Pileus Git - ~andy/linux/blob - drivers/net/bonding/bond_main.c

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