Pileus Git - ~andy/linux/blob - 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 < BOND_MODE_ROUNDROBIN || 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 #else
1249 static inline int slave_enable_netpoll(struct slave *slave)
1250 {
1251         return 0;
1252 }
1253 static inline void slave_disable_netpoll(struct slave *slave)
1254 {
1255 }
1256 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1257 {
1258 }
1259 #endif
1260
1261 /*---------------------------------- IOCTL ----------------------------------*/
1262
1263 static netdev_features_t bond_fix_features(struct net_device *dev,
1264         netdev_features_t features)
1265 {
1266         struct slave *slave;
1267         struct bonding *bond = netdev_priv(dev);
1268         netdev_features_t mask;
1269         int i;
1270
1271         read_lock(&bond->lock);
1272
1273         if (!bond->first_slave) {
1274                 /* Disable adding VLANs to empty bond. But why? --mq */
1275                 features |= NETIF_F_VLAN_CHALLENGED;
1276                 goto out;
1277         }
1278
1279         mask = features;
1280         features &= ~NETIF_F_ONE_FOR_ALL;
1281         features |= NETIF_F_ALL_FOR_ALL;
1282
1283         bond_for_each_slave(bond, slave, i) {
1284                 features = netdev_increment_features(features,
1285                                                      slave->dev->features,
1286                                                      mask);
1287         }
1288         features = netdev_add_tso_features(features, mask);
1289
1290 out:
1291         read_unlock(&bond->lock);
1292         return features;
1293 }
1294
1295 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1296                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1297                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1298
1299 static void bond_compute_features(struct bonding *bond)
1300 {
1301         struct slave *slave;
1302         struct net_device *bond_dev = bond->dev;
1303         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1304         unsigned short max_hard_header_len = ETH_HLEN;
1305         unsigned int gso_max_size = GSO_MAX_SIZE;
1306         u16 gso_max_segs = GSO_MAX_SEGS;
1307         int i;
1308         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1309
1310         read_lock(&bond->lock);
1311
1312         if (!bond->first_slave)
1313                 goto done;
1314
1315         bond_for_each_slave(bond, slave, i) {
1316                 vlan_features = netdev_increment_features(vlan_features,
1317                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1318
1319                 dst_release_flag &= slave->dev->priv_flags;
1320                 if (slave->dev->hard_header_len > max_hard_header_len)
1321                         max_hard_header_len = slave->dev->hard_header_len;
1322
1323                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1324                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1325         }
1326
1327 done:
1328         bond_dev->vlan_features = vlan_features;
1329         bond_dev->hard_header_len = max_hard_header_len;
1330         bond_dev->gso_max_segs = gso_max_segs;
1331         netif_set_gso_max_size(bond_dev, gso_max_size);
1332
1333         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1334         bond_dev->priv_flags = flags | dst_release_flag;
1335
1336         read_unlock(&bond->lock);
1337
1338         netdev_change_features(bond_dev);
1339 }
1340
1341 static void bond_setup_by_slave(struct net_device *bond_dev,
1342                                 struct net_device *slave_dev)
1343 {
1344         bond_dev->header_ops        = slave_dev->header_ops;
1345
1346         bond_dev->type              = slave_dev->type;
1347         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1348         bond_dev->addr_len          = slave_dev->addr_len;
1349
1350         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1351                 slave_dev->addr_len);
1352 }
1353
1354 /* On bonding slaves other than the currently active slave, suppress
1355  * duplicates except for alb non-mcast/bcast.
1356  */
1357 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1358                                             struct slave *slave,
1359                                             struct bonding *bond)
1360 {
1361         if (bond_is_slave_inactive(slave)) {
1362                 if (bond->params.mode == BOND_MODE_ALB &&
1363                     skb->pkt_type != PACKET_BROADCAST &&
1364                     skb->pkt_type != PACKET_MULTICAST)
1365                         return false;
1366                 return true;
1367         }
1368         return false;
1369 }
1370
1371 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1372 {
1373         struct sk_buff *skb = *pskb;
1374         struct slave *slave;
1375         struct bonding *bond;
1376         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1377                           struct slave *);
1378         int ret = RX_HANDLER_ANOTHER;
1379
1380         skb = skb_share_check(skb, GFP_ATOMIC);
1381         if (unlikely(!skb))
1382                 return RX_HANDLER_CONSUMED;
1383
1384         *pskb = skb;
1385
1386         slave = bond_slave_get_rcu(skb->dev);
1387         bond = slave->bond;
1388
1389         if (bond->params.arp_interval)
1390                 slave->dev->last_rx = jiffies;
1391
1392         recv_probe = ACCESS_ONCE(bond->recv_probe);
1393         if (recv_probe) {
1394                 ret = recv_probe(skb, bond, slave);
1395                 if (ret == RX_HANDLER_CONSUMED) {
1396                         consume_skb(skb);
1397                         return ret;
1398                 }
1399         }
1400
1401         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1402                 return RX_HANDLER_EXACT;
1403         }
1404
1405         skb->dev = bond->dev;
1406
1407         if (bond->params.mode == BOND_MODE_ALB &&
1408             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1409             skb->pkt_type == PACKET_HOST) {
1410
1411                 if (unlikely(skb_cow_head(skb,
1412                                           skb->data - skb_mac_header(skb)))) {
1413                         kfree_skb(skb);
1414                         return RX_HANDLER_CONSUMED;
1415                 }
1416                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1417         }
1418
1419         return ret;
1420 }
1421
1422 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1423                                       struct net_device *slave_dev)
1424 {
1425         int err;
1426
1427         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1428         if (err)
1429                 return err;
1430         slave_dev->flags |= IFF_SLAVE;
1431         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1432         return 0;
1433 }
1434
1435 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1436                                   struct net_device *slave_dev)
1437 {
1438         netdev_upper_dev_unlink(slave_dev, bond_dev);
1439         slave_dev->flags &= ~IFF_SLAVE;
1440         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1441 }
1442
1443 /* enslave device <slave> to bond device <master> */
1444 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1445 {
1446         struct bonding *bond = netdev_priv(bond_dev);
1447         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1448         struct slave *new_slave = NULL;
1449         struct sockaddr addr;
1450         int link_reporting;
1451         int res = 0, i;
1452
1453         if (!bond->params.use_carrier &&
1454             slave_dev->ethtool_ops->get_link == NULL &&
1455             slave_ops->ndo_do_ioctl == NULL) {
1456                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1457                            bond_dev->name, slave_dev->name);
1458         }
1459
1460         /* already enslaved */
1461         if (slave_dev->flags & IFF_SLAVE) {
1462                 pr_debug("Error, Device was already enslaved\n");
1463                 return -EBUSY;
1464         }
1465
1466         /* vlan challenged mutual exclusion */
1467         /* no need to lock since we're protected by rtnl_lock */
1468         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1469                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1470                 if (vlan_uses_dev(bond_dev)) {
1471                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1472                                bond_dev->name, slave_dev->name, bond_dev->name);
1473                         return -EPERM;
1474                 } else {
1475                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1476                                    bond_dev->name, slave_dev->name,
1477                                    slave_dev->name, bond_dev->name);
1478                 }
1479         } else {
1480                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1481         }
1482
1483         /*
1484          * Old ifenslave binaries are no longer supported.  These can
1485          * be identified with moderate accuracy by the state of the slave:
1486          * the current ifenslave will set the interface down prior to
1487          * enslaving it; the old ifenslave will not.
1488          */
1489         if ((slave_dev->flags & IFF_UP)) {
1490                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1491                        slave_dev->name);
1492                 res = -EPERM;
1493                 goto err_undo_flags;
1494         }
1495
1496         /* set bonding device ether type by slave - bonding netdevices are
1497          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1498          * there is a need to override some of the type dependent attribs/funcs.
1499          *
1500          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1501          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1502          */
1503         if (bond->slave_cnt == 0) {
1504                 if (bond_dev->type != slave_dev->type) {
1505                         pr_debug("%s: change device type from %d to %d\n",
1506                                  bond_dev->name,
1507                                  bond_dev->type, slave_dev->type);
1508
1509                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1510                                                        bond_dev);
1511                         res = notifier_to_errno(res);
1512                         if (res) {
1513                                 pr_err("%s: refused to change device type\n",
1514                                        bond_dev->name);
1515                                 res = -EBUSY;
1516                                 goto err_undo_flags;
1517                         }
1518
1519                         /* Flush unicast and multicast addresses */
1520                         dev_uc_flush(bond_dev);
1521                         dev_mc_flush(bond_dev);
1522
1523                         if (slave_dev->type != ARPHRD_ETHER)
1524                                 bond_setup_by_slave(bond_dev, slave_dev);
1525                         else {
1526                                 ether_setup(bond_dev);
1527                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1528                         }
1529
1530                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1531                                                  bond_dev);
1532                 }
1533         } else if (bond_dev->type != slave_dev->type) {
1534                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1535                        slave_dev->name,
1536                        slave_dev->type, bond_dev->type);
1537                 res = -EINVAL;
1538                 goto err_undo_flags;
1539         }
1540
1541         if (slave_ops->ndo_set_mac_address == NULL) {
1542                 if (bond->slave_cnt == 0) {
1543                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1544                                    bond_dev->name);
1545                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1546                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1547                         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",
1548                                bond_dev->name);
1549                         res = -EOPNOTSUPP;
1550                         goto err_undo_flags;
1551                 }
1552         }
1553
1554         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1555
1556         /* If this is the first slave, then we need to set the master's hardware
1557          * address to be the same as the slave's. */
1558         if (!bond->slave_cnt && bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1559                 bond_set_dev_addr(bond->dev, slave_dev);
1560
1561         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1562         if (!new_slave) {
1563                 res = -ENOMEM;
1564                 goto err_undo_flags;
1565         }
1566
1567         /*
1568          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1569          * is set via sysfs or module option if desired.
1570          */
1571         new_slave->queue_id = 0;
1572
1573         /* Save slave's original mtu and then set it to match the bond */
1574         new_slave->original_mtu = slave_dev->mtu;
1575         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1576         if (res) {
1577                 pr_debug("Error %d calling dev_set_mtu\n", res);
1578                 goto err_free;
1579         }
1580
1581         /*
1582          * Save slave's original ("permanent") mac address for modes
1583          * that need it, and for restoring it upon release, and then
1584          * set it to the master's address
1585          */
1586         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1587
1588         if (!bond->params.fail_over_mac) {
1589                 /*
1590                  * Set slave to master's mac address.  The application already
1591                  * set the master's mac address to that of the first slave
1592                  */
1593                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1594                 addr.sa_family = slave_dev->type;
1595                 res = dev_set_mac_address(slave_dev, &addr);
1596                 if (res) {
1597                         pr_debug("Error %d calling set_mac_address\n", res);
1598                         goto err_restore_mtu;
1599                 }
1600         }
1601
1602         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1603         if (res) {
1604                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1605                 goto err_restore_mac;
1606         }
1607
1608         /* open the slave since the application closed it */
1609         res = dev_open(slave_dev);
1610         if (res) {
1611                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1612                 goto err_unset_master;
1613         }
1614
1615         new_slave->bond = bond;
1616         new_slave->dev = slave_dev;
1617         slave_dev->priv_flags |= IFF_BONDING;
1618
1619         if (bond_is_lb(bond)) {
1620                 /* bond_alb_init_slave() must be called before all other stages since
1621                  * it might fail and we do not want to have to undo everything
1622                  */
1623                 res = bond_alb_init_slave(bond, new_slave);
1624                 if (res)
1625                         goto err_close;
1626         }
1627
1628         /* If the mode USES_PRIMARY, then the following is handled by
1629          * bond_change_active_slave().
1630          */
1631         if (!USES_PRIMARY(bond->params.mode)) {
1632                 /* set promiscuity level to new slave */
1633                 if (bond_dev->flags & IFF_PROMISC) {
1634                         res = dev_set_promiscuity(slave_dev, 1);
1635                         if (res)
1636                                 goto err_close;
1637                 }
1638
1639                 /* set allmulti level to new slave */
1640                 if (bond_dev->flags & IFF_ALLMULTI) {
1641                         res = dev_set_allmulti(slave_dev, 1);
1642                         if (res)
1643                                 goto err_close;
1644                 }
1645
1646                 netif_addr_lock_bh(bond_dev);
1647
1648                 dev_mc_sync_multiple(slave_dev, bond_dev);
1649                 dev_uc_sync_multiple(slave_dev, bond_dev);
1650
1651                 netif_addr_unlock_bh(bond_dev);
1652         }
1653
1654         if (bond->params.mode == BOND_MODE_8023AD) {
1655                 /* add lacpdu mc addr to mc list */
1656                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1657
1658                 dev_mc_add(slave_dev, lacpdu_multicast);
1659         }
1660
1661         bond_add_vlans_on_slave(bond, slave_dev);
1662
1663         write_lock_bh(&bond->lock);
1664
1665         bond_attach_slave(bond, new_slave);
1666
1667         new_slave->delay = 0;
1668         new_slave->link_failure_count = 0;
1669
1670         write_unlock_bh(&bond->lock);
1671
1672         bond_compute_features(bond);
1673
1674         bond_update_speed_duplex(new_slave);
1675
1676         read_lock(&bond->lock);
1677
1678         new_slave->last_arp_rx = jiffies -
1679                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1680         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1681                 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1682
1683         if (bond->params.miimon && !bond->params.use_carrier) {
1684                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1685
1686                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1687                         /*
1688                          * miimon is set but a bonded network driver
1689                          * does not support ETHTOOL/MII and
1690                          * arp_interval is not set.  Note: if
1691                          * use_carrier is enabled, we will never go
1692                          * here (because netif_carrier is always
1693                          * supported); thus, we don't need to change
1694                          * the messages for netif_carrier.
1695                          */
1696                         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",
1697                                bond_dev->name, slave_dev->name);
1698                 } else if (link_reporting == -1) {
1699                         /* unable get link status using mii/ethtool */
1700                         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",
1701                                    bond_dev->name, slave_dev->name);
1702                 }
1703         }
1704
1705         /* check for initial state */
1706         if (bond->params.miimon) {
1707                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1708                         if (bond->params.updelay) {
1709                                 new_slave->link = BOND_LINK_BACK;
1710                                 new_slave->delay = bond->params.updelay;
1711                         } else {
1712                                 new_slave->link = BOND_LINK_UP;
1713                         }
1714                 } else {
1715                         new_slave->link = BOND_LINK_DOWN;
1716                 }
1717         } else if (bond->params.arp_interval) {
1718                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1719                         BOND_LINK_UP : BOND_LINK_DOWN);
1720         } else {
1721                 new_slave->link = BOND_LINK_UP;
1722         }
1723
1724         if (new_slave->link != BOND_LINK_DOWN)
1725                 new_slave->jiffies = jiffies;
1726         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1727                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1728                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1729
1730         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1731                 /* if there is a primary slave, remember it */
1732                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1733                         bond->primary_slave = new_slave;
1734                         bond->force_primary = true;
1735                 }
1736         }
1737
1738         write_lock_bh(&bond->curr_slave_lock);
1739
1740         switch (bond->params.mode) {
1741         case BOND_MODE_ACTIVEBACKUP:
1742                 bond_set_slave_inactive_flags(new_slave);
1743                 bond_select_active_slave(bond);
1744                 break;
1745         case BOND_MODE_8023AD:
1746                 /* in 802.3ad mode, the internal mechanism
1747                  * will activate the slaves in the selected
1748                  * aggregator
1749                  */
1750                 bond_set_slave_inactive_flags(new_slave);
1751                 /* if this is the first slave */
1752                 if (bond->slave_cnt == 1) {
1753                         SLAVE_AD_INFO(new_slave).id = 1;
1754                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1755                          * can be called only after the mac address of the bond is set
1756                          */
1757                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1758                 } else {
1759                         SLAVE_AD_INFO(new_slave).id =
1760                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1761                 }
1762
1763                 bond_3ad_bind_slave(new_slave);
1764                 break;
1765         case BOND_MODE_TLB:
1766         case BOND_MODE_ALB:
1767                 bond_set_active_slave(new_slave);
1768                 bond_set_slave_inactive_flags(new_slave);
1769                 bond_select_active_slave(bond);
1770                 break;
1771         default:
1772                 pr_debug("This slave is always active in trunk mode\n");
1773
1774                 /* always active in trunk mode */
1775                 bond_set_active_slave(new_slave);
1776
1777                 /* In trunking mode there is little meaning to curr_active_slave
1778                  * anyway (it holds no special properties of the bond device),
1779                  * so we can change it without calling change_active_interface()
1780                  */
1781                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1782                         bond->curr_active_slave = new_slave;
1783
1784                 break;
1785         } /* switch(bond_mode) */
1786
1787         write_unlock_bh(&bond->curr_slave_lock);
1788
1789         bond_set_carrier(bond);
1790
1791 #ifdef CONFIG_NET_POLL_CONTROLLER
1792         slave_dev->npinfo = bond->dev->npinfo;
1793         if (slave_dev->npinfo) {
1794                 if (slave_enable_netpoll(new_slave)) {
1795                         read_unlock(&bond->lock);
1796                         pr_info("Error, %s: master_dev is using netpoll, "
1797                                  "but new slave device does not support netpoll.\n",
1798                                  bond_dev->name);
1799                         res = -EBUSY;
1800                         goto err_detach;
1801                 }
1802         }
1803 #endif
1804
1805         read_unlock(&bond->lock);
1806
1807         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1808         if (res)
1809                 goto err_detach;
1810
1811         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1812                                          new_slave);
1813         if (res) {
1814                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1815                 goto err_dest_symlinks;
1816         }
1817
1818         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1819                 bond_dev->name, slave_dev->name,
1820                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1821                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1822
1823         /* enslave is successful */
1824         return 0;
1825
1826 /* Undo stages on error */
1827 err_dest_symlinks:
1828         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1829
1830 err_detach:
1831         if (!USES_PRIMARY(bond->params.mode))
1832                 bond_hw_addr_flush(bond_dev, slave_dev);
1833
1834         bond_del_vlans_from_slave(bond, slave_dev);
1835         write_lock_bh(&bond->lock);
1836         bond_detach_slave(bond, new_slave);
1837         if (bond->primary_slave == new_slave)
1838                 bond->primary_slave = NULL;
1839         if (bond->curr_active_slave == new_slave) {
1840                 bond_change_active_slave(bond, NULL);
1841                 write_unlock_bh(&bond->lock);
1842                 read_lock(&bond->lock);
1843                 write_lock_bh(&bond->curr_slave_lock);
1844                 bond_select_active_slave(bond);
1845                 write_unlock_bh(&bond->curr_slave_lock);
1846                 read_unlock(&bond->lock);
1847         } else {
1848                 write_unlock_bh(&bond->lock);
1849         }
1850         slave_disable_netpoll(new_slave);
1851
1852 err_close:
1853         slave_dev->priv_flags &= ~IFF_BONDING;
1854         dev_close(slave_dev);
1855
1856 err_unset_master:
1857         bond_upper_dev_unlink(bond_dev, slave_dev);
1858
1859 err_restore_mac:
1860         if (!bond->params.fail_over_mac) {
1861                 /* XXX TODO - fom follow mode needs to change master's
1862                  * MAC if this slave's MAC is in use by the bond, or at
1863                  * least print a warning.
1864                  */
1865                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1866                 addr.sa_family = slave_dev->type;
1867                 dev_set_mac_address(slave_dev, &addr);
1868         }
1869
1870 err_restore_mtu:
1871         dev_set_mtu(slave_dev, new_slave->original_mtu);
1872
1873 err_free:
1874         kfree(new_slave);
1875
1876 err_undo_flags:
1877         bond_compute_features(bond);
1878         /* Enslave of first slave has failed and we need to fix master's mac */
1879         if (bond->slave_cnt == 0 &&
1880             ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1881                 eth_hw_addr_random(bond_dev);
1882
1883         return res;
1884 }
1885
1886 /*
1887  * Try to release the slave device <slave> from the bond device <master>
1888  * It is legal to access curr_active_slave without a lock because all the function
1889  * is write-locked. If "all" is true it means that the function is being called
1890  * while destroying a bond interface and all slaves are being released.
1891  *
1892  * The rules for slave state should be:
1893  *   for Active/Backup:
1894  *     Active stays on all backups go down
1895  *   for Bonded connections:
1896  *     The first up interface should be left on and all others downed.
1897  */
1898 static int __bond_release_one(struct net_device *bond_dev,
1899                               struct net_device *slave_dev,
1900                               bool all)
1901 {
1902         struct bonding *bond = netdev_priv(bond_dev);
1903         struct slave *slave, *oldcurrent;
1904         struct sockaddr addr;
1905         netdev_features_t old_features = bond_dev->features;
1906
1907         /* slave is not a slave or master is not master of this slave */
1908         if (!(slave_dev->flags & IFF_SLAVE) ||
1909             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1910                 pr_err("%s: Error: cannot release %s.\n",
1911                        bond_dev->name, slave_dev->name);
1912                 return -EINVAL;
1913         }
1914
1915         block_netpoll_tx();
1916         write_lock_bh(&bond->lock);
1917
1918         slave = bond_get_slave_by_dev(bond, slave_dev);
1919         if (!slave) {
1920                 /* not a slave of this bond */
1921                 pr_info("%s: %s not enslaved\n",
1922                         bond_dev->name, slave_dev->name);
1923                 write_unlock_bh(&bond->lock);
1924                 unblock_netpoll_tx();
1925                 return -EINVAL;
1926         }
1927
1928         write_unlock_bh(&bond->lock);
1929         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1930          * for this slave anymore.
1931          */
1932         netdev_rx_handler_unregister(slave_dev);
1933         write_lock_bh(&bond->lock);
1934
1935         if (!all && !bond->params.fail_over_mac) {
1936                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1937                     bond->slave_cnt > 1)
1938                         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",
1939                                    bond_dev->name, slave_dev->name,
1940                                    slave->perm_hwaddr,
1941                                    bond_dev->name, slave_dev->name);
1942         }
1943
1944         /* Inform AD package of unbinding of slave. */
1945         if (bond->params.mode == BOND_MODE_8023AD) {
1946                 /* must be called before the slave is
1947                  * detached from the list
1948                  */
1949                 bond_3ad_unbind_slave(slave);
1950         }
1951
1952         pr_info("%s: releasing %s interface %s\n",
1953                 bond_dev->name,
1954                 bond_is_active_slave(slave) ? "active" : "backup",
1955                 slave_dev->name);
1956
1957         oldcurrent = bond->curr_active_slave;
1958
1959         bond->current_arp_slave = NULL;
1960
1961         /* release the slave from its bond */
1962         bond_detach_slave(bond, slave);
1963
1964         if (bond->primary_slave == slave)
1965                 bond->primary_slave = NULL;
1966
1967         if (oldcurrent == slave)
1968                 bond_change_active_slave(bond, NULL);
1969
1970         if (bond_is_lb(bond)) {
1971                 /* Must be called only after the slave has been
1972                  * detached from the list and the curr_active_slave
1973                  * has been cleared (if our_slave == old_current),
1974                  * but before a new active slave is selected.
1975                  */
1976                 write_unlock_bh(&bond->lock);
1977                 bond_alb_deinit_slave(bond, slave);
1978                 write_lock_bh(&bond->lock);
1979         }
1980
1981         if (all) {
1982                 bond->curr_active_slave = NULL;
1983         } else if (oldcurrent == slave) {
1984                 /*
1985                  * Note that we hold RTNL over this sequence, so there
1986                  * is no concern that another slave add/remove event
1987                  * will interfere.
1988                  */
1989                 write_unlock_bh(&bond->lock);
1990                 read_lock(&bond->lock);
1991                 write_lock_bh(&bond->curr_slave_lock);
1992
1993                 bond_select_active_slave(bond);
1994
1995                 write_unlock_bh(&bond->curr_slave_lock);
1996                 read_unlock(&bond->lock);
1997                 write_lock_bh(&bond->lock);
1998         }
1999
2000         if (bond->slave_cnt == 0) {
2001                 bond_set_carrier(bond);
2002                 eth_hw_addr_random(bond_dev);
2003
2004                 if (bond_vlan_used(bond)) {
2005                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2006                                    bond_dev->name, bond_dev->name);
2007                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2008                                    bond_dev->name);
2009                 }
2010         }
2011
2012         write_unlock_bh(&bond->lock);
2013         unblock_netpoll_tx();
2014
2015         if (bond->slave_cnt == 0) {
2016                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2017                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2018         }
2019
2020         bond_compute_features(bond);
2021         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2022             (old_features & NETIF_F_VLAN_CHALLENGED))
2023                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2024                         bond_dev->name, slave_dev->name, bond_dev->name);
2025
2026         /* must do this from outside any spinlocks */
2027         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2028
2029         bond_del_vlans_from_slave(bond, slave_dev);
2030
2031         /* If the mode USES_PRIMARY, then this cases was handled above by
2032          * bond_change_active_slave(..., NULL)
2033          */
2034         if (!USES_PRIMARY(bond->params.mode)) {
2035                 /* unset promiscuity level from slave */
2036                 if (bond_dev->flags & IFF_PROMISC)
2037                         dev_set_promiscuity(slave_dev, -1);
2038
2039                 /* unset allmulti level from slave */
2040                 if (bond_dev->flags & IFF_ALLMULTI)
2041                         dev_set_allmulti(slave_dev, -1);
2042
2043                 bond_hw_addr_flush(bond_dev, slave_dev);
2044         }
2045
2046         bond_upper_dev_unlink(bond_dev, slave_dev);
2047
2048         slave_disable_netpoll(slave);
2049
2050         /* close slave before restoring its mac address */
2051         dev_close(slave_dev);
2052
2053         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2054                 /* restore original ("permanent") mac address */
2055                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2056                 addr.sa_family = slave_dev->type;
2057                 dev_set_mac_address(slave_dev, &addr);
2058         }
2059
2060         dev_set_mtu(slave_dev, slave->original_mtu);
2061
2062         slave_dev->priv_flags &= ~IFF_BONDING;
2063
2064         kfree(slave);
2065
2066         return 0;  /* deletion OK */
2067 }
2068
2069 /* A wrapper used because of ndo_del_link */
2070 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2071 {
2072         return __bond_release_one(bond_dev, slave_dev, false);
2073 }
2074
2075 /*
2076 * First release a slave and then destroy the bond if no more slaves are left.
2077 * Must be under rtnl_lock when this function is called.
2078 */
2079 static int  bond_release_and_destroy(struct net_device *bond_dev,
2080                                      struct net_device *slave_dev)
2081 {
2082         struct bonding *bond = netdev_priv(bond_dev);
2083         int ret;
2084
2085         ret = bond_release(bond_dev, slave_dev);
2086         if ((ret == 0) && (bond->slave_cnt == 0)) {
2087                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2088                 pr_info("%s: destroying bond %s.\n",
2089                         bond_dev->name, bond_dev->name);
2090                 unregister_netdevice(bond_dev);
2091         }
2092         return ret;
2093 }
2094
2095 /*
2096  * This function changes the active slave to slave <slave_dev>.
2097  * It returns -EINVAL in the following cases.
2098  *  - <slave_dev> is not found in the list.
2099  *  - There is not active slave now.
2100  *  - <slave_dev> is already active.
2101  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2102  *  - <slave_dev> is not running.
2103  * In these cases, this function does nothing.
2104  * In the other cases, current_slave pointer is changed and 0 is returned.
2105  */
2106 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2107 {
2108         struct bonding *bond = netdev_priv(bond_dev);
2109         struct slave *old_active = NULL;
2110         struct slave *new_active = NULL;
2111         int res = 0;
2112
2113         if (!USES_PRIMARY(bond->params.mode))
2114                 return -EINVAL;
2115
2116         /* Verify that bond_dev is indeed the master of slave_dev */
2117         if (!(slave_dev->flags & IFF_SLAVE) ||
2118             !netdev_has_upper_dev(slave_dev, bond_dev))
2119                 return -EINVAL;
2120
2121         read_lock(&bond->lock);
2122
2123         read_lock(&bond->curr_slave_lock);
2124         old_active = bond->curr_active_slave;
2125         read_unlock(&bond->curr_slave_lock);
2126
2127         new_active = bond_get_slave_by_dev(bond, slave_dev);
2128
2129         /*
2130          * Changing to the current active: do nothing; return success.
2131          */
2132         if (new_active && (new_active == old_active)) {
2133                 read_unlock(&bond->lock);
2134                 return 0;
2135         }
2136
2137         if ((new_active) &&
2138             (old_active) &&
2139             (new_active->link == BOND_LINK_UP) &&
2140             IS_UP(new_active->dev)) {
2141                 block_netpoll_tx();
2142                 write_lock_bh(&bond->curr_slave_lock);
2143                 bond_change_active_slave(bond, new_active);
2144                 write_unlock_bh(&bond->curr_slave_lock);
2145                 unblock_netpoll_tx();
2146         } else
2147                 res = -EINVAL;
2148
2149         read_unlock(&bond->lock);
2150
2151         return res;
2152 }
2153
2154 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2155 {
2156         struct bonding *bond = netdev_priv(bond_dev);
2157
2158         info->bond_mode = bond->params.mode;
2159         info->miimon = bond->params.miimon;
2160
2161         read_lock(&bond->lock);
2162         info->num_slaves = bond->slave_cnt;
2163         read_unlock(&bond->lock);
2164
2165         return 0;
2166 }
2167
2168 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2169 {
2170         struct bonding *bond = netdev_priv(bond_dev);
2171         struct slave *slave;
2172         int i, res = -ENODEV;
2173
2174         read_lock(&bond->lock);
2175
2176         bond_for_each_slave(bond, slave, i) {
2177                 if (i == (int)info->slave_id) {
2178                         res = 0;
2179                         strcpy(info->slave_name, slave->dev->name);
2180                         info->link = slave->link;
2181                         info->state = bond_slave_state(slave);
2182                         info->link_failure_count = slave->link_failure_count;
2183                         break;
2184                 }
2185         }
2186
2187         read_unlock(&bond->lock);
2188
2189         return res;
2190 }
2191
2192 /*-------------------------------- Monitoring -------------------------------*/
2193
2194
2195 static int bond_miimon_inspect(struct bonding *bond)
2196 {
2197         struct slave *slave;
2198         int i, link_state, commit = 0;
2199         bool ignore_updelay;
2200
2201         ignore_updelay = !bond->curr_active_slave ? true : false;
2202
2203         bond_for_each_slave(bond, slave, i) {
2204                 slave->new_link = BOND_LINK_NOCHANGE;
2205
2206                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2207
2208                 switch (slave->link) {
2209                 case BOND_LINK_UP:
2210                         if (link_state)
2211                                 continue;
2212
2213                         slave->link = BOND_LINK_FAIL;
2214                         slave->delay = bond->params.downdelay;
2215                         if (slave->delay) {
2216                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2217                                         bond->dev->name,
2218                                         (bond->params.mode ==
2219                                          BOND_MODE_ACTIVEBACKUP) ?
2220                                         (bond_is_active_slave(slave) ?
2221                                          "active " : "backup ") : "",
2222                                         slave->dev->name,
2223                                         bond->params.downdelay * bond->params.miimon);
2224                         }
2225                         /*FALLTHRU*/
2226                 case BOND_LINK_FAIL:
2227                         if (link_state) {
2228                                 /*
2229                                  * recovered before downdelay expired
2230                                  */
2231                                 slave->link = BOND_LINK_UP;
2232                                 slave->jiffies = jiffies;
2233                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2234                                         bond->dev->name,
2235                                         (bond->params.downdelay - slave->delay) *
2236                                         bond->params.miimon,
2237                                         slave->dev->name);
2238                                 continue;
2239                         }
2240
2241                         if (slave->delay <= 0) {
2242                                 slave->new_link = BOND_LINK_DOWN;
2243                                 commit++;
2244                                 continue;
2245                         }
2246
2247                         slave->delay--;
2248                         break;
2249
2250                 case BOND_LINK_DOWN:
2251                         if (!link_state)
2252                                 continue;
2253
2254                         slave->link = BOND_LINK_BACK;
2255                         slave->delay = bond->params.updelay;
2256
2257                         if (slave->delay) {
2258                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2259                                         bond->dev->name, slave->dev->name,
2260                                         ignore_updelay ? 0 :
2261                                         bond->params.updelay *
2262                                         bond->params.miimon);
2263                         }
2264                         /*FALLTHRU*/
2265                 case BOND_LINK_BACK:
2266                         if (!link_state) {
2267                                 slave->link = BOND_LINK_DOWN;
2268                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2269                                         bond->dev->name,
2270                                         (bond->params.updelay - slave->delay) *
2271                                         bond->params.miimon,
2272                                         slave->dev->name);
2273
2274                                 continue;
2275                         }
2276
2277                         if (ignore_updelay)
2278                                 slave->delay = 0;
2279
2280                         if (slave->delay <= 0) {
2281                                 slave->new_link = BOND_LINK_UP;
2282                                 commit++;
2283                                 ignore_updelay = false;
2284                                 continue;
2285                         }
2286
2287                         slave->delay--;
2288                         break;
2289                 }
2290         }
2291
2292         return commit;
2293 }
2294
2295 static void bond_miimon_commit(struct bonding *bond)
2296 {
2297         struct slave *slave;
2298         int i;
2299
2300         bond_for_each_slave(bond, slave, i) {
2301                 switch (slave->new_link) {
2302                 case BOND_LINK_NOCHANGE:
2303                         continue;
2304
2305                 case BOND_LINK_UP:
2306                         slave->link = BOND_LINK_UP;
2307                         slave->jiffies = jiffies;
2308
2309                         if (bond->params.mode == BOND_MODE_8023AD) {
2310                                 /* prevent it from being the active one */
2311                                 bond_set_backup_slave(slave);
2312                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2313                                 /* make it immediately active */
2314                                 bond_set_active_slave(slave);
2315                         } else if (slave != bond->primary_slave) {
2316                                 /* prevent it from being the active one */
2317                                 bond_set_backup_slave(slave);
2318                         }
2319
2320                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2321                                 bond->dev->name, slave->dev->name,
2322                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2323                                 slave->duplex ? "full" : "half");
2324
2325                         /* notify ad that the link status has changed */
2326                         if (bond->params.mode == BOND_MODE_8023AD)
2327                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2328
2329                         if (bond_is_lb(bond))
2330                                 bond_alb_handle_link_change(bond, slave,
2331                                                             BOND_LINK_UP);
2332
2333                         if (!bond->curr_active_slave ||
2334                             (slave == bond->primary_slave))
2335                                 goto do_failover;
2336
2337                         continue;
2338
2339                 case BOND_LINK_DOWN:
2340                         if (slave->link_failure_count < UINT_MAX)
2341                                 slave->link_failure_count++;
2342
2343                         slave->link = BOND_LINK_DOWN;
2344
2345                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2346                             bond->params.mode == BOND_MODE_8023AD)
2347                                 bond_set_slave_inactive_flags(slave);
2348
2349                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2350                                 bond->dev->name, slave->dev->name);
2351
2352                         if (bond->params.mode == BOND_MODE_8023AD)
2353                                 bond_3ad_handle_link_change(slave,
2354                                                             BOND_LINK_DOWN);
2355
2356                         if (bond_is_lb(bond))
2357                                 bond_alb_handle_link_change(bond, slave,
2358                                                             BOND_LINK_DOWN);
2359
2360                         if (slave == bond->curr_active_slave)
2361                                 goto do_failover;
2362
2363                         continue;
2364
2365                 default:
2366                         pr_err("%s: invalid new link %d on slave %s\n",
2367                                bond->dev->name, slave->new_link,
2368                                slave->dev->name);
2369                         slave->new_link = BOND_LINK_NOCHANGE;
2370
2371                         continue;
2372                 }
2373
2374 do_failover:
2375                 ASSERT_RTNL();
2376                 block_netpoll_tx();
2377                 write_lock_bh(&bond->curr_slave_lock);
2378                 bond_select_active_slave(bond);
2379                 write_unlock_bh(&bond->curr_slave_lock);
2380                 unblock_netpoll_tx();
2381         }
2382
2383         bond_set_carrier(bond);
2384 }
2385
2386 /*
2387  * bond_mii_monitor
2388  *
2389  * Really a wrapper that splits the mii monitor into two phases: an
2390  * inspection, then (if inspection indicates something needs to be done)
2391  * an acquisition of appropriate locks followed by a commit phase to
2392  * implement whatever link state changes are indicated.
2393  */
2394 void bond_mii_monitor(struct work_struct *work)
2395 {
2396         struct bonding *bond = container_of(work, struct bonding,
2397                                             mii_work.work);
2398         bool should_notify_peers = false;
2399         unsigned long delay;
2400
2401         read_lock(&bond->lock);
2402
2403         delay = msecs_to_jiffies(bond->params.miimon);
2404
2405         if (bond->slave_cnt == 0)
2406                 goto re_arm;
2407
2408         should_notify_peers = bond_should_notify_peers(bond);
2409
2410         if (bond_miimon_inspect(bond)) {
2411                 read_unlock(&bond->lock);
2412
2413                 /* Race avoidance with bond_close cancel of workqueue */
2414                 if (!rtnl_trylock()) {
2415                         read_lock(&bond->lock);
2416                         delay = 1;
2417                         should_notify_peers = false;
2418                         goto re_arm;
2419                 }
2420
2421                 read_lock(&bond->lock);
2422
2423                 bond_miimon_commit(bond);
2424
2425                 read_unlock(&bond->lock);
2426                 rtnl_unlock();  /* might sleep, hold no other locks */
2427                 read_lock(&bond->lock);
2428         }
2429
2430 re_arm:
2431         if (bond->params.miimon)
2432                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2433
2434         read_unlock(&bond->lock);
2435
2436         if (should_notify_peers) {
2437                 if (!rtnl_trylock()) {
2438                         read_lock(&bond->lock);
2439                         bond->send_peer_notif++;
2440                         read_unlock(&bond->lock);
2441                         return;
2442                 }
2443                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2444                 rtnl_unlock();
2445         }
2446 }
2447
2448 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2449 {
2450         struct vlan_entry *vlan;
2451         struct net_device *vlan_dev;
2452
2453         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2454                 return 1;
2455
2456         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2457                 rcu_read_lock();
2458                 vlan_dev = __vlan_find_dev_deep(bond->dev, htons(ETH_P_8021Q),
2459                                                 vlan->vlan_id);
2460                 rcu_read_unlock();
2461                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2462                         return 1;
2463         }
2464
2465         return 0;
2466 }
2467
2468 /*
2469  * We go to the (large) trouble of VLAN tagging ARP frames because
2470  * switches in VLAN mode (especially if ports are configured as
2471  * "native" to a VLAN) might not pass non-tagged frames.
2472  */
2473 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2474 {
2475         struct sk_buff *skb;
2476
2477         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2478                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2479
2480         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2481                          NULL, slave_dev->dev_addr, NULL);
2482
2483         if (!skb) {
2484                 pr_err("ARP packet allocation failed\n");
2485                 return;
2486         }
2487         if (vlan_id) {
2488                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2489                 if (!skb) {
2490                         pr_err("failed to insert VLAN tag\n");
2491                         return;
2492                 }
2493         }
2494         arp_xmit(skb);
2495 }
2496
2497
2498 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2499 {
2500         int i, vlan_id;
2501         __be32 *targets = bond->params.arp_targets;
2502         struct vlan_entry *vlan;
2503         struct net_device *vlan_dev = NULL;
2504         struct rtable *rt;
2505
2506         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2507                 __be32 addr;
2508                 if (!targets[i])
2509                         break;
2510                 pr_debug("basa: target %pI4\n", &targets[i]);
2511                 if (!bond_vlan_used(bond)) {
2512                         pr_debug("basa: empty vlan: arp_send\n");
2513                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2514                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2515                                       addr, 0);
2516                         continue;
2517                 }
2518
2519                 /*
2520                  * If VLANs are configured, we do a route lookup to
2521                  * determine which VLAN interface would be used, so we
2522                  * can tag the ARP with the proper VLAN tag.
2523                  */
2524                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2525                                      RTO_ONLINK, 0);
2526                 if (IS_ERR(rt)) {
2527                         if (net_ratelimit()) {
2528                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2529                                            bond->dev->name, &targets[i]);
2530                         }
2531                         continue;
2532                 }
2533
2534                 /*
2535                  * This target is not on a VLAN
2536                  */
2537                 if (rt->dst.dev == bond->dev) {
2538                         ip_rt_put(rt);
2539                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2540                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2541                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2542                                       addr, 0);
2543                         continue;
2544                 }
2545
2546                 vlan_id = 0;
2547                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2548                         rcu_read_lock();
2549                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2550                                                         htons(ETH_P_8021Q),
2551                                                         vlan->vlan_id);
2552                         rcu_read_unlock();
2553                         if (vlan_dev == rt->dst.dev) {
2554                                 vlan_id = vlan->vlan_id;
2555                                 pr_debug("basa: vlan match on %s %d\n",
2556                                        vlan_dev->name, vlan_id);
2557                                 break;
2558                         }
2559                 }
2560
2561                 if (vlan_id && vlan_dev) {
2562                         ip_rt_put(rt);
2563                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2564                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2565                                       addr, vlan_id);
2566                         continue;
2567                 }
2568
2569                 if (net_ratelimit()) {
2570                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2571                                    bond->dev->name, &targets[i],
2572                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2573                 }
2574                 ip_rt_put(rt);
2575         }
2576 }
2577
2578 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2579 {
2580         int i;
2581
2582         if (!sip || !bond_has_this_ip(bond, tip)) {
2583                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2584                 return;
2585         }
2586
2587         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2588         if (i == -1) {
2589                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2590                 return;
2591         }
2592         slave->last_arp_rx = jiffies;
2593         slave->target_last_arp_rx[i] = jiffies;
2594 }
2595
2596 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2597                         struct slave *slave)
2598 {
2599         struct arphdr *arp = (struct arphdr *)skb->data;
2600         unsigned char *arp_ptr;
2601         __be32 sip, tip;
2602         int alen;
2603
2604         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2605                 return RX_HANDLER_ANOTHER;
2606
2607         read_lock(&bond->lock);
2608
2609         if (!slave_do_arp_validate(bond, slave))
2610                 goto out_unlock;
2611
2612         alen = arp_hdr_len(bond->dev);
2613
2614         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2615                  bond->dev->name, skb->dev->name);
2616
2617         if (alen > skb_headlen(skb)) {
2618                 arp = kmalloc(alen, GFP_ATOMIC);
2619                 if (!arp)
2620                         goto out_unlock;
2621                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2622                         goto out_unlock;
2623         }
2624
2625         if (arp->ar_hln != bond->dev->addr_len ||
2626             skb->pkt_type == PACKET_OTHERHOST ||
2627             skb->pkt_type == PACKET_LOOPBACK ||
2628             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2629             arp->ar_pro != htons(ETH_P_IP) ||
2630             arp->ar_pln != 4)
2631                 goto out_unlock;
2632
2633         arp_ptr = (unsigned char *)(arp + 1);
2634         arp_ptr += bond->dev->addr_len;
2635         memcpy(&sip, arp_ptr, 4);
2636         arp_ptr += 4 + bond->dev->addr_len;
2637         memcpy(&tip, arp_ptr, 4);
2638
2639         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2640                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2641                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2642                  &sip, &tip);
2643
2644         /*
2645          * Backup slaves won't see the ARP reply, but do come through
2646          * here for each ARP probe (so we swap the sip/tip to validate
2647          * the probe).  In a "redundant switch, common router" type of
2648          * configuration, the ARP probe will (hopefully) travel from
2649          * the active, through one switch, the router, then the other
2650          * switch before reaching the backup.
2651          *
2652          * We 'trust' the arp requests if there is an active slave and
2653          * it received valid arp reply(s) after it became active. This
2654          * is done to avoid endless looping when we can't reach the
2655          * arp_ip_target and fool ourselves with our own arp requests.
2656          */
2657         if (bond_is_active_slave(slave))
2658                 bond_validate_arp(bond, slave, sip, tip);
2659         else if (bond->curr_active_slave &&
2660                  time_after(slave_last_rx(bond, bond->curr_active_slave),
2661                             bond->curr_active_slave->jiffies))
2662                 bond_validate_arp(bond, slave, tip, sip);
2663
2664 out_unlock:
2665         read_unlock(&bond->lock);
2666         if (arp != (struct arphdr *)skb->data)
2667                 kfree(arp);
2668         return RX_HANDLER_ANOTHER;
2669 }
2670
2671 /*
2672  * this function is called regularly to monitor each slave's link
2673  * ensuring that traffic is being sent and received when arp monitoring
2674  * is used in load-balancing mode. if the adapter has been dormant, then an
2675  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2676  * arp monitoring in active backup mode.
2677  */
2678 void bond_loadbalance_arp_mon(struct work_struct *work)
2679 {
2680         struct bonding *bond = container_of(work, struct bonding,
2681                                             arp_work.work);
2682         struct slave *slave, *oldcurrent;
2683         int do_failover = 0;
2684         int delta_in_ticks, extra_ticks;
2685         int i;
2686
2687         read_lock(&bond->lock);
2688
2689         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2690         extra_ticks = delta_in_ticks / 2;
2691
2692         if (bond->slave_cnt == 0)
2693                 goto re_arm;
2694
2695         read_lock(&bond->curr_slave_lock);
2696         oldcurrent = bond->curr_active_slave;
2697         read_unlock(&bond->curr_slave_lock);
2698
2699         /* see if any of the previous devices are up now (i.e. they have
2700          * xmt and rcv traffic). the curr_active_slave does not come into
2701          * the picture unless it is null. also, slave->jiffies is not needed
2702          * here because we send an arp on each slave and give a slave as
2703          * long as it needs to get the tx/rx within the delta.
2704          * TODO: what about up/down delay in arp mode? it wasn't here before
2705          *       so it can wait
2706          */
2707         bond_for_each_slave(bond, slave, i) {
2708                 unsigned long trans_start = dev_trans_start(slave->dev);
2709
2710                 if (slave->link != BOND_LINK_UP) {
2711                         if (time_in_range(jiffies,
2712                                 trans_start - delta_in_ticks,
2713                                 trans_start + delta_in_ticks + extra_ticks) &&
2714                             time_in_range(jiffies,
2715                                 slave->dev->last_rx - delta_in_ticks,
2716                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2717
2718                                 slave->link  = BOND_LINK_UP;
2719                                 bond_set_active_slave(slave);
2720
2721                                 /* primary_slave has no meaning in round-robin
2722                                  * mode. the window of a slave being up and
2723                                  * curr_active_slave being null after enslaving
2724                                  * is closed.
2725                                  */
2726                                 if (!oldcurrent) {
2727                                         pr_info("%s: link status definitely up for interface %s, ",
2728                                                 bond->dev->name,
2729                                                 slave->dev->name);
2730                                         do_failover = 1;
2731                                 } else {
2732                                         pr_info("%s: interface %s is now up\n",
2733                                                 bond->dev->name,
2734                                                 slave->dev->name);
2735                                 }
2736                         }
2737                 } else {
2738                         /* slave->link == BOND_LINK_UP */
2739
2740                         /* not all switches will respond to an arp request
2741                          * when the source ip is 0, so don't take the link down
2742                          * if we don't know our ip yet
2743                          */
2744                         if (!time_in_range(jiffies,
2745                                 trans_start - delta_in_ticks,
2746                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2747                             !time_in_range(jiffies,
2748                                 slave->dev->last_rx - delta_in_ticks,
2749                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2750
2751                                 slave->link  = BOND_LINK_DOWN;
2752                                 bond_set_backup_slave(slave);
2753
2754                                 if (slave->link_failure_count < UINT_MAX)
2755                                         slave->link_failure_count++;
2756
2757                                 pr_info("%s: interface %s is now down.\n",
2758                                         bond->dev->name,
2759                                         slave->dev->name);
2760
2761                                 if (slave == oldcurrent)
2762                                         do_failover = 1;
2763                         }
2764                 }
2765
2766                 /* note: if switch is in round-robin mode, all links
2767                  * must tx arp to ensure all links rx an arp - otherwise
2768                  * links may oscillate or not come up at all; if switch is
2769                  * in something like xor mode, there is nothing we can
2770                  * do - all replies will be rx'ed on same link causing slaves
2771                  * to be unstable during low/no traffic periods
2772                  */
2773                 if (IS_UP(slave->dev))
2774                         bond_arp_send_all(bond, slave);
2775         }
2776
2777         if (do_failover) {
2778                 block_netpoll_tx();
2779                 write_lock_bh(&bond->curr_slave_lock);
2780
2781                 bond_select_active_slave(bond);
2782
2783                 write_unlock_bh(&bond->curr_slave_lock);
2784                 unblock_netpoll_tx();
2785         }
2786
2787 re_arm:
2788         if (bond->params.arp_interval)
2789                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2790
2791         read_unlock(&bond->lock);
2792 }
2793
2794 /*
2795  * Called to inspect slaves for active-backup mode ARP monitor link state
2796  * changes.  Sets new_link in slaves to specify what action should take
2797  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2798  * to link states must be committed.
2799  *
2800  * Called with bond->lock held for read.
2801  */
2802 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2803 {
2804         struct slave *slave;
2805         int i, commit = 0;
2806         unsigned long trans_start;
2807         int extra_ticks;
2808
2809         /* All the time comparisons below need some extra time. Otherwise, on
2810          * fast networks the ARP probe/reply may arrive within the same jiffy
2811          * as it was sent.  Then, the next time the ARP monitor is run, one
2812          * arp_interval will already have passed in the comparisons.
2813          */
2814         extra_ticks = delta_in_ticks / 2;
2815
2816         bond_for_each_slave(bond, slave, i) {
2817                 slave->new_link = BOND_LINK_NOCHANGE;
2818
2819                 if (slave->link != BOND_LINK_UP) {
2820                         if (time_in_range(jiffies,
2821                                 slave_last_rx(bond, slave) - delta_in_ticks,
2822                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2823
2824                                 slave->new_link = BOND_LINK_UP;
2825                                 commit++;
2826                         }
2827
2828                         continue;
2829                 }
2830
2831                 /*
2832                  * Give slaves 2*delta after being enslaved or made
2833                  * active.  This avoids bouncing, as the last receive
2834                  * times need a full ARP monitor cycle to be updated.
2835                  */
2836                 if (time_in_range(jiffies,
2837                                   slave->jiffies - delta_in_ticks,
2838                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2839                         continue;
2840
2841                 /*
2842                  * Backup slave is down if:
2843                  * - No current_arp_slave AND
2844                  * - more than 3*delta since last receive AND
2845                  * - the bond has an IP address
2846                  *
2847                  * Note: a non-null current_arp_slave indicates
2848                  * the curr_active_slave went down and we are
2849                  * searching for a new one; under this condition
2850                  * we only take the curr_active_slave down - this
2851                  * gives each slave a chance to tx/rx traffic
2852                  * before being taken out
2853                  */
2854                 if (!bond_is_active_slave(slave) &&
2855                     !bond->current_arp_slave &&
2856                     !time_in_range(jiffies,
2857                         slave_last_rx(bond, slave) - delta_in_ticks,
2858                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2859
2860                         slave->new_link = BOND_LINK_DOWN;
2861                         commit++;
2862                 }
2863
2864                 /*
2865                  * Active slave is down if:
2866                  * - more than 2*delta since transmitting OR
2867                  * - (more than 2*delta since receive AND
2868                  *    the bond has an IP address)
2869                  */
2870                 trans_start = dev_trans_start(slave->dev);
2871                 if (bond_is_active_slave(slave) &&
2872                     (!time_in_range(jiffies,
2873                         trans_start - delta_in_ticks,
2874                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
2875                      !time_in_range(jiffies,
2876                         slave_last_rx(bond, slave) - delta_in_ticks,
2877                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2878
2879                         slave->new_link = BOND_LINK_DOWN;
2880                         commit++;
2881                 }
2882         }
2883
2884         return commit;
2885 }
2886
2887 /*
2888  * Called to commit link state changes noted by inspection step of
2889  * active-backup mode ARP monitor.
2890  *
2891  * Called with RTNL and bond->lock for read.
2892  */
2893 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2894 {
2895         struct slave *slave;
2896         int i;
2897         unsigned long trans_start;
2898
2899         bond_for_each_slave(bond, slave, i) {
2900                 switch (slave->new_link) {
2901                 case BOND_LINK_NOCHANGE:
2902                         continue;
2903
2904                 case BOND_LINK_UP:
2905                         trans_start = dev_trans_start(slave->dev);
2906                         if ((!bond->curr_active_slave &&
2907                              time_in_range(jiffies,
2908                                            trans_start - delta_in_ticks,
2909                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2910                             bond->curr_active_slave != slave) {
2911                                 slave->link = BOND_LINK_UP;
2912                                 if (bond->current_arp_slave) {
2913                                         bond_set_slave_inactive_flags(
2914                                                 bond->current_arp_slave);
2915                                         bond->current_arp_slave = NULL;
2916                                 }
2917
2918                                 pr_info("%s: link status definitely up for interface %s.\n",
2919                                         bond->dev->name, slave->dev->name);
2920
2921                                 if (!bond->curr_active_slave ||
2922                                     (slave == bond->primary_slave))
2923                                         goto do_failover;
2924
2925                         }
2926
2927                         continue;
2928
2929                 case BOND_LINK_DOWN:
2930                         if (slave->link_failure_count < UINT_MAX)
2931                                 slave->link_failure_count++;
2932
2933                         slave->link = BOND_LINK_DOWN;
2934                         bond_set_slave_inactive_flags(slave);
2935
2936                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2937                                 bond->dev->name, slave->dev->name);
2938
2939                         if (slave == bond->curr_active_slave) {
2940                                 bond->current_arp_slave = NULL;
2941                                 goto do_failover;
2942                         }
2943
2944                         continue;
2945
2946                 default:
2947                         pr_err("%s: impossible: new_link %d on slave %s\n",
2948                                bond->dev->name, slave->new_link,
2949                                slave->dev->name);
2950                         continue;
2951                 }
2952
2953 do_failover:
2954                 ASSERT_RTNL();
2955                 block_netpoll_tx();
2956                 write_lock_bh(&bond->curr_slave_lock);
2957                 bond_select_active_slave(bond);
2958                 write_unlock_bh(&bond->curr_slave_lock);
2959                 unblock_netpoll_tx();
2960         }
2961
2962         bond_set_carrier(bond);
2963 }
2964
2965 /*
2966  * Send ARP probes for active-backup mode ARP monitor.
2967  *
2968  * Called with bond->lock held for read.
2969  */
2970 static void bond_ab_arp_probe(struct bonding *bond)
2971 {
2972         struct slave *slave;
2973         int i;
2974
2975         read_lock(&bond->curr_slave_lock);
2976
2977         if (bond->current_arp_slave && bond->curr_active_slave)
2978                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2979                         bond->current_arp_slave->dev->name,
2980                         bond->curr_active_slave->dev->name);
2981
2982         if (bond->curr_active_slave) {
2983                 bond_arp_send_all(bond, bond->curr_active_slave);
2984                 read_unlock(&bond->curr_slave_lock);
2985                 return;
2986         }
2987
2988         read_unlock(&bond->curr_slave_lock);
2989
2990         /* if we don't have a curr_active_slave, search for the next available
2991          * backup slave from the current_arp_slave and make it the candidate
2992          * for becoming the curr_active_slave
2993          */
2994
2995         if (!bond->current_arp_slave) {
2996                 bond->current_arp_slave = bond->first_slave;
2997                 if (!bond->current_arp_slave)
2998                         return;
2999         }
3000
3001         bond_set_slave_inactive_flags(bond->current_arp_slave);
3002
3003         /* search for next candidate */
3004         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3005                 if (IS_UP(slave->dev)) {
3006                         slave->link = BOND_LINK_BACK;
3007                         bond_set_slave_active_flags(slave);
3008                         bond_arp_send_all(bond, slave);
3009                         slave->jiffies = jiffies;
3010                         bond->current_arp_slave = slave;
3011                         break;
3012                 }
3013
3014                 /* if the link state is up at this point, we
3015                  * mark it down - this can happen if we have
3016                  * simultaneous link failures and
3017                  * reselect_active_interface doesn't make this
3018                  * one the current slave so it is still marked
3019                  * up when it is actually down
3020                  */
3021                 if (slave->link == BOND_LINK_UP) {
3022                         slave->link = BOND_LINK_DOWN;
3023                         if (slave->link_failure_count < UINT_MAX)
3024                                 slave->link_failure_count++;
3025
3026                         bond_set_slave_inactive_flags(slave);
3027
3028                         pr_info("%s: backup interface %s is now down.\n",
3029                                 bond->dev->name, slave->dev->name);
3030                 }
3031         }
3032 }
3033
3034 void bond_activebackup_arp_mon(struct work_struct *work)
3035 {
3036         struct bonding *bond = container_of(work, struct bonding,
3037                                             arp_work.work);
3038         bool should_notify_peers = false;
3039         int delta_in_ticks;
3040
3041         read_lock(&bond->lock);
3042
3043         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3044
3045         if (bond->slave_cnt == 0)
3046                 goto re_arm;
3047
3048         should_notify_peers = bond_should_notify_peers(bond);
3049
3050         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3051                 read_unlock(&bond->lock);
3052
3053                 /* Race avoidance with bond_close flush of workqueue */
3054                 if (!rtnl_trylock()) {
3055                         read_lock(&bond->lock);
3056                         delta_in_ticks = 1;
3057                         should_notify_peers = false;
3058                         goto re_arm;
3059                 }
3060
3061                 read_lock(&bond->lock);
3062
3063                 bond_ab_arp_commit(bond, delta_in_ticks);
3064
3065                 read_unlock(&bond->lock);
3066                 rtnl_unlock();
3067                 read_lock(&bond->lock);
3068         }
3069
3070         bond_ab_arp_probe(bond);
3071
3072 re_arm:
3073         if (bond->params.arp_interval)
3074                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3075
3076         read_unlock(&bond->lock);
3077
3078         if (should_notify_peers) {
3079                 if (!rtnl_trylock()) {
3080                         read_lock(&bond->lock);
3081                         bond->send_peer_notif++;
3082                         read_unlock(&bond->lock);
3083                         return;
3084                 }
3085                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3086                 rtnl_unlock();
3087         }
3088 }
3089
3090 /*-------------------------- netdev event handling --------------------------*/
3091
3092 /*
3093  * Change device name
3094  */
3095 static int bond_event_changename(struct bonding *bond)
3096 {
3097         bond_remove_proc_entry(bond);
3098         bond_create_proc_entry(bond);
3099
3100         bond_debug_reregister(bond);
3101
3102         return NOTIFY_DONE;
3103 }
3104
3105 static int bond_master_netdev_event(unsigned long event,
3106                                     struct net_device *bond_dev)
3107 {
3108         struct bonding *event_bond = netdev_priv(bond_dev);
3109
3110         switch (event) {
3111         case NETDEV_CHANGENAME:
3112                 return bond_event_changename(event_bond);
3113         case NETDEV_UNREGISTER:
3114                 bond_remove_proc_entry(event_bond);
3115                 break;
3116         case NETDEV_REGISTER:
3117                 bond_create_proc_entry(event_bond);
3118                 break;
3119         default:
3120                 break;
3121         }
3122
3123         return NOTIFY_DONE;
3124 }
3125
3126 static int bond_slave_netdev_event(unsigned long event,
3127                                    struct net_device *slave_dev)
3128 {
3129         struct slave *slave = bond_slave_get_rtnl(slave_dev);
3130         struct bonding *bond;
3131         struct net_device *bond_dev;
3132         u32 old_speed;
3133         u8 old_duplex;
3134
3135         /* A netdev event can be generated while enslaving a device
3136          * before netdev_rx_handler_register is called in which case
3137          * slave will be NULL
3138          */
3139         if (!slave)
3140                 return NOTIFY_DONE;
3141         bond_dev = slave->bond->dev;
3142         bond = slave->bond;
3143
3144         switch (event) {
3145         case NETDEV_UNREGISTER:
3146                 if (bond_dev->type != ARPHRD_ETHER)
3147                         bond_release_and_destroy(bond_dev, slave_dev);
3148                 else
3149                         bond_release(bond_dev, slave_dev);
3150                 break;
3151         case NETDEV_UP:
3152         case NETDEV_CHANGE:
3153                 old_speed = slave->speed;
3154                 old_duplex = slave->duplex;
3155
3156                 bond_update_speed_duplex(slave);
3157
3158                 if (bond->params.mode == BOND_MODE_8023AD) {
3159                         if (old_speed != slave->speed)
3160                                 bond_3ad_adapter_speed_changed(slave);
3161                         if (old_duplex != slave->duplex)
3162                                 bond_3ad_adapter_duplex_changed(slave);
3163                 }
3164                 break;
3165         case NETDEV_DOWN:
3166                 /*
3167                  * ... Or is it this?
3168                  */
3169                 break;
3170         case NETDEV_CHANGEMTU:
3171                 /*
3172                  * TODO: Should slaves be allowed to
3173                  * independently alter their MTU?  For
3174                  * an active-backup bond, slaves need
3175                  * not be the same type of device, so
3176                  * MTUs may vary.  For other modes,
3177                  * slaves arguably should have the
3178                  * same MTUs. To do this, we'd need to
3179                  * take over the slave's change_mtu
3180                  * function for the duration of their
3181                  * servitude.
3182                  */
3183                 break;
3184         case NETDEV_CHANGENAME:
3185                 /*
3186                  * TODO: handle changing the primary's name
3187                  */
3188                 break;
3189         case NETDEV_FEAT_CHANGE:
3190                 bond_compute_features(bond);
3191                 break;
3192         case NETDEV_RESEND_IGMP:
3193                 /* Propagate to master device */
3194                 call_netdevice_notifiers(event, slave->bond->dev);
3195                 break;
3196         default:
3197                 break;
3198         }
3199
3200         return NOTIFY_DONE;
3201 }
3202
3203 /*
3204  * bond_netdev_event: handle netdev notifier chain events.
3205  *
3206  * This function receives events for the netdev chain.  The caller (an
3207  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3208  * locks for us to safely manipulate the slave devices (RTNL lock,
3209  * dev_probe_lock).
3210  */
3211 static int bond_netdev_event(struct notifier_block *this,
3212                              unsigned long event, void *ptr)
3213 {
3214         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3215
3216         pr_debug("event_dev: %s, event: %lx\n",
3217                  event_dev ? event_dev->name : "None",
3218                  event);
3219
3220         if (!(event_dev->priv_flags & IFF_BONDING))
3221                 return NOTIFY_DONE;
3222
3223         if (event_dev->flags & IFF_MASTER) {
3224                 pr_debug("IFF_MASTER\n");
3225                 return bond_master_netdev_event(event, event_dev);
3226         }
3227
3228         if (event_dev->flags & IFF_SLAVE) {
3229                 pr_debug("IFF_SLAVE\n");
3230                 return bond_slave_netdev_event(event, event_dev);
3231         }
3232
3233         return NOTIFY_DONE;
3234 }
3235
3236 static struct notifier_block bond_netdev_notifier = {
3237         .notifier_call = bond_netdev_event,
3238 };
3239
3240 /*---------------------------- Hashing Policies -----------------------------*/
3241
3242 /*
3243  * Hash for the output device based upon layer 2 data
3244  */
3245 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3246 {
3247         struct ethhdr *data = (struct ethhdr *)skb->data;
3248
3249         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3250                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3251
3252         return 0;
3253 }
3254
3255 /*
3256  * Hash for the output device based upon layer 2 and layer 3 data. If
3257  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3258  */
3259 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3260 {
3261         const struct ethhdr *data;
3262         const struct iphdr *iph;
3263         const struct ipv6hdr *ipv6h;
3264         u32 v6hash;
3265         const __be32 *s, *d;
3266
3267         if (skb->protocol == htons(ETH_P_IP) &&
3268             pskb_network_may_pull(skb, sizeof(*iph))) {
3269                 iph = ip_hdr(skb);
3270                 data = (struct ethhdr *)skb->data;
3271                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3272                         (data->h_dest[5] ^ data->h_source[5])) % count;
3273         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3274                    pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3275                 ipv6h = ipv6_hdr(skb);
3276                 data = (struct ethhdr *)skb->data;
3277                 s = &ipv6h->saddr.s6_addr32[0];
3278                 d = &ipv6h->daddr.s6_addr32[0];
3279                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3280                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3281                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3282         }
3283
3284         return bond_xmit_hash_policy_l2(skb, count);
3285 }
3286
3287 /*
3288  * Hash for the output device based upon layer 3 and layer 4 data. If
3289  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3290  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3291  */
3292 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3293 {
3294         u32 layer4_xor = 0;
3295         const struct iphdr *iph;
3296         const struct ipv6hdr *ipv6h;
3297         const __be32 *s, *d;
3298         const __be16 *l4 = NULL;
3299         __be16 _l4[2];
3300         int noff = skb_network_offset(skb);
3301         int poff;
3302
3303         if (skb->protocol == htons(ETH_P_IP) &&
3304             pskb_may_pull(skb, noff + sizeof(*iph))) {
3305                 iph = ip_hdr(skb);
3306                 poff = proto_ports_offset(iph->protocol);
3307
3308                 if (!ip_is_fragment(iph) && poff >= 0) {
3309                         l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3310                                                 sizeof(_l4), &_l4);
3311                         if (l4)
3312                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3313                 }
3314                 return (layer4_xor ^
3315                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3316         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3317                    pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3318                 ipv6h = ipv6_hdr(skb);
3319                 poff = proto_ports_offset(ipv6h->nexthdr);
3320                 if (poff >= 0) {
3321                         l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3322                                                 sizeof(_l4), &_l4);
3323                         if (l4)
3324                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3325                 }
3326                 s = &ipv6h->saddr.s6_addr32[0];
3327                 d = &ipv6h->daddr.s6_addr32[0];
3328                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3329                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3330                                (layer4_xor >> 8);
3331                 return layer4_xor % count;
3332         }
3333
3334         return bond_xmit_hash_policy_l2(skb, count);
3335 }
3336
3337 /*-------------------------- Device entry points ----------------------------*/
3338
3339 static void bond_work_init_all(struct bonding *bond)
3340 {
3341         INIT_DELAYED_WORK(&bond->mcast_work,
3342                           bond_resend_igmp_join_requests_delayed);
3343         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3344         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3345         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3346                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3347         else
3348                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3349         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3350 }
3351
3352 static void bond_work_cancel_all(struct bonding *bond)
3353 {
3354         cancel_delayed_work_sync(&bond->mii_work);
3355         cancel_delayed_work_sync(&bond->arp_work);
3356         cancel_delayed_work_sync(&bond->alb_work);
3357         cancel_delayed_work_sync(&bond->ad_work);
3358         cancel_delayed_work_sync(&bond->mcast_work);
3359 }
3360
3361 static int bond_open(struct net_device *bond_dev)
3362 {
3363         struct bonding *bond = netdev_priv(bond_dev);
3364         struct slave *slave;
3365         int i;
3366
3367         /* reset slave->backup and slave->inactive */
3368         read_lock(&bond->lock);
3369         if (bond->slave_cnt > 0) {
3370                 read_lock(&bond->curr_slave_lock);
3371                 bond_for_each_slave(bond, slave, i) {
3372                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3373                                 && (slave != bond->curr_active_slave)) {
3374                                 bond_set_slave_inactive_flags(slave);
3375                         } else {
3376                                 bond_set_slave_active_flags(slave);
3377                         }
3378                 }
3379                 read_unlock(&bond->curr_slave_lock);
3380         }
3381         read_unlock(&bond->lock);
3382
3383         bond_work_init_all(bond);
3384
3385         if (bond_is_lb(bond)) {
3386                 /* bond_alb_initialize must be called before the timer
3387                  * is started.
3388                  */
3389                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3390                         return -ENOMEM;
3391                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3392         }
3393
3394         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3395                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3396
3397         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3398                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3399                 if (bond->params.arp_validate)
3400                         bond->recv_probe = bond_arp_rcv;
3401         }
3402
3403         if (bond->params.mode == BOND_MODE_8023AD) {
3404                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3405                 /* register to receive LACPDUs */
3406                 bond->recv_probe = bond_3ad_lacpdu_recv;
3407                 bond_3ad_initiate_agg_selection(bond, 1);
3408         }
3409
3410         return 0;
3411 }
3412
3413 static int bond_close(struct net_device *bond_dev)
3414 {
3415         struct bonding *bond = netdev_priv(bond_dev);
3416
3417         write_lock_bh(&bond->lock);
3418         bond->send_peer_notif = 0;
3419         write_unlock_bh(&bond->lock);
3420
3421         bond_work_cancel_all(bond);
3422         if (bond_is_lb(bond)) {
3423                 /* Must be called only after all
3424                  * slaves have been released
3425                  */
3426                 bond_alb_deinitialize(bond);
3427         }
3428         bond->recv_probe = NULL;
3429
3430         return 0;
3431 }
3432
3433 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3434                                                 struct rtnl_link_stats64 *stats)
3435 {
3436         struct bonding *bond = netdev_priv(bond_dev);
3437         struct rtnl_link_stats64 temp;
3438         struct slave *slave;
3439         int i;
3440
3441         memset(stats, 0, sizeof(*stats));
3442
3443         read_lock_bh(&bond->lock);
3444
3445         bond_for_each_slave(bond, slave, i) {
3446                 const struct rtnl_link_stats64 *sstats =
3447                         dev_get_stats(slave->dev, &temp);
3448
3449                 stats->rx_packets += sstats->rx_packets;
3450                 stats->rx_bytes += sstats->rx_bytes;
3451                 stats->rx_errors += sstats->rx_errors;
3452                 stats->rx_dropped += sstats->rx_dropped;
3453
3454                 stats->tx_packets += sstats->tx_packets;
3455                 stats->tx_bytes += sstats->tx_bytes;
3456                 stats->tx_errors += sstats->tx_errors;
3457                 stats->tx_dropped += sstats->tx_dropped;
3458
3459                 stats->multicast += sstats->multicast;
3460                 stats->collisions += sstats->collisions;
3461
3462                 stats->rx_length_errors += sstats->rx_length_errors;
3463                 stats->rx_over_errors += sstats->rx_over_errors;
3464                 stats->rx_crc_errors += sstats->rx_crc_errors;
3465                 stats->rx_frame_errors += sstats->rx_frame_errors;
3466                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3467                 stats->rx_missed_errors += sstats->rx_missed_errors;
3468
3469                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3470                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3471                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3472                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3473                 stats->tx_window_errors += sstats->tx_window_errors;
3474         }
3475
3476         read_unlock_bh(&bond->lock);
3477
3478         return stats;
3479 }
3480
3481 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3482 {
3483         struct net_device *slave_dev = NULL;
3484         struct ifbond k_binfo;
3485         struct ifbond __user *u_binfo = NULL;
3486         struct ifslave k_sinfo;
3487         struct ifslave __user *u_sinfo = NULL;
3488         struct mii_ioctl_data *mii = NULL;
3489         struct net *net;
3490         int res = 0;
3491
3492         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3493
3494         switch (cmd) {
3495         case SIOCGMIIPHY:
3496                 mii = if_mii(ifr);
3497                 if (!mii)
3498                         return -EINVAL;
3499
3500                 mii->phy_id = 0;
3501                 /* Fall Through */
3502         case SIOCGMIIREG:
3503                 /*
3504                  * We do this again just in case we were called by SIOCGMIIREG
3505                  * instead of SIOCGMIIPHY.
3506                  */
3507                 mii = if_mii(ifr);
3508                 if (!mii)
3509                         return -EINVAL;
3510
3511
3512                 if (mii->reg_num == 1) {
3513                         struct bonding *bond = netdev_priv(bond_dev);
3514                         mii->val_out = 0;
3515                         read_lock(&bond->lock);
3516                         read_lock(&bond->curr_slave_lock);
3517                         if (netif_carrier_ok(bond->dev))
3518                                 mii->val_out = BMSR_LSTATUS;
3519
3520                         read_unlock(&bond->curr_slave_lock);
3521                         read_unlock(&bond->lock);
3522                 }
3523
3524                 return 0;
3525         case BOND_INFO_QUERY_OLD:
3526         case SIOCBONDINFOQUERY:
3527                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3528
3529                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3530                         return -EFAULT;
3531
3532                 res = bond_info_query(bond_dev, &k_binfo);
3533                 if (res == 0 &&
3534                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3535                         return -EFAULT;
3536
3537                 return res;
3538         case BOND_SLAVE_INFO_QUERY_OLD:
3539         case SIOCBONDSLAVEINFOQUERY:
3540                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3541
3542                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3543                         return -EFAULT;
3544
3545                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3546                 if (res == 0 &&
3547                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3548                         return -EFAULT;
3549
3550                 return res;
3551         default:
3552                 /* Go on */
3553                 break;
3554         }
3555
3556         net = dev_net(bond_dev);
3557
3558         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3559                 return -EPERM;
3560
3561         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3562
3563         pr_debug("slave_dev=%p:\n", slave_dev);
3564
3565         if (!slave_dev)
3566                 res = -ENODEV;
3567         else {
3568                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3569                 switch (cmd) {
3570                 case BOND_ENSLAVE_OLD:
3571                 case SIOCBONDENSLAVE:
3572                         res = bond_enslave(bond_dev, slave_dev);
3573                         break;
3574                 case BOND_RELEASE_OLD:
3575                 case SIOCBONDRELEASE:
3576                         res = bond_release(bond_dev, slave_dev);
3577                         break;
3578                 case BOND_SETHWADDR_OLD:
3579                 case SIOCBONDSETHWADDR:
3580                         bond_set_dev_addr(bond_dev, slave_dev);
3581                         res = 0;
3582                         break;
3583                 case BOND_CHANGE_ACTIVE_OLD:
3584                 case SIOCBONDCHANGEACTIVE:
3585                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3586                         break;
3587                 default:
3588                         res = -EOPNOTSUPP;
3589                 }
3590
3591                 dev_put(slave_dev);
3592         }
3593
3594         return res;
3595 }
3596
3597 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3598 {
3599         struct bonding *bond = netdev_priv(bond_dev);
3600
3601         if (change & IFF_PROMISC)
3602                 bond_set_promiscuity(bond,
3603                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3604
3605         if (change & IFF_ALLMULTI)
3606                 bond_set_allmulti(bond,
3607                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3608 }
3609
3610 static void bond_set_rx_mode(struct net_device *bond_dev)
3611 {
3612         struct bonding *bond = netdev_priv(bond_dev);
3613         struct slave *slave;
3614         int i;
3615
3616         read_lock(&bond->lock);
3617
3618         if (USES_PRIMARY(bond->params.mode)) {
3619                 read_lock(&bond->curr_slave_lock);
3620                 slave = bond->curr_active_slave;
3621                 if (slave) {
3622                         dev_uc_sync(slave->dev, bond_dev);
3623                         dev_mc_sync(slave->dev, bond_dev);
3624                 }
3625                 read_unlock(&bond->curr_slave_lock);
3626         } else {
3627                 bond_for_each_slave(bond, slave, i) {
3628                         dev_uc_sync_multiple(slave->dev, bond_dev);
3629                         dev_mc_sync_multiple(slave->dev, bond_dev);
3630                 }
3631         }
3632
3633         read_unlock(&bond->lock);
3634 }
3635
3636 static int bond_neigh_init(struct neighbour *n)
3637 {
3638         struct bonding *bond = netdev_priv(n->dev);
3639         struct slave *slave = bond->first_slave;
3640         const struct net_device_ops *slave_ops;
3641         struct neigh_parms parms;
3642         int ret;
3643
3644         if (!slave)
3645                 return 0;
3646
3647         slave_ops = slave->dev->netdev_ops;
3648
3649         if (!slave_ops->ndo_neigh_setup)
3650                 return 0;
3651
3652         parms.neigh_setup = NULL;
3653         parms.neigh_cleanup = NULL;
3654         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3655         if (ret)
3656                 return ret;
3657
3658         /*
3659          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3660          * after the last slave has been detached.  Assumes that all slaves
3661          * utilize the same neigh_cleanup (true at this writing as only user
3662          * is ipoib).
3663          */
3664         n->parms->neigh_cleanup = parms.neigh_cleanup;
3665
3666         if (!parms.neigh_setup)
3667                 return 0;
3668
3669         return parms.neigh_setup(n);
3670 }
3671
3672 /*
3673  * The bonding ndo_neigh_setup is called at init time beofre any
3674  * slave exists. So we must declare proxy setup function which will
3675  * be used at run time to resolve the actual slave neigh param setup.
3676  */
3677 static int bond_neigh_setup(struct net_device *dev,
3678                             struct neigh_parms *parms)
3679 {
3680         parms->neigh_setup   = bond_neigh_init;
3681
3682         return 0;
3683 }
3684
3685 /*
3686  * Change the MTU of all of a master's slaves to match the master
3687  */
3688 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3689 {
3690         struct bonding *bond = netdev_priv(bond_dev);
3691         struct slave *slave, *stop_at;
3692         int res = 0;
3693         int i;
3694
3695         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3696                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3697
3698         /* Can't hold bond->lock with bh disabled here since
3699          * some base drivers panic. On the other hand we can't
3700          * hold bond->lock without bh disabled because we'll
3701          * deadlock. The only solution is to rely on the fact
3702          * that we're under rtnl_lock here, and the slaves
3703          * list won't change. This doesn't solve the problem
3704          * of setting the slave's MTU while it is
3705          * transmitting, but the assumption is that the base
3706          * driver can handle that.
3707          *
3708          * TODO: figure out a way to safely iterate the slaves
3709          * list, but without holding a lock around the actual
3710          * call to the base driver.
3711          */
3712
3713         bond_for_each_slave(bond, slave, i) {
3714                 pr_debug("s %p s->p %p c_m %p\n",
3715                          slave,
3716                          slave->prev,
3717                          slave->dev->netdev_ops->ndo_change_mtu);
3718
3719                 res = dev_set_mtu(slave->dev, new_mtu);
3720
3721                 if (res) {
3722                         /* If we failed to set the slave's mtu to the new value
3723                          * we must abort the operation even in ACTIVE_BACKUP
3724                          * mode, because if we allow the backup slaves to have
3725                          * different mtu values than the active slave we'll
3726                          * need to change their mtu when doing a failover. That
3727                          * means changing their mtu from timer context, which
3728                          * is probably not a good idea.
3729                          */
3730                         pr_debug("err %d %s\n", res, slave->dev->name);
3731                         goto unwind;
3732                 }
3733         }
3734
3735         bond_dev->mtu = new_mtu;
3736
3737         return 0;
3738
3739 unwind:
3740         /* unwind from head to the slave that failed */
3741         stop_at = slave;
3742         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3743                 int tmp_res;
3744
3745                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3746                 if (tmp_res) {
3747                         pr_debug("unwind err %d dev %s\n",
3748                                  tmp_res, slave->dev->name);
3749                 }
3750         }
3751
3752         return res;
3753 }
3754
3755 /*
3756  * Change HW address
3757  *
3758  * Note that many devices must be down to change the HW address, and
3759  * downing the master releases all slaves.  We can make bonds full of
3760  * bonding devices to test this, however.
3761  */
3762 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3763 {
3764         struct bonding *bond = netdev_priv(bond_dev);
3765         struct sockaddr *sa = addr, tmp_sa;
3766         struct slave *slave, *stop_at;
3767         int res = 0;
3768         int i;
3769
3770         if (bond->params.mode == BOND_MODE_ALB)
3771                 return bond_alb_set_mac_address(bond_dev, addr);
3772
3773
3774         pr_debug("bond=%p, name=%s\n",
3775                  bond, bond_dev ? bond_dev->name : "None");
3776
3777         /* If fail_over_mac is enabled, do nothing and return success.
3778          * Returning an error causes ifenslave to fail.
3779          */
3780         if (bond->params.fail_over_mac)
3781                 return 0;
3782
3783         if (!is_valid_ether_addr(sa->sa_data))
3784                 return -EADDRNOTAVAIL;
3785
3786         /* Can't hold bond->lock with bh disabled here since
3787          * some base drivers panic. On the other hand we can't
3788          * hold bond->lock without bh disabled because we'll
3789          * deadlock. The only solution is to rely on the fact
3790          * that we're under rtnl_lock here, and the slaves
3791          * list won't change. This doesn't solve the problem
3792          * of setting the slave's hw address while it is
3793          * transmitting, but the assumption is that the base
3794          * driver can handle that.
3795          *
3796          * TODO: figure out a way to safely iterate the slaves
3797          * list, but without holding a lock around the actual
3798          * call to the base driver.
3799          */
3800
3801         bond_for_each_slave(bond, slave, i) {
3802                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3803                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3804
3805                 if (slave_ops->ndo_set_mac_address == NULL) {
3806                         res = -EOPNOTSUPP;
3807                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3808                         goto unwind;
3809                 }
3810
3811                 res = dev_set_mac_address(slave->dev, addr);
3812                 if (res) {
3813                         /* TODO: consider downing the slave
3814                          * and retry ?
3815                          * User should expect communications
3816                          * breakage anyway until ARP finish
3817                          * updating, so...
3818                          */
3819                         pr_debug("err %d %s\n", res, slave->dev->name);
3820                         goto unwind;
3821                 }
3822         }
3823
3824         /* success */
3825         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3826         return 0;
3827
3828 unwind:
3829         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3830         tmp_sa.sa_family = bond_dev->type;
3831
3832         /* unwind from head to the slave that failed */
3833         stop_at = slave;
3834         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3835                 int tmp_res;
3836
3837                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3838                 if (tmp_res) {
3839                         pr_debug("unwind err %d dev %s\n",
3840                                  tmp_res, slave->dev->name);
3841                 }
3842         }
3843
3844         return res;
3845 }
3846
3847 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3848 {
3849         struct bonding *bond = netdev_priv(bond_dev);
3850         struct slave *slave, *start_at;
3851         int i, slave_no, res = 1;
3852         struct iphdr *iph = ip_hdr(skb);
3853
3854         /*
3855          * Start with the curr_active_slave that joined the bond as the
3856          * default for sending IGMP traffic.  For failover purposes one
3857          * needs to maintain some consistency for the interface that will
3858          * send the join/membership reports.  The curr_active_slave found
3859          * will send all of this type of traffic.
3860          */
3861         if ((iph->protocol == IPPROTO_IGMP) &&
3862             (skb->protocol == htons(ETH_P_IP))) {
3863
3864                 read_lock(&bond->curr_slave_lock);
3865                 slave = bond->curr_active_slave;
3866                 read_unlock(&bond->curr_slave_lock);
3867
3868                 if (!slave)
3869                         goto out;
3870         } else {
3871                 /*
3872                  * Concurrent TX may collide on rr_tx_counter; we accept
3873                  * that as being rare enough not to justify using an
3874                  * atomic op here.
3875                  */
3876                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3877
3878                 bond_for_each_slave(bond, slave, i) {
3879                         slave_no--;
3880                         if (slave_no < 0)
3881                                 break;
3882                 }
3883         }
3884
3885         start_at = slave;
3886         bond_for_each_slave_from(bond, slave, i, start_at) {
3887                 if (IS_UP(slave->dev) &&
3888                     (slave->link == BOND_LINK_UP) &&
3889                     bond_is_active_slave(slave)) {
3890                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3891                         break;
3892                 }
3893         }
3894
3895 out:
3896         if (res) {
3897                 /* no suitable interface, frame not sent */
3898                 kfree_skb(skb);
3899         }
3900
3901         return NETDEV_TX_OK;
3902 }
3903
3904
3905 /*
3906  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3907  * the bond has a usable interface.
3908  */
3909 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3910 {
3911         struct bonding *bond = netdev_priv(bond_dev);
3912         int res = 1;
3913
3914         read_lock(&bond->curr_slave_lock);
3915
3916         if (bond->curr_active_slave)
3917                 res = bond_dev_queue_xmit(bond, skb,
3918                         bond->curr_active_slave->dev);
3919
3920         read_unlock(&bond->curr_slave_lock);
3921
3922         if (res)
3923                 /* no suitable interface, frame not sent */
3924                 kfree_skb(skb);
3925
3926         return NETDEV_TX_OK;
3927 }
3928
3929 /*
3930  * In bond_xmit_xor() , we determine the output device by using a pre-
3931  * determined xmit_hash_policy(), If the selected device is not enabled,
3932  * find the next active slave.
3933  */
3934 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3935 {
3936         struct bonding *bond = netdev_priv(bond_dev);
3937         struct slave *slave, *start_at;
3938         int slave_no;
3939         int i;
3940         int res = 1;
3941
3942         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
3943
3944         bond_for_each_slave(bond, slave, i) {
3945                 slave_no--;
3946                 if (slave_no < 0)
3947                         break;
3948         }
3949
3950         start_at = slave;
3951
3952         bond_for_each_slave_from(bond, slave, i, start_at) {
3953                 if (IS_UP(slave->dev) &&
3954                     (slave->link == BOND_LINK_UP) &&
3955                     bond_is_active_slave(slave)) {
3956                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3957                         break;
3958                 }
3959         }
3960
3961         if (res) {
3962                 /* no suitable interface, frame not sent */
3963                 kfree_skb(skb);
3964         }
3965
3966         return NETDEV_TX_OK;
3967 }
3968
3969 /*
3970  * in broadcast mode, we send everything to all usable interfaces.
3971  */
3972 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3973 {
3974         struct bonding *bond = netdev_priv(bond_dev);
3975         struct slave *slave, *start_at;
3976         struct net_device *tx_dev = NULL;
3977         int i;
3978         int res = 1;
3979
3980         read_lock(&bond->curr_slave_lock);
3981         start_at = bond->curr_active_slave;
3982         read_unlock(&bond->curr_slave_lock);
3983
3984         if (!start_at)
3985                 goto out;
3986
3987         bond_for_each_slave_from(bond, slave, i, start_at) {
3988                 if (IS_UP(slave->dev) &&
3989                     (slave->link == BOND_LINK_UP) &&
3990                     bond_is_active_slave(slave)) {
3991                         if (tx_dev) {
3992                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3993                                 if (!skb2) {
3994                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3995                                                bond_dev->name);
3996                                         continue;
3997                                 }
3998
3999                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4000                                 if (res) {
4001                                         kfree_skb(skb2);
4002                                         continue;
4003                                 }
4004                         }
4005                         tx_dev = slave->dev;
4006                 }
4007         }
4008
4009         if (tx_dev)
4010                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4011
4012 out:
4013         if (res)
4014                 /* no suitable interface, frame not sent */
4015                 kfree_skb(skb);
4016
4017         /* frame sent to all suitable interfaces */
4018         return NETDEV_TX_OK;
4019 }
4020
4021 /*------------------------- Device initialization ---------------------------*/
4022
4023 static void bond_set_xmit_hash_policy(struct bonding *bond)
4024 {
4025         switch (bond->params.xmit_policy) {
4026         case BOND_XMIT_POLICY_LAYER23:
4027                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4028                 break;
4029         case BOND_XMIT_POLICY_LAYER34:
4030                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4031                 break;
4032         case BOND_XMIT_POLICY_LAYER2:
4033         default:
4034                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4035                 break;
4036         }
4037 }
4038
4039 /*
4040  * Lookup the slave that corresponds to a qid
4041  */
4042 static inline int bond_slave_override(struct bonding *bond,
4043                                       struct sk_buff *skb)
4044 {
4045         int i, res = 1;
4046         struct slave *slave = NULL;
4047         struct slave *check_slave;
4048
4049         if (!skb->queue_mapping)
4050                 return 1;
4051
4052         /* Find out if any slaves have the same mapping as this skb. */
4053         bond_for_each_slave(bond, check_slave, i) {
4054                 if (check_slave->queue_id == skb->queue_mapping) {
4055                         slave = check_slave;
4056                         break;
4057                 }
4058         }
4059
4060         /* If the slave isn't UP, use default transmit policy. */
4061         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4062             (slave->link == BOND_LINK_UP)) {
4063                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4064         }
4065
4066         return res;
4067 }
4068
4069
4070 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4071 {
4072         /*
4073          * This helper function exists to help dev_pick_tx get the correct
4074          * destination queue.  Using a helper function skips a call to
4075          * skb_tx_hash and will put the skbs in the queue we expect on their
4076          * way down to the bonding driver.
4077          */
4078         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4079
4080         /*
4081          * Save the original txq to restore before passing to the driver
4082          */
4083         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4084
4085         if (unlikely(txq >= dev->real_num_tx_queues)) {
4086                 do {
4087                         txq -= dev->real_num_tx_queues;
4088                 } while (txq >= dev->real_num_tx_queues);
4089         }
4090         return txq;
4091 }
4092
4093 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4094 {
4095         struct bonding *bond = netdev_priv(dev);
4096
4097         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4098                 if (!bond_slave_override(bond, skb))
4099                         return NETDEV_TX_OK;
4100         }
4101
4102         switch (bond->params.mode) {
4103         case BOND_MODE_ROUNDROBIN:
4104                 return bond_xmit_roundrobin(skb, dev);
4105         case BOND_MODE_ACTIVEBACKUP:
4106                 return bond_xmit_activebackup(skb, dev);
4107         case BOND_MODE_XOR:
4108                 return bond_xmit_xor(skb, dev);
4109         case BOND_MODE_BROADCAST:
4110                 return bond_xmit_broadcast(skb, dev);
4111         case BOND_MODE_8023AD:
4112                 return bond_3ad_xmit_xor(skb, dev);
4113         case BOND_MODE_ALB:
4114         case BOND_MODE_TLB:
4115                 return bond_alb_xmit(skb, dev);
4116         default:
4117                 /* Should never happen, mode already checked */
4118                 pr_err("%s: Error: Unknown bonding mode %d\n",
4119                        dev->name, bond->params.mode);
4120                 WARN_ON_ONCE(1);
4121                 kfree_skb(skb);
4122                 return NETDEV_TX_OK;
4123         }
4124 }
4125
4126 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4127 {
4128         struct bonding *bond = netdev_priv(dev);
4129         netdev_tx_t ret = NETDEV_TX_OK;
4130
4131         /*
4132          * If we risk deadlock from transmitting this in the
4133          * netpoll path, tell netpoll to queue the frame for later tx
4134          */
4135         if (is_netpoll_tx_blocked(dev))
4136                 return NETDEV_TX_BUSY;
4137
4138         read_lock(&bond->lock);
4139
4140         if (bond->slave_cnt)
4141                 ret = __bond_start_xmit(skb, dev);
4142         else
4143                 kfree_skb(skb);
4144
4145         read_unlock(&bond->lock);
4146
4147         return ret;
4148 }
4149
4150 /*
4151  * set bond mode specific net device operations
4152  */
4153 void bond_set_mode_ops(struct bonding *bond, int mode)
4154 {
4155         struct net_device *bond_dev = bond->dev;
4156
4157         switch (mode) {
4158         case BOND_MODE_ROUNDROBIN:
4159                 break;
4160         case BOND_MODE_ACTIVEBACKUP:
4161                 break;
4162         case BOND_MODE_XOR:
4163                 bond_set_xmit_hash_policy(bond);
4164                 break;
4165         case BOND_MODE_BROADCAST:
4166                 break;
4167         case BOND_MODE_8023AD:
4168                 bond_set_xmit_hash_policy(bond);
4169                 break;
4170         case BOND_MODE_ALB:
4171                 /* FALLTHRU */
4172         case BOND_MODE_TLB:
4173                 break;
4174         default:
4175                 /* Should never happen, mode already checked */
4176                 pr_err("%s: Error: Unknown bonding mode %d\n",
4177                        bond_dev->name, mode);
4178                 break;
4179         }
4180 }
4181
4182 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4183                                      struct ethtool_cmd *ecmd)
4184 {
4185         struct bonding *bond = netdev_priv(bond_dev);
4186         struct slave *slave;
4187         int i;
4188         unsigned long speed = 0;
4189
4190         ecmd->duplex = DUPLEX_UNKNOWN;
4191         ecmd->port = PORT_OTHER;
4192
4193         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4194          * do not need to check mode.  Though link speed might not represent
4195          * the true receive or transmit bandwidth (not all modes are symmetric)
4196          * this is an accurate maximum.
4197          */
4198         read_lock(&bond->lock);
4199         bond_for_each_slave(bond, slave, i) {
4200                 if (SLAVE_IS_OK(slave)) {
4201                         if (slave->speed != SPEED_UNKNOWN)
4202                                 speed += slave->speed;
4203                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4204                             slave->duplex != DUPLEX_UNKNOWN)
4205                                 ecmd->duplex = slave->duplex;
4206                 }
4207         }
4208         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4209         read_unlock(&bond->lock);
4210         return 0;
4211 }
4212
4213 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4214                                      struct ethtool_drvinfo *drvinfo)
4215 {
4216         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4217         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4218         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4219                  BOND_ABI_VERSION);
4220 }
4221
4222 static const struct ethtool_ops bond_ethtool_ops = {
4223         .get_drvinfo            = bond_ethtool_get_drvinfo,
4224         .get_settings           = bond_ethtool_get_settings,
4225         .get_link               = ethtool_op_get_link,
4226 };
4227
4228 static const struct net_device_ops bond_netdev_ops = {
4229         .ndo_init               = bond_init,
4230         .ndo_uninit             = bond_uninit,
4231         .ndo_open               = bond_open,
4232         .ndo_stop               = bond_close,
4233         .ndo_start_xmit         = bond_start_xmit,
4234         .ndo_select_queue       = bond_select_queue,
4235         .ndo_get_stats64        = bond_get_stats,
4236         .ndo_do_ioctl           = bond_do_ioctl,
4237         .ndo_change_rx_flags    = bond_change_rx_flags,
4238         .ndo_set_rx_mode        = bond_set_rx_mode,
4239         .ndo_change_mtu         = bond_change_mtu,
4240         .ndo_set_mac_address    = bond_set_mac_address,
4241         .ndo_neigh_setup        = bond_neigh_setup,
4242         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4243         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4244 #ifdef CONFIG_NET_POLL_CONTROLLER
4245         .ndo_netpoll_setup      = bond_netpoll_setup,
4246         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4247         .ndo_poll_controller    = bond_poll_controller,
4248 #endif
4249         .ndo_add_slave          = bond_enslave,
4250         .ndo_del_slave          = bond_release,
4251         .ndo_fix_features       = bond_fix_features,
4252 };
4253
4254 static const struct device_type bond_type = {
4255         .name = "bond",
4256 };
4257
4258 static void bond_destructor(struct net_device *bond_dev)
4259 {
4260         struct bonding *bond = netdev_priv(bond_dev);
4261         if (bond->wq)
4262                 destroy_workqueue(bond->wq);
4263         free_netdev(bond_dev);
4264 }
4265
4266 static void bond_setup(struct net_device *bond_dev)
4267 {
4268         struct bonding *bond = netdev_priv(bond_dev);
4269
4270         /* initialize rwlocks */
4271         rwlock_init(&bond->lock);
4272         rwlock_init(&bond->curr_slave_lock);
4273
4274         bond->params = bonding_defaults;
4275
4276         /* Initialize pointers */
4277         bond->dev = bond_dev;
4278         INIT_LIST_HEAD(&bond->vlan_list);
4279
4280         /* Initialize the device entry points */
4281         ether_setup(bond_dev);
4282         bond_dev->netdev_ops = &bond_netdev_ops;
4283         bond_dev->ethtool_ops = &bond_ethtool_ops;
4284         bond_set_mode_ops(bond, bond->params.mode);
4285
4286         bond_dev->destructor = bond_destructor;
4287
4288         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4289
4290         /* Initialize the device options */
4291         bond_dev->tx_queue_len = 0;
4292         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4293         bond_dev->priv_flags |= IFF_BONDING;
4294         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4295
4296         /* At first, we block adding VLANs. That's the only way to
4297          * prevent problems that occur when adding VLANs over an
4298          * empty bond. The block will be removed once non-challenged
4299          * slaves are enslaved.
4300          */
4301         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4302
4303         /* don't acquire bond device's netif_tx_lock when
4304          * transmitting */
4305         bond_dev->features |= NETIF_F_LLTX;
4306
4307         /* By default, we declare the bond to be fully
4308          * VLAN hardware accelerated capable. Special
4309          * care is taken in the various xmit functions
4310          * when there are slaves that are not hw accel
4311          * capable
4312          */
4313
4314         bond_dev->hw_features = BOND_VLAN_FEATURES |
4315                                 NETIF_F_HW_VLAN_CTAG_TX |
4316                                 NETIF_F_HW_VLAN_CTAG_RX |
4317                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4318
4319         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4320         bond_dev->features |= bond_dev->hw_features;
4321 }
4322
4323 /*
4324 * Destroy a bonding device.
4325 * Must be under rtnl_lock when this function is called.
4326 */
4327 static void bond_uninit(struct net_device *bond_dev)
4328 {
4329         struct bonding *bond = netdev_priv(bond_dev);
4330         struct vlan_entry *vlan, *tmp;
4331
4332         bond_netpoll_cleanup(bond_dev);
4333
4334         /* Release the bonded slaves */
4335         while (bond->first_slave != NULL)
4336                 __bond_release_one(bond_dev, bond->first_slave->dev, true);
4337         pr_info("%s: released all slaves\n", bond_dev->name);
4338
4339         list_del(&bond->bond_list);
4340
4341         bond_debug_unregister(bond);
4342
4343         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4344                 list_del(&vlan->vlan_list);
4345                 kfree(vlan);
4346         }
4347 }
4348
4349 /*------------------------- Module initialization ---------------------------*/
4350
4351 /*
4352  * Convert string input module parms.  Accept either the
4353  * number of the mode or its string name.  A bit complicated because
4354  * some mode names are substrings of other names, and calls from sysfs
4355  * may have whitespace in the name (trailing newlines, for example).
4356  */
4357 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4358 {
4359         int modeint = -1, i, rv;
4360         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4361
4362         for (p = (char *)buf; *p; p++)
4363                 if (!(isdigit(*p) || isspace(*p)))
4364                         break;
4365
4366         if (*p)
4367                 rv = sscanf(buf, "%20s", modestr);
4368         else
4369                 rv = sscanf(buf, "%d", &modeint);
4370
4371         if (!rv)
4372                 return -1;
4373
4374         for (i = 0; tbl[i].modename; i++) {
4375                 if (modeint == tbl[i].mode)
4376                         return tbl[i].mode;
4377                 if (strcmp(modestr, tbl[i].modename) == 0)
4378                         return tbl[i].mode;
4379         }
4380
4381         return -1;
4382 }
4383
4384 static int bond_check_params(struct bond_params *params)
4385 {
4386         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4387         int arp_all_targets_value;
4388
4389         /*
4390          * Convert string parameters.
4391          */
4392         if (mode) {
4393                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4394                 if (bond_mode == -1) {
4395                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4396                                mode == NULL ? "NULL" : mode);
4397                         return -EINVAL;
4398                 }
4399         }
4400
4401         if (xmit_hash_policy) {
4402                 if ((bond_mode != BOND_MODE_XOR) &&
4403                     (bond_mode != BOND_MODE_8023AD)) {
4404                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4405                                bond_mode_name(bond_mode));
4406                 } else {
4407                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4408                                                         xmit_hashtype_tbl);
4409                         if (xmit_hashtype == -1) {
4410                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4411                                        xmit_hash_policy == NULL ? "NULL" :
4412                                        xmit_hash_policy);
4413                                 return -EINVAL;
4414                         }
4415                 }
4416         }
4417
4418         if (lacp_rate) {
4419                 if (bond_mode != BOND_MODE_8023AD) {
4420                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4421                                 bond_mode_name(bond_mode));
4422                 } else {
4423                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4424                         if (lacp_fast == -1) {
4425                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4426                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4427                                 return -EINVAL;
4428                         }
4429                 }
4430         }
4431
4432         if (ad_select) {
4433                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4434                 if (params->ad_select == -1) {
4435                         pr_err("Error: Invalid ad_select \"%s\"\n",
4436                                ad_select == NULL ? "NULL" : ad_select);
4437                         return -EINVAL;
4438                 }
4439
4440                 if (bond_mode != BOND_MODE_8023AD) {
4441                         pr_warning("ad_select param only affects 802.3ad mode\n");
4442                 }
4443         } else {
4444                 params->ad_select = BOND_AD_STABLE;
4445         }
4446
4447         if (max_bonds < 0) {
4448                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4449                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4450                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4451         }
4452
4453         if (miimon < 0) {
4454                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4455                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4456                 miimon = BOND_LINK_MON_INTERV;
4457         }
4458
4459         if (updelay < 0) {
4460                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4461                            updelay, INT_MAX);
4462                 updelay = 0;
4463         }
4464
4465         if (downdelay < 0) {
4466                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4467                            downdelay, INT_MAX);
4468                 downdelay = 0;
4469         }
4470
4471         if ((use_carrier != 0) && (use_carrier != 1)) {
4472                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4473                            use_carrier);
4474                 use_carrier = 1;
4475         }
4476
4477         if (num_peer_notif < 0 || num_peer_notif > 255) {
4478                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4479                            num_peer_notif);
4480                 num_peer_notif = 1;
4481         }
4482
4483         /* reset values for 802.3ad */
4484         if (bond_mode == BOND_MODE_8023AD) {
4485                 if (!miimon) {
4486                         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");
4487                         pr_warning("Forcing miimon to 100msec\n");
4488                         miimon = 100;
4489                 }
4490         }
4491
4492         if (tx_queues < 1 || tx_queues > 255) {
4493                 pr_warning("Warning: tx_queues (%d) should be between "
4494                            "1 and 255, resetting to %d\n",
4495                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4496                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4497         }
4498
4499         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4500                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4501                            "not of valid value (0/1), so it was set to "
4502                            "0\n", all_slaves_active);
4503                 all_slaves_active = 0;
4504         }
4505
4506         if (resend_igmp < 0 || resend_igmp > 255) {
4507                 pr_warning("Warning: resend_igmp (%d) should be between "
4508                            "0 and 255, resetting to %d\n",
4509                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4510                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4511         }
4512
4513         /* reset values for TLB/ALB */
4514         if ((bond_mode == BOND_MODE_TLB) ||
4515             (bond_mode == BOND_MODE_ALB)) {
4516                 if (!miimon) {
4517                         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");
4518                         pr_warning("Forcing miimon to 100msec\n");
4519                         miimon = 100;
4520                 }
4521         }
4522
4523         if (bond_mode == BOND_MODE_ALB) {
4524                 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",
4525                           updelay);
4526         }
4527
4528         if (!miimon) {
4529                 if (updelay || downdelay) {
4530                         /* just warn the user the up/down delay will have
4531                          * no effect since miimon is zero...
4532                          */
4533                         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",
4534                                    updelay, downdelay);
4535                 }
4536         } else {
4537                 /* don't allow arp monitoring */
4538                 if (arp_interval) {
4539                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4540                                    miimon, arp_interval);
4541                         arp_interval = 0;
4542                 }
4543
4544                 if ((updelay % miimon) != 0) {
4545                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4546                                    updelay, miimon,
4547                                    (updelay / miimon) * miimon);
4548                 }
4549
4550                 updelay /= miimon;
4551
4552                 if ((downdelay % miimon) != 0) {
4553                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4554                                    downdelay, miimon,
4555                                    (downdelay / miimon) * miimon);
4556                 }
4557
4558                 downdelay /= miimon;
4559         }
4560
4561         if (arp_interval < 0) {
4562                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4563                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4564                 arp_interval = BOND_LINK_ARP_INTERV;
4565         }
4566
4567         for (arp_ip_count = 0, i = 0;
4568              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4569                 /* not complete check, but should be good enough to
4570                    catch mistakes */
4571                 __be32 ip = in_aton(arp_ip_target[i]);
4572                 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4573                     ip == htonl(INADDR_BROADCAST)) {
4574                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4575                                    arp_ip_target[i]);
4576                         arp_interval = 0;
4577                 } else {
4578                         if (bond_get_targets_ip(arp_target, ip) == -1)
4579                                 arp_target[arp_ip_count++] = ip;
4580                         else
4581                                 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4582                                            &ip);
4583                 }
4584         }
4585
4586         if (arp_interval && !arp_ip_count) {
4587                 /* don't allow arping if no arp_ip_target given... */
4588                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4589                            arp_interval);
4590                 arp_interval = 0;
4591         }
4592
4593         if (arp_validate) {
4594                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4595                         pr_err("arp_validate only supported in active-backup mode\n");
4596                         return -EINVAL;
4597                 }
4598                 if (!arp_interval) {
4599                         pr_err("arp_validate requires arp_interval\n");
4600                         return -EINVAL;
4601                 }
4602
4603                 arp_validate_value = bond_parse_parm(arp_validate,
4604                                                      arp_validate_tbl);
4605                 if (arp_validate_value == -1) {
4606                         pr_err("Error: invalid arp_validate \"%s\"\n",
4607                                arp_validate == NULL ? "NULL" : arp_validate);
4608                         return -EINVAL;
4609                 }
4610         } else
4611                 arp_validate_value = 0;
4612
4613         arp_all_targets_value = 0;
4614         if (arp_all_targets) {
4615                 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4616                                                         arp_all_targets_tbl);
4617
4618                 if (arp_all_targets_value == -1) {
4619                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4620                                arp_all_targets);
4621                         arp_all_targets_value = 0;
4622                 }
4623         }
4624
4625         if (miimon) {
4626                 pr_info("MII link monitoring set to %d ms\n", miimon);
4627         } else if (arp_interval) {
4628                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4629                         arp_interval,
4630                         arp_validate_tbl[arp_validate_value].modename,
4631                         arp_ip_count);
4632
4633                 for (i = 0; i < arp_ip_count; i++)
4634                         pr_info(" %s", arp_ip_target[i]);
4635
4636                 pr_info("\n");
4637
4638         } else if (max_bonds) {
4639                 /* miimon and arp_interval not set, we need one so things
4640                  * work as expected, see bonding.txt for details
4641                  */
4642                 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");
4643         }
4644
4645         if (primary && !USES_PRIMARY(bond_mode)) {
4646                 /* currently, using a primary only makes sense
4647                  * in active backup, TLB or ALB modes
4648                  */
4649                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4650                            primary, bond_mode_name(bond_mode));
4651                 primary = NULL;
4652         }
4653
4654         if (primary && primary_reselect) {
4655                 primary_reselect_value = bond_parse_parm(primary_reselect,
4656                                                          pri_reselect_tbl);
4657                 if (primary_reselect_value == -1) {
4658                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4659                                primary_reselect ==
4660                                         NULL ? "NULL" : primary_reselect);
4661                         return -EINVAL;
4662                 }
4663         } else {
4664                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4665         }
4666
4667         if (fail_over_mac) {
4668                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4669                                                       fail_over_mac_tbl);
4670                 if (fail_over_mac_value == -1) {
4671                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4672                                arp_validate == NULL ? "NULL" : arp_validate);
4673                         return -EINVAL;
4674                 }
4675
4676                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4677                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4678         } else {
4679                 fail_over_mac_value = BOND_FOM_NONE;
4680         }
4681
4682         /* fill params struct with the proper values */
4683         params->mode = bond_mode;
4684         params->xmit_policy = xmit_hashtype;
4685         params->miimon = miimon;
4686         params->num_peer_notif = num_peer_notif;
4687         params->arp_interval = arp_interval;
4688         params->arp_validate = arp_validate_value;
4689         params->arp_all_targets = arp_all_targets_value;
4690         params->updelay = updelay;
4691         params->downdelay = downdelay;
4692         params->use_carrier = use_carrier;
4693         params->lacp_fast = lacp_fast;
4694         params->primary[0] = 0;
4695         params->primary_reselect = primary_reselect_value;
4696         params->fail_over_mac = fail_over_mac_value;
4697         params->tx_queues = tx_queues;
4698         params->all_slaves_active = all_slaves_active;
4699         params->resend_igmp = resend_igmp;
4700         params->min_links = min_links;
4701
4702         if (primary) {
4703                 strncpy(params->primary, primary, IFNAMSIZ);
4704                 params->primary[IFNAMSIZ - 1] = 0;
4705         }
4706
4707         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4708
4709         return 0;
4710 }
4711
4712 static struct lock_class_key bonding_netdev_xmit_lock_key;
4713 static struct lock_class_key bonding_netdev_addr_lock_key;
4714 static struct lock_class_key bonding_tx_busylock_key;
4715
4716 static void bond_set_lockdep_class_one(struct net_device *dev,
4717                                        struct netdev_queue *txq,
4718                                        void *_unused)
4719 {
4720         lockdep_set_class(&txq->_xmit_lock,
4721                           &bonding_netdev_xmit_lock_key);
4722 }
4723
4724 static void bond_set_lockdep_class(struct net_device *dev)
4725 {
4726         lockdep_set_class(&dev->addr_list_lock,
4727                           &bonding_netdev_addr_lock_key);
4728         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4729         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4730 }
4731
4732 /*
4733  * Called from registration process
4734  */
4735 static int bond_init(struct net_device *bond_dev)
4736 {
4737         struct bonding *bond = netdev_priv(bond_dev);
4738         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4739         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4740
4741         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4742
4743         /*
4744          * Initialize locks that may be required during
4745          * en/deslave operations.  All of the bond_open work
4746          * (of which this is part) should really be moved to
4747          * a phase prior to dev_open
4748          */
4749         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4750         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4751
4752         bond->wq = create_singlethread_workqueue(bond_dev->name);
4753         if (!bond->wq)
4754                 return -ENOMEM;
4755
4756         bond_set_lockdep_class(bond_dev);
4757
4758         list_add_tail(&bond->bond_list, &bn->dev_list);
4759
4760         bond_prepare_sysfs_group(bond);
4761
4762         bond_debug_register(bond);
4763
4764         /* Ensure valid dev_addr */
4765         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4766             bond_dev->addr_assign_type == NET_ADDR_PERM)
4767                 eth_hw_addr_random(bond_dev);
4768
4769         return 0;
4770 }
4771
4772 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4773 {
4774         if (tb[IFLA_ADDRESS]) {
4775                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4776                         return -EINVAL;
4777                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4778                         return -EADDRNOTAVAIL;
4779         }
4780         return 0;
4781 }
4782
4783 static unsigned int bond_get_num_tx_queues(void)
4784 {
4785         return tx_queues;
4786 }
4787
4788 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4789         .kind                   = "bond",
4790         .priv_size              = sizeof(struct bonding),
4791         .setup                  = bond_setup,
4792         .validate               = bond_validate,
4793         .get_num_tx_queues      = bond_get_num_tx_queues,
4794         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4795                                                              as for TX queues */
4796 };
4797
4798 /* Create a new bond based on the specified name and bonding parameters.
4799  * If name is NULL, obtain a suitable "bond%d" name for us.
4800  * Caller must NOT hold rtnl_lock; we need to release it here before we
4801  * set up our sysfs entries.
4802  */
4803 int bond_create(struct net *net, const char *name)
4804 {
4805         struct net_device *bond_dev;
4806         int res;
4807
4808         rtnl_lock();
4809
4810         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4811                                    name ? name : "bond%d",
4812                                    bond_setup, tx_queues);
4813         if (!bond_dev) {
4814                 pr_err("%s: eek! can't alloc netdev!\n", name);
4815                 rtnl_unlock();
4816                 return -ENOMEM;
4817         }
4818
4819         dev_net_set(bond_dev, net);
4820         bond_dev->rtnl_link_ops = &bond_link_ops;
4821
4822         res = register_netdevice(bond_dev);
4823
4824         netif_carrier_off(bond_dev);
4825
4826         rtnl_unlock();
4827         if (res < 0)
4828                 bond_destructor(bond_dev);
4829         return res;
4830 }
4831
4832 static int __net_init bond_net_init(struct net *net)
4833 {
4834         struct bond_net *bn = net_generic(net, bond_net_id);
4835
4836         bn->net = net;
4837         INIT_LIST_HEAD(&bn->dev_list);
4838
4839         bond_create_proc_dir(bn);
4840         bond_create_sysfs(bn);
4841
4842         return 0;
4843 }
4844
4845 static void __net_exit bond_net_exit(struct net *net)
4846 {
4847         struct bond_net *bn = net_generic(net, bond_net_id);
4848         struct bonding *bond, *tmp_bond;
4849         LIST_HEAD(list);
4850
4851         bond_destroy_sysfs(bn);
4852         bond_destroy_proc_dir(bn);
4853
4854         /* Kill off any bonds created after unregistering bond rtnl ops */
4855         rtnl_lock();
4856         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4857                 unregister_netdevice_queue(bond->dev, &list);
4858         unregister_netdevice_many(&list);
4859         rtnl_unlock();
4860 }
4861
4862 static struct pernet_operations bond_net_ops = {
4863         .init = bond_net_init,
4864         .exit = bond_net_exit,
4865         .id   = &bond_net_id,
4866         .size = sizeof(struct bond_net),
4867 };
4868
4869 static int __init bonding_init(void)
4870 {
4871         int i;
4872         int res;
4873
4874         pr_info("%s", bond_version);
4875
4876         res = bond_check_params(&bonding_defaults);
4877         if (res)
4878                 goto out;
4879
4880         res = register_pernet_subsys(&bond_net_ops);
4881         if (res)
4882                 goto out;
4883
4884         res = rtnl_link_register(&bond_link_ops);
4885         if (res)
4886                 goto err_link;
4887
4888         bond_create_debugfs();
4889
4890         for (i = 0; i < max_bonds; i++) {
4891                 res = bond_create(&init_net, NULL);
4892                 if (res)
4893                         goto err;
4894         }
4895
4896         register_netdevice_notifier(&bond_netdev_notifier);
4897 out:
4898         return res;
4899 err:
4900         rtnl_link_unregister(&bond_link_ops);
4901 err_link:
4902         unregister_pernet_subsys(&bond_net_ops);
4903         goto out;
4904
4905 }
4906
4907 static void __exit bonding_exit(void)
4908 {
4909         unregister_netdevice_notifier(&bond_netdev_notifier);
4910
4911         bond_destroy_debugfs();
4912
4913         rtnl_link_unregister(&bond_link_ops);
4914         unregister_pernet_subsys(&bond_net_ops);
4915
4916 #ifdef CONFIG_NET_POLL_CONTROLLER
4917         /*
4918          * Make sure we don't have an imbalance on our netpoll blocking
4919          */
4920         WARN_ON(atomic_read(&netpoll_block_tx));
4921 #endif
4922 }
4923
4924 module_init(bonding_init);
4925 module_exit(bonding_exit);
4926 MODULE_LICENSE("GPL");
4927 MODULE_VERSION(DRV_VERSION);
4928 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4929 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4930 MODULE_ALIAS_RTNL_LINK("bond");