i]);
2536                         }
2537                         continue;
2538                 }
2539
2540                 /*
2541                  * This target is not on a VLAN
2542                  */
2543                 if (rt->dst.dev == bond->dev) {
2544                         ip_rt_put(rt);
2545                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2546                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2547                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2548                                       addr, 0);
2549                         continue;
2550                 }
2551
2552                 vlan_id = 0;
2553                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2554                         rcu_read_lock();
2555                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2556                                                         htons(ETH_P_8021Q),
2557                                                         vlan->vlan_id);
2558                         rcu_read_unlock();
2559                         if (vlan_dev == rt->dst.dev) {
2560                                 vlan_id = vlan->vlan_id;
2561                                 pr_debug("basa: vlan match on %s %d\n",
2562                                        vlan_dev->name, vlan_id);
2563                                 break;
2564                         }
2565                 }
2566
2567                 if (vlan_id && vlan_dev) {
2568                         ip_rt_put(rt);
2569                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2570                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2571                                       addr, vlan_id);
2572                         continue;
2573                 }
2574
2575                 if (net_ratelimit()) {
2576                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2577                                    bond->dev->name, &targets[i],
2578                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2579                 }
2580                 ip_rt_put(rt);
2581         }
2582 }
2583
2584 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2585 {
2586         int i;
2587
2588         if (!sip || !bond_has_this_ip(bond, tip)) {
2589                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2590                 return;
2591         }
2592
2593         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2594         if (i == -1) {
2595                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2596                 return;
2597         }
2598         slave->last_arp_rx = jiffies;
2599         slave->target_last_arp_rx[i] = jiffies;
2600 }
2601
2602 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2603                         struct slave *slave)
2604 {
2605         struct arphdr *arp = (struct arphdr *)skb->data;
2606         unsigned char *arp_ptr;
2607         __be32 sip, tip;
2608         int alen;
2609
2610         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2611                 return RX_HANDLER_ANOTHER;
2612
2613         read_lock(&bond->lock);
2614
2615         if (!slave_do_arp_validate(bond, slave))
2616                 goto out_unlock;
2617
2618         alen = arp_hdr_len(bond->dev);
2619
2620         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2621                  bond->dev->name, skb->dev->name);
2622
2623         if (alen > skb_headlen(skb)) {
2624                 arp = kmalloc(alen, GFP_ATOMIC);
2625                 if (!arp)
2626                         goto out_unlock;
2627                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2628                         goto out_unlock;
2629         }
2630
2631         if (arp->ar_hln != bond->dev->addr_len ||
2632             skb->pkt_type == PACKET_OTHERHOST ||
2633             skb->pkt_type == PACKET_LOOPBACK ||
2634             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2635             arp->ar_pro != htons(ETH_P_IP) ||
2636             arp->ar_pln != 4)
2637                 goto out_unlock;
2638
2639         arp_ptr = (unsigned char *)(arp + 1);
2640         arp_ptr += bond->dev->addr_len;
2641         memcpy(&sip, arp_ptr, 4);
2642         arp_ptr += 4 + bond->dev->addr_len;
2643         memcpy(&tip, arp_ptr, 4);
2644
2645         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2646                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2647                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2648                  &sip, &tip);
2649
2650         /*
2651          * Backup slaves won't see the ARP reply, but do come through
2652          * here for each ARP probe (so we swap the sip/tip to validate
2653          * the probe).  In a "redundant switch, common router" type of
2654          * configuration, the ARP probe will (hopefully) travel from
2655          * the active, through one switch, the router, then the other
2656          * switch before reaching the backup.
2657          *
2658          * We 'trust' the arp requests if there is an active slave and
2659          * it received valid arp reply(s) after it became active. This
2660          * is done to avoid endless looping when we can't reach the
2661          * arp_ip_target and fool ourselves with our own arp requests.
2662          */
2663         if (bond_is_active_slave(slave))
2664                 bond_validate_arp(bond, slave, sip, tip);
2665         else if (bond->curr_active_slave &&
2666                  time_after(slave_last_rx(bond, bond->curr_active_slave),
2667                             bond->curr_active_slave->jiffies))
2668                 bond_validate_arp(bond, slave, tip, sip);
2669
2670 out_unlock:
2671         read_unlock(&bond->lock);
2672         if (arp != (struct arphdr *)skb->data)
2673                 kfree(arp);
2674         return RX_HANDLER_ANOTHER;
2675 }
2676
2677 /*
2678  * this function is called regularly to monitor each slave's link
2679  * ensuring that traffic is being sent and received when arp monitoring
2680  * is used in load-balancing mode. if the adapter has been dormant, then an
2681  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2682  * arp monitoring in active backup mode.
2683  */
2684 void bond_loadbalance_arp_mon(struct work_struct *work)
2685 {
2686         struct bonding *bond = container_of(work, struct bonding,
2687                                             arp_work.work);
2688         struct slave *slave, *oldcurrent;
2689         int do_failover = 0;
2690         int delta_in_ticks, extra_ticks;
2691         int i;
2692
2693         read_lock(&bond->lock);
2694
2695         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2696         extra_ticks = delta_in_ticks / 2;
2697
2698         if (bond->slave_cnt == 0)
2699                 goto re_arm;
2700
2701         read_lock(&bond->curr_slave_lock);
2702         oldcurrent = bond->curr_active_slave;
2703         read_unlock(&bond->curr_slave_lock);
2704
2705         /* see if any of the previous devices are up now (i.e. they have
2706          * xmt and rcv traffic). the curr_active_slave does not come into
2707          * the picture unless it is null. also, slave->jiffies is not needed
2708          * here because we send an arp on each slave and give a slave as
2709          * long as it needs to get the tx/rx within the delta.
2710          * TODO: what about up/down delay in arp mode? it wasn't here before
2711          *       so it can wait
2712          */
2713         bond_for_each_slave(bond, slave, i) {
2714                 unsigned long trans_start = dev_trans_start(slave->dev);
2715
2716                 if (slave->link != BOND_LINK_UP) {
2717                         if (time_in_range(jiffies,
2718                                 trans_start - delta_in_ticks,
2719                                 trans_start + delta_in_ticks + extra_ticks) &&
2720                             time_in_range(jiffies,
2721                                 slave->dev->last_rx - delta_in_ticks,
2722                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2723
2724                                 slave->link  = BOND_LINK_UP;
2725                                 bond_set_active_slave(slave);
2726
2727                                 /* primary_slave has no meaning in round-robin
2728                                  * mode. the window of a slave being up and
2729                                  * curr_active_slave being null after enslaving
2730                                  * is closed.
2731                                  */
2732                                 if (!oldcurrent) {
2733                                         pr_info("%s: link status definitely up for interface %s, ",
2734                                                 bond->dev->name,
2735                                                 slave->dev->name);
2736                                         do_failover = 1;
2737                                 } else {
2738                                         pr_info("%s: interface %s is now up\n",
2739                                                 bond->dev->name,
2740                                                 slave->dev->name);
2741                                 }
2742                         }
2743                 } else {
2744                         /* slave->link == BOND_LINK_UP */
2745
2746                         /* not all switches will respond to an arp request
2747                          * when the source ip is 0, so don't take the link down
2748                          * if we don't know our ip yet
2749                          */
2750                         if (!time_in_range(jiffies,
2751                                 trans_start - delta_in_ticks,
2752                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2753                             !time_in_range(jiffies,
2754                                 slave->dev->last_rx - delta_in_ticks,
2755                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2756
2757                                 slave->link  = BOND_LINK_DOWN;
2758                                 bond_set_backup_slave(slave);
2759
2760                                 if (slave->link_failure_count < UINT_MAX)
2761                                         slave->link_failure_count++;
2762
2763                                 pr_info("%s: interface %s is now down.\n",
2764                                         bond->dev->name,
2765                                         slave->dev->name);
2766
2767                                 if (slave == oldcurrent)
2768                                         do_failover = 1;
2769                         }
2770                 }
2771
2772                 /* note: if switch is in round-robin mode, all links
2773                  * must tx arp to ensure all links rx an arp - otherwise
2774                  * links may oscillate or not come up at all; if switch is
2775                  * in something like xor mode, there is nothing we can
2776                  * do - all replies will be rx'ed on same link causing slaves
2777                  * to be unstable during low/no traffic periods
2778                  */
2779                 if (IS_UP(slave->dev))
2780                         bond_arp_send_all(bond, slave);
2781         }
2782
2783         if (do_failover) {
2784                 block_netpoll_tx();
2785                 write_lock_bh(&bond->curr_slave_lock);
2786
2787                 bond_select_active_slave(bond);
2788
2789                 write_unlock_bh(&bond->curr_slave_lock);
2790                 unblock_netpoll_tx();
2791         }
2792
2793 re_arm:
2794         if (bond->params.arp_interval)
2795                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2796
2797         read_unlock(&bond->lock);
2798 }
2799
2800 /*
2801  * Called to inspect slaves for active-backup mode ARP monitor link state
2802  * changes.  Sets new_link in slaves to specify what action should take
2803  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2804  * to link states must be committed.
2805  *
2806  * Called with bond->lock held for read.
2807  */
2808 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2809 {
2810         struct slave *slave;
2811         int i, commit = 0;
2812         unsigned long trans_start;
2813         int extra_ticks;
2814
2815         /* All the time comparisons below need some extra time. Otherwise, on
2816          * fast networks the ARP probe/reply may arrive within the same jiffy
2817          * as it was sent.  Then, the next time the ARP monitor is run, one
2818          * arp_interval will already have passed in the comparisons.
2819          */
2820         extra_ticks = delta_in_ticks / 2;
2821
2822         bond_for_each_slave(bond, slave, i) {
2823                 slave->new_link = BOND_LINK_NOCHANGE;
2824
2825                 if (slave->link != BOND_LINK_UP) {
2826                         if (time_in_range(jiffies,
2827                                 slave_last_rx(bond, slave) - delta_in_ticks,
2828                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2829
2830                                 slave->new_link = BOND_LINK_UP;
2831                                 commit++;
2832                         }
2833
2834                         continue;
2835                 }
2836
2837                 /*
2838                  * Give slaves 2*delta after being enslaved or made
2839                  * active.  This avoids bouncing, as the last receive
2840                  * times need a full ARP monitor cycle to be updated.
2841                  */
2842                 if (time_in_range(jiffies,
2843                                   slave->jiffies - delta_in_ticks,
2844                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2845                         continue;
2846
2847                 /*
2848                  * Backup slave is down if:
2849                  * - No current_arp_slave AND
2850                  * - more than 3*delta since last receive AND
2851                  * - the bond has an IP address
2852                  *
2853                  * Note: a non-null current_arp_slave indicates
2854                  * the curr_active_slave went down and we are
2855                  * searching for a new one; under this condition
2856                  * we only take the curr_active_slave down - this
2857                  * gives each slave a chance to tx/rx traffic
2858                  * before being taken out
2859                  */
2860                 if (!bond_is_active_slave(slave) &&
2861                     !bond->current_arp_slave &&
2862                     !time_in_range(jiffies,
2863                         slave_last_rx(bond, slave) - delta_in_ticks,
2864                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2865
2866                         slave->new_link = BOND_LINK_DOWN;
2867                         commit++;
2868                 }
2869
2870                 /*
2871                  * Active slave is down if:
2872                  * - more than 2*delta since transmitting OR
2873                  * - (more than 2*delta since receive AND
2874                  *    the bond has an IP address)
2875                  */
2876                 trans_start = dev_trans_start(slave->dev);
2877                 if (bond_is_active_slave(slave) &&
2878                     (!time_in_range(jiffies,
2879                         trans_start - delta_in_ticks,
2880                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
2881                      !time_in_range(jiffies,
2882                         slave_last_rx(bond, slave) - delta_in_ticks,
2883                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2884
2885                         slave->new_link = BOND_LINK_DOWN;
2886                         commit++;
2887                 }
2888         }
2889
2890         return commit;
2891 }
2892
2893 /*
2894  * Called to commit link state changes noted by inspection step of
2895  * active-backup mode ARP monitor.
2896  *
2897  * Called with RTNL and bond->lock for read.
2898  */
2899 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2900 {
2901         struct slave *slave;
2902         int i;
2903         unsigned long trans_start;
2904
2905         bond_for_each_slave(bond, slave, i) {
2906                 switch (slave->new_link) {
2907                 case BOND_LINK_NOCHANGE:
2908                         continue;
2909
2910                 case BOND_LINK_UP:
2911                         trans_start = dev_trans_start(slave->dev);
2912                         if ((!bond->curr_active_slave &&
2913                              time_in_range(jiffies,
2914                                            trans_start - delta_in_ticks,
2915                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2916                             bond->curr_active_slave != slave) {
2917                                 slave->link = BOND_LINK_UP;
2918                                 if (bond->current_arp_slave) {
2919                                         bond_set_slave_inactive_flags(
2920                                                 bond->current_arp_slave);
2921                                         bond->current_arp_slave = NULL;
2922                                 }
2923
2924                                 pr_info("%s: link status definitely up for interface %s.\n",
2925                                         bond->dev->name, slave->dev->name);
2926
2927                                 if (!bond->curr_active_slave ||
2928                                     (slave == bond->primary_slave))
2929                                         goto do_failover;
2930
2931                         }
2932
2933                         continue;
2934
2935                 case BOND_LINK_DOWN:
2936                         if (slave->link_failure_count < UINT_MAX)
2937                                 slave->link_failure_count++;
2938
2939                         slave->link = BOND_LINK_DOWN;
2940                         bond_set_slave_inactive_flags(slave);
2941
2942                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2943                                 bond->dev->name, slave->dev->name);
2944
2945                         if (slave == bond->curr_active_slave) {
2946                                 bond->current_arp_slave = NULL;
2947                                 goto do_failover;
2948                         }
2949
2950                         continue;
2951
2952                 default:
2953                         pr_err("%s: impossible: new_link %d on slave %s\n",
2954                                bond->dev->name, slave->new_link,
2955                                slave->dev->name);
2956                         continue;
2957                 }
2958
2959 do_failover:
2960                 ASSERT_RTNL();
2961                 block_netpoll_tx();
2962                 write_lock_bh(&bond->curr_slave_lock);
2963                 bond_select_active_slave(bond);
2964                 write_unlock_bh(&bond->curr_slave_lock);
2965                 unblock_netpoll_tx();
2966         }
2967
2968         bond_set_carrier(bond);
2969 }
2970
2971 /*
2972  * Send ARP probes for active-backup mode ARP monitor.
2973  *
2974  * Called with bond->lock held for read.
2975  */
2976 static void bond_ab_arp_probe(struct bonding *bond)
2977 {
2978         struct slave *slave;
2979         int i;
2980
2981         read_lock(&bond->curr_slave_lock);
2982
2983         if (bond->current_arp_slave && bond->curr_active_slave)
2984                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2985                         bond->current_arp_slave->dev->name,
2986                         bond->curr_active_slave->dev->name);
2987
2988         if (bond->curr_active_slave) {
2989                 bond_arp_send_all(bond, bond->curr_active_slave);
2990                 read_unlock(&bond->curr_slave_lock);
2991                 return;
2992         }
2993
2994         read_unlock(&bond->curr_slave_lock);
2995
2996         /* if we don't have a curr_active_slave, search for the next available
2997          * backup slave from the current_arp_slave and make it the candidate
2998          * for becoming the curr_active_slave
2999          */
3000
3001         if (!bond->current_arp_slave) {
3002                 bond->current_arp_slave = bond->first_slave;
3003                 if (!bond->current_arp_slave)
3004                         return;
3005         }
3006
3007         bond_set_slave_inactive_flags(bond->current_arp_slave);
3008
3009         /* search for next candidate */
3010         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3011                 if (IS_UP(slave->dev)) {
3012                         slave->link = BOND_LINK_BACK;
3013                         bond_set_slave_active_flags(slave);
3014                         bond_arp_send_all(bond, slave);
3015                         slave->jiffies = jiffies;
3016                         bond->current_arp_slave = slave;
3017                         break;
3018                 }
3019
3020                 /* if the link state is up at this point, we
3021                  * mark it down - this can happen if we have
3022                  * simultaneous link failures and
3023                  * reselect_active_interface doesn't make this
3024                  * one the current slave so it is still marked
3025                  * up when it is actually down
3026                  */
3027                 if (slave->link == BOND_LINK_UP) {
3028                         slave->link = BOND_LINK_DOWN;
3029                         if (slave->link_failure_count < UINT_MAX)
3030                                 slave->link_failure_count++;
3031
3032                         bond_set_slave_inactive_flags(slave);
3033
3034                         pr_info("%s: backup interface %s is now down.\n",
3035                                 bond->dev->name, slave->dev->name);
3036                 }
3037         }
3038 }
3039
3040 void bond_activebackup_arp_mon(struct work_struct *work)
3041 {
3042         struct bonding *bond = container_of(work, struct bonding,
3043                                             arp_work.work);
3044         bool should_notify_peers = false;
3045         int delta_in_ticks;
3046
3047         read_lock(&bond->lock);
3048
3049         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3050
3051         if (bond->slave_cnt == 0)
3052                 goto re_arm;
3053
3054         should_notify_peers = bond_should_notify_peers(bond);
3055
3056         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3057                 read_unlock(&bond->lock);
3058
3059                 /* Race avoidance with bond_close flush of workqueue */
3060                 if (!rtnl_trylock()) {
3061                         read_lock(&bond->lock);
3062                         delta_in_ticks = 1;
3063                         should_notify_peers = false;
3064                         goto re_arm;
3065                 }
3066
3067                 read_lock(&bond->lock);
3068
3069                 bond_ab_arp_commit(bond, delta_in_ticks);
3070
3071                 read_unlock(&bond->lock);
3072                 rtnl_unlock();
3073                 read_lock(&bond->lock);
3074         }
3075
3076         bond_ab_arp_probe(bond);
3077
3078 re_arm:
3079         if (bond->params.arp_interval)
3080                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3081
3082         read_unlock(&bond->lock);
3083
3084         if (should_notify_peers) {
3085                 if (!rtnl_trylock()) {
3086                         read_lock(&bond->lock);
3087                         bond->send_peer_notif++;
3088                         read_unlock(&bond->lock);
3089                         return;
3090                 }
3091                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3092                 rtnl_unlock();
3093         }
3094 }
3095
3096 /*-------------------------- netdev event handling --------------------------*/
3097
3098 /*
3099  * Change device name
3100  */
3101 static int bond_event_changename(struct bonding *bond)
3102 {
3103         bond_remove_proc_entry(bond);
3104         bond_create_proc_entry(bond);
3105
3106         bond_debug_reregister(bond);
3107
3108         return NOTIFY_DONE;
3109 }
3110
3111 static int bond_master_netdev_event(unsigned long event,
3112                                     struct net_device *bond_dev)
3113 {
3114         struct bonding *event_bond = netdev_priv(bond_dev);
3115
3116         switch (event) {
3117         case NETDEV_CHANGENAME:
3118                 return bond_event_changename(event_bond);
3119         case NETDEV_UNREGISTER:
3120                 bond_remove_proc_entry(event_bond);
3121                 break;
3122         case NETDEV_REGISTER:
3123                 bond_create_proc_entry(event_bond);
3124                 break;
3125         default:
3126                 break;
3127         }
3128
3129         return NOTIFY_DONE;
3130 }
3131
3132 static int bond_slave_netdev_event(unsigned long event,
3133                                    struct net_device *slave_dev)
3134 {
3135         struct slave *slave = bond_slave_get_rtnl(slave_dev);
3136         struct bonding *bond;
3137         struct net_device *bond_dev;
3138         u32 old_speed;
3139         u8 old_duplex;
3140
3141         /* A netdev event can be generated while enslaving a device
3142          * before netdev_rx_handler_register is called in which case
3143          * slave will be NULL
3144          */
3145         if (!slave)
3146                 return NOTIFY_DONE;
3147         bond_dev = slave->bond->dev;
3148         bond = slave->bond;
3149
3150         switch (event) {
3151         case NETDEV_UNREGISTER:
3152                 if (bond_dev->type != ARPHRD_ETHER)
3153                         bond_release_and_destroy(bond_dev, slave_dev);
3154                 else
3155                         bond_release(bond_dev, slave_dev);
3156                 break;
3157         case NETDEV_UP:
3158         case NETDEV_CHANGE:
3159                 old_speed = slave->speed;
3160                 old_duplex = slave->duplex;
3161
3162                 bond_update_speed_duplex(slave);
3163
3164                 if (bond->params.mode == BOND_MODE_8023AD) {
3165                         if (old_speed != slave->speed)
3166                                 bond_3ad_adapter_speed_changed(slave);
3167                         if (old_duplex != slave->duplex)
3168                                 bond_3ad_adapter_duplex_changed(slave);
3169                 }
3170                 break;
3171         case NETDEV_DOWN:
3172                 /*
3173                  * ... Or is it this?
3174                  */
3175                 break;
3176         case NETDEV_CHANGEMTU:
3177                 /*
3178                  * TODO: Should slaves be allowed to
3179                  * independently alter their MTU?  For
3180                  * an active-backup bond, slaves need
3181                  * not be the same type of device, so
3182                  * MTUs may vary.  For other modes,
3183                  * slaves arguably should have the
3184                  * same MTUs. To do this, we'd need to
3185                  * take over the slave's change_mtu
3186                  * function for the duration of their
3187                  * servitude.
3188                  */
3189                 break;
3190         case NETDEV_CHANGENAME:
3191                 /*
3192                  * TODO: handle changing the primary's name
3193                  */
3194                 break;
3195         case NETDEV_FEAT_CHANGE:
3196                 bond_compute_features(bond);
3197                 break;
3198         case NETDEV_RESEND_IGMP:
3199                 /* Propagate to master device */
3200                 call_netdevice_notifiers(event, slave->bond->dev);
3201                 break;
3202         default:
3203                 break;
3204         }
3205
3206         return NOTIFY_DONE;
3207 }
3208
3209 /*
3210  * bond_netdev_event: handle netdev notifier chain events.
3211  *
3212  * This function receives events for the netdev chain.  The caller (an
3213  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3214  * locks for us to safely manipulate the slave devices (RTNL lock,
3215  * dev_probe_lock).
3216  */
3217 static int bond_netdev_event(struct notifier_block *this,
3218                              unsigned long event, void *ptr)
3219 {
3220         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3221
3222         pr_debug("event_dev: %s, event: %lx\n",
3223                  event_dev ? event_dev->name : "None",
3224                  event);
3225
3226         if (!(event_dev->priv_flags & IFF_BONDING))
3227                 return NOTIFY_DONE;
3228
3229         if (event_dev->flags & IFF_MASTER) {
3230                 pr_debug("IFF_MASTER\n");
3231                 return bond_master_netdev_event(event, event_dev);
3232         }
3233
3234         if (event_dev->flags & IFF_SLAVE) {
3235                 pr_debug("IFF_SLAVE\n");
3236                 return bond_slave_netdev_event(event, event_dev);
3237         }
3238
3239         return NOTIFY_DONE;
3240 }
3241
3242 static struct notifier_block bond_netdev_notifier = {
3243         .notifier_call = bond_netdev_event,
3244 };
3245
3246 /*---------------------------- Hashing Policies -----------------------------*/
3247
3248 /*
3249  * Hash for the output device based upon layer 2 data
3250  */
3251 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3252 {
3253         struct ethhdr *data = (struct ethhdr *)skb->data;
3254
3255         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3256                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3257
3258         return 0;
3259 }
3260
3261 /*
3262  * Hash for the output device based upon layer 2 and layer 3 data. If
3263  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3264  */
3265 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3266 {
3267         const struct ethhdr *data;
3268         const struct iphdr *iph;
3269         const struct ipv6hdr *ipv6h;
3270         u32 v6hash;
3271         const __be32 *s, *d;
3272
3273         if (skb->protocol == htons(ETH_P_IP) &&
3274             pskb_network_may_pull(skb, sizeof(*iph))) {
3275                 iph = ip_hdr(skb);
3276                 data = (struct ethhdr *)skb->data;
3277                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3278                         (data->h_dest[5] ^ data->h_source[5])) % count;
3279         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3280                    pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3281                 ipv6h = ipv6_hdr(skb);
3282                 data = (struct ethhdr *)skb->data;
3283                 s = &ipv6h->saddr.s6_addr32[0];
3284                 d = &ipv6h->daddr.s6_addr32[0];
3285                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3286                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3287                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3288         }
3289
3290         return bond_xmit_hash_policy_l2(skb, count);
3291 }
3292
3293 /*
3294  * Hash for the output device based upon layer 3 and layer 4 data. If
3295  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3296  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3297  */
3298 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3299 {
3300         u32 layer4_xor = 0;
3301         const struct iphdr *iph;
3302         const struct ipv6hdr *ipv6h;
3303         const __be32 *s, *d;
3304         const __be16 *l4 = NULL;
3305         __be16 _l4[2];
3306         int noff = skb_network_offset(skb);
3307         int poff;
3308
3309         if (skb->protocol == htons(ETH_P_IP) &&
3310             pskb_may_pull(skb, noff + sizeof(*iph))) {
3311                 iph = ip_hdr(skb);
3312                 poff = proto_ports_offset(iph->protocol);
3313
3314                 if (!ip_is_fragment(iph) && poff >= 0) {
3315                         l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3316                                                 sizeof(_l4), &_l4);
3317                         if (l4)
3318                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3319                 }
3320                 return (layer4_xor ^
3321                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3322         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3323                    pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3324                 ipv6h = ipv6_hdr(skb);
3325                 poff = proto_ports_offset(ipv6h->nexthdr);
3326                 if (poff >= 0) {
3327                         l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3328                                                 sizeof(_l4), &_l4);
3329                         if (l4)
3330                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3331                 }
3332                 s = &ipv6h->saddr.s6_addr32[0];
3333                 d = &ipv6h->daddr.s6_addr32[0];
3334                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3335                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3336                                (layer4_xor >> 8);
3337                 return layer4_xor % count;
3338         }
3339
3340         return bond_xmit_hash_policy_l2(skb, count);
3341 }
3342
3343 /*-------------------------- Device entry points ----------------------------*/
3344
3345 static void bond_work_init_all(struct bonding *bond)
3346 {
3347         INIT_DELAYED_WORK(&bond->mcast_work,
3348                           bond_resend_igmp_join_requests_delayed);
3349         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3350         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3351         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3352                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3353         else
3354                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3355         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3356 }
3357
3358 static void bond_work_cancel_all(struct bonding *bond)
3359 {
3360         cancel_delayed_work_sync(&bond->mii_work);
3361         cancel_delayed_work_sync(&bond->arp_work);
3362         cancel_delayed_work_sync(&bond->alb_work);
3363         cancel_delayed_work_sync(&bond->ad_work);
3364         cancel_delayed_work_sync(&bond->mcast_work);
3365 }
3366
3367 static int bond_open(struct net_device *bond_dev)
3368 {
3369         struct bonding *bond = netdev_priv(bond_dev);
3370         struct slave *slave;
3371         int i;
3372
3373         /* reset slave->backup and slave->inactive */
3374         read_lock(&bond->lock);
3375         if (bond->slave_cnt > 0) {
3376                 read_lock(&bond->curr_slave_lock);
3377                 bond_for_each_slave(bond, slave, i) {
3378                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3379                                 && (slave != bond->curr_active_slave)) {
3380                                 bond_set_slave_inactive_flags(slave);
3381                         } else {
3382                                 bond_set_slave_active_flags(slave);
3383                         }
3384                 }
3385                 read_unlock(&bond->curr_slave_lock);
3386         }
3387         read_unlock(&bond->lock);
3388
3389         bond_work_init_all(bond);
3390
3391         if (bond_is_lb(bond)) {
3392                 /* bond_alb_initialize must be called before the timer
3393                  * is started.
3394                  */
3395                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3396                         return -ENOMEM;
3397                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3398         }
3399
3400         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3401                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3402
3403         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3404                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3405                 if (bond->params.arp_validate)
3406                         bond->recv_probe = bond_arp_rcv;
3407         }
3408
3409         if (bond->params.mode == BOND_MODE_8023AD) {
3410                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3411                 /* register to receive LACPDUs */
3412                 bond->recv_probe = bond_3ad_lacpdu_recv;
3413                 bond_3ad_initiate_agg_selection(bond, 1);
3414         }
3415
3416         return 0;
3417 }
3418
3419 static int bond_close(struct net_device *bond_dev)
3420 {
3421         struct bonding *bond = netdev_priv(bond_dev);
3422
3423         write_lock_bh(&bond->lock);
3424         bond->send_peer_notif = 0;
3425         write_unlock_bh(&bond->lock);
3426
3427         bond_work_cancel_all(bond);
3428         if (bond_is_lb(bond)) {
3429                 /* Must be called only after all
3430                  * slaves have been released
3431                  */
3432                 bond_alb_deinitialize(bond);
3433         }
3434         bond->recv_probe = NULL;
3435
3436         return 0;
3437 }
3438
3439 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3440                                                 struct rtnl_link_stats64 *stats)
3441 {
3442         struct bonding *bond = netdev_priv(bond_dev);
3443         struct rtnl_link_stats64 temp;
3444         struct slave *slave;
3445         int i;
3446
3447         memset(stats, 0, sizeof(*stats));
3448
3449         read_lock_bh(&bond->lock);
3450
3451         bond_for_each_slave(bond, slave, i) {
3452                 const struct rtnl_link_stats64 *sstats =
3453                         dev_get_stats(slave->dev, &temp);
3454
3455                 stats->rx_packets += sstats->rx_packets;
3456                 stats->rx_bytes += sstats->rx_bytes;
3457                 stats->rx_errors += sstats->rx_errors;
3458                 stats->rx_dropped += sstats->rx_dropped;
3459
3460                 stats->tx_packets += sstats->tx_packets;
3461                 stats->tx_bytes += sstats->tx_bytes;
3462                 stats->tx_errors += sstats->tx_errors;
3463                 stats->tx_dropped += sstats->tx_dropped;
3464
3465                 stats->multicast += sstats->multicast;
3466                 stats->collisions += sstats->collisions;
3467
3468                 stats->rx_length_errors += sstats->rx_length_errors;
3469                 stats->rx_over_errors += sstats->rx_over_errors;
3470                 stats->rx_crc_errors += sstats->rx_crc_errors;
3471                 stats->rx_frame_errors += sstats->rx_frame_errors;
3472                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3473                 stats->rx_missed_errors += sstats->rx_missed_errors;
3474
3475                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3476                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3477                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3478                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3479                 stats->tx_window_errors += sstats->tx_window_errors;
3480         }
3481
3482         read_unlock_bh(&bond->lock);
3483
3484         return stats;
3485 }
3486
3487 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3488 {
3489         struct net_device *slave_dev = NULL;
3490         struct ifbond k_binfo;
3491         struct ifbond __user *u_binfo = NULL;
3492         struct ifslave k_sinfo;
3493         struct ifslave __user *u_sinfo = NULL;
3494         struct mii_ioctl_data *mii = NULL;
3495         struct net *net;
3496         int res = 0;
3497
3498         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3499
3500         switch (cmd) {
3501         case SIOCGMIIPHY:
3502                 mii = if_mii(ifr);
3503                 if (!mii)
3504                         return -EINVAL;
3505
3506                 mii->phy_id = 0;
3507                 /* Fall Through */
3508         case SIOCGMIIREG:
3509                 /*
3510                  * We do this again just in case we were called by SIOCGMIIREG
3511                  * instead of SIOCGMIIPHY.
3512                  */
3513                 mii = if_mii(ifr);
3514                 if (!mii)
3515                         return -EINVAL;
3516
3517
3518                 if (mii->reg_num == 1) {
3519                         struct bonding *bond = netdev_priv(bond_dev);
3520                         mii->val_out = 0;
3521                         read_lock(&bond->lock);
3522                         read_lock(&bond->curr_slave_lock);
3523                         if (netif_carrier_ok(bond->dev))
3524                                 mii->val_out = BMSR_LSTATUS;
3525
3526                         read_unlock(&bond->curr_slave_lock);
3527                         read_unlock(&bond->lock);
3528                 }
3529
3530                 return 0;
3531         case BOND_INFO_QUERY_OLD:
3532         case SIOCBONDINFOQUERY:
3533                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3534
3535                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3536                         return -EFAULT;
3537
3538                 res = bond_info_query(bond_dev, &k_binfo);
3539                 if (res == 0 &&
3540                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3541                         return -EFAULT;
3542
3543                 return res;
3544         case BOND_SLAVE_INFO_QUERY_OLD:
3545         case SIOCBONDSLAVEINFOQUERY:
3546                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3547
3548                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3549                         return -EFAULT;
3550
3551                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3552                 if (res == 0 &&
3553                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3554                         return -EFAULT;
3555
3556                 return res;
3557         default:
3558                 /* Go on */
3559                 break;
3560         }
3561
3562         net = dev_net(bond_dev);
3563
3564         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3565                 return -EPERM;
3566
3567         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3568
3569         pr_debug("slave_dev=%p:\n", slave_dev);
3570
3571         if (!slave_dev)
3572                 res = -ENODEV;
3573         else {
3574                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3575                 switch (cmd) {
3576                 case BOND_ENSLAVE_OLD:
3577                 case SIOCBONDENSLAVE:
3578                         res = bond_enslave(bond_dev, slave_dev);
3579                         break;
3580                 case BOND_RELEASE_OLD:
3581                 case SIOCBONDRELEASE:
3582                         res = bond_release(bond_dev, slave_dev);
3583                         break;
3584                 case BOND_SETHWADDR_OLD:
3585                 case SIOCBONDSETHWADDR:
3586                         bond_set_dev_addr(bond_dev, slave_dev);
3587                         res = 0;
3588                         break;
3589                 case BOND_CHANGE_ACTIVE_OLD:
3590                 case SIOCBONDCHANGEACTIVE:
3591                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3592                         break;
3593                 default:
3594                         res = -EOPNOTSUPP;
3595                 }
3596
3597                 dev_put(slave_dev);
3598         }
3599
3600         return res;
3601 }
3602
3603 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3604 {
3605         struct bonding *bond = netdev_priv(bond_dev);
3606
3607         if (change & IFF_PROMISC)
3608                 bond_set_promiscuity(bond,
3609                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3610
3611         if (change & IFF_ALLMULTI)
3612                 bond_set_allmulti(bond,
3613                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3614 }
3615
3616 static void bond_set_rx_mode(struct net_device *bond_dev)
3617 {
3618         struct bonding *bond = netdev_priv(bond_dev);
3619         struct slave *slave;
3620         int i;
3621
3622         read_lock(&bond->lock);
3623
3624         if (USES_PRIMARY(bond->params.mode)) {
3625                 read_lock(&bond->curr_slave_lock);
3626                 slave = bond->curr_active_slave;
3627                 if (slave) {
3628                         dev_uc_sync(slave->dev, bond_dev);
3629                         dev_mc_sync(slave->dev, bond_dev);
3630                 }
3631                 read_unlock(&bond->curr_slave_lock);
3632         } else {
3633                 bond_for_each_slave(bond, slave, i) {
3634                         dev_uc_sync_multiple(slave->dev, bond_dev);
3635                         dev_mc_sync_multiple(slave->dev, bond_dev);
3636                 }
3637         }
3638
3639         read_unlock(&bond->lock);
3640 }
3641
3642 static int bond_neigh_init(struct neighbour *n)
3643 {
3644         struct bonding *bond = netdev_priv(n->dev);
3645         struct slave *slave = bond->first_slave;
3646         const struct net_device_ops *slave_ops;
3647         struct neigh_parms parms;
3648         int ret;
3649
3650         if (!slave)
3651                 return 0;
3652
3653         slave_ops = slave->dev->netdev_ops;
3654
3655         if (!slave_ops->ndo_neigh_setup)
3656                 return 0;
3657
3658         parms.neigh_setup = NULL;
3659         parms.neigh_cleanup = NULL;
3660         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3661         if (ret)
3662                 return ret;
3663
3664         /*
3665          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3666          * after the last slave has been detached.  Assumes that all slaves
3667          * utilize the same neigh_cleanup (true at this writing as only user
3668          * is ipoib).
3669          */
3670         n->parms->neigh_cleanup = parms.neigh_cleanup;
3671
3672         if (!parms.neigh_setup)
3673                 return 0;
3674
3675         return parms.neigh_setup(n);
3676 }
3677
3678 /*
3679  * The bonding ndo_neigh_setup is called at init time beofre any
3680  * slave exists. So we must declare proxy setup function which will
3681  * be used at run time to resolve the actual slave neigh param setup.
3682  */
3683 static int bond_neigh_setup(struct net_device *dev,
3684                             struct neigh_parms *parms)
3685 {
3686         parms->neigh_setup   = bond_neigh_init;
3687
3688         return 0;
3689 }
3690
3691 /*
3692  * Change the MTU of all of a master's slaves to match the master
3693  */
3694 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3695 {
3696         struct bonding *bond = netdev_priv(bond_dev);
3697         struct slave *slave, *stop_at;
3698         int res = 0;
3699         int i;
3700
3701         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3702                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3703
3704         /* Can't hold bond->lock with bh disabled here since
3705          * some base drivers panic. On the other hand we can't
3706          * hold bond->lock without bh disabled because we'll
3707          * deadlock. The only solution is to rely on the fact
3708          * that we're under rtnl_lock here, and the slaves
3709          * list won't change. This doesn't solve the problem
3710          * of setting the slave's MTU while it is
3711          * transmitting, but the assumption is that the base
3712          * driver can handle that.
3713          *
3714          * TODO: figure out a way to safely iterate the slaves
3715          * list, but without holding a lock around the actual
3716          * call to the base driver.
3717          */
3718
3719         bond_for_each_slave(bond, slave, i) {
3720                 pr_debug("s %p s->p %p c_m %p\n",
3721                          slave,
3722                          slave->prev,
3723                          slave->dev->netdev_ops->ndo_change_mtu);
3724
3725                 res = dev_set_mtu(slave->dev, new_mtu);
3726
3727                 if (res) {
3728                         /* If we failed to set the slave's mtu to the new value
3729                          * we must abort the operation even in ACTIVE_BACKUP
3730                          * mode, because if we allow the backup slaves to have
3731                          * different mtu values than the active slave we'll
3732                          * need to change their mtu when doing a failover. That
3733                          * means changing their mtu from timer context, which
3734                          * is probably not a good idea.
3735                          */
3736                         pr_debug("err %d %s\n", res, slave->dev->name);
3737                         goto unwind;
3738                 }
3739         }
3740
3741         bond_dev->mtu = new_mtu;
3742
3743         return 0;
3744
3745 unwind:
3746         /* unwind from head to the slave that failed */
3747         stop_at = slave;
3748         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3749                 int tmp_res;
3750
3751                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3752                 if (tmp_res) {
3753                         pr_debug("unwind err %d dev %s\n",
3754                                  tmp_res, slave->dev->name);
3755                 }
3756         }
3757
3758         return res;
3759 }
3760
3761 /*
3762  * Change HW address
3763  *
3764  * Note that many devices must be down to change the HW address, and
3765  * downing the master releases all slaves.  We can make bonds full of
3766  * bonding devices to test this, however.
3767  */
3768 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3769 {
3770         struct bonding *bond = netdev_priv(bond_dev);
3771         struct sockaddr *sa = addr, tmp_sa;
3772         struct slave *slave, *stop_at;
3773         int res = 0;
3774         int i;
3775
3776         if (bond->params.mode == BOND_MODE_ALB)
3777                 return bond_alb_set_mac_address(bond_dev, addr);
3778
3779
3780         pr_debug("bond=%p, name=%s\n",
3781                  bond, bond_dev ? bond_dev->name : "None");
3782
3783         /* If fail_over_mac is enabled, do nothing and return success.
3784          * Returning an error causes ifenslave to fail.
3785          */
3786         if (bond->params.fail_over_mac)
3787                 return 0;
3788
3789         if (!is_valid_ether_addr(sa->sa_data))
3790                 return -EADDRNOTAVAIL;
3791
3792         /* Can't hold bond->lock with bh disabled here since
3793          * some base drivers panic. On the other hand we can't
3794          * hold bond->lock without bh disabled because we'll
3795          * deadlock. The only solution is to rely on the fact
3796          * that we're under rtnl_lock here, and the slaves
3797          * list won't change. This doesn't solve the problem
3798          * of setting the slave's hw address while it is
3799          * transmitting, but the assumption is that the base
3800          * driver can handle that.
3801          *
3802          * TODO: figure out a way to safely iterate the slaves
3803          * list, but without holding a lock around the actual
3804          * call to the base driver.
3805          */
3806
3807         bond_for_each_slave(bond, slave, i) {
3808                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3809                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3810
3811                 if (slave_ops->ndo_set_mac_address == NULL) {
3812                         res = -EOPNOTSUPP;
3813                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3814                         goto unwind;
3815                 }
3816
3817                 res = dev_set_mac_address(slave->dev, addr);
3818                 if (res) {
3819                         /* TODO: consider downing the slave
3820                          * and retry ?
3821                          * User should expect communications
3822                          * breakage anyway until ARP finish
3823                          * updating, so...
3824                          */
3825                         pr_debug("err %d %s\n", res, slave->dev->name);
3826                         goto unwind;
3827                 }
3828         }
3829
3830         /* success */
3831         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3832         return 0;
3833
3834 unwind:
3835         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3836         tmp_sa.sa_family = bond_dev->type;
3837
3838         /* unwind from head to the slave that failed */
3839         stop_at = slave;
3840         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3841                 int tmp_res;
3842
3843                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3844                 if (tmp_res) {
3845                         pr_debug("unwind err %d dev %s\n",
3846                                  tmp_res, slave->dev->name);
3847                 }
3848         }
3849
3850         return res;
3851 }
3852
3853 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3854 {
3855         struct bonding *bond = netdev_priv(bond_dev);
3856         struct slave *slave, *start_at;
3857         int i, slave_no, res = 1;
3858         struct iphdr *iph = ip_hdr(skb);
3859
3860         /*
3861          * Start with the curr_active_slave that joined the bond as the
3862          * default for sending IGMP traffic.  For failover purposes one
3863          * needs to maintain some consistency for the interface that will
3864          * send the join/membership reports.  The curr_active_slave found
3865          * will send all of this type of traffic.
3866          */
3867         if ((iph->protocol == IPPROTO_IGMP) &&
3868             (skb->protocol == htons(ETH_P_IP))) {
3869
3870                 read_lock(&bond->curr_slave_lock);
3871                 slave = bond->curr_active_slave;
3872                 read_unlock(&bond->curr_slave_lock);
3873
3874                 if (!slave)
3875                         goto out;
3876         } else {
3877                 /*
3878                  * Concurrent TX may collide on rr_tx_counter; we accept
3879                  * that as being rare enough not to justify using an
3880                  * atomic op here.
3881                  */
3882                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3883
3884                 bond_for_each_slave(bond, slave, i) {
3885                         slave_no--;
3886                         if (slave_no < 0)
3887                                 break;
3888                 }
3889         }
3890
3891         start_at = slave;
3892         bond_for_each_slave_from(bond, slave, i, start_at) {
3893                 if (IS_UP(slave->dev) &&
3894                     (slave->link == BOND_LINK_UP) &&
3895                     bond_is_active_slave(slave)) {
3896                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3897                         break;
3898                 }
3899         }
3900
3901 out:
3902         if (res) {
3903                 /* no suitable interface, frame not sent */
3904                 kfree_skb(skb);
3905         }
3906
3907         return NETDEV_TX_OK;
3908 }
3909
3910
3911 /*
3912  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3913  * the bond has a usable interface.
3914  */
3915 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3916 {
3917         struct bonding *bond = netdev_priv(bond_dev);
3918         int res = 1;
3919
3920         read_lock(&bond->curr_slave_lock);
3921
3922         if (bond->curr_active_slave)
3923                 res = bond_dev_queue_xmit(bond, skb,
3924                         bond->curr_active_slave->dev);
3925
3926         read_unlock(&bond->curr_slave_lock);
3927
3928         if (res)
3929                 /* no suitable interface, frame not sent */
3930                 kfree_skb(skb);
3931
3932         return NETDEV_TX_OK;
3933 }
3934
3935 /*
3936  * In bond_xmit_xor() , we determine the output device by using a pre-
3937  * determined xmit_hash_policy(), If the selected device is not enabled,
3938  * find the next active slave.
3939  */
3940 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3941 {
3942         struct bonding *bond = netdev_priv(bond_dev);
3943         struct slave *slave, *start_at;
3944         int slave_no;
3945         int i;
3946         int res = 1;
3947
3948         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
3949
3950         bond_for_each_slave(bond, slave, i) {
3951                 slave_no--;
3952                 if (slave_no < 0)
3953                         break;
3954         }
3955
3956         start_at = slave;
3957
3958         bond_for_each_slave_from(bond, slave, i, start_at) {
3959                 if (IS_UP(slave->dev) &&
3960                     (slave->link == BOND_LINK_UP) &&
3961                     bond_is_active_slave(slave)) {
3962                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3963                         break;
3964                 }
3965         }
3966
3967         if (res) {
3968                 /* no suitable interface, frame not sent */
3969                 kfree_skb(skb);
3970         }
3971
3972         return NETDEV_TX_OK;
3973 }
3974
3975 /*
3976  * in broadcast mode, we send everything to all usable interfaces.
3977  */
3978 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3979 {
3980         struct bonding *bond = netdev_priv(bond_dev);
3981         struct slave *slave, *start_at;
3982         struct net_device *tx_dev = NULL;
3983         int i;
3984         int res = 1;
3985
3986         read_lock(&bond->curr_slave_lock);
3987         start_at = bond->curr_active_slave;
3988         read_unlock(&bond->curr_slave_lock);
3989
3990         if (!start_at)
3991                 goto out;
3992
3993         bond_for_each_slave_from(bond, slave, i, start_at) {
3994                 if (IS_UP(slave->dev) &&
3995                     (slave->link == BOND_LINK_UP) &&
3996                     bond_is_active_slave(slave)) {
3997                         if (tx_dev) {
3998                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3999                                 if (!skb2) {
4000                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4001                                                bond_dev->name);
4002                                         continue;
4003                                 }
4004
4005                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4006                                 if (res) {
4007                                         kfree_skb(skb2);
4008                                         continue;
4009                                 }
4010                         }
4011                         tx_dev = slave->dev;
4012                 }
4013         }
4014
4015         if (tx_dev)
4016                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4017
4018 out:
4019         if (res)
4020                 /* no suitable interface, frame not sent */
4021                 kfree_skb(skb);
4022
4023         /* frame sent to all suitable interfaces */
4024         return NETDEV_TX_OK;
4025 }
4026
4027 /*------------------------- Device initialization ---------------------------*/
4028
4029 static void bond_set_xmit_hash_policy(struct bonding *bond)
4030 {
4031         switch (bond->params.xmit_policy) {
4032         case BOND_XMIT_POLICY_LAYER23:
4033                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4034                 break;
4035         case BOND_XMIT_POLICY_LAYER34:
4036                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4037                 break;
4038         case BOND_XMIT_POLICY_LAYER2:
4039         default:
4040                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4041                 break;
4042         }
4043 }
4044
4045 /*
4046  * Lookup the slave that corresponds to a qid
4047  */
4048 static inline int bond_slave_override(struct bonding *bond,
4049                                       struct sk_buff *skb)
4050 {
4051         int i, res = 1;
4052         struct slave *slave = NULL;
4053         struct slave *check_slave;
4054
4055         if (!skb->queue_mapping)
4056                 return 1;
4057
4058         /* Find out if any slaves have the same mapping as this skb. */
4059         bond_for_each_slave(bond, check_slave, i) {
4060                 if (check_slave->queue_id == skb->queue_mapping) {
4061                         slave = check_slave;
4062                         break;
4063                 }
4064         }
4065
4066         /* If the slave isn't UP, use default transmit policy. */
4067         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4068             (slave->link == BOND_LINK_UP)) {
4069                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4070         }
4071
4072         return res;
4073 }
4074
4075
4076 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4077 {
4078         /*
4079          * This helper function exists to help dev_pick_tx get the correct
4080          * destination queue.  Using a helper function skips a call to
4081          * skb_tx_hash and will put the skbs in the queue we expect on their
4082          * way down to the bonding driver.
4083          */
4084         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4085
4086         /*
4087          * Save the original txq to restore before passing to the driver
4088          */
4089         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4090
4091         if (unlikely(txq >= dev->real_num_tx_queues)) {
4092                 do {
4093                         txq -= dev->real_num_tx_queues;
4094                 } while (txq >= dev->real_num_tx_queues);
4095         }
4096         return txq;
4097 }
4098
4099 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4100 {
4101         struct bonding *bond = netdev_priv(dev);
4102
4103         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4104                 if (!bond_slave_override(bond, skb))
4105                         return NETDEV_TX_OK;
4106         }
4107
4108         switch (bond->params.mode) {
4109         case BOND_MODE_ROUNDROBIN:
4110                 return bond_xmit_roundrobin(skb, dev);
4111         case BOND_MODE_ACTIVEBACKUP:
4112                 return bond_xmit_activebackup(skb, dev);
4113         case BOND_MODE_XOR:
4114                 return bond_xmit_xor(skb, dev);
4115         case BOND_MODE_BROADCAST:
4116                 return bond_xmit_broadcast(skb, dev);
4117         case BOND_MODE_8023AD:
4118                 return bond_3ad_xmit_xor(skb, dev);
4119         case BOND_MODE_ALB:
4120         case BOND_MODE_TLB:
4121                 return bond_alb_xmit(skb, dev);
4122         default:
4123                 /* Should never happen, mode already checked */
4124                 pr_err("%s: Error: Unknown bonding mode %d\n",
4125                        dev->name, bond->params.mode);
4126                 WARN_ON_ONCE(1);
4127                 kfree_skb(skb);
4128                 return NETDEV_TX_OK;
4129         }
4130 }
4131
4132 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4133 {
4134         struct bonding *bond = netdev_priv(dev);
4135         netdev_tx_t ret = NETDEV_TX_OK;
4136
4137         /*
4138          * If we risk deadlock from transmitting this in the
4139          * netpoll path, tell netpoll to queue the frame for later tx
4140          */
4141         if (is_netpoll_tx_blocked(dev))
4142                 return NETDEV_TX_BUSY;
4143
4144         read_lock(&bond->lock);
4145
4146         if (bond->slave_cnt)
4147                 ret = __bond_start_xmit(skb, dev);
4148         else
4149                 kfree_skb(skb);
4150
4151         read_unlock(&bond->lock);
4152
4153         return ret;
4154 }
4155
4156 /*
4157  * set bond mode specific net device operations
4158  */
4159 void bond_set_mode_ops(struct bonding *bond, int mode)
4160 {
4161         struct net_device *bond_dev = bond->dev;
4162
4163         switch (mode) {
4164         case BOND_MODE_ROUNDROBIN:
4165                 break;
4166         case BOND_MODE_ACTIVEBACKUP:
4167                 break;
4168         case BOND_MODE_XOR:
4169                 bond_set_xmit_hash_policy(bond);
4170                 break;
4171         case BOND_MODE_BROADCAST:
4172                 break;
4173         case BOND_MODE_8023AD:
4174                 bond_set_xmit_hash_policy(bond);
4175                 break;
4176         case BOND_MODE_ALB:
4177                 /* FALLTHRU */
4178         case BOND_MODE_TLB:
4179                 break;
4180         default:
4181                 /* Should never happen, mode already checked */
4182                 pr_err("%s: Error: Unknown bonding mode %d\n",
4183                        bond_dev->name, mode);
4184                 break;
4185         }
4186 }
4187
4188 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4189                                      struct ethtool_cmd *ecmd)
4190 {
4191         struct bonding *bond = netdev_priv(bond_dev);
4192         struct slave *slave;
4193         int i;
4194         unsigned long speed = 0;
4195
4196         ecmd->duplex = DUPLEX_UNKNOWN;
4197         ecmd->port = PORT_OTHER;
4198
4199         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4200          * do not need to check mode.  Though link speed might not represent
4201          * the true receive or transmit bandwidth (not all modes are symmetric)
4202          * this is an accurate maximum.
4203          */
4204         read_lock(&bond->lock);
4205         bond_for_each_slave(bond, slave, i) {
4206                 if (SLAVE_IS_OK(slave)) {
4207                         if (slave->speed != SPEED_UNKNOWN)
4208                                 speed += slave->speed;
4209                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4210                             slave->duplex != DUPLEX_UNKNOWN)
4211                                 ecmd->duplex = slave->duplex;
4212                 }
4213         }
4214         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4215         read_unlock(&bond->lock);
4216         return 0;
4217 }
4218
4219 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4220                                      struct ethtool_drvinfo *drvinfo)
4221 {
4222         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4223         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4224         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4225                  BOND_ABI_VERSION);
4226 }
4227
4228 static const struct ethtool_ops bond_ethtool_ops = {
4229         .get_drvinfo            = bond_ethtool_get_drvinfo,
4230         .get_settings           = bond_ethtool_get_settings,
4231         .get_link               = ethtool_op_get_link,
4232 };
4233
4234 static const struct net_device_ops bond_netdev_ops = {
4235         .ndo_init               = bond_init,
4236         .ndo_uninit             = bond_uninit,
4237         .ndo_open               = bond_open,
4238         .ndo_stop               = bond_close,
4239         .ndo_start_xmit         = bond_start_xmit,
4240         .ndo_select_queue       = bond_select_queue,
4241         .ndo_get_stats64        = bond_get_stats,
4242         .ndo_do_ioctl           = bond_do_ioctl,
4243         .ndo_change_rx_flags    = bond_change_rx_flags,
4244         .ndo_set_rx_mode        = bond_set_rx_mode,
4245         .ndo_change_mtu         = bond_change_mtu,
4246         .ndo_set_mac_address    = bond_set_mac_address,
4247         .ndo_neigh_setup        = bond_neigh_setup,
4248         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4249         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4250 #ifdef CONFIG_NET_POLL_CONTROLLER
4251         .ndo_netpoll_setup      = bond_netpoll_setup,
4252         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4253         .ndo_poll_controller    = bond_poll_controller,
4254 #endif
4255         .ndo_add_slave          = bond_enslave,
4256         .ndo_del_slave          = bond_release,
4257         .ndo_fix_features       = bond_fix_features,
4258 };
4259
4260 static const struct device_type bond_type = {
4261         .name = "bond",
4262 };
4263
4264 static void bond_destructor(struct net_device *bond_dev)
4265 {
4266         struct bonding *bond = netdev_priv(bond_dev);
4267         if (bond->wq)
4268                 destroy_workqueue(bond->wq);
4269         free_netdev(bond_dev);
4270 }
4271
4272 static void bond_setup(struct net_device *bond_dev)
4273 {
4274         struct bonding *bond = netdev_priv(bond_dev);
4275
4276         /* initialize rwlocks */
4277         rwlock_init(&bond->lock);
4278         rwlock_init(&bond->curr_slave_lock);
4279
4280         bond->params = bonding_defaults;
4281
4282         /* Initialize pointers */
4283         bond->dev = bond_dev;
4284         INIT_LIST_HEAD(&bond->vlan_list);
4285
4286         /* Initialize the device entry points */
4287         ether_setup(bond_dev);
4288         bond_dev->netdev_ops = &bond_netdev_ops;
4289         bond_dev->ethtool_ops = &bond_ethtool_ops;
4290         bond_set_mode_ops(bond, bond->params.mode);
4291
4292         bond_dev->destructor = bond_destructor;
4293
4294         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4295
4296         /* Initialize the device options */
4297         bond_dev->tx_queue_len = 0;
4298         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4299         bond_dev->priv_flags |= IFF_BONDING;
4300         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4301
4302         /* At first, we block adding VLANs. That's the only way to
4303          * prevent problems that occur when adding VLANs over an
4304          * empty bond. The block will be removed once non-challenged
4305          * slaves are enslaved.
4306          */
4307         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4308
4309         /* don't acquire bond device's netif_tx_lock when
4310          * transmitting */
4311         bond_dev->features |= NETIF_F_LLTX;
4312
4313         /* By default, we declare the bond to be fully
4314          * VLAN hardware accelerated capable. Special
4315          * care is taken in the various xmit functions
4316          * when there are slaves that are not hw accel
4317          * capable
4318          */
4319
4320         bond_dev->hw_features = BOND_VLAN_FEATURES |
4321                                 NETIF_F_HW_VLAN_CTAG_TX |
4322                                 NETIF_F_HW_VLAN_CTAG_RX |
4323                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4324
4325         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4326         bond_dev->features |= bond_dev->hw_features;
4327 }
4328
4329 /*
4330 * Destroy a bonding device.
4331 * Must be under rtnl_lock when this function is called.
4332 */
4333 static void bond_uninit(struct net_device *bond_dev)
4334 {
4335         struct bonding *bond = netdev_priv(bond_dev);
4336         struct vlan_entry *vlan, *tmp;
4337
4338         bond_netpoll_cleanup(bond_dev);
4339
4340         /* Release the bonded slaves */
4341         while (bond->first_slave != NULL)
4342                 __bond_release_one(bond_dev, bond->first_slave->dev, true);
4343         pr_info("%s: released all slaves\n", bond_dev->name);
4344
4345         list_del(&bond->bond_list);
4346
4347         bond_debug_unregister(bond);
4348
4349         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4350                 list_del(&vlan->vlan_list);
4351                 kfree(vlan);
4352         }
4353 }
4354
4355 /*------------------------- Module initialization ---------------------------*/
4356
4357 /*
4358  * Convert string input module parms.  Accept either the
4359  * number of the mode or its string name.  A bit complicated because
4360  * some mode names are substrings of other names, and calls from sysfs
4361  * may have whitespace in the name (trailing newlines, for example).
4362  */
4363 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4364 {
4365         int modeint = -1, i, rv;
4366         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4367
4368         for (p = (char *)buf; *p; p++)
4369                 if (!(isdigit(*p) || isspace(*p)))
4370                         break;
4371
4372         if (*p)
4373                 rv = sscanf(buf, "%20s", modestr);
4374         else
4375                 rv = sscanf(buf, "%d", &modeint);
4376
4377         if (!rv)
4378                 return -1;
4379
4380         for (i = 0; tbl[i].modename; i++) {
4381                 if (modeint == tbl[i].mode)
4382                         return tbl[i].mode;
4383                 if (strcmp(modestr, tbl[i].modename) == 0)
4384                         return tbl[i].mode;
4385         }
4386
4387         return -1;
4388 }
4389
4390 static int bond_check_params(struct bond_params *params)
4391 {
4392         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4393         int arp_all_targets_value;
4394
4395         /*
4396          * Convert string parameters.
4397          */
4398         if (mode) {
4399                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4400                 if (bond_mode == -1) {
4401                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4402                                mode == NULL ? "NULL" : mode);
4403                         return -EINVAL;
4404                 }
4405         }
4406
4407         if (xmit_hash_policy) {
4408                 if ((bond_mode != BOND_MODE_XOR) &&
4409                     (bond_mode != BOND_MODE_8023AD)) {
4410                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4411                                bond_mode_name(bond_mode));
4412                 } else {
4413                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4414                                                         xmit_hashtype_tbl);
4415                         if (xmit_hashtype == -1) {
4416                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4417                                        xmit_hash_policy == NULL ? "NULL" :
4418                                        xmit_hash_policy);
4419                                 return -EINVAL;
4420                         }
4421                 }
4422         }
4423
4424         if (lacp_rate) {
4425                 if (bond_mode != BOND_MODE_8023AD) {
4426                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4427                                 bond_mode_name(bond_mode));
4428                 } else {
4429                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4430                         if (lacp_fast == -1) {
4431                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4432                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4433                                 return -EINVAL;
4434                         }
4435                 }
4436         }
4437
4438         if (ad_select) {
4439                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4440                 if (params->ad_select == -1) {
4441                         pr_err("Error: Invalid ad_select \"%s\"\n",
4442                                ad_select == NULL ? "NULL" : ad_select);
4443                         return -EINVAL;
4444                 }
4445
4446                 if (bond_mode != BOND_MODE_8023AD) {
4447                         pr_warning("ad_select param only affects 802.3ad mode\n");
4448                 }
4449         } else {
4450                 params->ad_select = BOND_AD_STABLE;
4451         }
4452
4453         if (max_bonds < 0) {
4454                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4455                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4456                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4457         }
4458
4459         if (miimon < 0) {
4460                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4461                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4462                 miimon = BOND_LINK_MON_INTERV;
4463         }
4464
4465         if (updelay < 0) {
4466                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4467                            updelay, INT_MAX);
4468                 updelay = 0;
4469         }
4470
4471         if (downdelay < 0) {
4472                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4473                            downdelay, INT_MAX);
4474                 downdelay = 0;
4475         }
4476
4477         if ((use_carrier != 0) && (use_carrier != 1)) {
4478                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4479                            use_carrier);
4480                 use_carrier = 1;
4481         }
4482
4483         if (num_peer_notif < 0 || num_peer_notif > 255) {
4484                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4485                            num_peer_notif);
4486                 num_peer_notif = 1;
4487         }
4488
4489         /* reset values for 802.3ad */
4490         if (bond_mode == BOND_MODE_8023AD) {
4491                 if (!miimon) {
4492                         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");
4493                         pr_warning("Forcing miimon to 100msec\n");
4494                         miimon = 100;
4495                 }
4496         }
4497
4498         if (tx_queues < 1 || tx_queues > 255) {
4499                 pr_warning("Warning: tx_queues (%d) should be between "
4500                            "1 and 255, resetting to %d\n",
4501                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4502                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4503         }
4504
4505         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4506                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4507                            "not of valid value (0/1), so it was set to "
4508                            "0\n", all_slaves_active);
4509                 all_slaves_active = 0;
4510         }
4511
4512         if (resend_igmp < 0 || resend_igmp > 255) {
4513                 pr_warning("Warning: resend_igmp (%d) should be between "
4514                            "0 and 255, resetting to %d\n",
4515                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4516                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4517         }
4518
4519         /* reset values for TLB/ALB */
4520         if ((bond_mode == BOND_MODE_TLB) ||
4521             (bond_mode == BOND_MODE_ALB)) {
4522                 if (!miimon) {
4523                         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");
4524                         pr_warning("Forcing miimon to 100msec\n");
4525                         miimon = 100;
4526                 }
4527         }
4528
4529         if (bond_mode == BOND_MODE_ALB) {
4530                 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",
4531                           updelay);
4532         }
4533
4534         if (!miimon) {
4535                 if (updelay || downdelay) {
4536                         /* just warn the user the up/down delay will have
4537                          * no effect since miimon is zero...
4538                          */
4539                         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",
4540                                    updelay, downdelay);
4541                 }
4542         } else {
4543                 /* don't allow arp monitoring */
4544                 if (arp_interval) {
4545                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4546                                    miimon, arp_interval);
4547                         arp_interval = 0;
4548                 }
4549
4550                 if ((updelay % miimon) != 0) {
4551                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4552                                    updelay, miimon,
4553                                    (updelay / miimon) * miimon);
4554                 }
4555
4556                 updelay /= miimon;
4557
4558                 if ((downdelay % miimon) != 0) {
4559                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4560                                    downdelay, miimon,
4561                                    (downdelay / miimon) * miimon);
4562                 }
4563
4564                 downdelay /= miimon;
4565         }
4566
4567         if (arp_interval < 0) {
4568                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4569                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4570                 arp_interval = BOND_LINK_ARP_INTERV;
4571         }
4572
4573         for (arp_ip_count = 0, i = 0;
4574              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4575                 /* not complete check, but should be good enough to
4576                    catch mistakes */
4577                 __be32 ip = in_aton(arp_ip_target[i]);
4578                 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4579                     ip == htonl(INADDR_BROADCAST)) {
4580                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4581                                    arp_ip_target[i]);
4582                         arp_interval = 0;
4583                 } else {
4584                         if (bond_get_targets_ip(arp_target, ip) == -1)
4585                                 arp_target[arp_ip_count++] = ip;
4586                         else
4587                                 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4588                                            &ip);
4589                 }
4590         }
4591
4592         if (arp_interval && !arp_ip_count) {
4593                 /* don't allow arping if no arp_ip_target given... */
4594                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4595                            arp_interval);
4596                 arp_interval = 0;
4597         }
4598
4599         if (arp_validate) {
4600                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4601                         pr_err("arp_validate only supported in active-backup mode\n");
4602                         return -EINVAL;
4603                 }
4604                 if (!arp_interval) {
4605                         pr_err("arp_validate requires arp_interval\n");
4606                         return -EINVAL;
4607                 }
4608
4609                 arp_validate_value = bond_parse_parm(arp_validate,
4610                                                      arp_validate_tbl);
4611                 if (arp_validate_value == -1) {
4612                         pr_err("Error: invalid arp_validate \"%s\"\n",
4613                                arp_validate == NULL ? "NULL" : arp_validate);
4614                         return -EINVAL;
4615                 }
4616         } else
4617                 arp_validate_value = 0;
4618
4619         arp_all_targets_value = 0;
4620         if (arp_all_targets) {
4621                 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4622                                                         arp_all_targets_tbl);
4623
4624                 if (arp_all_targets_value == -1) {
4625                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4626                                arp_all_targets);
4627                         arp_all_targets_value = 0;
4628                 }
4629         }
4630
4631         if (miimon) {
4632                 pr_info("MII link monitoring set to %d ms\n", miimon);
4633         } else if (arp_interval) {
4634                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4635                         arp_interval,
4636                         arp_validate_tbl[arp_validate_value].modename,
4637                         arp_ip_count);
4638
4639                 for (i = 0; i < arp_ip_count; i++)
4640                         pr_info(" %s", arp_ip_target[i]);
4641
4642                 pr_info("\n");
4643
4644         } else if (max_bonds) {
4645                 /* miimon and arp_interval not set, we need one so things
4646                  * work as expected, see bonding.txt for details
4647                  */
4648                 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");
4649         }
4650
4651         if (primary && !USES_PRIMARY(bond_mode)) {
4652                 /* currently, using a primary only makes sense
4653                  * in active backup, TLB or ALB modes
4654                  */
4655                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4656                            primary, bond_mode_name(bond_mode));
4657                 primary = NULL;
4658         }
4659
4660         if (primary && primary_reselect) {
4661                 primary_reselect_value = bond_parse_parm(primary_reselect,
4662                                                          pri_reselect_tbl);
4663                 if (primary_reselect_value == -1) {
4664                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4665                                primary_reselect ==
4666                                         NULL ? "NULL" : primary_reselect);
4667                         return -EINVAL;
4668                 }
4669         } else {
4670                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4671         }
4672
4673         if (fail_over_mac) {
4674                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4675                                                       fail_over_mac_tbl);
4676                 if (fail_over_mac_value == -1) {
4677                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4678                                arp_validate == NULL ? "NULL" : arp_validate);
4679                         return -EINVAL;
4680                 }
4681
4682                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4683                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4684         } else {
4685                 fail_over_mac_value = BOND_FOM_NONE;
4686         }
4687
4688         /* fill params struct with the proper values */
4689         params->mode = bond_mode;
4690         params->xmit_policy = xmit_hashtype;
4691         params->miimon = miimon;
4692         params->num_peer_notif = num_peer_notif;
4693         params->arp_interval = arp_interval;
4694         params->arp_validate = arp_validate_value;
4695         params->arp_all_targets = arp_all_targets_value;
4696         params->updelay = updelay;
4697         params->downdelay = downdelay;
4698         params->use_carrier = use_carrier;
4699         params->lacp_fast = lacp_fast;
4700         params->primary[0] = 0;
4701         params->primary_reselect = primary_reselect_value;
4702         params->fail_over_mac = fail_over_mac_value;
4703         params->tx_queues = tx_queues;
4704         params->all_slaves_active = all_slaves_active;
4705         params->resend_igmp = resend_igmp;
4706         params->min_links = min_links;
4707
4708         if (primary) {
4709                 strncpy(params->primary, primary, IFNAMSIZ);
4710                 params->primary[IFNAMSIZ - 1] = 0;
4711         }
4712
4713         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4714
4715         return 0;
4716 }
4717
4718 static struct lock_class_key bonding_netdev_xmit_lock_key;
4719 static struct lock_class_key bonding_netdev_addr_lock_key;
4720 static struct lock_class_key bonding_tx_busylock_key;
4721
4722 static void bond_set_lockdep_class_one(struct net_device *dev,
4723                                        struct netdev_queue *txq,
4724                                        void *_unused)
4725 {
4726         lockdep_set_class(&txq->_xmit_lock,
4727                           &bonding_netdev_xmit_lock_key);
4728 }
4729
4730 static void bond_set_lockdep_class(struct net_device *dev)
4731 {
4732         lockdep_set_class(&dev->addr_list_lock,
4733                           &bonding_netdev_addr_lock_key);
4734         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4735         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4736 }
4737
4738 /*
4739  * Called from registration process
4740  */
4741 static int bond_init(struct net_device *bond_dev)
4742 {
4743         struct bonding *bond = netdev_priv(bond_dev);
4744         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4745         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4746
4747         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4748
4749         /*
4750          * Initialize locks that may be required during
4751          * en/deslave operations.  All of the bond_open work
4752          * (of which this is part) should really be moved to
4753          * a phase prior to dev_open
4754          */
4755         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4756         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4757
4758         bond->wq = create_singlethread_workqueue(bond_dev->name);
4759         if (!bond->wq)
4760                 return -ENOMEM;
4761
4762         bond_set_lockdep_class(bond_dev);
4763
4764         list_add_tail(&bond->bond_list, &bn->dev_list);
4765
4766         bond_prepare_sysfs_group(bond);
4767
4768         bond_debug_register(bond);
4769
4770         /* Ensure valid dev_addr */
4771         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4772             bond_dev->addr_assign_type == NET_ADDR_PERM)
4773                 eth_hw_addr_random(bond_dev);
4774
4775         return 0;
4776 }
4777
4778 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4779 {
4780         if (tb[IFLA_ADDRESS]) {
4781                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4782                         return -EINVAL;
4783                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4784                         return -EADDRNOTAVAIL;
4785         }
4786         return 0;
4787 }
4788
4789 static unsigned int bond_get_num_tx_queues(void)
4790 {
4791         return tx_queues;
4792 }
4793
4794 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4795         .kind                   = "bond",
4796         .priv_size              = sizeof(struct bonding),
4797         .setup                  = bond_setup,
4798         .validate               = bond_validate,
4799         .get_num_tx_queues      = bond_get_num_tx_queues,
4800         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4801                                                              as for TX queues */
4802 };
4803
4804 /* Create a new bond based on the specified name and bonding parameters.
4805  * If name is NULL, obtain a suitable "bond%d" name for us.
4806  * Caller must NOT hold rtnl_lock; we need to release it here before we
4807  * set up our sysfs entries.
4808  */
4809 int bond_create(struct net *net, const char *name)
4810 {
4811         struct net_device *bond_dev;
4812         int res;
4813
4814         rtnl_lock();
4815
4816         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4817                                    name ? name : "bond%d",
4818                                    bond_setup, tx_queues);
4819         if (!bond_dev) {
4820                 pr_err("%s: eek! can't alloc netdev!\n", name);
4821                 rtnl_unlock();
4822                 return -ENOMEM;
4823         }
4824
4825         dev_net_set(bond_dev, net);
4826         bond_dev->rtnl_link_ops = &bond_link_ops;
4827
4828         res = register_netdevice(bond_dev);
4829
4830         netif_carrier_off(bond_dev);
4831
4832         rtnl_unlock();
4833         if (res < 0)
4834                 bond_destructor(bond_dev);
4835         return res;
4836 }
4837
4838 static int __net_init bond_net_init(struct net *net)
4839 {
4840         struct bond_net *bn = net_generic(net, bond_net_id);
4841
4842         bn->net = net;
4843         INIT_LIST_HEAD(&bn->dev_list);
4844
4845         bond_create_proc_dir(bn);
4846         bond_create_sysfs(bn);
4847
4848         return 0;
4849 }
4850
4851 static void __net_exit bond_net_exit(struct net *net)
4852 {
4853         struct bond_net *bn = net_generic(net, bond_net_id);
4854         struct bonding *bond, *tmp_bond;
4855         LIST_HEAD(list);
4856
4857         bond_destroy_sysfs(bn);
4858         bond_destroy_proc_dir(bn);
4859
4860         /* Kill off any bonds created after unregistering bond rtnl ops */
4861         rtnl_lock();
4862         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4863                 unregister_netdevice_queue(bond->dev, &list);
4864         unregister_netdevice_many(&list);
4865         rtnl_unlock();
4866 }
4867
4868 static struct pernet_operations bond_net_ops = {
4869         .init = bond_net_init,
4870         .exit = bond_net_exit,
4871         .id   = &bond_net_id,
4872         .size = sizeof(struct bond_net),
4873 };
4874
4875 static int __init bonding_init(void)
4876 {
4877         int i;
4878         int res;
4879
4880         pr_info("%s", bond_version);
4881
4882         res = bond_check_params(&bonding_defaults);
4883         if (res)
4884                 goto out;
4885
4886         res = register_pernet_subsys(&bond_net_ops);
4887         if (res)
4888                 goto out;
4889
4890         res = rtnl_link_register(&bond_link_ops);
4891         if (res)
4892                 goto err_link;
4893
4894         bond_create_debugfs();
4895
4896         for (i = 0; i < max_bonds; i++) {
4897                 res = bond_create(&init_net, NULL);
4898                 if (res)
4899                         goto err;
4900         }
4901
4902         register_netdevice_notifier(&bond_netdev_notifier);
4903 out:
4904         return res;
4905 err:
4906         rtnl_link_unregister(&bond_link_ops);
4907 err_link:
4908         unregister_pernet_subsys(&bond_net_ops);
4909         goto out;
4910
4911 }
4912
4913 static void __exit bonding_exit(void)
4914 {
4915         unregister_netdevice_notifier(&bond_netdev_notifier);
4916
4917         bond_destroy_debugfs();
4918
4919         rtnl_link_unregister(&bond_link_ops);
4920         unregister_pernet_subsys(&bond_net_ops);
4921
4922 #ifdef CONFIG_NET_POLL_CONTROLLER
4923         /*
4924          * Make sure we don't have an imbalance on our netpoll blocking
4925          */
4926         WARN_ON(atomic_read(&netpoll_block_tx));
4927 #endif
4928 }
4929
4930 module_init(bonding_init);
4931 module_exit(bonding_exit);
4932 MODULE_LICENSE("GPL");
4933 MODULE_VERSION(DRV_VERSION);
4934 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4935 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4936 MODULE_ALIAS_RTNL_LINK("bond");