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 = 0;
 110 static struct bond_params bonding_defaults;
 111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 112
 113 module_param(max_bonds, int, 0);
 114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 115 module_param(tx_queues, int, 0);
 116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 117 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 118 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 119                                "failover event (alias of num_unsol_na)");
 120 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 122                                "failover event (alias of num_grat_arp)");
 123 module_param(miimon, int, 0);
 124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 125 module_param(updelay, int, 0);
 126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 127 module_param(downdelay, int, 0);
 128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 129                             "in milliseconds");
 130 module_param(use_carrier, int, 0);
 131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 132                               "0 for off, 1 for on (default)");
 133 module_param(mode, charp, 0);
 134 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 135                        "1 for active-backup, 2 for balance-xor, "
 136                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 137                        "6 for balance-alb");
 138 module_param(primary, charp, 0);
 139 MODULE_PARM_DESC(primary, "Primary network device to use");
 140 module_param(primary_reselect, charp, 0);
 141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 142                                    "once it comes up; "
 143                                    "0 for always (default), "
 144                                    "1 for only if speed of primary is "
 145                                    "better, "
 146                                    "2 for only on active slave "
 147                                    "failure");
 148 module_param(lacp_rate, charp, 0);
 149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 150                             "0 for slow, 1 for fast");
 151 module_param(ad_select, charp, 0);
 152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 153                             "0 for stable (default), 1 for bandwidth, "
 154                             "2 for count");
 155 module_param(min_links, int, 0);
 156 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 157
 158 module_param(xmit_hash_policy, charp, 0);
 159 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 160                                    "0 for layer 2 (default), 1 for layer 3+4, "
 161                                    "2 for layer 2+3");
 162 module_param(arp_interval, int, 0);
 163 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 164 module_param_array(arp_ip_target, charp, NULL, 0);
 165 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 166 module_param(arp_validate, charp, 0);
 167 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 168                                "0 for none (default), 1 for active, "
 169                                "2 for backup, 3 for all");
 170 module_param(arp_all_targets, charp, 0);
 171 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 172 module_param(fail_over_mac, charp, 0);
 173 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 174                                 "the same MAC; 0 for none (default), "
 175                                 "1 for active, 2 for follow");
 176 module_param(all_slaves_active, int, 0);
 177 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 178                                      "by setting active flag for all slaves; "
 179                                      "0 for never (default), 1 for always.");
 180 module_param(resend_igmp, int, 0);
 181 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 182                               "link failure");
 183
 184 /*----------------------------- Global variables ----------------------------*/
 185
 186 #ifdef CONFIG_NET_POLL_CONTROLLER
 187 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 188 #endif
 189
 190 int bond_net_id __read_mostly;
 191
 192 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 193 static int arp_ip_count;
 194 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 195 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 196 static int lacp_fast;
 197
 198 const struct bond_parm_tbl bond_lacp_tbl[] = {
 199 {       "slow",         AD_LACP_SLOW},
 200 {       "fast",         AD_LACP_FAST},
 201 {       NULL,           -1},
 202 };
 203
 204 const struct bond_parm_tbl bond_mode_tbl[] = {
 205 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 206 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 207 {       "balance-xor",          BOND_MODE_XOR},
 208 {       "broadcast",            BOND_MODE_BROADCAST},
 209 {       "802.3ad",              BOND_MODE_8023AD},
 210 {       "balance-tlb",          BOND_MODE_TLB},
 211 {       "balance-alb",          BOND_MODE_ALB},
 212 {       NULL,                   -1},
 213 };
 214
 215 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 216 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 217 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 218 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 219 {       NULL,                   -1},
 220 };
 221
 222 const struct bond_parm_tbl arp_all_targets_tbl[] = {
 223 {       "any",                  BOND_ARP_TARGETS_ANY},
 224 {       "all",                  BOND_ARP_TARGETS_ALL},
 225 {       NULL,                   -1},
 226 };
 227
 228 const struct bond_parm_tbl arp_validate_tbl[] = {
 229 {       "none",                 BOND_ARP_VALIDATE_NONE},
 230 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 231 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 232 {       "all",                  BOND_ARP_VALIDATE_ALL},
 233 {       NULL,                   -1},
 234 };
 235
 236 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 237 {       "none",                 BOND_FOM_NONE},
 238 {       "active",               BOND_FOM_ACTIVE},
 239 {       "follow",               BOND_FOM_FOLLOW},
 240 {       NULL,                   -1},
 241 };
 242
 243 const struct bond_parm_tbl pri_reselect_tbl[] = {
 244 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 245 {       "better",               BOND_PRI_RESELECT_BETTER},
 246 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 247 {       NULL,                   -1},
 248 };
 249
 250 struct bond_parm_tbl ad_select_tbl[] = {
 251 {       "stable",       BOND_AD_STABLE},
 252 {       "bandwidth",    BOND_AD_BANDWIDTH},
 253 {       "count",        BOND_AD_COUNT},
 254 {       NULL,           -1},
 255 };
 256
 257 /*-------------------------- Forward declarations ---------------------------*/
 258
 259 static int bond_init(struct net_device *bond_dev);
 260 static void bond_uninit(struct net_device *bond_dev);
 261
 262 /*---------------------------- General routines -----------------------------*/
 263
 264 const char *bond_mode_name(int mode)
 265 {
 266         static const char *names[] = {
 267                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 268                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 269                 [BOND_MODE_XOR] = "load balancing (xor)",
 270                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 271                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 272                 [BOND_MODE_TLB] = "transmit load balancing",
 273                 [BOND_MODE_ALB] = "adaptive load balancing",
 274         };
 275
 276         if (mode < 0 || mode > BOND_MODE_ALB)
 277                 return "unknown";
 278
 279         return names[mode];
 280 }
 281
 282 /*---------------------------------- VLAN -----------------------------------*/
 283
 284 /**
 285  * bond_add_vlan - add a new vlan id on bond
 286  * @bond: bond that got the notification
 287  * @vlan_id: the vlan id to add
 288  *
 289  * Returns -ENOMEM if allocation failed.
 290  */
 291 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 292 {
 293         struct vlan_entry *vlan;
 294
 295         pr_debug("bond: %s, vlan id %d\n",
 296                  (bond ? bond->dev->name : "None"), vlan_id);
 297
 298         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 299         if (!vlan)
 300                 return -ENOMEM;
 301
 302         INIT_LIST_HEAD(&vlan->vlan_list);
 303         vlan->vlan_id = vlan_id;
 304
 305         write_lock_bh(&bond->lock);
 306
 307         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 308
 309         write_unlock_bh(&bond->lock);
 310
 311         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 312
 313         return 0;
 314 }
 315
 316 /**
 317  * bond_del_vlan - delete a vlan id from bond
 318  * @bond: bond that got the notification
 319  * @vlan_id: the vlan id to delete
 320  *
 321  * returns -ENODEV if @vlan_id was not found in @bond.
 322  */
 323 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 324 {
 325         struct vlan_entry *vlan;
 326         int res = -ENODEV;
 327
 328         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 329
 330         block_netpoll_tx();
 331         write_lock_bh(&bond->lock);
 332
 333         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 334                 if (vlan->vlan_id == vlan_id) {
 335                         list_del(&vlan->vlan_list);
 336
 337                         if (bond_is_lb(bond))
 338                                 bond_alb_clear_vlan(bond, vlan_id);
 339
 340                         pr_debug("removed VLAN ID %d from bond %s\n",
 341                                  vlan_id, bond->dev->name);
 342
 343                         kfree(vlan);
 344
 345                         res = 0;
 346                         goto out;
 347                 }
 348         }
 349
 350         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 351                  vlan_id, bond->dev->name);
 352
 353 out:
 354         write_unlock_bh(&bond->lock);
 355         unblock_netpoll_tx();
 356         return res;
 357 }
 358
 359 /**
 360  * bond_next_vlan - safely skip to the next item in the vlans list.
 361  * @bond: the bond we're working on
 362  * @curr: item we're advancing from
 363  *
 364  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 365  * or @curr->next otherwise (even if it is @curr itself again).
 366  *
 367  * Caller must hold bond->lock
 368  */
 369 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 370 {
 371         struct vlan_entry *next, *last;
 372
 373         if (list_empty(&bond->vlan_list))
 374                 return NULL;
 375
 376         if (!curr) {
 377                 next = list_entry(bond->vlan_list.next,
 378                                   struct vlan_entry, vlan_list);
 379         } else {
 380                 last = list_entry(bond->vlan_list.prev,
 381                                   struct vlan_entry, vlan_list);
 382                 if (last == curr) {
 383                         next = list_entry(bond->vlan_list.next,
 384                                           struct vlan_entry, vlan_list);
 385                 } else {
 386                         next = list_entry(curr->vlan_list.next,
 387                                           struct vlan_entry, vlan_list);
 388                 }
 389         }
 390
 391         return next;
 392 }
 393
 394 /**
 395  * bond_dev_queue_xmit - Prepare skb for xmit.
 396  *
 397  * @bond: bond device that got this skb for tx.
 398  * @skb: hw accel VLAN tagged skb to transmit
 399  * @slave_dev: slave that is supposed to xmit this skbuff
 400  */
 401 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 402                         struct net_device *slave_dev)
 403 {
 404         skb->dev = slave_dev;
 405
 406         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 407                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 408         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 409
 410         if (unlikely(netpoll_tx_running(bond->dev)))
 411                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 412         else
 413                 dev_queue_xmit(skb);
 414
 415         return 0;
 416 }
 417
 418 /*
 419  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 420  * We don't protect the slave list iteration with a lock because:
 421  * a. This operation is performed in IOCTL context,
 422  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 423  * c. Holding a lock with BH disabled while directly calling a base driver
 424  *    entry point is generally a BAD idea.
 425  *
 426  * The design of synchronization/protection for this operation in the 8021q
 427  * module is good for one or more VLAN devices over a single physical device
 428  * and cannot be extended for a teaming solution like bonding, so there is a
 429  * potential race condition here where a net device from the vlan group might
 430  * be referenced (either by a base driver or the 8021q code) while it is being
 431  * removed from the system. However, it turns out we're not making matters
 432  * worse, and if it works for regular VLAN usage it will work here too.
 433 */
 434
 435 /**
 436  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 437  * @bond_dev: bonding net device that got called
 438  * @vid: vlan id being added
 439  */
 440 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 441                                 __be16 proto, u16 vid)
 442 {
 443         struct bonding *bond = netdev_priv(bond_dev);
 444         struct slave *slave, *stop_at;
 445         int i, res;
 446
 447         bond_for_each_slave(bond, slave, i) {
 448                 res = vlan_vid_add(slave->dev, proto, vid);
 449                 if (res)
 450                         goto unwind;
 451         }
 452
 453         res = bond_add_vlan(bond, vid);
 454         if (res) {
 455                 pr_err("%s: Error: Failed to add vlan id %d\n",
 456                        bond_dev->name, vid);
 457                 return res;
 458         }
 459
 460         return 0;
 461
 462 unwind:
 463         /* unwind from head to the slave that failed */
 464         stop_at = slave;
 465         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
 466                 vlan_vid_del(slave->dev, proto, vid);
 467
 468         return res;
 469 }
 470
 471 /**
 472  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 473  * @bond_dev: bonding net device that got called
 474  * @vid: vlan id being removed
 475  */
 476 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 477                                  __be16 proto, u16 vid)
 478 {
 479         struct bonding *bond = netdev_priv(bond_dev);
 480         struct slave *slave;
 481         int i, res;
 482
 483         bond_for_each_slave(bond, slave, i)
 484                 vlan_vid_del(slave->dev, proto, vid);
 485
 486         res = bond_del_vlan(bond, vid);
 487         if (res) {
 488                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 489                        bond_dev->name, vid);
 490                 return res;
 491         }
 492
 493         return 0;
 494 }
 495
 496 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 497 {
 498         struct vlan_entry *vlan;
 499         int res;
 500
 501         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 502                 res = vlan_vid_add(slave_dev, htons(ETH_P_8021Q),
 503                                    vlan->vlan_id);
 504                 if (res)
 505                         pr_warning("%s: Failed to add vlan id %d to device %s\n",
 506                                    bond->dev->name, vlan->vlan_id,
 507                                    slave_dev->name);
 508         }
 509 }
 510
 511 static void bond_del_vlans_from_slave(struct bonding *bond,
 512                                       struct net_device *slave_dev)
 513 {
 514         struct vlan_entry *vlan;
 515
 516         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 517                 if (!vlan->vlan_id)
 518                         continue;
 519                 vlan_vid_del(slave_dev, htons(ETH_P_8021Q), vlan->vlan_id);
 520         }
 521 }
 522
 523 /*------------------------------- Link status -------------------------------*/
 524
 525 /*
 526  * Set the carrier state for the master according to the state of its
 527  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 528  * do special 802.3ad magic.
 529  *
 530  * Returns zero if carrier state does not change, nonzero if it does.
 531  */
 532 static int bond_set_carrier(struct bonding *bond)
 533 {
 534         struct slave *slave;
 535         int i;
 536
 537         if (bond->slave_cnt == 0)
 538                 goto down;
 539
 540         if (bond->params.mode == BOND_MODE_8023AD)
 541                 return bond_3ad_set_carrier(bond);
 542
 543         bond_for_each_slave(bond, slave, i) {
 544                 if (slave->link == BOND_LINK_UP) {
 545                         if (!netif_carrier_ok(bond->dev)) {
 546                                 netif_carrier_on(bond->dev);
 547                                 return 1;
 548                         }
 549                         return 0;
 550                 }
 551         }
 552
 553 down:
 554         if (netif_carrier_ok(bond->dev)) {
 555                 netif_carrier_off(bond->dev);
 556                 return 1;
 557         }
 558         return 0;
 559 }
 560
 561 /*
 562  * Get link speed and duplex from the slave's base driver
 563  * using ethtool. If for some reason the call fails or the
 564  * values are invalid, set speed and duplex to -1,
 565  * and return.
 566  */
 567 static void bond_update_speed_duplex(struct slave *slave)
 568 {
 569         struct net_device *slave_dev = slave->dev;
 570         struct ethtool_cmd ecmd;
 571         u32 slave_speed;
 572         int res;
 573
 574         slave->speed = SPEED_UNKNOWN;
 575         slave->duplex = DUPLEX_UNKNOWN;
 576
 577         res = __ethtool_get_settings(slave_dev, &ecmd);
 578         if (res < 0)
 579                 return;
 580
 581         slave_speed = ethtool_cmd_speed(&ecmd);
 582         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 583                 return;
 584
 585         switch (ecmd.duplex) {
 586         case DUPLEX_FULL:
 587         case DUPLEX_HALF:
 588                 break;
 589         default:
 590                 return;
 591         }
 592
 593         slave->speed = slave_speed;
 594         slave->duplex = ecmd.duplex;
 595
 596         return;
 597 }
 598
 599 /*
 600  * if <dev> supports MII link status reporting, check its link status.
 601  *
 602  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 603  * depending upon the setting of the use_carrier parameter.
 604  *
 605  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 606  * can't tell and just pretend it is), or 0, meaning that the link is
 607  * down.
 608  *
 609  * If reporting is non-zero, instead of faking link up, return -1 if
 610  * both ETHTOOL and MII ioctls fail (meaning the device does not
 611  * support them).  If use_carrier is set, return whatever it says.
 612  * It'd be nice if there was a good way to tell if a driver supports
 613  * netif_carrier, but there really isn't.
 614  */
 615 static int bond_check_dev_link(struct bonding *bond,
 616                                struct net_device *slave_dev, int reporting)
 617 {
 618         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 619         int (*ioctl)(struct net_device *, struct ifreq *, int);
 620         struct ifreq ifr;
 621         struct mii_ioctl_data *mii;
 622
 623         if (!reporting && !netif_running(slave_dev))
 624                 return 0;
 625
 626         if (bond->params.use_carrier)
 627                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 628
 629         /* Try to get link status using Ethtool first. */
 630         if (slave_dev->ethtool_ops->get_link)
 631                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 632                         BMSR_LSTATUS : 0;
 633
 634         /* Ethtool can't be used, fallback to MII ioctls. */
 635         ioctl = slave_ops->ndo_do_ioctl;
 636         if (ioctl) {
 637                 /* TODO: set pointer to correct ioctl on a per team member */
 638                 /*       bases to make this more efficient. that is, once  */
 639                 /*       we determine the correct ioctl, we will always    */
 640                 /*       call it and not the others for that team          */
 641                 /*       member.                                           */
 642
 643                 /*
 644                  * We cannot assume that SIOCGMIIPHY will also read a
 645                  * register; not all network drivers (e.g., e100)
 646                  * support that.
 647                  */
 648
 649                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 650                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 651                 mii = if_mii(&ifr);
 652                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 653                         mii->reg_num = MII_BMSR;
 654                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 655                                 return mii->val_out & BMSR_LSTATUS;
 656                 }
 657         }
 658
 659         /*
 660          * If reporting, report that either there's no dev->do_ioctl,
 661          * or both SIOCGMIIREG and get_link failed (meaning that we
 662          * cannot report link status).  If not reporting, pretend
 663          * we're ok.
 664          */
 665         return reporting ? -1 : BMSR_LSTATUS;
 666 }
 667
 668 /*----------------------------- Multicast list ------------------------------*/
 669
 670 /*
 671  * Push the promiscuity flag down to appropriate slaves
 672  */
 673 static int bond_set_promiscuity(struct bonding *bond, int inc)
 674 {
 675         int err = 0;
 676         if (USES_PRIMARY(bond->params.mode)) {
 677                 /* write lock already acquired */
 678                 if (bond->curr_active_slave) {
 679                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 680                                                   inc);
 681                 }
 682         } else {
 683                 struct slave *slave;
 684                 int i;
 685                 bond_for_each_slave(bond, slave, i) {
 686                         err = dev_set_promiscuity(slave->dev, inc);
 687                         if (err)
 688                                 return err;
 689                 }
 690         }
 691         return err;
 692 }
 693
 694 /*
 695  * Push the allmulti flag down to all slaves
 696  */
 697 static int bond_set_allmulti(struct bonding *bond, int inc)
 698 {
 699         int err = 0;
 700         if (USES_PRIMARY(bond->params.mode)) {
 701                 /* write lock already acquired */
 702                 if (bond->curr_active_slave) {
 703                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 704                                                inc);
 705                 }
 706         } else {
 707                 struct slave *slave;
 708                 int i;
 709                 bond_for_each_slave(bond, slave, i) {
 710                         err = dev_set_allmulti(slave->dev, inc);
 711                         if (err)
 712                                 return err;
 713                 }
 714         }
 715         return err;
 716 }
 717
 718 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 719 {
 720         struct in_device *in_dev;
 721
 722         in_dev = __in_dev_get_rcu(dev);
 723         if (in_dev)
 724                 ip_mc_rejoin_groups(in_dev);
 725 }
 726
 727 /*
 728  * Retrieve the list of registered multicast addresses for the bonding
 729  * device and retransmit an IGMP JOIN request to the current active
 730  * slave.
 731  */
 732 static void bond_resend_igmp_join_requests(struct bonding *bond)
 733 {
 734         struct net_device *bond_dev, *vlan_dev, *upper_dev;
 735         struct vlan_entry *vlan;
 736
 737         read_lock(&bond->lock);
 738         rcu_read_lock();
 739
 740         bond_dev = bond->dev;
 741
 742         /* rejoin all groups on bond device */
 743         __bond_resend_igmp_join_requests(bond_dev);
 744
 745         /*
 746          * if bond is enslaved to a bridge,
 747          * then rejoin all groups on its master
 748          */
 749         upper_dev = netdev_master_upper_dev_get_rcu(bond_dev);
 750         if (upper_dev && upper_dev->priv_flags & IFF_EBRIDGE)
 751                 __bond_resend_igmp_join_requests(upper_dev);
 752
 753         /* rejoin all groups on vlan devices */
 754         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 755                 vlan_dev = __vlan_find_dev_deep(bond_dev, htons(ETH_P_8021Q),
 756                                                 vlan->vlan_id);
 757                 if (vlan_dev)
 758                         __bond_resend_igmp_join_requests(vlan_dev);
 759         }
 760         rcu_read_unlock();
 761
 762         /* We use curr_slave_lock to protect against concurrent access to
 763          * igmp_retrans from multiple running instances of this function and
 764          * bond_change_active_slave
 765          */
 766         write_lock_bh(&bond->curr_slave_lock);
 767         if (bond->igmp_retrans > 1) {
 768                 bond->igmp_retrans--;
 769                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 770         }
 771         write_unlock_bh(&bond->curr_slave_lock);
 772         read_unlock(&bond->lock);
 773 }
 774
 775 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 776 {
 777         struct bonding *bond = container_of(work, struct bonding,
 778                                             mcast_work.work);
 779
 780         bond_resend_igmp_join_requests(bond);
 781 }
 782
 783 /* Flush bond's hardware addresses from slave
 784  */
 785 static void bond_hw_addr_flush(struct net_device *bond_dev,
 786                                struct net_device *slave_dev)
 787 {
 788         struct bonding *bond = netdev_priv(bond_dev);
 789
 790         dev_uc_unsync(slave_dev, bond_dev);
 791         dev_mc_unsync(slave_dev, bond_dev);
 792
 793         if (bond->params.mode == BOND_MODE_8023AD) {
 794                 /* del lacpdu mc addr from mc list */
 795                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 796
 797                 dev_mc_del(slave_dev, lacpdu_multicast);
 798         }
 799 }
 800
 801 /*--------------------------- Active slave change ---------------------------*/
 802
 803 /* Update the hardware address list and promisc/allmulti for the new and
 804  * old active slaves (if any).  Modes that are !USES_PRIMARY keep all
 805  * slaves up date at all times; only the USES_PRIMARY modes need to call
 806  * this function to swap these settings during a failover.
 807  */
 808 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 809                               struct slave *old_active)
 810 {
 811         if (old_active) {
 812                 if (bond->dev->flags & IFF_PROMISC)
 813                         dev_set_promiscuity(old_active->dev, -1);
 814
 815                 if (bond->dev->flags & IFF_ALLMULTI)
 816                         dev_set_allmulti(old_active->dev, -1);
 817
 818                 bond_hw_addr_flush(bond->dev, old_active->dev);
 819         }
 820
 821         if (new_active) {
 822                 /* FIXME: Signal errors upstream. */
 823                 if (bond->dev->flags & IFF_PROMISC)
 824                         dev_set_promiscuity(new_active->dev, 1);
 825
 826                 if (bond->dev->flags & IFF_ALLMULTI)
 827                         dev_set_allmulti(new_active->dev, 1);
 828
 829                 netif_addr_lock_bh(bond->dev);
 830                 dev_uc_sync(new_active->dev, bond->dev);
 831                 dev_mc_sync(new_active->dev, bond->dev);
 832                 netif_addr_unlock_bh(bond->dev);
 833         }
 834 }
 835
 836 /**
 837  * bond_set_dev_addr - clone slave's address to bond
 838  * @bond_dev: bond net device
 839  * @slave_dev: slave net device
 840  *
 841  * Should be called with RTNL held.
 842  */
 843 static void bond_set_dev_addr(struct net_device *bond_dev,
 844                               struct net_device *slave_dev)
 845 {
 846         pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 847                  bond_dev, slave_dev, slave_dev->addr_len);
 848         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 849         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 850         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 851 }
 852
 853 /*
 854  * bond_do_fail_over_mac
 855  *
 856  * Perform special MAC address swapping for fail_over_mac settings
 857  *
 858  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 859  */
 860 static void bond_do_fail_over_mac(struct bonding *bond,
 861                                   struct slave *new_active,
 862                                   struct slave *old_active)
 863         __releases(&bond->curr_slave_lock)
 864         __releases(&bond->lock)
 865         __acquires(&bond->lock)
 866         __acquires(&bond->curr_slave_lock)
 867 {
 868         u8 tmp_mac[ETH_ALEN];
 869         struct sockaddr saddr;
 870         int rv;
 871
 872         switch (bond->params.fail_over_mac) {
 873         case BOND_FOM_ACTIVE:
 874                 if (new_active) {
 875                         write_unlock_bh(&bond->curr_slave_lock);
 876                         read_unlock(&bond->lock);
 877                         bond_set_dev_addr(bond->dev, new_active->dev);
 878                         read_lock(&bond->lock);
 879                         write_lock_bh(&bond->curr_slave_lock);
 880                 }
 881                 break;
 882         case BOND_FOM_FOLLOW:
 883                 /*
 884                  * if new_active && old_active, swap them
 885                  * if just old_active, do nothing (going to no active slave)
 886                  * if just new_active, set new_active to bond's MAC
 887                  */
 888                 if (!new_active)
 889                         return;
 890
 891                 write_unlock_bh(&bond->curr_slave_lock);
 892                 read_unlock(&bond->lock);
 893
 894                 if (old_active) {
 895                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 896                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 897                                ETH_ALEN);
 898                         saddr.sa_family = new_active->dev->type;
 899                 } else {
 900                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 901                         saddr.sa_family = bond->dev->type;
 902                 }
 903
 904                 rv = dev_set_mac_address(new_active->dev, &saddr);
 905                 if (rv) {
 906                         pr_err("%s: Error %d setting MAC of slave %s\n",
 907                                bond->dev->name, -rv, new_active->dev->name);
 908                         goto out;
 909                 }
 910
 911                 if (!old_active)
 912                         goto out;
 913
 914                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 915                 saddr.sa_family = old_active->dev->type;
 916
 917                 rv = dev_set_mac_address(old_active->dev, &saddr);
 918                 if (rv)
 919                         pr_err("%s: Error %d setting MAC of slave %s\n",
 920                                bond->dev->name, -rv, new_active->dev->name);
 921 out:
 922                 read_lock(&bond->lock);
 923                 write_lock_bh(&bond->curr_slave_lock);
 924                 break;
 925         default:
 926                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 927                        bond->dev->name, bond->params.fail_over_mac);
 928                 break;
 929         }
 930
 931 }
 932
 933 static bool bond_should_change_active(struct bonding *bond)
 934 {
 935         struct slave *prim = bond->primary_slave;
 936         struct slave *curr = bond->curr_active_slave;
 937
 938         if (!prim || !curr || curr->link != BOND_LINK_UP)
 939                 return true;
 940         if (bond->force_primary) {
 941                 bond->force_primary = false;
 942                 return true;
 943         }
 944         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 945             (prim->speed < curr->speed ||
 946              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 947                 return false;
 948         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 949                 return false;
 950         return true;
 951 }
 952
 953 /**
 954  * find_best_interface - select the best available slave to be the active one
 955  * @bond: our bonding struct
 956  *
 957  * Warning: Caller must hold curr_slave_lock for writing.
 958  */
 959 static struct slave *bond_find_best_slave(struct bonding *bond)
 960 {
 961         struct slave *new_active, *old_active;
 962         struct slave *bestslave = NULL;
 963         int mintime = bond->params.updelay;
 964         int i;
 965
 966         new_active = bond->curr_active_slave;
 967
 968         if (!new_active) { /* there were no active slaves left */
 969                 if (bond->slave_cnt > 0)   /* found one slave */
 970                         new_active = bond->first_slave;
 971                 else
 972                         return NULL; /* still no slave, return NULL */
 973         }
 974
 975         if ((bond->primary_slave) &&
 976             bond->primary_slave->link == BOND_LINK_UP &&
 977             bond_should_change_active(bond)) {
 978                 new_active = bond->primary_slave;
 979         }
 980
 981         /* remember where to stop iterating over the slaves */
 982         old_active = new_active;
 983
 984         bond_for_each_slave_from(bond, new_active, i, old_active) {
 985                 if (new_active->link == BOND_LINK_UP) {
 986                         return new_active;
 987                 } else if (new_active->link == BOND_LINK_BACK &&
 988                            IS_UP(new_active->dev)) {
 989                         /* link up, but waiting for stabilization */
 990                         if (new_active->delay < mintime) {
 991                                 mintime = new_active->delay;
 992                                 bestslave = new_active;
 993                         }
 994                 }
 995         }
 996
 997         return bestslave;
 998 }
 999
1000 static bool bond_should_notify_peers(struct bonding *bond)
1001 {
1002         struct slave *slave = bond->curr_active_slave;
1003
1004         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1005                  bond->dev->name, slave ? slave->dev->name : "NULL");
1006
1007         if (!slave || !bond->send_peer_notif ||
1008             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1009                 return false;
1010
1011         bond->send_peer_notif--;
1012         return true;
1013 }
1014
1015 /**
1016  * change_active_interface - change the active slave into the specified one
1017  * @bond: our bonding struct
1018  * @new: the new slave to make the active one
1019  *
1020  * Set the new slave to the bond's settings and unset them on the old
1021  * curr_active_slave.
1022  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1023  *
1024  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1025  * because it is apparently the best available slave we have, even though its
1026  * updelay hasn't timed out yet.
1027  *
1028  * If new_active is not NULL, caller must hold bond->lock for read and
1029  * curr_slave_lock for write_bh.
1030  */
1031 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1032 {
1033         struct slave *old_active = bond->curr_active_slave;
1034
1035         if (old_active == new_active)
1036                 return;
1037
1038         if (new_active) {
1039                 new_active->jiffies = jiffies;
1040
1041                 if (new_active->link == BOND_LINK_BACK) {
1042                         if (USES_PRIMARY(bond->params.mode)) {
1043                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1044                                         bond->dev->name, new_active->dev->name,
1045                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1046                         }
1047
1048                         new_active->delay = 0;
1049                         new_active->link = BOND_LINK_UP;
1050
1051                         if (bond->params.mode == BOND_MODE_8023AD)
1052                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1053
1054                         if (bond_is_lb(bond))
1055                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1056                 } else {
1057                         if (USES_PRIMARY(bond->params.mode)) {
1058                                 pr_info("%s: making interface %s the new active one.\n",
1059                                         bond->dev->name, new_active->dev->name);
1060                         }
1061                 }
1062         }
1063
1064         if (USES_PRIMARY(bond->params.mode))
1065                 bond_hw_addr_swap(bond, new_active, old_active);
1066
1067         if (bond_is_lb(bond)) {
1068                 bond_alb_handle_active_change(bond, new_active);
1069                 if (old_active)
1070                         bond_set_slave_inactive_flags(old_active);
1071                 if (new_active)
1072                         bond_set_slave_active_flags(new_active);
1073         } else {
1074                 bond->curr_active_slave = new_active;
1075         }
1076
1077         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1078                 if (old_active)
1079                         bond_set_slave_inactive_flags(old_active);
1080
1081                 if (new_active) {
1082                         bool should_notify_peers = false;
1083
1084                         bond_set_slave_active_flags(new_active);
1085
1086                         if (bond->params.fail_over_mac)
1087                                 bond_do_fail_over_mac(bond, new_active,
1088                                                       old_active);
1089
1090                         if (netif_running(bond->dev)) {
1091                                 bond->send_peer_notif =
1092                                         bond->params.num_peer_notif;
1093                                 should_notify_peers =
1094                                         bond_should_notify_peers(bond);
1095                         }
1096
1097                         write_unlock_bh(&bond->curr_slave_lock);
1098                         read_unlock(&bond->lock);
1099
1100                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1101                         if (should_notify_peers)
1102                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1103                                                          bond->dev);
1104
1105                         read_lock(&bond->lock);
1106                         write_lock_bh(&bond->curr_slave_lock);
1107                 }
1108         }
1109
1110         /* resend IGMP joins since active slave has changed or
1111          * all were sent on curr_active_slave.
1112          * resend only if bond is brought up with the affected
1113          * bonding modes and the retransmission is enabled */
1114         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1115             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1116              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1117                 bond->igmp_retrans = bond->params.resend_igmp;
1118                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1119         }
1120 }
1121
1122 /**
1123  * bond_select_active_slave - select a new active slave, if needed
1124  * @bond: our bonding struct
1125  *
1126  * This functions should be called when one of the following occurs:
1127  * - The old curr_active_slave has been released or lost its link.
1128  * - The primary_slave has got its link back.
1129  * - A slave has got its link back and there's no old curr_active_slave.
1130  *
1131  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1132  */
1133 void bond_select_active_slave(struct bonding *bond)
1134 {
1135         struct slave *best_slave;
1136         int rv;
1137
1138         best_slave = bond_find_best_slave(bond);
1139         if (best_slave != bond->curr_active_slave) {
1140                 bond_change_active_slave(bond, best_slave);
1141                 rv = bond_set_carrier(bond);
1142                 if (!rv)
1143                         return;
1144
1145                 if (netif_carrier_ok(bond->dev)) {
1146                         pr_info("%s: first active interface up!\n",
1147                                 bond->dev->name);
1148                 } else {
1149                         pr_info("%s: now running without any active interface !\n",
1150                                 bond->dev->name);
1151                 }
1152         }
1153 }
1154
1155 /*--------------------------- slave list handling ---------------------------*/
1156
1157 /*
1158  * This function attaches the slave to the end of list.
1159  *
1160  * bond->lock held for writing by caller.
1161  */
1162 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1163 {
1164         if (bond->first_slave == NULL) { /* attaching the first slave */
1165                 new_slave->next = new_slave;
1166                 new_slave->prev = new_slave;
1167                 bond->first_slave = new_slave;
1168         } else {
1169                 new_slave->next = bond->first_slave;
1170                 new_slave->prev = bond->first_slave->prev;
1171                 new_slave->next->prev = new_slave;
1172                 new_slave->prev->next = new_slave;
1173         }
1174
1175         bond->slave_cnt++;
1176 }
1177
1178 /*
1179  * This function detaches the slave from the list.
1180  * WARNING: no check is made to verify if the slave effectively
1181  * belongs to <bond>.
1182  * Nothing is freed on return, structures are just unchained.
1183  * If any slave pointer in bond was pointing to <slave>,
1184  * it should be changed by the calling function.
1185  *
1186  * bond->lock held for writing by caller.
1187  */
1188 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1189 {
1190         if (slave->next)
1191                 slave->next->prev = slave->prev;
1192
1193         if (slave->prev)
1194                 slave->prev->next = slave->next;
1195
1196         if (bond->first_slave == slave) { /* slave is the first slave */
1197                 if (bond->slave_cnt > 1) { /* there are more slave */
1198                         bond->first_slave = slave->next;
1199                 } else {
1200                         bond->first_slave = NULL; /* slave was the last one */
1201                 }
1202         }
1203
1204         slave->next = NULL;
1205         slave->prev = NULL;
1206         bond->slave_cnt--;
1207 }
1208
1209 #ifdef CONFIG_NET_POLL_CONTROLLER
1210 static inline int slave_enable_netpoll(struct slave *slave)
1211 {
1212         struct netpoll *np;
1213         int err = 0;
1214
1215         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1216         err = -ENOMEM;
1217         if (!np)
1218                 goto out;
1219
1220         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1221         if (err) {
1222                 kfree(np);
1223                 goto out;
1224         }
1225         slave->np = np;
1226 out:
1227         return err;
1228 }
1229 static inline void slave_disable_netpoll(struct slave *slave)
1230 {
1231         struct netpoll *np = slave->np;
1232
1233         if (!np)
1234                 return;
1235
1236         slave->np = NULL;
1237         __netpoll_free_async(np);
1238 }
1239 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1240 {
1241         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1242                 return false;
1243         if (!slave_dev->netdev_ops->ndo_poll_controller)
1244                 return false;
1245         return true;
1246 }
1247
1248 static void bond_poll_controller(struct net_device *bond_dev)
1249 {
1250 }
1251
1252 static void __bond_netpoll_cleanup(struct bonding *bond)
1253 {
1254         struct slave *slave;
1255         int i;
1256
1257         bond_for_each_slave(bond, slave, i)
1258                 if (IS_UP(slave->dev))
1259                         slave_disable_netpoll(slave);
1260 }
1261 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1262 {
1263         struct bonding *bond = netdev_priv(bond_dev);
1264
1265         read_lock(&bond->lock);
1266         __bond_netpoll_cleanup(bond);
1267         read_unlock(&bond->lock);
1268 }
1269
1270 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1271 {
1272         struct bonding *bond = netdev_priv(dev);
1273         struct slave *slave;
1274         int i, err = 0;
1275
1276         read_lock(&bond->lock);
1277         bond_for_each_slave(bond, slave, i) {
1278                 err = slave_enable_netpoll(slave);
1279                 if (err) {
1280                         __bond_netpoll_cleanup(bond);
1281                         break;
1282                 }
1283         }
1284         read_unlock(&bond->lock);
1285         return err;
1286 }
1287
1288 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1289 {
1290         return bond->dev->npinfo;
1291 }
1292
1293 #else
1294 static inline int slave_enable_netpoll(struct slave *slave)
1295 {
1296         return 0;
1297 }
1298 static inline void slave_disable_netpoll(struct slave *slave)
1299 {
1300 }
1301 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1302 {
1303 }
1304 #endif
1305
1306 /*---------------------------------- IOCTL ----------------------------------*/
1307
1308 static netdev_features_t bond_fix_features(struct net_device *dev,
1309         netdev_features_t features)
1310 {
1311         struct slave *slave;
1312         struct bonding *bond = netdev_priv(dev);
1313         netdev_features_t mask;
1314         int i;
1315
1316         read_lock(&bond->lock);
1317
1318         if (!bond->first_slave) {
1319                 /* Disable adding VLANs to empty bond. But why? --mq */
1320                 features |= NETIF_F_VLAN_CHALLENGED;
1321                 goto out;
1322         }
1323
1324         mask = features;
1325         features &= ~NETIF_F_ONE_FOR_ALL;
1326         features |= NETIF_F_ALL_FOR_ALL;
1327
1328         bond_for_each_slave(bond, slave, i) {
1329                 features = netdev_increment_features(features,
1330                                                      slave->dev->features,
1331                                                      mask);
1332         }
1333         features = netdev_add_tso_features(features, mask);
1334
1335 out:
1336         read_unlock(&bond->lock);
1337         return features;
1338 }
1339
1340 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1341                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1342                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1343
1344 static void bond_compute_features(struct bonding *bond)
1345 {
1346         struct slave *slave;
1347         struct net_device *bond_dev = bond->dev;
1348         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1349         unsigned short max_hard_header_len = ETH_HLEN;
1350         unsigned int gso_max_size = GSO_MAX_SIZE;
1351         u16 gso_max_segs = GSO_MAX_SEGS;
1352         int i;
1353         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1354
1355         read_lock(&bond->lock);
1356
1357         if (!bond->first_slave)
1358                 goto done;
1359
1360         bond_for_each_slave(bond, slave, i) {
1361                 vlan_features = netdev_increment_features(vlan_features,
1362                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1363
1364                 dst_release_flag &= slave->dev->priv_flags;
1365                 if (slave->dev->hard_header_len > max_hard_header_len)
1366                         max_hard_header_len = slave->dev->hard_header_len;
1367
1368                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1369                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1370         }
1371
1372 done:
1373         bond_dev->vlan_features = vlan_features;
1374         bond_dev->hard_header_len = max_hard_header_len;
1375         bond_dev->gso_max_segs = gso_max_segs;
1376         netif_set_gso_max_size(bond_dev, gso_max_size);
1377
1378         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1379         bond_dev->priv_flags = flags | dst_release_flag;
1380
1381         read_unlock(&bond->lock);
1382
1383         netdev_change_features(bond_dev);
1384 }
1385
1386 static void bond_setup_by_slave(struct net_device *bond_dev,
1387                                 struct net_device *slave_dev)
1388 {
1389         bond_dev->header_ops        = slave_dev->header_ops;
1390
1391         bond_dev->type              = slave_dev->type;
1392         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1393         bond_dev->addr_len          = slave_dev->addr_len;
1394
1395         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1396                 slave_dev->addr_len);
1397 }
1398
1399 /* On bonding slaves other than the currently active slave, suppress
1400  * duplicates except for alb non-mcast/bcast.
1401  */
1402 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1403                                             struct slave *slave,
1404                                             struct bonding *bond)
1405 {
1406         if (bond_is_slave_inactive(slave)) {
1407                 if (bond->params.mode == BOND_MODE_ALB &&
1408                     skb->pkt_type != PACKET_BROADCAST &&
1409                     skb->pkt_type != PACKET_MULTICAST)
1410                         return false;
1411                 return true;
1412         }
1413         return false;
1414 }
1415
1416 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1417 {
1418         struct sk_buff *skb = *pskb;
1419         struct slave *slave;
1420         struct bonding *bond;
1421         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1422                           struct slave *);
1423         int ret = RX_HANDLER_ANOTHER;
1424
1425         skb = skb_share_check(skb, GFP_ATOMIC);
1426         if (unlikely(!skb))
1427                 return RX_HANDLER_CONSUMED;
1428
1429         *pskb = skb;
1430
1431         slave = bond_slave_get_rcu(skb->dev);
1432         bond = slave->bond;
1433
1434         if (bond->params.arp_interval)
1435                 slave->dev->last_rx = jiffies;
1436
1437         recv_probe = ACCESS_ONCE(bond->recv_probe);
1438         if (recv_probe) {
1439                 ret = recv_probe(skb, bond, slave);
1440                 if (ret == RX_HANDLER_CONSUMED) {
1441                         consume_skb(skb);
1442                         return ret;
1443                 }
1444         }
1445
1446         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1447                 return RX_HANDLER_EXACT;
1448         }
1449
1450         skb->dev = bond->dev;
1451
1452         if (bond->params.mode == BOND_MODE_ALB &&
1453             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1454             skb->pkt_type == PACKET_HOST) {
1455
1456                 if (unlikely(skb_cow_head(skb,
1457                                           skb->data - skb_mac_header(skb)))) {
1458                         kfree_skb(skb);
1459                         return RX_HANDLER_CONSUMED;
1460                 }
1461                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1462         }
1463
1464         return ret;
1465 }
1466
1467 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1468                                       struct net_device *slave_dev)
1469 {
1470         int err;
1471
1472         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1473         if (err)
1474                 return err;
1475         slave_dev->flags |= IFF_SLAVE;
1476         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1477         return 0;
1478 }
1479
1480 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1481                                   struct net_device *slave_dev)
1482 {
1483         netdev_upper_dev_unlink(slave_dev, bond_dev);
1484         slave_dev->flags &= ~IFF_SLAVE;
1485         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1486 }
1487
1488 /* enslave device <slave> to bond device <master> */
1489 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1490 {
1491         struct bonding *bond = netdev_priv(bond_dev);
1492         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1493         struct slave *new_slave = NULL;
1494         struct sockaddr addr;
1495         int link_reporting;
1496         int res = 0, i;
1497
1498         if (!bond->params.use_carrier &&
1499             slave_dev->ethtool_ops->get_link == NULL &&
1500             slave_ops->ndo_do_ioctl == NULL) {
1501                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1502                            bond_dev->name, slave_dev->name);
1503         }
1504
1505         /* already enslaved */
1506         if (slave_dev->flags & IFF_SLAVE) {
1507                 pr_debug("Error, Device was already enslaved\n");
1508                 return -EBUSY;
1509         }
1510
1511         /* vlan challenged mutual exclusion */
1512         /* no need to lock since we're protected by rtnl_lock */
1513         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1514                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1515                 if (vlan_uses_dev(bond_dev)) {
1516                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1517                                bond_dev->name, slave_dev->name, bond_dev->name);
1518                         return -EPERM;
1519                 } else {
1520                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1521                                    bond_dev->name, slave_dev->name,
1522                                    slave_dev->name, bond_dev->name);
1523                 }
1524         } else {
1525                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1526         }
1527
1528         /*
1529          * Old ifenslave binaries are no longer supported.  These can
1530          * be identified with moderate accuracy by the state of the slave:
1531          * the current ifenslave will set the interface down prior to
1532          * enslaving it; the old ifenslave will not.
1533          */
1534         if ((slave_dev->flags & IFF_UP)) {
1535                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1536                        slave_dev->name);
1537                 res = -EPERM;
1538                 goto err_undo_flags;
1539         }
1540
1541         /* set bonding device ether type by slave - bonding netdevices are
1542          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1543          * there is a need to override some of the type dependent attribs/funcs.
1544          *
1545          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1546          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1547          */
1548         if (bond->slave_cnt == 0) {
1549                 if (bond_dev->type != slave_dev->type) {
1550                         pr_debug("%s: change device type from %d to %d\n",
1551                                  bond_dev->name,
1552                                  bond_dev->type, slave_dev->type);
1553
1554                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1555                                                        bond_dev);
1556                         res = notifier_to_errno(res);
1557                         if (res) {
1558                                 pr_err("%s: refused to change device type\n",
1559                                        bond_dev->name);
1560                                 res = -EBUSY;
1561                                 goto err_undo_flags;
1562                         }
1563
1564                         /* Flush unicast and multicast addresses */
1565                         dev_uc_flush(bond_dev);
1566                         dev_mc_flush(bond_dev);
1567
1568                         if (slave_dev->type != ARPHRD_ETHER)
1569                                 bond_setup_by_slave(bond_dev, slave_dev);
1570                         else {
1571                                 ether_setup(bond_dev);
1572                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1573                         }
1574
1575                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1576                                                  bond_dev);
1577                 }
1578         } else if (bond_dev->type != slave_dev->type) {
1579                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1580                        slave_dev->name,
1581                        slave_dev->type, bond_dev->type);
1582                 res = -EINVAL;
1583                 goto err_undo_flags;
1584         }
1585
1586         if (slave_ops->ndo_set_mac_address == NULL) {
1587                 if (bond->slave_cnt == 0) {
1588                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1589                                    bond_dev->name);
1590                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1591                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1592                         pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1593                                bond_dev->name);
1594                         res = -EOPNOTSUPP;
1595                         goto err_undo_flags;
1596                 }
1597         }
1598
1599         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1600
1601         /* If this is the first slave, then we need to set the master's hardware
1602          * address to be the same as the slave's. */
1603         if (!bond->slave_cnt && bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1604                 bond_set_dev_addr(bond->dev, slave_dev);
1605
1606         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1607         if (!new_slave) {
1608                 res = -ENOMEM;
1609                 goto err_undo_flags;
1610         }
1611
1612         /*
1613          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1614          * is set via sysfs or module option if desired.
1615          */
1616         new_slave->queue_id = 0;
1617
1618         /* Save slave's original mtu and then set it to match the bond */
1619         new_slave->original_mtu = slave_dev->mtu;
1620         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1621         if (res) {
1622                 pr_debug("Error %d calling dev_set_mtu\n", res);
1623                 goto err_free;
1624         }
1625
1626         /*
1627          * Save slave's original ("permanent") mac address for modes
1628          * that need it, and for restoring it upon release, and then
1629          * set it to the master's address
1630          */
1631         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1632
1633         if (!bond->params.fail_over_mac) {
1634                 /*
1635                  * Set slave to master's mac address.  The application already
1636                  * set the master's mac address to that of the first slave
1637                  */
1638                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1639                 addr.sa_family = slave_dev->type;
1640                 res = dev_set_mac_address(slave_dev, &addr);
1641                 if (res) {
1642                         pr_debug("Error %d calling set_mac_address\n", res);
1643                         goto err_restore_mtu;
1644                 }
1645         }
1646
1647         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1648         if (res) {
1649                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1650                 goto err_restore_mac;
1651         }
1652
1653         /* open the slave since the application closed it */
1654         res = dev_open(slave_dev);
1655         if (res) {
1656                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1657                 goto err_unset_master;
1658         }
1659
1660         new_slave->bond = bond;
1661         new_slave->dev = slave_dev;
1662         slave_dev->priv_flags |= IFF_BONDING;
1663
1664         if (bond_is_lb(bond)) {
1665                 /* bond_alb_init_slave() must be called before all other stages since
1666                  * it might fail and we do not want to have to undo everything
1667                  */
1668                 res = bond_alb_init_slave(bond, new_slave);
1669                 if (res)
1670                         goto err_close;
1671         }
1672
1673         /* If the mode USES_PRIMARY, then the following is handled by
1674          * bond_change_active_slave().
1675          */
1676         if (!USES_PRIMARY(bond->params.mode)) {
1677                 /* set promiscuity level to new slave */
1678                 if (bond_dev->flags & IFF_PROMISC) {
1679                         res = dev_set_promiscuity(slave_dev, 1);
1680                         if (res)
1681                                 goto err_close;
1682                 }
1683
1684                 /* set allmulti level to new slave */
1685                 if (bond_dev->flags & IFF_ALLMULTI) {
1686                         res = dev_set_allmulti(slave_dev, 1);
1687                         if (res)
1688                                 goto err_close;
1689                 }
1690
1691                 netif_addr_lock_bh(bond_dev);
1692
1693                 dev_mc_sync_multiple(slave_dev, bond_dev);
1694                 dev_uc_sync_multiple(slave_dev, bond_dev);
1695
1696                 netif_addr_unlock_bh(bond_dev);
1697         }
1698
1699         if (bond->params.mode == BOND_MODE_8023AD) {
1700                 /* add lacpdu mc addr to mc list */
1701                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1702
1703                 dev_mc_add(slave_dev, lacpdu_multicast);
1704         }
1705
1706         bond_add_vlans_on_slave(bond, slave_dev);
1707
1708         write_lock_bh(&bond->lock);
1709
1710         bond_attach_slave(bond, new_slave);
1711
1712         new_slave->delay = 0;
1713         new_slave->link_failure_count = 0;
1714
1715         write_unlock_bh(&bond->lock);
1716
1717         bond_compute_features(bond);
1718
1719         bond_update_speed_duplex(new_slave);
1720
1721         read_lock(&bond->lock);
1722
1723         new_slave->last_arp_rx = jiffies -
1724                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1725         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1726                 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1727
1728         if (bond->params.miimon && !bond->params.use_carrier) {
1729                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1730
1731                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1732                         /*
1733                          * miimon is set but a bonded network driver
1734                          * does not support ETHTOOL/MII and
1735                          * arp_interval is not set.  Note: if
1736                          * use_carrier is enabled, we will never go
1737                          * here (because netif_carrier is always
1738                          * supported); thus, we don't need to change
1739                          * the messages for netif_carrier.
1740                          */
1741                         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",
1742                                bond_dev->name, slave_dev->name);
1743                 } else if (link_reporting == -1) {
1744                         /* unable get link status using mii/ethtool */
1745                         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",
1746                                    bond_dev->name, slave_dev->name);
1747                 }
1748         }
1749
1750         /* check for initial state */
1751         if (bond->params.miimon) {
1752                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1753                         if (bond->params.updelay) {
1754                                 new_slave->link = BOND_LINK_BACK;
1755                                 new_slave->delay = bond->params.updelay;
1756                         } else {
1757                                 new_slave->link = BOND_LINK_UP;
1758                         }
1759                 } else {
1760                         new_slave->link = BOND_LINK_DOWN;
1761                 }
1762         } else if (bond->params.arp_interval) {
1763                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1764                         BOND_LINK_UP : BOND_LINK_DOWN);
1765         } else {
1766                 new_slave->link = BOND_LINK_UP;
1767         }
1768
1769         if (new_slave->link != BOND_LINK_DOWN)
1770                 new_slave->jiffies = jiffies;
1771         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1772                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1773                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1774
1775         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1776                 /* if there is a primary slave, remember it */
1777                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1778                         bond->primary_slave = new_slave;
1779                         bond->force_primary = true;
1780                 }
1781         }
1782
1783         write_lock_bh(&bond->curr_slave_lock);
1784
1785         switch (bond->params.mode) {
1786         case BOND_MODE_ACTIVEBACKUP:
1787                 bond_set_slave_inactive_flags(new_slave);
1788                 bond_select_active_slave(bond);
1789                 break;
1790         case BOND_MODE_8023AD:
1791                 /* in 802.3ad mode, the internal mechanism
1792                  * will activate the slaves in the selected
1793                  * aggregator
1794                  */
1795                 bond_set_slave_inactive_flags(new_slave);
1796                 /* if this is the first slave */
1797                 if (bond->slave_cnt == 1) {
1798                         SLAVE_AD_INFO(new_slave).id = 1;
1799                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1800                          * can be called only after the mac address of the bond is set
1801                          */
1802                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1803                 } else {
1804                         SLAVE_AD_INFO(new_slave).id =
1805                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1806                 }
1807
1808                 bond_3ad_bind_slave(new_slave);
1809                 break;
1810         case BOND_MODE_TLB:
1811         case BOND_MODE_ALB:
1812                 bond_set_active_slave(new_slave);
1813                 bond_set_slave_inactive_flags(new_slave);
1814                 bond_select_active_slave(bond);
1815                 break;
1816         default:
1817                 pr_debug("This slave is always active in trunk mode\n");
1818
1819                 /* always active in trunk mode */
1820                 bond_set_active_slave(new_slave);
1821
1822                 /* In trunking mode there is little meaning to curr_active_slave
1823                  * anyway (it holds no special properties of the bond device),
1824                  * so we can change it without calling change_active_interface()
1825                  */
1826                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1827                         bond->curr_active_slave = new_slave;
1828
1829                 break;
1830         } /* switch(bond_mode) */
1831
1832         write_unlock_bh(&bond->curr_slave_lock);
1833
1834         bond_set_carrier(bond);
1835
1836 #ifdef CONFIG_NET_POLL_CONTROLLER
1837         slave_dev->npinfo = bond_netpoll_info(bond);
1838         if (slave_dev->npinfo) {
1839                 if (slave_enable_netpoll(new_slave)) {
1840                         read_unlock(&bond->lock);
1841                         pr_info("Error, %s: master_dev is using netpoll, "
1842                                  "but new slave device does not support netpoll.\n",
1843                                  bond_dev->name);
1844                         res = -EBUSY;
1845                         goto err_detach;
1846                 }
1847         }
1848 #endif
1849
1850         read_unlock(&bond->lock);
1851
1852         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1853         if (res)
1854                 goto err_detach;
1855
1856         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1857                                          new_slave);
1858         if (res) {
1859                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1860                 goto err_dest_symlinks;
1861         }
1862
1863         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1864                 bond_dev->name, slave_dev->name,
1865                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1866                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1867
1868         /* enslave is successful */
1869         return 0;
1870
1871 /* Undo stages on error */
1872 err_dest_symlinks:
1873         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1874
1875 err_detach:
1876         if (!USES_PRIMARY(bond->params.mode))
1877                 bond_hw_addr_flush(bond_dev, slave_dev);
1878
1879         bond_del_vlans_from_slave(bond, slave_dev);
1880         write_lock_bh(&bond->lock);
1881         bond_detach_slave(bond, new_slave);
1882         if (bond->primary_slave == new_slave)
1883                 bond->primary_slave = NULL;
1884         if (bond->curr_active_slave == new_slave) {
1885                 bond_change_active_slave(bond, NULL);
1886                 write_unlock_bh(&bond->lock);
1887                 read_lock(&bond->lock);
1888                 write_lock_bh(&bond->curr_slave_lock);
1889                 bond_select_active_slave(bond);
1890                 write_unlock_bh(&bond->curr_slave_lock);
1891                 read_unlock(&bond->lock);
1892         } else {
1893                 write_unlock_bh(&bond->lock);
1894         }
1895         slave_disable_netpoll(new_slave);
1896
1897 err_close:
1898         slave_dev->priv_flags &= ~IFF_BONDING;
1899         dev_close(slave_dev);
1900
1901 err_unset_master:
1902         bond_upper_dev_unlink(bond_dev, slave_dev);
1903
1904 err_restore_mac:
1905         if (!bond->params.fail_over_mac) {
1906                 /* XXX TODO - fom follow mode needs to change master's
1907                  * MAC if this slave's MAC is in use by the bond, or at
1908                  * least print a warning.
1909                  */
1910                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1911                 addr.sa_family = slave_dev->type;
1912                 dev_set_mac_address(slave_dev, &addr);
1913         }
1914
1915 err_restore_mtu:
1916         dev_set_mtu(slave_dev, new_slave->original_mtu);
1917
1918 err_free:
1919         kfree(new_slave);
1920
1921 err_undo_flags:
1922         bond_compute_features(bond);
1923         /* Enslave of first slave has failed and we need to fix master's mac */
1924         if (bond->slave_cnt == 0 &&
1925             ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1926                 eth_hw_addr_random(bond_dev);
1927
1928         return res;
1929 }
1930
1931 /*
1932  * Try to release the slave device <slave> from the bond device <master>
1933  * It is legal to access curr_active_slave without a lock because all the function
1934  * is write-locked. If "all" is true it means that the function is being called
1935  * while destroying a bond interface and all slaves are being released.
1936  *
1937  * The rules for slave state should be:
1938  *   for Active/Backup:
1939  *     Active stays on all backups go down
1940  *   for Bonded connections:
1941  *     The first up interface should be left on and all others downed.
1942  */
1943 static int __bond_release_one(struct net_device *bond_dev,
1944                               struct net_device *slave_dev,
1945                               bool all)
1946 {
1947         struct bonding *bond = netdev_priv(bond_dev);
1948         struct slave *slave, *oldcurrent;
1949         struct sockaddr addr;
1950         netdev_features_t old_features = bond_dev->features;
1951
1952         /* slave is not a slave or master is not master of this slave */
1953         if (!(slave_dev->flags & IFF_SLAVE) ||
1954             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1955                 pr_err("%s: Error: cannot release %s.\n",
1956                        bond_dev->name, slave_dev->name);
1957                 return -EINVAL;
1958         }
1959
1960         block_netpoll_tx();
1961         write_lock_bh(&bond->lock);
1962
1963         slave = bond_get_slave_by_dev(bond, slave_dev);
1964         if (!slave) {
1965                 /* not a slave of this bond */
1966                 pr_info("%s: %s not enslaved\n",
1967                         bond_dev->name, slave_dev->name);
1968                 write_unlock_bh(&bond->lock);
1969                 unblock_netpoll_tx();
1970                 return -EINVAL;
1971         }
1972
1973         write_unlock_bh(&bond->lock);
1974         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1975          * for this slave anymore.
1976          */
1977         netdev_rx_handler_unregister(slave_dev);
1978         write_lock_bh(&bond->lock);
1979
1980         if (!all && !bond->params.fail_over_mac) {
1981                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1982                     bond->slave_cnt > 1)
1983                         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",
1984                                    bond_dev->name, slave_dev->name,
1985                                    slave->perm_hwaddr,
1986                                    bond_dev->name, slave_dev->name);
1987         }
1988
1989         /* Inform AD package of unbinding of slave. */
1990         if (bond->params.mode == BOND_MODE_8023AD) {
1991                 /* must be called before the slave is
1992                  * detached from the list
1993                  */
1994                 bond_3ad_unbind_slave(slave);
1995         }
1996
1997         pr_info("%s: releasing %s interface %s\n",
1998                 bond_dev->name,
1999                 bond_is_active_slave(slave) ? "active" : "backup",
2000                 slave_dev->name);
2001
2002         oldcurrent = bond->curr_active_slave;
2003
2004         bond->current_arp_slave = NULL;
2005
2006         /* release the slave from its bond */
2007         bond_detach_slave(bond, slave);
2008
2009         if (bond->primary_slave == slave)
2010                 bond->primary_slave = NULL;
2011
2012         if (oldcurrent == slave)
2013                 bond_change_active_slave(bond, NULL);
2014
2015         if (bond_is_lb(bond)) {
2016                 /* Must be called only after the slave has been
2017                  * detached from the list and the curr_active_slave
2018                  * has been cleared (if our_slave == old_current),
2019                  * but before a new active slave is selected.
2020                  */
2021                 write_unlock_bh(&bond->lock);
2022                 bond_alb_deinit_slave(bond, slave);
2023                 write_lock_bh(&bond->lock);
2024         }
2025
2026         if (all) {
2027                 bond->curr_active_slave = NULL;
2028         } else if (oldcurrent == slave) {
2029                 /*
2030                  * Note that we hold RTNL over this sequence, so there
2031                  * is no concern that another slave add/remove event
2032                  * will interfere.
2033                  */
2034                 write_unlock_bh(&bond->lock);
2035                 read_lock(&bond->lock);
2036                 write_lock_bh(&bond->curr_slave_lock);
2037
2038                 bond_select_active_slave(bond);
2039
2040                 write_unlock_bh(&bond->curr_slave_lock);
2041                 read_unlock(&bond->lock);
2042                 write_lock_bh(&bond->lock);
2043         }
2044
2045         if (bond->slave_cnt == 0) {
2046                 bond_set_carrier(bond);
2047                 eth_hw_addr_random(bond_dev);
2048
2049                 if (bond_vlan_used(bond)) {
2050                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2051                                    bond_dev->name, bond_dev->name);
2052                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2053                                    bond_dev->name);
2054                 }
2055         }
2056
2057         write_unlock_bh(&bond->lock);
2058         unblock_netpoll_tx();
2059
2060         if (bond->slave_cnt == 0) {
2061                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2062                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2063         }
2064
2065         bond_compute_features(bond);
2066         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2067             (old_features & NETIF_F_VLAN_CHALLENGED))
2068                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2069                         bond_dev->name, slave_dev->name, bond_dev->name);
2070
2071         /* must do this from outside any spinlocks */
2072         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2073
2074         bond_del_vlans_from_slave(bond, slave_dev);
2075
2076         /* If the mode USES_PRIMARY, then this cases was handled above by
2077          * bond_change_active_slave(..., NULL)
2078          */
2079         if (!USES_PRIMARY(bond->params.mode)) {
2080                 /* unset promiscuity level from slave */
2081                 if (bond_dev->flags & IFF_PROMISC)
2082                         dev_set_promiscuity(slave_dev, -1);
2083
2084                 /* unset allmulti level from slave */
2085                 if (bond_dev->flags & IFF_ALLMULTI)
2086                         dev_set_allmulti(slave_dev, -1);
2087
2088                 bond_hw_addr_flush(bond_dev, slave_dev);
2089         }
2090
2091         bond_upper_dev_unlink(bond_dev, slave_dev);
2092
2093         slave_disable_netpoll(slave);
2094
2095         /* close slave before restoring its mac address */
2096         dev_close(slave_dev);
2097
2098         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2099                 /* restore original ("permanent") mac address */
2100                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2101                 addr.sa_family = slave_dev->type;
2102                 dev_set_mac_address(slave_dev, &addr);
2103         }
2104
2105         dev_set_mtu(slave_dev, slave->original_mtu);
2106
2107         slave_dev->priv_flags &= ~IFF_BONDING;
2108
2109         kfree(slave);
2110
2111         return 0;  /* deletion OK */
2112 }
2113
2114 /* A wrapper used because of ndo_del_link */
2115 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2116 {
2117         return __bond_release_one(bond_dev, slave_dev, false);
2118 }
2119
2120 /*
2121 * First release a slave and then destroy the bond if no more slaves are left.
2122 * Must be under rtnl_lock when this function is called.
2123 */
2124 static int  bond_release_and_destroy(struct net_device *bond_dev,
2125                                      struct net_device *slave_dev)
2126 {
2127         struct bonding *bond = netdev_priv(bond_dev);
2128         int ret;
2129
2130         ret = bond_release(bond_dev, slave_dev);
2131         if ((ret == 0) && (bond->slave_cnt == 0)) {
2132                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2133                 pr_info("%s: destroying bond %s.\n",
2134                         bond_dev->name, bond_dev->name);
2135                 unregister_netdevice(bond_dev);
2136         }
2137         return ret;
2138 }
2139
2140 /*
2141  * This function changes the active slave to slave <slave_dev>.
2142  * It returns -EINVAL in the following cases.
2143  *  - <slave_dev> is not found in the list.
2144  *  - There is not active slave now.
2145  *  - <slave_dev> is already active.
2146  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2147  *  - <slave_dev> is not running.
2148  * In these cases, this function does nothing.
2149  * In the other cases, current_slave pointer is changed and 0 is returned.
2150  */
2151 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2152 {
2153         struct bonding *bond = netdev_priv(bond_dev);
2154         struct slave *old_active = NULL;
2155         struct slave *new_active = NULL;
2156         int res = 0;
2157
2158         if (!USES_PRIMARY(bond->params.mode))
2159                 return -EINVAL;
2160
2161         /* Verify that bond_dev is indeed the master of slave_dev */
2162         if (!(slave_dev->flags & IFF_SLAVE) ||
2163             !netdev_has_upper_dev(slave_dev, bond_dev))
2164                 return -EINVAL;
2165
2166         read_lock(&bond->lock);
2167
2168         read_lock(&bond->curr_slave_lock);
2169         old_active = bond->curr_active_slave;
2170         read_unlock(&bond->curr_slave_lock);
2171
2172         new_active = bond_get_slave_by_dev(bond, slave_dev);
2173
2174         /*
2175          * Changing to the current active: do nothing; return success.
2176          */
2177         if (new_active && (new_active == old_active)) {
2178                 read_unlock(&bond->lock);
2179                 return 0;
2180         }
2181
2182         if ((new_active) &&
2183             (old_active) &&
2184             (new_active->link == BOND_LINK_UP) &&
2185             IS_UP(new_active->dev)) {
2186                 block_netpoll_tx();
2187                 write_lock_bh(&bond->curr_slave_lock);
2188                 bond_change_active_slave(bond, new_active);
2189                 write_unlock_bh(&bond->curr_slave_lock);
2190                 unblock_netpoll_tx();
2191         } else
2192                 res = -EINVAL;
2193
2194         read_unlock(&bond->lock);
2195
2196         return res;
2197 }
2198
2199 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2200 {
2201         struct bonding *bond = netdev_priv(bond_dev);
2202
2203         info->bond_mode = bond->params.mode;
2204         info->miimon = bond->params.miimon;
2205
2206         read_lock(&bond->lock);
2207         info->num_slaves = bond->slave_cnt;
2208         read_unlock(&bond->lock);
2209
2210         return 0;
2211 }
2212
2213 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2214 {
2215         struct bonding *bond = netdev_priv(bond_dev);
2216         struct slave *slave;
2217         int i, res = -ENODEV;
2218
2219         read_lock(&bond->lock);
2220
2221         bond_for_each_slave(bond, slave, i) {
2222                 if (i == (int)info->slave_id) {
2223                         res = 0;
2224                         strcpy(info->slave_name, slave->dev->name);
2225                         info->link = slave->link;
2226                         info->state = bond_slave_state(slave);
2227                         info->link_failure_count = slave->link_failure_count;
2228                         break;
2229                 }
2230         }
2231
2232         read_unlock(&bond->lock);
2233
2234         return res;
2235 }
2236
2237 /*-------------------------------- Monitoring -------------------------------*/
2238
2239
2240 static int bond_miimon_inspect(struct bonding *bond)
2241 {
2242         struct slave *slave;
2243         int i, link_state, commit = 0;
2244         bool ignore_updelay;
2245
2246         ignore_updelay = !bond->curr_active_slave ? true : false;
2247
2248         bond_for_each_slave(bond, slave, i) {
2249                 slave->new_link = BOND_LINK_NOCHANGE;
2250
2251                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2252
2253                 switch (slave->link) {
2254                 case BOND_LINK_UP:
2255                         if (link_state)
2256                                 continue;
2257
2258                         slave->link = BOND_LINK_FAIL;
2259                         slave->delay = bond->params.downdelay;
2260                         if (slave->delay) {
2261                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2262                                         bond->dev->name,
2263                                         (bond->params.mode ==
2264                                          BOND_MODE_ACTIVEBACKUP) ?
2265                                         (bond_is_active_slave(slave) ?
2266                                          "active " : "backup ") : "",
2267                                         slave->dev->name,
2268                                         bond->params.downdelay * bond->params.miimon);
2269                         }
2270                         /*FALLTHRU*/
2271                 case BOND_LINK_FAIL:
2272                         if (link_state) {
2273                                 /*
2274                                  * recovered before downdelay expired
2275                                  */
2276                                 slave->link = BOND_LINK_UP;
2277                                 slave->jiffies = jiffies;
2278                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2279                                         bond->dev->name,
2280                                         (bond->params.downdelay - slave->delay) *
2281                                         bond->params.miimon,
2282                                         slave->dev->name);
2283                                 continue;
2284                         }
2285
2286                         if (slave->delay <= 0) {
2287                                 slave->new_link = BOND_LINK_DOWN;
2288                                 commit++;
2289                                 continue;
2290                         }
2291
2292                         slave->delay--;
2293                         break;
2294
2295                 case BOND_LINK_DOWN:
2296                         if (!link_state)
2297                                 continue;
2298
2299                         slave->link = BOND_LINK_BACK;
2300                         slave->delay = bond->params.updelay;
2301
2302                         if (slave->delay) {
2303                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2304                                         bond->dev->name, slave->dev->name,
2305                                         ignore_updelay ? 0 :
2306                                         bond->params.updelay *
2307                                         bond->params.miimon);
2308                         }
2309                         /*FALLTHRU*/
2310                 case BOND_LINK_BACK:
2311                         if (!link_state) {
2312                                 slave->link = BOND_LINK_DOWN;
2313                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2314                                         bond->dev->name,
2315                                         (bond->params.updelay - slave->delay) *
2316                                         bond->params.miimon,
2317                                         slave->dev->name);
2318
2319                                 continue;
2320                         }
2321
2322                         if (ignore_updelay)
2323                                 slave->delay = 0;
2324
2325                         if (slave->delay <= 0) {
2326                                 slave->new_link = BOND_LINK_UP;
2327                                 commit++;
2328                                 ignore_updelay = false;
2329                                 continue;
2330                         }
2331
2332                         slave->delay--;
2333                         break;
2334                 }
2335         }
2336
2337         return commit;
2338 }
2339
2340 static void bond_miimon_commit(struct bonding *bond)
2341 {
2342         struct slave *slave;
2343         int i;
2344
2345         bond_for_each_slave(bond, slave, i) {
2346                 switch (slave->new_link) {
2347                 case BOND_LINK_NOCHANGE:
2348                         continue;
2349
2350                 case BOND_LINK_UP:
2351                         slave->link = BOND_LINK_UP;
2352                         slave->jiffies = jiffies;
2353
2354                         if (bond->params.mode == BOND_MODE_8023AD) {
2355                                 /* prevent it from being the active one */
2356                                 bond_set_backup_slave(slave);
2357                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2358                                 /* make it immediately active */
2359                                 bond_set_active_slave(slave);
2360                         } else if (slave != bond->primary_slave) {
2361                                 /* prevent it from being the active one */
2362                                 bond_set_backup_slave(slave);
2363                         }
2364
2365                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2366                                 bond->dev->name, slave->dev->name,
2367                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2368                                 slave->duplex ? "full" : "half");
2369
2370                         /* notify ad that the link status has changed */
2371                         if (bond->params.mode == BOND_MODE_8023AD)
2372                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2373
2374                         if (bond_is_lb(bond))
2375                                 bond_alb_handle_link_change(bond, slave,
2376                                                             BOND_LINK_UP);
2377
2378                         if (!bond->curr_active_slave ||
2379                             (slave == bond->primary_slave))
2380                                 goto do_failover;
2381
2382                         continue;
2383
2384                 case BOND_LINK_DOWN:
2385                         if (slave->link_failure_count < UINT_MAX)
2386                                 slave->link_failure_count++;
2387
2388                         slave->link = BOND_LINK_DOWN;
2389
2390                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2391                             bond->params.mode == BOND_MODE_8023AD)
2392                                 bond_set_slave_inactive_flags(slave);
2393
2394                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2395                                 bond->dev->name, slave->dev->name);
2396
2397                         if (bond->params.mode == BOND_MODE_8023AD)
2398                                 bond_3ad_handle_link_change(slave,
2399                                                             BOND_LINK_DOWN);
2400
2401                         if (bond_is_lb(bond))
2402                                 bond_alb_handle_link_change(bond, slave,
2403                                                             BOND_LINK_DOWN);
2404
2405                         if (slave == bond->curr_active_slave)
2406                                 goto do_failover;
2407
2408                         continue;
2409
2410                 default:
2411                         pr_err("%s: invalid new link %d on slave %s\n",
2412                                bond->dev->name, slave->new_link,
2413                                slave->dev->name);
2414                         slave->new_link = BOND_LINK_NOCHANGE;
2415
2416                         continue;
2417                 }
2418
2419 do_failover:
2420                 ASSERT_RTNL();
2421                 block_netpoll_tx();
2422                 write_lock_bh(&bond->curr_slave_lock);
2423                 bond_select_active_slave(bond);
2424                 write_unlock_bh(&bond->curr_slave_lock);
2425                 unblock_netpoll_tx();
2426         }
2427
2428         bond_set_carrier(bond);
2429 }
2430
2431 /*
2432  * bond_mii_monitor
2433  *
2434  * Really a wrapper that splits the mii monitor into two phases: an
2435  * inspection, then (if inspection indicates something needs to be done)
2436  * an acquisition of appropriate locks followed by a commit phase to
2437  * implement whatever link state changes are indicated.
2438  */
2439 void bond_mii_monitor(struct work_struct *work)
2440 {
2441         struct bonding *bond = container_of(work, struct bonding,
2442                                             mii_work.work);
2443         bool should_notify_peers = false;
2444         unsigned long delay;
2445
2446         read_lock(&bond->lock);
2447
2448         delay = msecs_to_jiffies(bond->params.miimon);
2449
2450         if (bond->slave_cnt == 0)
2451                 goto re_arm;
2452
2453         should_notify_peers = bond_should_notify_peers(bond);
2454
2455         if (bond_miimon_inspect(bond)) {
2456                 read_unlock(&bond->lock);
2457
2458                 /* Race avoidance with bond_close cancel of workqueue */
2459                 if (!rtnl_trylock()) {
2460                         read_lock(&bond->lock);
2461                         delay = 1;
2462                         should_notify_peers = false;
2463                         goto re_arm;
2464                 }
2465
2466                 read_lock(&bond->lock);
2467
2468                 bond_miimon_commit(bond);
2469
2470                 read_unlock(&bond->lock);
2471                 rtnl_unlock();  /* might sleep, hold no other locks */
2472                 read_lock(&bond->lock);
2473         }
2474
2475 re_arm:
2476         if (bond->params.miimon)
2477                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2478
2479         read_unlock(&bond->lock);
2480
2481         if (should_notify_peers) {
2482                 if (!rtnl_trylock()) {
2483                         read_lock(&bond->lock);
2484                         bond->send_peer_notif++;
2485                         read_unlock(&bond->lock);
2486                         return;
2487                 }
2488                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2489                 rtnl_unlock();
2490         }
2491 }
2492
2493 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2494 {
2495         struct vlan_entry *vlan;
2496         struct net_device *vlan_dev;
2497
2498         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2499                 return 1;
2500
2501         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2502                 rcu_read_lock();
2503                 vlan_dev = __vlan_find_dev_deep(bond->dev, htons(ETH_P_8021Q),
2504                                                 vlan->vlan_id);
2505                 rcu_read_unlock();
2506                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2507                         return 1;
2508         }
2509
2510         return 0;
2511 }
2512
2513 /*
2514  * We go to the (large) trouble of VLAN tagging ARP frames because
2515  * switches in VLAN mode (especially if ports are configured as
2516  * "native" to a VLAN) might not pass non-tagged frames.
2517  */
2518 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2519 {
2520         struct sk_buff *skb;
2521
2522         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2523                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2524
2525         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2526                          NULL, slave_dev->dev_addr, NULL);
2527
2528         if (!skb) {
2529                 pr_err("ARP packet allocation failed\n");
2530                 return;
2531         }
2532         if (vlan_id) {
2533                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2534                 if (!skb) {
2535                         pr_err("failed to insert VLAN tag\n");
2536                         return;
2537                 }
2538         }
2539         arp_xmit(skb);
2540 }
2541
2542
2543 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2544 {
2545         int i, vlan_id;
2546         __be32 *targets = bond->params.arp_targets;
2547         struct vlan_entry *vlan;
2548         struct net_device *vlan_dev = NULL;
2549         struct rtable *rt;
2550
2551         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2552                 __be32 addr;
2553                 if (!targets[i])
2554                         break;
2555                 pr_debug("basa: target %pI4\n", &targets[i]);
2556                 if (!bond_vlan_used(bond)) {
2557                         pr_debug("basa: empty vlan: arp_send\n");
2558                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2559                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2560                                       addr, 0);
2561                         continue;
2562                 }
2563
2564                 /*
2565                  * If VLANs are configured, we do a route lookup to
2566                  * determine which VLAN interface would be used, so we
2567                  * can tag the ARP with the proper VLAN tag.
2568                  */
2569                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2570                                      RTO_ONLINK, 0);
2571                 if (IS_ERR(rt)) {
2572                         if (net_ratelimit()) {
2573                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2574                                            bond->dev->name, &targets[i]);
2575                         }
2576                         continue;
2577                 }
2578
2579                 /*
2580                  * This target is not on a VLAN
2581                  */
2582                 if (rt->dst.dev == bond->dev) {
2583                         ip_rt_put(rt);
2584                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2585                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2586                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2587                                       addr, 0);
2588                         continue;
2589                 }
2590
2591                 vlan_id = 0;
2592                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2593                         rcu_read_lock();
2594                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2595                                                         htons(ETH_P_8021Q),
2596                                                         vlan->vlan_id);
2597                         rcu_read_unlock();
2598                         if (vlan_dev == rt->dst.dev) {
2599                                 vlan_id = vlan->vlan_id;
2600                                 pr_debug("basa: vlan match on %s %d\n",
2601                                        vlan_dev->name, vlan_id);
2602                                 break;
2603                         }
2604                 }
2605
2606                 if (vlan_id && vlan_dev) {
2607                         ip_rt_put(rt);
2608                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2609                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2610                                       addr, vlan_id);
2611                         continue;
2612                 }
2613
2614                 if (net_ratelimit()) {
2615                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2616                                    bond->dev->name, &targets[i],
2617                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2618                 }
2619                 ip_rt_put(rt);
2620         }
2621 }
2622
2623 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2624 {
2625         int i;
2626
2627         if (!sip || !bond_has_this_ip(bond, tip)) {
2628                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2629                 return;
2630         }
2631
2632         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2633         if (i == -1) {
2634                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2635                 return;
2636         }
2637         slave->last_arp_rx = jiffies;
2638         slave->target_last_arp_rx[i] = jiffies;
2639 }
2640
2641 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2642                         struct slave *slave)
2643 {
2644         struct arphdr *arp = (struct arphdr *)skb->data;
2645         unsigned char *arp_ptr;
2646         __be32 sip, tip;
2647         int alen;
2648
2649         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2650                 return RX_HANDLER_ANOTHER;
2651
2652         read_lock(&bond->lock);
2653
2654         if (!slave_do_arp_validate(bond, slave))
2655                 goto out_unlock;
2656
2657         alen = arp_hdr_len(bond->dev);
2658
2659         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2660                  bond->dev->name, skb->dev->name);
2661
2662         if (alen > skb_headlen(skb)) {
2663                 arp = kmalloc(alen, GFP_ATOMIC);
2664                 if (!arp)
2665                         goto out_unlock;
2666                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2667                         goto out_unlock;
2668         }
2669
2670         if (arp->ar_hln != bond->dev->addr_len ||
2671             skb->pkt_type == PACKET_OTHERHOST ||
2672             skb->pkt_type == PACKET_LOOPBACK ||
2673             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2674             arp->ar_pro != htons(ETH_P_IP) ||
2675             arp->ar_pln != 4)
2676                 goto out_unlock;
2677
2678         arp_ptr = (unsigned char *)(arp + 1);
2679         arp_ptr += bond->dev->addr_len;
2680         memcpy(&sip, arp_ptr, 4);
2681         arp_ptr += 4 + bond->dev->addr_len;
2682         memcpy(&tip, arp_ptr, 4);
2683
2684         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2685                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2686                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2687                  &sip, &tip);
2688
2689         /*
2690          * Backup slaves won't see the ARP reply, but do come through
2691          * here for each ARP probe (so we swap the sip/tip to validate
2692          * the probe).  In a "redundant switch, common router" type of
2693          * configuration, the ARP probe will (hopefully) travel from
2694          * the active, through one switch, the router, then the other
2695          * switch before reaching the backup.
2696          *
2697          * We 'trust' the arp requests if there is an active slave and
2698          * it received valid arp reply(s) after it became active. This
2699          * is done to avoid endless looping when we can't reach the
2700          * arp_ip_target and fool ourselves with our own arp requests.
2701          */
2702         if (bond_is_active_slave(slave))
2703                 bond_validate_arp(bond, slave, sip, tip);
2704         else if (bond->curr_active_slave &&
2705                  time_after(slave_last_rx(bond, bond->curr_active_slave),
2706                             bond->curr_active_slave->jiffies))
2707                 bond_validate_arp(bond, slave, tip, sip);
2708
2709 out_unlock:
2710         read_unlock(&bond->lock);
2711         if (arp != (struct arphdr *)skb->data)
2712                 kfree(arp);
2713         return RX_HANDLER_ANOTHER;
2714 }
2715
2716 /*
2717  * this function is called regularly to monitor each slave's link
2718  * ensuring that traffic is being sent and received when arp monitoring
2719  * is used in load-balancing mode. if the adapter has been dormant, then an
2720  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2721  * arp monitoring in active backup mode.
2722  */
2723 void bond_loadbalance_arp_mon(struct work_struct *work)
2724 {
2725         struct bonding *bond = container_of(work, struct bonding,
2726                                             arp_work.work);
2727         struct slave *slave, *oldcurrent;
2728         int do_failover = 0;
2729         int delta_in_ticks, extra_ticks;
2730         int i;
2731
2732         read_lock(&bond->lock);
2733
2734         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2735         extra_ticks = delta_in_ticks / 2;
2736
2737         if (bond->slave_cnt == 0)
2738                 goto re_arm;
2739
2740         read_lock(&bond->curr_slave_lock);
2741         oldcurrent = bond->curr_active_slave;
2742         read_unlock(&bond->curr_slave_lock);
2743
2744         /* see if any of the previous devices are up now (i.e. they have
2745          * xmt and rcv traffic). the curr_active_slave does not come into
2746          * the picture unless it is null. also, slave->jiffies is not needed
2747          * here because we send an arp on each slave and give a slave as
2748          * long as it needs to get the tx/rx within the delta.
2749          * TODO: what about up/down delay in arp mode? it wasn't here before
2750          *       so it can wait
2751          */
2752         bond_for_each_slave(bond, slave, i) {
2753                 unsigned long trans_start = dev_trans_start(slave->dev);
2754
2755                 if (slave->link != BOND_LINK_UP) {
2756                         if (time_in_range(jiffies,
2757                                 trans_start - delta_in_ticks,
2758                                 trans_start + delta_in_ticks + extra_ticks) &&
2759                             time_in_range(jiffies,
2760                                 slave->dev->last_rx - delta_in_ticks,
2761                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2762
2763                                 slave->link  = BOND_LINK_UP;
2764                                 bond_set_active_slave(slave);
2765
2766                                 /* primary_slave has no meaning in round-robin
2767                                  * mode. the window of a slave being up and
2768                                  * curr_active_slave being null after enslaving
2769                                  * is closed.
2770                                  */
2771                                 if (!oldcurrent) {
2772                                         pr_info("%s: link status definitely up for interface %s, ",
2773                                                 bond->dev->name,
2774                                                 slave->dev->name);
2775                                         do_failover = 1;
2776                                 } else {
2777                                         pr_info("%s: interface %s is now up\n",
2778                                                 bond->dev->name,
2779                                                 slave->dev->name);
2780                                 }
2781                         }
2782                 } else {
2783                         /* slave->link == BOND_LINK_UP */
2784
2785                         /* not all switches will respond to an arp request
2786                          * when the source ip is 0, so don't take the link down
2787                          * if we don't know our ip yet
2788                          */
2789                         if (!time_in_range(jiffies,
2790                                 trans_start - delta_in_ticks,
2791                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2792                             !time_in_range(jiffies,
2793                                 slave->dev->last_rx - delta_in_ticks,
2794                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2795
2796                                 slave->link  = BOND_LINK_DOWN;
2797                                 bond_set_backup_slave(slave);
2798
2799                                 if (slave->link_failure_count < UINT_MAX)
2800                                         slave->link_failure_count++;
2801
2802                                 pr_info("%s: interface %s is now down.\n",
2803                                         bond->dev->name,
2804                                         slave->dev->name);
2805
2806                                 if (slave == oldcurrent)
2807                                         do_failover = 1;
2808                         }
2809                 }
2810
2811                 /* note: if switch is in round-robin mode, all links
2812                  * must tx arp to ensure all links rx an arp - otherwise
2813                  * links may oscillate or not come up at all; if switch is
2814                  * in something like xor mode, there is nothing we can
2815                  * do - all replies will be rx'ed on same link causing slaves
2816                  * to be unstable during low/no traffic periods
2817                  */
2818                 if (IS_UP(slave->dev))
2819                         bond_arp_send_all(bond, slave);
2820         }
2821
2822         if (do_failover) {
2823                 block_netpoll_tx();
2824                 write_lock_bh(&bond->curr_slave_lock);
2825
2826                 bond_select_active_slave(bond);
2827
2828                 write_unlock_bh(&bond->curr_slave_lock);
2829                 unblock_netpoll_tx();
2830         }
2831
2832 re_arm:
2833         if (bond->params.arp_interval)
2834                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2835
2836         read_unlock(&bond->lock);
2837 }
2838
2839 /*
2840  * Called to inspect slaves for active-backup mode ARP monitor link state
2841  * changes.  Sets new_link in slaves to specify what action should take
2842  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2843  * to link states must be committed.
2844  *
2845  * Called with bond->lock held for read.
2846  */
2847 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2848 {
2849         struct slave *slave;
2850         int i, commit = 0;
2851         unsigned long trans_start;
2852         int extra_ticks;
2853
2854         /* All the time comparisons below need some extra time. Otherwise, on
2855          * fast networks the ARP probe/reply may arrive within the same jiffy
2856          * as it was sent.  Then, the next time the ARP monitor is run, one
2857          * arp_interval will already have passed in the comparisons.
2858          */
2859         extra_ticks = delta_in_ticks / 2;
2860
2861         bond_for_each_slave(bond, slave, i) {
2862                 slave->new_link = BOND_LINK_NOCHANGE;
2863
2864                 if (slave->link != BOND_LINK_UP) {
2865                         if (time_in_range(jiffies,
2866                                 slave_last_rx(bond, slave) - delta_in_ticks,
2867                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2868
2869                                 slave->new_link = BOND_LINK_UP;
2870                                 commit++;
2871                         }
2872
2873                         continue;
2874                 }
2875
2876                 /*
2877                  * Give slaves 2*delta after being enslaved or made
2878                  * active.  This avoids bouncing, as the last receive
2879                  * times need a full ARP monitor cycle to be updated.
2880                  */
2881                 if (time_in_range(jiffies,
2882                                   slave->jiffies - delta_in_ticks,
2883                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2884                         continue;
2885
2886                 /*
2887                  * Backup slave is down if:
2888                  * - No current_arp_slave AND
2889                  * - more than 3*delta since last receive AND
2890                  * - the bond has an IP address
2891                  *
2892                  * Note: a non-null current_arp_slave indicates
2893                  * the curr_active_slave went down and we are
2894                  * searching for a new one; under this condition
2895                  * we only take the curr_active_slave down - this
2896                  * gives each slave a chance to tx/rx traffic
2897                  * before being taken out
2898                  */
2899                 if (!bond_is_active_slave(slave) &&
2900                     !bond->current_arp_slave &&
2901                     !time_in_range(jiffies,
2902                         slave_last_rx(bond, slave) - delta_in_ticks,
2903                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2904
2905                         slave->new_link = BOND_LINK_DOWN;
2906                         commit++;
2907                 }
2908
2909                 /*
2910                  * Active slave is down if:
2911                  * - more than 2*delta since transmitting OR
2912                  * - (more than 2*delta since receive AND
2913                  *    the bond has an IP address)
2914                  */
2915                 trans_start = dev_trans_start(slave->dev);
2916                 if (bond_is_active_slave(slave) &&
2917                     (!time_in_range(jiffies,
2918                         trans_start - delta_in_ticks,
2919                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
2920                      !time_in_range(jiffies,
2921                         slave_last_rx(bond, slave) - delta_in_ticks,
2922                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2923
2924                         slave->new_link = BOND_LINK_DOWN;
2925                         commit++;
2926                 }
2927         }
2928
2929         return commit;
2930 }
2931
2932 /*
2933  * Called to commit link state changes noted by inspection step of
2934  * active-backup mode ARP monitor.
2935  *
2936  * Called with RTNL and bond->lock for read.
2937  */
2938 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2939 {
2940         struct slave *slave;
2941         int i;
2942         unsigned long trans_start;
2943
2944         bond_for_each_slave(bond, slave, i) {
2945                 switch (slave->new_link) {
2946                 case BOND_LINK_NOCHANGE:
2947                         continue;
2948
2949                 case BOND_LINK_UP:
2950                         trans_start = dev_trans_start(slave->dev);
2951                         if ((!bond->curr_active_slave &&
2952                              time_in_range(jiffies,
2953                                            trans_start - delta_in_ticks,
2954                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2955                             bond->curr_active_slave != slave) {
2956                                 slave->link = BOND_LINK_UP;
2957                                 if (bond->current_arp_slave) {
2958                                         bond_set_slave_inactive_flags(
2959                                                 bond->current_arp_slave);
2960                                         bond->current_arp_slave = NULL;
2961                                 }
2962
2963                                 pr_info("%s: link status definitely up for interface %s.\n",
2964                                         bond->dev->name, slave->dev->name);
2965
2966                                 if (!bond->curr_active_slave ||
2967                                     (slave == bond->primary_slave))
2968                                         goto do_failover;
2969
2970                         }
2971
2972                         continue;
2973
2974                 case BOND_LINK_DOWN:
2975                         if (slave->link_failure_count < UINT_MAX)
2976                                 slave->link_failure_count++;
2977
2978                         slave->link = BOND_LINK_DOWN;
2979                         bond_set_slave_inactive_flags(slave);
2980
2981                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2982                                 bond->dev->name, slave->dev->name);
2983
2984                         if (slave == bond->curr_active_slave) {
2985                                 bond->current_arp_slave = NULL;
2986                                 goto do_failover;
2987                         }
2988
2989                         continue;
2990
2991                 default:
2992                         pr_err("%s: impossible: new_link %d on slave %s\n",
2993                                bond->dev->name, slave->new_link,
2994                                slave->dev->name);
2995                         continue;
2996                 }
2997
2998 do_failover:
2999                 ASSERT_RTNL();
3000                 block_netpoll_tx();
3001                 write_lock_bh(&bond->curr_slave_lock);
3002                 bond_select_active_slave(bond);
3003                 write_unlock_bh(&bond->curr_slave_lock);
3004                 unblock_netpoll_tx();
3005         }
3006
3007         bond_set_carrier(bond);
3008 }
3009
3010 /*
3011  * Send ARP probes for active-backup mode ARP monitor.
3012  *
3013  * Called with bond->lock held for read.
3014  */
3015 static void bond_ab_arp_probe(struct bonding *bond)
3016 {
3017         struct slave *slave;
3018         int i;
3019
3020         read_lock(&bond->curr_slave_lock);
3021
3022         if (bond->current_arp_slave && bond->curr_active_slave)
3023                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3024                         bond->current_arp_slave->dev->name,
3025                         bond->curr_active_slave->dev->name);
3026
3027         if (bond->curr_active_slave) {
3028                 bond_arp_send_all(bond, bond->curr_active_slave);
3029                 read_unlock(&bond->curr_slave_lock);
3030                 return;
3031         }
3032
3033         read_unlock(&bond->curr_slave_lock);
3034
3035         /* if we don't have a curr_active_slave, search for the next available
3036          * backup slave from the current_arp_slave and make it the candidate
3037          * for becoming the curr_active_slave
3038          */
3039
3040         if (!bond->current_arp_slave) {
3041                 bond->current_arp_slave = bond->first_slave;
3042                 if (!bond->current_arp_slave)
3043                         return;
3044         }
3045
3046         bond_set_slave_inactive_flags(bond->current_arp_slave);
3047
3048         /* search for next candidate */
3049         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3050                 if (IS_UP(slave->dev)) {
3051                         slave->link = BOND_LINK_BACK;
3052                         bond_set_slave_active_flags(slave);
3053                         bond_arp_send_all(bond, slave);
3054                         slave->jiffies = jiffies;
3055                         bond->current_arp_slave = slave;
3056                         break;
3057                 }
3058
3059                 /* if the link state is up at this point, we
3060                  * mark it down - this can happen if we have
3061                  * simultaneous link failures and
3062                  * reselect_active_interface doesn't make this
3063                  * one the current slave so it is still marked
3064                  * up when it is actually down
3065                  */
3066                 if (slave->link == BOND_LINK_UP) {
3067                         slave->link = BOND_LINK_DOWN;
3068                         if (slave->link_failure_count < UINT_MAX)
3069                                 slave->link_failure_count++;
3070
3071                         bond_set_slave_inactive_flags(slave);
3072
3073                         pr_info("%s: backup interface %s is now down.\n",
3074                                 bond->dev->name, slave->dev->name);
3075                 }
3076         }
3077 }
3078
3079 void bond_activebackup_arp_mon(struct work_struct *work)
3080 {
3081         struct bonding *bond = container_of(work, struct bonding,
3082                                             arp_work.work);
3083         bool should_notify_peers = false;
3084         int delta_in_ticks;
3085
3086         read_lock(&bond->lock);
3087
3088         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3089
3090         if (bond->slave_cnt == 0)
3091                 goto re_arm;
3092
3093         should_notify_peers = bond_should_notify_peers(bond);
3094
3095         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3096                 read_unlock(&bond->lock);
3097
3098                 /* Race avoidance with bond_close flush of workqueue */
3099                 if (!rtnl_trylock()) {
3100                         read_lock(&bond->lock);
3101                         delta_in_ticks = 1;
3102                         should_notify_peers = false;
3103                         goto re_arm;
3104                 }
3105
3106                 read_lock(&bond->lock);
3107
3108                 bond_ab_arp_commit(bond, delta_in_ticks);
3109
3110                 read_unlock(&bond->lock);
3111                 rtnl_unlock();
3112                 read_lock(&bond->lock);
3113         }
3114
3115         bond_ab_arp_probe(bond);
3116
3117 re_arm:
3118         if (bond->params.arp_interval)
3119                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3120
3121         read_unlock(&bond->lock);
3122
3123         if (should_notify_peers) {
3124                 if (!rtnl_trylock()) {
3125                         read_lock(&bond->lock);
3126                         bond->send_peer_notif++;
3127                         read_unlock(&bond->lock);
3128                         return;
3129                 }
3130                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3131                 rtnl_unlock();
3132         }
3133 }
3134
3135 /*-------------------------- netdev event handling --------------------------*/
3136
3137 /*
3138  * Change device name
3139  */
3140 static int bond_event_changename(struct bonding *bond)
3141 {
3142         bond_remove_proc_entry(bond);
3143         bond_create_proc_entry(bond);
3144
3145         bond_debug_reregister(bond);
3146
3147         return NOTIFY_DONE;
3148 }
3149
3150 static int bond_master_netdev_event(unsigned long event,
3151                                     struct net_device *bond_dev)
3152 {
3153         struct bonding *event_bond = netdev_priv(bond_dev);
3154
3155         switch (event) {
3156         case NETDEV_CHANGENAME:
3157                 return bond_event_changename(event_bond);
3158         case NETDEV_UNREGISTER:
3159                 bond_remove_proc_entry(event_bond);
3160                 break;
3161         case NETDEV_REGISTER:
3162                 bond_create_proc_entry(event_bond);
3163                 break;
3164         default:
3165                 break;
3166         }
3167
3168         return NOTIFY_DONE;
3169 }
3170
3171 static int bond_slave_netdev_event(unsigned long event,
3172                                    struct net_device *slave_dev)
3173 {
3174         struct slave *slave = bond_slave_get_rtnl(slave_dev);
3175         struct bonding *bond;
3176         struct net_device *bond_dev;
3177         u32 old_speed;
3178         u8 old_duplex;
3179
3180         /* A netdev event can be generated while enslaving a device
3181          * before netdev_rx_handler_register is called in which case
3182          * slave will be NULL
3183          */
3184         if (!slave)
3185                 return NOTIFY_DONE;
3186         bond_dev = slave->bond->dev;
3187         bond = slave->bond;
3188
3189         switch (event) {
3190         case NETDEV_UNREGISTER:
3191                 if (bond_dev->type != ARPHRD_ETHER)
3192                         bond_release_and_destroy(bond_dev, slave_dev);
3193                 else
3194                         bond_release(bond_dev, slave_dev);
3195                 break;
3196         case NETDEV_UP:
3197         case NETDEV_CHANGE:
3198                 old_speed = slave->speed;
3199                 old_duplex = slave->duplex;
3200
3201                 bond_update_speed_duplex(slave);
3202
3203                 if (bond->params.mode == BOND_MODE_8023AD) {
3204                         if (old_speed != slave->speed)
3205                                 bond_3ad_adapter_speed_changed(slave);
3206                         if (old_duplex != slave->duplex)
3207                                 bond_3ad_adapter_duplex_changed(slave);
3208                 }
3209                 break;
3210         case NETDEV_DOWN:
3211                 /*
3212                  * ... Or is it this?
3213                  */
3214                 break;
3215         case NETDEV_CHANGEMTU:
3216                 /*
3217                  * TODO: Should slaves be allowed to
3218                  * independently alter their MTU?  For
3219                  * an active-backup bond, slaves need
3220                  * not be the same type of device, so
3221                  * MTUs may vary.  For other modes,
3222                  * slaves arguably should have the
3223                  * same MTUs. To do this, we'd need to
3224                  * take over the slave's change_mtu
3225                  * function for the duration of their
3226                  * servitude.
3227                  */
3228                 break;
3229         case NETDEV_CHANGENAME:
3230                 /*
3231                  * TODO: handle changing the primary's name
3232                  */
3233                 break;
3234         case NETDEV_FEAT_CHANGE:
3235                 bond_compute_features(bond);
3236                 break;
3237         default:
3238                 break;
3239         }
3240
3241         return NOTIFY_DONE;
3242 }
3243
3244 /*
3245  * bond_netdev_event: handle netdev notifier chain events.
3246  *
3247  * This function receives events for the netdev chain.  The caller (an
3248  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3249  * locks for us to safely manipulate the slave devices (RTNL lock,
3250  * dev_probe_lock).
3251  */
3252 static int bond_netdev_event(struct notifier_block *this,
3253                              unsigned long event, void *ptr)
3254 {
3255         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3256
3257         pr_debug("event_dev: %s, event: %lx\n",
3258                  event_dev ? event_dev->name : "None",
3259                  event);
3260
3261         if (!(event_dev->priv_flags & IFF_BONDING))
3262                 return NOTIFY_DONE;
3263
3264         if (event_dev->flags & IFF_MASTER) {
3265                 pr_debug("IFF_MASTER\n");
3266                 return bond_master_netdev_event(event, event_dev);
3267         }
3268
3269         if (event_dev->flags & IFF_SLAVE) {
3270                 pr_debug("IFF_SLAVE\n");
3271                 return bond_slave_netdev_event(event, event_dev);
3272         }
3273
3274         return NOTIFY_DONE;
3275 }
3276
3277 static struct notifier_block bond_netdev_notifier = {
3278         .notifier_call = bond_netdev_event,
3279 };
3280
3281 /*---------------------------- Hashing Policies -----------------------------*/
3282
3283 /*
3284  * Hash for the output device based upon layer 2 data
3285  */
3286 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3287 {
3288         struct ethhdr *data = (struct ethhdr *)skb->data;
3289
3290         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3291                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3292
3293         return 0;
3294 }
3295
3296 /*
3297  * Hash for the output device based upon layer 2 and layer 3 data. If
3298  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3299  */
3300 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3301 {
3302         const struct ethhdr *data;
3303         const struct iphdr *iph;
3304         const struct ipv6hdr *ipv6h;
3305         u32 v6hash;
3306         const __be32 *s, *d;
3307
3308         if (skb->protocol == htons(ETH_P_IP) &&
3309             pskb_network_may_pull(skb, sizeof(*iph))) {
3310                 iph = ip_hdr(skb);
3311                 data = (struct ethhdr *)skb->data;
3312                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3313                         (data->h_dest[5] ^ data->h_source[5])) % count;
3314         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3315                    pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3316                 ipv6h = ipv6_hdr(skb);
3317                 data = (struct ethhdr *)skb->data;
3318                 s = &ipv6h->saddr.s6_addr32[0];
3319                 d = &ipv6h->daddr.s6_addr32[0];
3320                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3321                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3322                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3323         }
3324
3325         return bond_xmit_hash_policy_l2(skb, count);
3326 }
3327
3328 /*
3329  * Hash for the output device based upon layer 3 and layer 4 data. If
3330  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3331  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3332  */
3333 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3334 {
3335         u32 layer4_xor = 0;
3336         const struct iphdr *iph;
3337         const struct ipv6hdr *ipv6h;
3338         const __be32 *s, *d;
3339         const __be16 *l4 = NULL;
3340         __be16 _l4[2];
3341         int noff = skb_network_offset(skb);
3342         int poff;
3343
3344         if (skb->protocol == htons(ETH_P_IP) &&
3345             pskb_may_pull(skb, noff + sizeof(*iph))) {
3346                 iph = ip_hdr(skb);
3347                 poff = proto_ports_offset(iph->protocol);
3348
3349                 if (!ip_is_fragment(iph) && poff >= 0) {
3350                         l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3351                                                 sizeof(_l4), &_l4);
3352                         if (l4)
3353                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3354                 }
3355                 return (layer4_xor ^
3356                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3357         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3358                    pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3359                 ipv6h = ipv6_hdr(skb);
3360                 poff = proto_ports_offset(ipv6h->nexthdr);
3361                 if (poff >= 0) {
3362                         l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3363                                                 sizeof(_l4), &_l4);
3364                         if (l4)
3365                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3366                 }
3367                 s = &ipv6h->saddr.s6_addr32[0];
3368                 d = &ipv6h->daddr.s6_addr32[0];
3369                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3370                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3371                                (layer4_xor >> 8);
3372                 return layer4_xor % count;
3373         }
3374
3375         return bond_xmit_hash_policy_l2(skb, count);
3376 }
3377
3378 /*-------------------------- Device entry points ----------------------------*/
3379
3380 static void bond_work_init_all(struct bonding *bond)
3381 {
3382         INIT_DELAYED_WORK(&bond->mcast_work,
3383                           bond_resend_igmp_join_requests_delayed);
3384         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3385         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3386         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3387                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3388         else
3389                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3390         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3391 }
3392
3393 static void bond_work_cancel_all(struct bonding *bond)
3394 {
3395         cancel_delayed_work_sync(&bond->mii_work);
3396         cancel_delayed_work_sync(&bond->arp_work);
3397         cancel_delayed_work_sync(&bond->alb_work);
3398         cancel_delayed_work_sync(&bond->ad_work);
3399         cancel_delayed_work_sync(&bond->mcast_work);
3400 }
3401
3402 static int bond_open(struct net_device *bond_dev)
3403 {
3404         struct bonding *bond = netdev_priv(bond_dev);
3405         struct slave *slave;
3406         int i;
3407
3408         /* reset slave->backup and slave->inactive */
3409         read_lock(&bond->lock);
3410         if (bond->slave_cnt > 0) {
3411                 read_lock(&bond->curr_slave_lock);
3412                 bond_for_each_slave(bond, slave, i) {
3413                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3414                                 && (slave != bond->curr_active_slave)) {
3415                                 bond_set_slave_inactive_flags(slave);
3416                         } else {
3417                                 bond_set_slave_active_flags(slave);
3418                         }
3419                 }
3420                 read_unlock(&bond->curr_slave_lock);
3421         }
3422         read_unlock(&bond->lock);
3423
3424         bond_work_init_all(bond);
3425
3426         if (bond_is_lb(bond)) {
3427                 /* bond_alb_initialize must be called before the timer
3428                  * is started.
3429                  */
3430                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3431                         return -ENOMEM;
3432                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3433         }
3434
3435         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3436                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3437
3438         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3439                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3440                 if (bond->params.arp_validate)
3441                         bond->recv_probe = bond_arp_rcv;
3442         }
3443
3444         if (bond->params.mode == BOND_MODE_8023AD) {
3445                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3446                 /* register to receive LACPDUs */
3447                 bond->recv_probe = bond_3ad_lacpdu_recv;
3448                 bond_3ad_initiate_agg_selection(bond, 1);
3449         }
3450
3451         return 0;
3452 }
3453
3454 static int bond_close(struct net_device *bond_dev)
3455 {
3456         struct bonding *bond = netdev_priv(bond_dev);
3457
3458         write_lock_bh(&bond->lock);
3459         bond->send_peer_notif = 0;
3460         write_unlock_bh(&bond->lock);
3461
3462         bond_work_cancel_all(bond);
3463         if (bond_is_lb(bond)) {
3464                 /* Must be called only after all
3465                  * slaves have been released
3466                  */
3467                 bond_alb_deinitialize(bond);
3468         }
3469         bond->recv_probe = NULL;
3470
3471         return 0;
3472 }
3473
3474 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3475                                                 struct rtnl_link_stats64 *stats)
3476 {
3477         struct bonding *bond = netdev_priv(bond_dev);
3478         struct rtnl_link_stats64 temp;
3479         struct slave *slave;
3480         int i;
3481
3482         memset(stats, 0, sizeof(*stats));
3483
3484         read_lock_bh(&bond->lock);
3485
3486         bond_for_each_slave(bond, slave, i) {
3487                 const struct rtnl_link_stats64 *sstats =
3488                         dev_get_stats(slave->dev, &temp);
3489
3490                 stats->rx_packets += sstats->rx_packets;
3491                 stats->rx_bytes += sstats->rx_bytes;
3492                 stats->rx_errors += sstats->rx_errors;
3493                 stats->rx_dropped += sstats->rx_dropped;
3494
3495                 stats->tx_packets += sstats->tx_packets;
3496                 stats->tx_bytes += sstats->tx_bytes;
3497                 stats->tx_errors += sstats->tx_errors;
3498                 stats->tx_dropped += sstats->tx_dropped;
3499
3500                 stats->multicast += sstats->multicast;
3501                 stats->collisions += sstats->collisions;
3502
3503                 stats->rx_length_errors += sstats->rx_length_errors;
3504                 stats->rx_over_errors += sstats->rx_over_errors;
3505                 stats->rx_crc_errors += sstats->rx_crc_errors;
3506                 stats->rx_frame_errors += sstats->rx_frame_errors;
3507                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3508                 stats->rx_missed_errors += sstats->rx_missed_errors;
3509
3510                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3511                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3512                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3513                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3514                 stats->tx_window_errors += sstats->tx_window_errors;
3515         }
3516
3517         read_unlock_bh(&bond->lock);
3518
3519         return stats;
3520 }
3521
3522 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3523 {
3524         struct net_device *slave_dev = NULL;
3525         struct ifbond k_binfo;
3526         struct ifbond __user *u_binfo = NULL;
3527         struct ifslave k_sinfo;
3528         struct ifslave __user *u_sinfo = NULL;
3529         struct mii_ioctl_data *mii = NULL;
3530         struct net *net;
3531         int res = 0;
3532
3533         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3534
3535         switch (cmd) {
3536         case SIOCGMIIPHY:
3537                 mii = if_mii(ifr);
3538                 if (!mii)
3539                         return -EINVAL;
3540
3541                 mii->phy_id = 0;
3542                 /* Fall Through */
3543         case SIOCGMIIREG:
3544                 /*
3545                  * We do this again just in case we were called by SIOCGMIIREG
3546                  * instead of SIOCGMIIPHY.
3547                  */
3548                 mii = if_mii(ifr);
3549                 if (!mii)
3550                         return -EINVAL;
3551
3552
3553                 if (mii->reg_num == 1) {
3554                         struct bonding *bond = netdev_priv(bond_dev);
3555                         mii->val_out = 0;
3556                         read_lock(&bond->lock);
3557                         read_lock(&bond->curr_slave_lock);
3558                         if (netif_carrier_ok(bond->dev))
3559                                 mii->val_out = BMSR_LSTATUS;
3560
3561                         read_unlock(&bond->curr_slave_lock);
3562                         read_unlock(&bond->lock);
3563                 }
3564
3565                 return 0;
3566         case BOND_INFO_QUERY_OLD:
3567         case SIOCBONDINFOQUERY:
3568                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3569
3570                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3571                         return -EFAULT;
3572
3573                 res = bond_info_query(bond_dev, &k_binfo);
3574                 if (res == 0 &&
3575                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3576                         return -EFAULT;
3577
3578                 return res;
3579         case BOND_SLAVE_INFO_QUERY_OLD:
3580         case SIOCBONDSLAVEINFOQUERY:
3581                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3582
3583                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3584                         return -EFAULT;
3585
3586                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3587                 if (res == 0 &&
3588                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3589                         return -EFAULT;
3590
3591                 return res;
3592         default:
3593                 /* Go on */
3594                 break;
3595         }
3596
3597         net = dev_net(bond_dev);
3598
3599         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3600                 return -EPERM;
3601
3602         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3603
3604         pr_debug("slave_dev=%p:\n", slave_dev);
3605
3606         if (!slave_dev)
3607                 res = -ENODEV;
3608         else {
3609                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3610                 switch (cmd) {
3611                 case BOND_ENSLAVE_OLD:
3612                 case SIOCBONDENSLAVE:
3613                         res = bond_enslave(bond_dev, slave_dev);
3614                         break;
3615                 case BOND_RELEASE_OLD:
3616                 case SIOCBONDRELEASE:
3617                         res = bond_release(bond_dev, slave_dev);
3618                         break;
3619                 case BOND_SETHWADDR_OLD:
3620                 case SIOCBONDSETHWADDR:
3621                         bond_set_dev_addr(bond_dev, slave_dev);
3622                         res = 0;
3623                         break;
3624                 case BOND_CHANGE_ACTIVE_OLD:
3625                 case SIOCBONDCHANGEACTIVE:
3626                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3627                         break;
3628                 default:
3629                         res = -EOPNOTSUPP;
3630                 }
3631
3632                 dev_put(slave_dev);
3633         }
3634
3635         return res;
3636 }
3637
3638 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3639 {
3640         struct bonding *bond = netdev_priv(bond_dev);
3641
3642         if (change & IFF_PROMISC)
3643                 bond_set_promiscuity(bond,
3644                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3645
3646         if (change & IFF_ALLMULTI)
3647                 bond_set_allmulti(bond,
3648                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3649 }
3650
3651 static void bond_set_rx_mode(struct net_device *bond_dev)
3652 {
3653         struct bonding *bond = netdev_priv(bond_dev);
3654         struct slave *slave;
3655         int i;
3656
3657         read_lock(&bond->lock);
3658
3659         if (USES_PRIMARY(bond->params.mode)) {
3660                 read_lock(&bond->curr_slave_lock);
3661                 slave = bond->curr_active_slave;
3662                 if (slave) {
3663                         dev_uc_sync(slave->dev, bond_dev);
3664                         dev_mc_sync(slave->dev, bond_dev);
3665                 }
3666                 read_unlock(&bond->curr_slave_lock);
3667         } else {
3668                 bond_for_each_slave(bond, slave, i) {
3669                         dev_uc_sync_multiple(slave->dev, bond_dev);
3670                         dev_mc_sync_multiple(slave->dev, bond_dev);
3671                 }
3672         }
3673
3674         read_unlock(&bond->lock);
3675 }
3676
3677 static int bond_neigh_init(struct neighbour *n)
3678 {
3679         struct bonding *bond = netdev_priv(n->dev);
3680         struct slave *slave = bond->first_slave;
3681         const struct net_device_ops *slave_ops;
3682         struct neigh_parms parms;
3683         int ret;
3684
3685         if (!slave)
3686                 return 0;
3687
3688         slave_ops = slave->dev->netdev_ops;
3689
3690         if (!slave_ops->ndo_neigh_setup)
3691                 return 0;
3692
3693         parms.neigh_setup = NULL;
3694         parms.neigh_cleanup = NULL;
3695         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3696         if (ret)
3697                 return ret;
3698
3699         /*
3700          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3701          * after the last slave has been detached.  Assumes that all slaves
3702          * utilize the same neigh_cleanup (true at this writing as only user
3703          * is ipoib).
3704          */
3705         n->parms->neigh_cleanup = parms.neigh_cleanup;
3706
3707         if (!parms.neigh_setup)
3708                 return 0;
3709
3710         return parms.neigh_setup(n);
3711 }
3712
3713 /*
3714  * The bonding ndo_neigh_setup is called at init time beofre any
3715  * slave exists. So we must declare proxy setup function which will
3716  * be used at run time to resolve the actual slave neigh param setup.
3717  *
3718  * It's also called by master devices (such as vlans) to setup their
3719  * underlying devices. In that case - do nothing, we're already set up from
3720  * our init.
3721  */
3722 static int bond_neigh_setup(struct net_device *dev,
3723                             struct neigh_parms *parms)
3724 {
3725         /* modify only our neigh_parms */
3726         if (parms->dev == dev)
3727                 parms->neigh_setup = bond_neigh_init;
3728
3729         return 0;
3730 }
3731
3732 /*
3733  * Change the MTU of all of a master's slaves to match the master
3734  */
3735 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3736 {
3737         struct bonding *bond = netdev_priv(bond_dev);
3738         struct slave *slave, *stop_at;
3739         int res = 0;
3740         int i;
3741
3742         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3743                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3744
3745         /* Can't hold bond->lock with bh disabled here since
3746          * some base drivers panic. On the other hand we can't
3747          * hold bond->lock without bh disabled because we'll
3748          * deadlock. The only solution is to rely on the fact
3749          * that we're under rtnl_lock here, and the slaves
3750          * list won't change. This doesn't solve the problem
3751          * of setting the slave's MTU while it is
3752          * transmitting, but the assumption is that the base
3753          * driver can handle that.
3754          *
3755          * TODO: figure out a way to safely iterate the slaves
3756          * list, but without holding a lock around the actual
3757          * call to the base driver.
3758          */
3759
3760         bond_for_each_slave(bond, slave, i) {
3761                 pr_debug("s %p s->p %p c_m %p\n",
3762                          slave,
3763                          slave->prev,
3764                          slave->dev->netdev_ops->ndo_change_mtu);
3765
3766                 res = dev_set_mtu(slave->dev, new_mtu);
3767
3768                 if (res) {
3769                         /* If we failed to set the slave's mtu to the new value
3770                          * we must abort the operation even in ACTIVE_BACKUP
3771                          * mode, because if we allow the backup slaves to have
3772                          * different mtu values than the active slave we'll
3773                          * need to change their mtu when doing a failover. That
3774                          * means changing their mtu from timer context, which
3775                          * is probably not a good idea.
3776                          */
3777                         pr_debug("err %d %s\n", res, slave->dev->name);
3778                         goto unwind;
3779                 }
3780         }
3781
3782         bond_dev->mtu = new_mtu;
3783
3784         return 0;
3785
3786 unwind:
3787         /* unwind from head to the slave that failed */
3788         stop_at = slave;
3789         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3790                 int tmp_res;
3791
3792                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3793                 if (tmp_res) {
3794                         pr_debug("unwind err %d dev %s\n",
3795                                  tmp_res, slave->dev->name);
3796                 }
3797         }
3798
3799         return res;
3800 }
3801
3802 /*
3803  * Change HW address
3804  *
3805  * Note that many devices must be down to change the HW address, and
3806  * downing the master releases all slaves.  We can make bonds full of
3807  * bonding devices to test this, however.
3808  */
3809 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3810 {
3811         struct bonding *bond = netdev_priv(bond_dev);
3812         struct sockaddr *sa = addr, tmp_sa;
3813         struct slave *slave, *stop_at;
3814         int res = 0;
3815         int i;
3816
3817         if (bond->params.mode == BOND_MODE_ALB)
3818                 return bond_alb_set_mac_address(bond_dev, addr);
3819
3820
3821         pr_debug("bond=%p, name=%s\n",
3822                  bond, bond_dev ? bond_dev->name : "None");
3823
3824         /* If fail_over_mac is enabled, do nothing and return success.
3825          * Returning an error causes ifenslave to fail.
3826          */
3827         if (bond->params.fail_over_mac)
3828                 return 0;
3829
3830         if (!is_valid_ether_addr(sa->sa_data))
3831                 return -EADDRNOTAVAIL;
3832
3833         /* Can't hold bond->lock with bh disabled here since
3834          * some base drivers panic. On the other hand we can't
3835          * hold bond->lock without bh disabled because we'll
3836          * deadlock. The only solution is to rely on the fact
3837          * that we're under rtnl_lock here, and the slaves
3838          * list won't change. This doesn't solve the problem
3839          * of setting the slave's hw address while it is
3840          * transmitting, but the assumption is that the base
3841          * driver can handle that.
3842          *
3843          * TODO: figure out a way to safely iterate the slaves
3844          * list, but without holding a lock around the actual
3845          * call to the base driver.
3846          */
3847
3848         bond_for_each_slave(bond, slave, i) {
3849                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3850                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3851
3852                 if (slave_ops->ndo_set_mac_address == NULL) {
3853                         res = -EOPNOTSUPP;
3854                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3855                         goto unwind;
3856                 }
3857
3858                 res = dev_set_mac_address(slave->dev, addr);
3859                 if (res) {
3860                         /* TODO: consider downing the slave
3861                          * and retry ?
3862                          * User should expect communications
3863                          * breakage anyway until ARP finish
3864                          * updating, so...
3865                          */
3866                         pr_debug("err %d %s\n", res, slave->dev->name);
3867                         goto unwind;
3868                 }
3869         }
3870
3871         /* success */
3872         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3873         return 0;
3874
3875 unwind:
3876         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3877         tmp_sa.sa_family = bond_dev->type;
3878
3879         /* unwind from head to the slave that failed */
3880         stop_at = slave;
3881         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3882                 int tmp_res;
3883
3884                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3885                 if (tmp_res) {
3886                         pr_debug("unwind err %d dev %s\n",
3887                                  tmp_res, slave->dev->name);
3888                 }
3889         }
3890
3891         return res;
3892 }
3893
3894 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3895 {
3896         struct bonding *bond = netdev_priv(bond_dev);
3897         struct slave *slave, *start_at;
3898         int i, slave_no, res = 1;
3899         struct iphdr *iph = ip_hdr(skb);
3900
3901         /*
3902          * Start with the curr_active_slave that joined the bond as the
3903          * default for sending IGMP traffic.  For failover purposes one
3904          * needs to maintain some consistency for the interface that will
3905          * send the join/membership reports.  The curr_active_slave found
3906          * will send all of this type of traffic.
3907          */
3908         if ((iph->protocol == IPPROTO_IGMP) &&
3909             (skb->protocol == htons(ETH_P_IP))) {
3910
3911                 read_lock(&bond->curr_slave_lock);
3912                 slave = bond->curr_active_slave;
3913                 read_unlock(&bond->curr_slave_lock);
3914
3915                 if (!slave)
3916                         goto out;
3917         } else {
3918                 /*
3919                  * Concurrent TX may collide on rr_tx_counter; we accept
3920                  * that as being rare enough not to justify using an
3921                  * atomic op here.
3922                  */
3923                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3924
3925                 bond_for_each_slave(bond, slave, i) {
3926                         slave_no--;
3927                         if (slave_no < 0)
3928                                 break;
3929                 }
3930         }
3931
3932         start_at = slave;
3933         bond_for_each_slave_from(bond, slave, i, start_at) {
3934                 if (IS_UP(slave->dev) &&
3935                     (slave->link == BOND_LINK_UP) &&
3936                     bond_is_active_slave(slave)) {
3937                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3938                         break;
3939                 }
3940         }
3941
3942 out:
3943         if (res) {
3944                 /* no suitable interface, frame not sent */
3945                 kfree_skb(skb);
3946         }
3947
3948         return NETDEV_TX_OK;
3949 }
3950
3951
3952 /*
3953  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3954  * the bond has a usable interface.
3955  */
3956 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3957 {
3958         struct bonding *bond = netdev_priv(bond_dev);
3959         int res = 1;
3960
3961         read_lock(&bond->curr_slave_lock);
3962
3963         if (bond->curr_active_slave)
3964                 res = bond_dev_queue_xmit(bond, skb,
3965                         bond->curr_active_slave->dev);
3966
3967         read_unlock(&bond->curr_slave_lock);
3968
3969         if (res)
3970                 /* no suitable interface, frame not sent */
3971                 kfree_skb(skb);
3972
3973         return NETDEV_TX_OK;
3974 }
3975
3976 /*
3977  * In bond_xmit_xor() , we determine the output device by using a pre-
3978  * determined xmit_hash_policy(), If the selected device is not enabled,
3979  * find the next active slave.
3980  */
3981 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3982 {
3983         struct bonding *bond = netdev_priv(bond_dev);
3984         struct slave *slave, *start_at;
3985         int slave_no;
3986         int i;
3987         int res = 1;
3988
3989         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
3990
3991         bond_for_each_slave(bond, slave, i) {
3992                 slave_no--;
3993                 if (slave_no < 0)
3994                         break;
3995         }
3996
3997         start_at = slave;
3998
3999         bond_for_each_slave_from(bond, slave, i, start_at) {
4000                 if (IS_UP(slave->dev) &&
4001                     (slave->link == BOND_LINK_UP) &&
4002                     bond_is_active_slave(slave)) {
4003                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4004                         break;
4005                 }
4006         }
4007
4008         if (res) {
4009                 /* no suitable interface, frame not sent */
4010                 kfree_skb(skb);
4011         }
4012
4013         return NETDEV_TX_OK;
4014 }
4015
4016 /*
4017  * in broadcast mode, we send everything to all usable interfaces.
4018  */
4019 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4020 {
4021         struct bonding *bond = netdev_priv(bond_dev);
4022         struct slave *slave, *start_at;
4023         struct net_device *tx_dev = NULL;
4024         int i;
4025         int res = 1;
4026
4027         read_lock(&bond->curr_slave_lock);
4028         start_at = bond->curr_active_slave;
4029         read_unlock(&bond->curr_slave_lock);
4030
4031         if (!start_at)
4032                 goto out;
4033
4034         bond_for_each_slave_from(bond, slave, i, start_at) {
4035                 if (IS_UP(slave->dev) &&
4036                     (slave->link == BOND_LINK_UP) &&
4037                     bond_is_active_slave(slave)) {
4038                         if (tx_dev) {
4039                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4040                                 if (!skb2) {
4041                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4042                                                bond_dev->name);
4043                                         continue;
4044                                 }
4045
4046                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4047                                 if (res) {
4048                                         kfree_skb(skb2);
4049                                         continue;
4050                                 }
4051                         }
4052                         tx_dev = slave->dev;
4053                 }
4054         }
4055
4056         if (tx_dev)
4057                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4058
4059 out:
4060         if (res)
4061                 /* no suitable interface, frame not sent */
4062                 kfree_skb(skb);
4063
4064         /* frame sent to all suitable interfaces */
4065         return NETDEV_TX_OK;
4066 }
4067
4068 /*------------------------- Device initialization ---------------------------*/
4069
4070 static void bond_set_xmit_hash_policy(struct bonding *bond)
4071 {
4072         switch (bond->params.xmit_policy) {
4073         case BOND_XMIT_POLICY_LAYER23:
4074                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4075                 break;
4076         case BOND_XMIT_POLICY_LAYER34:
4077                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4078                 break;
4079         case BOND_XMIT_POLICY_LAYER2:
4080         default:
4081                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4082                 break;
4083         }
4084 }
4085
4086 /*
4087  * Lookup the slave that corresponds to a qid
4088  */
4089 static inline int bond_slave_override(struct bonding *bond,
4090                                       struct sk_buff *skb)
4091 {
4092         int i, res = 1;
4093         struct slave *slave = NULL;
4094         struct slave *check_slave;
4095
4096         if (!skb->queue_mapping)
4097                 return 1;
4098
4099         /* Find out if any slaves have the same mapping as this skb. */
4100         bond_for_each_slave(bond, check_slave, i) {
4101                 if (check_slave->queue_id == skb->queue_mapping) {
4102                         slave = check_slave;
4103                         break;
4104                 }
4105         }
4106
4107         /* If the slave isn't UP, use default transmit policy. */
4108         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4109             (slave->link == BOND_LINK_UP)) {
4110                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4111         }
4112
4113         return res;
4114 }
4115
4116
4117 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4118 {
4119         /*
4120          * This helper function exists to help dev_pick_tx get the correct
4121          * destination queue.  Using a helper function skips a call to
4122          * skb_tx_hash and will put the skbs in the queue we expect on their
4123          * way down to the bonding driver.
4124          */
4125         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4126
4127         /*
4128          * Save the original txq to restore before passing to the driver
4129          */
4130         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4131
4132         if (unlikely(txq >= dev->real_num_tx_queues)) {
4133                 do {
4134                         txq -= dev->real_num_tx_queues;
4135                 } while (txq >= dev->real_num_tx_queues);
4136         }
4137         return txq;
4138 }
4139
4140 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4141 {
4142         struct bonding *bond = netdev_priv(dev);
4143
4144         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4145                 if (!bond_slave_override(bond, skb))
4146                         return NETDEV_TX_OK;
4147         }
4148
4149         switch (bond->params.mode) {
4150         case BOND_MODE_ROUNDROBIN:
4151                 return bond_xmit_roundrobin(skb, dev);
4152         case BOND_MODE_ACTIVEBACKUP:
4153                 return bond_xmit_activebackup(skb, dev);
4154         case BOND_MODE_XOR:
4155                 return bond_xmit_xor(skb, dev);
4156         case BOND_MODE_BROADCAST:
4157                 return bond_xmit_broadcast(skb, dev);
4158         case BOND_MODE_8023AD:
4159                 return bond_3ad_xmit_xor(skb, dev);
4160         case BOND_MODE_ALB:
4161         case BOND_MODE_TLB:
4162                 return bond_alb_xmit(skb, dev);
4163         default:
4164                 /* Should never happen, mode already checked */
4165                 pr_err("%s: Error: Unknown bonding mode %d\n",
4166                        dev->name, bond->params.mode);
4167                 WARN_ON_ONCE(1);
4168                 kfree_skb(skb);
4169                 return NETDEV_TX_OK;
4170         }
4171 }
4172
4173 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4174 {
4175         struct bonding *bond = netdev_priv(dev);
4176         netdev_tx_t ret = NETDEV_TX_OK;
4177
4178         /*
4179          * If we risk deadlock from transmitting this in the
4180          * netpoll path, tell netpoll to queue the frame for later tx
4181          */
4182         if (is_netpoll_tx_blocked(dev))
4183                 return NETDEV_TX_BUSY;
4184
4185         read_lock(&bond->lock);
4186
4187         if (bond->slave_cnt)
4188                 ret = __bond_start_xmit(skb, dev);
4189         else
4190                 kfree_skb(skb);
4191
4192         read_unlock(&bond->lock);
4193
4194         return ret;
4195 }
4196
4197 /*
4198  * set bond mode specific net device operations
4199  */
4200 void bond_set_mode_ops(struct bonding *bond, int mode)
4201 {
4202         struct net_device *bond_dev = bond->dev;
4203
4204         switch (mode) {
4205         case BOND_MODE_ROUNDROBIN:
4206                 break;
4207         case BOND_MODE_ACTIVEBACKUP:
4208                 break;
4209         case BOND_MODE_XOR:
4210                 bond_set_xmit_hash_policy(bond);
4211                 break;
4212         case BOND_MODE_BROADCAST:
4213                 break;
4214         case BOND_MODE_8023AD:
4215                 bond_set_xmit_hash_policy(bond);
4216                 break;
4217         case BOND_MODE_ALB:
4218                 /* FALLTHRU */
4219         case BOND_MODE_TLB:
4220                 break;
4221         default:
4222                 /* Should never happen, mode already checked */
4223                 pr_err("%s: Error: Unknown bonding mode %d\n",
4224                        bond_dev->name, mode);
4225                 break;
4226         }
4227 }
4228
4229 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4230                                      struct ethtool_cmd *ecmd)
4231 {
4232         struct bonding *bond = netdev_priv(bond_dev);
4233         struct slave *slave;
4234         int i;
4235         unsigned long speed = 0;
4236
4237         ecmd->duplex = DUPLEX_UNKNOWN;
4238         ecmd->port = PORT_OTHER;
4239
4240         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4241          * do not need to check mode.  Though link speed might not represent
4242          * the true receive or transmit bandwidth (not all modes are symmetric)
4243          * this is an accurate maximum.
4244          */
4245         read_lock(&bond->lock);
4246         bond_for_each_slave(bond, slave, i) {
4247                 if (SLAVE_IS_OK(slave)) {
4248                         if (slave->speed != SPEED_UNKNOWN)
4249                                 speed += slave->speed;
4250                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4251                             slave->duplex != DUPLEX_UNKNOWN)
4252                                 ecmd->duplex = slave->duplex;
4253                 }
4254         }
4255         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4256         read_unlock(&bond->lock);
4257         return 0;
4258 }
4259
4260 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4261                                      struct ethtool_drvinfo *drvinfo)
4262 {
4263         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4264         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4265         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4266                  BOND_ABI_VERSION);
4267 }
4268
4269 static const struct ethtool_ops bond_ethtool_ops = {
4270         .get_drvinfo            = bond_ethtool_get_drvinfo,
4271         .get_settings           = bond_ethtool_get_settings,
4272         .get_link               = ethtool_op_get_link,
4273 };
4274
4275 static const struct net_device_ops bond_netdev_ops = {
4276         .ndo_init               = bond_init,
4277         .ndo_uninit             = bond_uninit,
4278         .ndo_open               = bond_open,
4279         .ndo_stop               = bond_close,
4280         .ndo_start_xmit         = bond_start_xmit,
4281         .ndo_select_queue       = bond_select_queue,
4282         .ndo_get_stats64        = bond_get_stats,
4283         .ndo_do_ioctl           = bond_do_ioctl,
4284         .ndo_change_rx_flags    = bond_change_rx_flags,
4285         .ndo_set_rx_mode        = bond_set_rx_mode,
4286         .ndo_change_mtu         = bond_change_mtu,
4287         .ndo_set_mac_address    = bond_set_mac_address,
4288         .ndo_neigh_setup        = bond_neigh_setup,
4289         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4290         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4291 #ifdef CONFIG_NET_POLL_CONTROLLER
4292         .ndo_netpoll_setup      = bond_netpoll_setup,
4293         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4294         .ndo_poll_controller    = bond_poll_controller,
4295 #endif
4296         .ndo_add_slave          = bond_enslave,
4297         .ndo_del_slave          = bond_release,
4298         .ndo_fix_features       = bond_fix_features,
4299 };
4300
4301 static const struct device_type bond_type = {
4302         .name = "bond",
4303 };
4304
4305 static void bond_destructor(struct net_device *bond_dev)
4306 {
4307         struct bonding *bond = netdev_priv(bond_dev);
4308         if (bond->wq)
4309                 destroy_workqueue(bond->wq);
4310         free_netdev(bond_dev);
4311 }
4312
4313 static void bond_setup(struct net_device *bond_dev)
4314 {
4315         struct bonding *bond = netdev_priv(bond_dev);
4316
4317         /* initialize rwlocks */
4318         rwlock_init(&bond->lock);
4319         rwlock_init(&bond->curr_slave_lock);
4320
4321         bond->params = bonding_defaults;
4322
4323         /* Initialize pointers */
4324         bond->dev = bond_dev;
4325         INIT_LIST_HEAD(&bond->vlan_list);
4326
4327         /* Initialize the device entry points */
4328         ether_setup(bond_dev);
4329         bond_dev->netdev_ops = &bond_netdev_ops;
4330         bond_dev->ethtool_ops = &bond_ethtool_ops;
4331         bond_set_mode_ops(bond, bond->params.mode);
4332
4333         bond_dev->destructor = bond_destructor;
4334
4335         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4336
4337         /* Initialize the device options */
4338         bond_dev->tx_queue_len = 0;
4339         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4340         bond_dev->priv_flags |= IFF_BONDING;
4341         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4342
4343         /* At first, we block adding VLANs. That's the only way to
4344          * prevent problems that occur when adding VLANs over an
4345          * empty bond. The block will be removed once non-challenged
4346          * slaves are enslaved.
4347          */
4348         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4349
4350         /* don't acquire bond device's netif_tx_lock when
4351          * transmitting */
4352         bond_dev->features |= NETIF_F_LLTX;
4353
4354         /* By default, we declare the bond to be fully
4355          * VLAN hardware accelerated capable. Special
4356          * care is taken in the various xmit functions
4357          * when there are slaves that are not hw accel
4358          * capable
4359          */
4360
4361         bond_dev->hw_features = BOND_VLAN_FEATURES |
4362                                 NETIF_F_HW_VLAN_CTAG_TX |
4363                                 NETIF_F_HW_VLAN_CTAG_RX |
4364                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4365
4366         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4367         bond_dev->features |= bond_dev->hw_features;
4368 }
4369
4370 /*
4371 * Destroy a bonding device.
4372 * Must be under rtnl_lock when this function is called.
4373 */
4374 static void bond_uninit(struct net_device *bond_dev)
4375 {
4376         struct bonding *bond = netdev_priv(bond_dev);
4377         struct vlan_entry *vlan, *tmp;
4378
4379         bond_netpoll_cleanup(bond_dev);
4380
4381         /* Release the bonded slaves */
4382         while (bond->first_slave != NULL)
4383                 __bond_release_one(bond_dev, bond->first_slave->dev, true);
4384         pr_info("%s: released all slaves\n", bond_dev->name);
4385
4386         list_del(&bond->bond_list);
4387
4388         bond_debug_unregister(bond);
4389
4390         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4391                 list_del(&vlan->vlan_list);
4392                 kfree(vlan);
4393         }
4394 }
4395
4396 /*------------------------- Module initialization ---------------------------*/
4397
4398 /*
4399  * Convert string input module parms.  Accept either the
4400  * number of the mode or its string name.  A bit complicated because
4401  * some mode names are substrings of other names, and calls from sysfs
4402  * may have whitespace in the name (trailing newlines, for example).
4403  */
4404 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4405 {
4406         int modeint = -1, i, rv;
4407         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4408
4409         for (p = (char *)buf; *p; p++)
4410                 if (!(isdigit(*p) || isspace(*p)))
4411                         break;
4412
4413         if (*p)
4414                 rv = sscanf(buf, "%20s", modestr);
4415         else
4416                 rv = sscanf(buf, "%d", &modeint);
4417
4418         if (!rv)
4419                 return -1;
4420
4421         for (i = 0; tbl[i].modename; i++) {
4422                 if (modeint == tbl[i].mode)
4423                         return tbl[i].mode;
4424                 if (strcmp(modestr, tbl[i].modename) == 0)
4425                         return tbl[i].mode;
4426         }
4427
4428         return -1;
4429 }
4430
4431 static int bond_check_params(struct bond_params *params)
4432 {
4433         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4434         int arp_all_targets_value;
4435
4436         /*
4437          * Convert string parameters.
4438          */
4439         if (mode) {
4440                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4441                 if (bond_mode == -1) {
4442                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4443                                mode == NULL ? "NULL" : mode);
4444                         return -EINVAL;
4445                 }
4446         }
4447
4448         if (xmit_hash_policy) {
4449                 if ((bond_mode != BOND_MODE_XOR) &&
4450                     (bond_mode != BOND_MODE_8023AD)) {
4451                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4452                                bond_mode_name(bond_mode));
4453                 } else {
4454                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4455                                                         xmit_hashtype_tbl);
4456                         if (xmit_hashtype == -1) {
4457                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4458                                        xmit_hash_policy == NULL ? "NULL" :
4459                                        xmit_hash_policy);
4460                                 return -EINVAL;
4461                         }
4462                 }
4463         }
4464
4465         if (lacp_rate) {
4466                 if (bond_mode != BOND_MODE_8023AD) {
4467                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4468                                 bond_mode_name(bond_mode));
4469                 } else {
4470                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4471                         if (lacp_fast == -1) {
4472                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4473                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4474                                 return -EINVAL;
4475                         }
4476                 }
4477         }
4478
4479         if (ad_select) {
4480                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4481                 if (params->ad_select == -1) {
4482                         pr_err("Error: Invalid ad_select \"%s\"\n",
4483                                ad_select == NULL ? "NULL" : ad_select);
4484                         return -EINVAL;
4485                 }
4486
4487                 if (bond_mode != BOND_MODE_8023AD) {
4488                         pr_warning("ad_select param only affects 802.3ad mode\n");
4489                 }
4490         } else {
4491                 params->ad_select = BOND_AD_STABLE;
4492         }
4493
4494         if (max_bonds < 0) {
4495                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4496                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4497                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4498         }
4499
4500         if (miimon < 0) {
4501                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4502                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4503                 miimon = BOND_LINK_MON_INTERV;
4504         }
4505
4506         if (updelay < 0) {
4507                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4508                            updelay, INT_MAX);
4509                 updelay = 0;
4510         }
4511
4512         if (downdelay < 0) {
4513                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4514                            downdelay, INT_MAX);
4515                 downdelay = 0;
4516         }
4517
4518         if ((use_carrier != 0) && (use_carrier != 1)) {
4519                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4520                            use_carrier);
4521                 use_carrier = 1;
4522         }
4523
4524         if (num_peer_notif < 0 || num_peer_notif > 255) {
4525                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4526                            num_peer_notif);
4527                 num_peer_notif = 1;
4528         }
4529
4530         /* reset values for 802.3ad */
4531         if (bond_mode == BOND_MODE_8023AD) {
4532                 if (!miimon) {
4533                         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");
4534                         pr_warning("Forcing miimon to 100msec\n");
4535                         miimon = 100;
4536                 }
4537         }
4538
4539         if (tx_queues < 1 || tx_queues > 255) {
4540                 pr_warning("Warning: tx_queues (%d) should be between "
4541                            "1 and 255, resetting to %d\n",
4542                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4543                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4544         }
4545
4546         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4547                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4548                            "not of valid value (0/1), so it was set to "
4549                            "0\n", all_slaves_active);
4550                 all_slaves_active = 0;
4551         }
4552
4553         if (resend_igmp < 0 || resend_igmp > 255) {
4554                 pr_warning("Warning: resend_igmp (%d) should be between "
4555                            "0 and 255, resetting to %d\n",
4556                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4557                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4558         }
4559
4560         /* reset values for TLB/ALB */
4561         if ((bond_mode == BOND_MODE_TLB) ||
4562             (bond_mode == BOND_MODE_ALB)) {
4563                 if (!miimon) {
4564                         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");
4565                         pr_warning("Forcing miimon to 100msec\n");
4566                         miimon = 100;
4567                 }
4568         }
4569
4570         if (bond_mode == BOND_MODE_ALB) {
4571                 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",
4572                           updelay);
4573         }
4574
4575         if (!miimon) {
4576                 if (updelay || downdelay) {
4577                         /* just warn the user the up/down delay will have
4578                          * no effect since miimon is zero...
4579                          */
4580                         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",
4581                                    updelay, downdelay);
4582                 }
4583         } else {
4584                 /* don't allow arp monitoring */
4585                 if (arp_interval) {
4586                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4587                                    miimon, arp_interval);
4588                         arp_interval = 0;
4589                 }
4590
4591                 if ((updelay % miimon) != 0) {
4592                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4593                                    updelay, miimon,
4594                                    (updelay / miimon) * miimon);
4595                 }
4596
4597                 updelay /= miimon;
4598
4599                 if ((downdelay % miimon) != 0) {
4600                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4601                                    downdelay, miimon,
4602                                    (downdelay / miimon) * miimon);
4603                 }
4604
4605                 downdelay /= miimon;
4606         }
4607
4608         if (arp_interval < 0) {
4609                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4610                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4611                 arp_interval = BOND_LINK_ARP_INTERV;
4612         }
4613
4614         for (arp_ip_count = 0, i = 0;
4615              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4616                 /* not complete check, but should be good enough to
4617                    catch mistakes */
4618                 __be32 ip = in_aton(arp_ip_target[i]);
4619                 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4620                     ip == htonl(INADDR_BROADCAST)) {
4621                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4622                                    arp_ip_target[i]);
4623                         arp_interval = 0;
4624                 } else {
4625                         if (bond_get_targets_ip(arp_target, ip) == -1)
4626                                 arp_target[arp_ip_count++] = ip;
4627                         else
4628                                 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4629                                            &ip);
4630                 }
4631         }
4632
4633         if (arp_interval && !arp_ip_count) {
4634                 /* don't allow arping if no arp_ip_target given... */
4635                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4636                            arp_interval);
4637                 arp_interval = 0;
4638         }
4639
4640         if (arp_validate) {
4641                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4642                         pr_err("arp_validate only supported in active-backup mode\n");
4643                         return -EINVAL;
4644                 }
4645                 if (!arp_interval) {
4646                         pr_err("arp_validate requires arp_interval\n");
4647                         return -EINVAL;
4648                 }
4649
4650                 arp_validate_value = bond_parse_parm(arp_validate,
4651                                                      arp_validate_tbl);
4652                 if (arp_validate_value == -1) {
4653                         pr_err("Error: invalid arp_validate \"%s\"\n",
4654                                arp_validate == NULL ? "NULL" : arp_validate);
4655                         return -EINVAL;
4656                 }
4657         } else
4658                 arp_validate_value = 0;
4659
4660         arp_all_targets_value = 0;
4661         if (arp_all_targets) {
4662                 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4663                                                         arp_all_targets_tbl);
4664
4665                 if (arp_all_targets_value == -1) {
4666                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4667                                arp_all_targets);
4668                         arp_all_targets_value = 0;
4669                 }
4670         }
4671
4672         if (miimon) {
4673                 pr_info("MII link monitoring set to %d ms\n", miimon);
4674         } else if (arp_interval) {
4675                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4676                         arp_interval,
4677                         arp_validate_tbl[arp_validate_value].modename,
4678                         arp_ip_count);
4679
4680                 for (i = 0; i < arp_ip_count; i++)
4681                         pr_info(" %s", arp_ip_target[i]);
4682
4683                 pr_info("\n");
4684
4685         } else if (max_bonds) {
4686                 /* miimon and arp_interval not set, we need one so things
4687                  * work as expected, see bonding.txt for details
4688                  */
4689                 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");
4690         }
4691
4692         if (primary && !USES_PRIMARY(bond_mode)) {
4693                 /* currently, using a primary only makes sense
4694                  * in active backup, TLB or ALB modes
4695                  */
4696                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4697                            primary, bond_mode_name(bond_mode));
4698                 primary = NULL;
4699         }
4700
4701         if (primary && primary_reselect) {
4702                 primary_reselect_value = bond_parse_parm(primary_reselect,
4703                                                          pri_reselect_tbl);
4704                 if (primary_reselect_value == -1) {
4705                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4706                                primary_reselect ==
4707                                         NULL ? "NULL" : primary_reselect);
4708                         return -EINVAL;
4709                 }
4710         } else {
4711                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4712         }
4713
4714         if (fail_over_mac) {
4715                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4716                                                       fail_over_mac_tbl);
4717                 if (fail_over_mac_value == -1) {
4718                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4719                                arp_validate == NULL ? "NULL" : arp_validate);
4720                         return -EINVAL;
4721                 }
4722
4723                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4724                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4725         } else {
4726                 fail_over_mac_value = BOND_FOM_NONE;
4727         }
4728
4729         /* fill params struct with the proper values */
4730         params->mode = bond_mode;
4731         params->xmit_policy = xmit_hashtype;
4732         params->miimon = miimon;
4733         params->num_peer_notif = num_peer_notif;
4734         params->arp_interval = arp_interval;
4735         params->arp_validate = arp_validate_value;
4736         params->arp_all_targets = arp_all_targets_value;
4737         params->updelay = updelay;
4738         params->downdelay = downdelay;
4739         params->use_carrier = use_carrier;
4740         params->lacp_fast = lacp_fast;
4741         params->primary[0] = 0;
4742         params->primary_reselect = primary_reselect_value;
4743         params->fail_over_mac = fail_over_mac_value;
4744         params->tx_queues = tx_queues;
4745         params->all_slaves_active = all_slaves_active;
4746         params->resend_igmp = resend_igmp;
4747         params->min_links = min_links;
4748
4749         if (primary) {
4750                 strncpy(params->primary, primary, IFNAMSIZ);
4751                 params->primary[IFNAMSIZ - 1] = 0;
4752         }
4753
4754         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4755
4756         return 0;
4757 }
4758
4759 static struct lock_class_key bonding_netdev_xmit_lock_key;
4760 static struct lock_class_key bonding_netdev_addr_lock_key;
4761 static struct lock_class_key bonding_tx_busylock_key;
4762
4763 static void bond_set_lockdep_class_one(struct net_device *dev,
4764                                        struct netdev_queue *txq,
4765                                        void *_unused)
4766 {
4767         lockdep_set_class(&txq->_xmit_lock,
4768                           &bonding_netdev_xmit_lock_key);
4769 }
4770
4771 static void bond_set_lockdep_class(struct net_device *dev)
4772 {
4773         lockdep_set_class(&dev->addr_list_lock,
4774                           &bonding_netdev_addr_lock_key);
4775         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4776         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4777 }
4778
4779 /*
4780  * Called from registration process
4781  */
4782 static int bond_init(struct net_device *bond_dev)
4783 {
4784         struct bonding *bond = netdev_priv(bond_dev);
4785         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4786         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4787
4788         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4789
4790         /*
4791          * Initialize locks that may be required during
4792          * en/deslave operations.  All of the bond_open work
4793          * (of which this is part) should really be moved to
4794          * a phase prior to dev_open
4795          */
4796         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4797         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4798
4799         bond->wq = create_singlethread_workqueue(bond_dev->name);
4800         if (!bond->wq)
4801                 return -ENOMEM;
4802
4803         bond_set_lockdep_class(bond_dev);
4804
4805         list_add_tail(&bond->bond_list, &bn->dev_list);
4806
4807         bond_prepare_sysfs_group(bond);
4808
4809         bond_debug_register(bond);
4810
4811         /* Ensure valid dev_addr */
4812         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4813             bond_dev->addr_assign_type == NET_ADDR_PERM)
4814                 eth_hw_addr_random(bond_dev);
4815
4816         return 0;
4817 }
4818
4819 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4820 {
4821         if (tb[IFLA_ADDRESS]) {
4822                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4823                         return -EINVAL;
4824                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4825                         return -EADDRNOTAVAIL;
4826         }
4827         return 0;
4828 }
4829
4830 static unsigned int bond_get_num_tx_queues(void)
4831 {
4832         return tx_queues;
4833 }
4834
4835 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4836         .kind                   = "bond",
4837         .priv_size              = sizeof(struct bonding),
4838         .setup                  = bond_setup,
4839         .validate               = bond_validate,
4840         .get_num_tx_queues      = bond_get_num_tx_queues,
4841         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4842                                                              as for TX queues */
4843 };
4844
4845 /* Create a new bond based on the specified name and bonding parameters.
4846  * If name is NULL, obtain a suitable "bond%d" name for us.
4847  * Caller must NOT hold rtnl_lock; we need to release it here before we
4848  * set up our sysfs entries.
4849  */
4850 int bond_create(struct net *net, const char *name)
4851 {
4852         struct net_device *bond_dev;
4853         int res;
4854
4855         rtnl_lock();
4856
4857         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4858                                    name ? name : "bond%d",
4859                                    bond_setup, tx_queues);
4860         if (!bond_dev) {
4861                 pr_err("%s: eek! can't alloc netdev!\n", name);
4862                 rtnl_unlock();
4863                 return -ENOMEM;
4864         }
4865
4866         dev_net_set(bond_dev, net);
4867         bond_dev->rtnl_link_ops = &bond_link_ops;
4868
4869         res = register_netdevice(bond_dev);
4870
4871         netif_carrier_off(bond_dev);
4872
4873         rtnl_unlock();
4874         if (res < 0)
4875                 bond_destructor(bond_dev);
4876         return res;
4877 }
4878
4879 static int __net_init bond_net_init(struct net *net)
4880 {
4881         struct bond_net *bn = net_generic(net, bond_net_id);
4882
4883         bn->net = net;
4884         INIT_LIST_HEAD(&bn->dev_list);
4885
4886         bond_create_proc_dir(bn);
4887         bond_create_sysfs(bn);
4888
4889         return 0;
4890 }
4891
4892 static void __net_exit bond_net_exit(struct net *net)
4893 {
4894         struct bond_net *bn = net_generic(net, bond_net_id);
4895         struct bonding *bond, *tmp_bond;
4896         LIST_HEAD(list);
4897
4898         bond_destroy_sysfs(bn);
4899         bond_destroy_proc_dir(bn);
4900
4901         /* Kill off any bonds created after unregistering bond rtnl ops */
4902         rtnl_lock();
4903         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4904                 unregister_netdevice_queue(bond->dev, &list);
4905         unregister_netdevice_many(&list);
4906         rtnl_unlock();
4907 }
4908
4909 static struct pernet_operations bond_net_ops = {
4910         .init = bond_net_init,
4911         .exit = bond_net_exit,
4912         .id   = &bond_net_id,
4913         .size = sizeof(struct bond_net),
4914 };
4915
4916 static int __init bonding_init(void)
4917 {
4918         int i;
4919         int res;
4920
4921         pr_info("%s", bond_version);
4922
4923         res = bond_check_params(&bonding_defaults);
4924         if (res)
4925                 goto out;
4926
4927         res = register_pernet_subsys(&bond_net_ops);
4928         if (res)
4929                 goto out;
4930
4931         res = rtnl_link_register(&bond_link_ops);
4932         if (res)
4933                 goto err_link;
4934
4935         bond_create_debugfs();
4936
4937         for (i = 0; i < max_bonds; i++) {
4938                 res = bond_create(&init_net, NULL);
4939                 if (res)
4940                         goto err;
4941         }
4942
4943         register_netdevice_notifier(&bond_netdev_notifier);
4944 out:
4945         return res;
4946 err:
4947         rtnl_link_unregister(&bond_link_ops);
4948 err_link:
4949         unregister_pernet_subsys(&bond_net_ops);
4950         goto out;
4951
4952 }
4953
4954 static void __exit bonding_exit(void)
4955 {
4956         unregister_netdevice_notifier(&bond_netdev_notifier);
4957
4958         bond_destroy_debugfs();
4959
4960         rtnl_link_unregister(&bond_link_ops);
4961         unregister_pernet_subsys(&bond_net_ops);
4962
4963 #ifdef CONFIG_NET_POLL_CONTROLLER
4964         /*
4965          * Make sure we don't have an imbalance on our netpoll blocking
4966          */
4967         WARN_ON(atomic_read(&netpoll_block_tx));
4968 #endif
4969 }
4970
4971 module_init(bonding_init);
4972 module_exit(bonding_exit);
4973 MODULE_LICENSE("GPL");
4974 MODULE_VERSION(DRV_VERSION);
4975 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4976 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4977 MODULE_ALIAS_RTNL_LINK("bond");