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 /*
 710  * Add a Multicast address to slaves
 711  * according to mode
 712  */
 713 static void bond_mc_add(struct bonding *bond, void *addr)
 714 {
 715         if (USES_PRIMARY(bond->params.mode)) {
 716                 /* write lock already acquired */
 717                 if (bond->curr_active_slave)
 718                         dev_mc_add(bond->curr_active_slave->dev, addr);
 719         } else {
 720                 struct slave *slave;
 721                 int i;
 722
 723                 bond_for_each_slave(bond, slave, i)
 724                         dev_mc_add(slave->dev, addr);
 725         }
 726 }
 727
 728 /*
 729  * Remove a multicast address from slave
 730  * according to mode
 731  */
 732 static void bond_mc_del(struct bonding *bond, void *addr)
 733 {
 734         if (USES_PRIMARY(bond->params.mode)) {
 735                 /* write lock already acquired */
 736                 if (bond->curr_active_slave)
 737                         dev_mc_del(bond->curr_active_slave->dev, addr);
 738         } else {
 739                 struct slave *slave;
 740                 int i;
 741                 bond_for_each_slave(bond, slave, i) {
 742                         dev_mc_del(slave->dev, addr);
 743                 }
 744         }
 745 }
 746
 747
 748 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 749 {
 750         struct in_device *in_dev;
 751
 752         in_dev = __in_dev_get_rcu(dev);
 753         if (in_dev)
 754                 ip_mc_rejoin_groups(in_dev);
 755 }
 756
 757 /*
 758  * Retrieve the list of registered multicast addresses for the bonding
 759  * device and retransmit an IGMP JOIN request to the current active
 760  * slave.
 761  */
 762 static void bond_resend_igmp_join_requests(struct bonding *bond)
 763 {
 764         struct net_device *bond_dev, *vlan_dev, *upper_dev;
 765         struct vlan_entry *vlan;
 766
 767         rcu_read_lock();
 768         read_lock(&bond->lock);
 769
 770         bond_dev = bond->dev;
 771
 772         /* rejoin all groups on bond device */
 773         __bond_resend_igmp_join_requests(bond_dev);
 774
 775         /*
 776          * if bond is enslaved to a bridge,
 777          * then rejoin all groups on its master
 778          */
 779         upper_dev = netdev_master_upper_dev_get_rcu(bond_dev);
 780         if (upper_dev && upper_dev->priv_flags & IFF_EBRIDGE)
 781                 __bond_resend_igmp_join_requests(upper_dev);
 782
 783         /* rejoin all groups on vlan devices */
 784         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 785                 vlan_dev = __vlan_find_dev_deep(bond_dev, htons(ETH_P_8021Q),
 786                                                 vlan->vlan_id);
 787                 if (vlan_dev)
 788                         __bond_resend_igmp_join_requests(vlan_dev);
 789         }
 790
 791         if (--bond->igmp_retrans > 0)
 792                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 793
 794         read_unlock(&bond->lock);
 795         rcu_read_unlock();
 796 }
 797
 798 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 799 {
 800         struct bonding *bond = container_of(work, struct bonding,
 801                                             mcast_work.work);
 802
 803         bond_resend_igmp_join_requests(bond);
 804 }
 805
 806 /*
 807  * flush all members of flush->mc_list from device dev->mc_list
 808  */
 809 static void bond_mc_list_flush(struct net_device *bond_dev,
 810                                struct net_device *slave_dev)
 811 {
 812         struct bonding *bond = netdev_priv(bond_dev);
 813         struct netdev_hw_addr *ha;
 814
 815         netdev_for_each_mc_addr(ha, bond_dev)
 816                 dev_mc_del(slave_dev, ha->addr);
 817
 818         if (bond->params.mode == BOND_MODE_8023AD) {
 819                 /* del lacpdu mc addr from mc list */
 820                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 821
 822                 dev_mc_del(slave_dev, lacpdu_multicast);
 823         }
 824 }
 825
 826 /*--------------------------- Active slave change ---------------------------*/
 827
 828 /*
 829  * Update the mc list and multicast-related flags for the new and
 830  * old active slaves (if any) according to the multicast mode, and
 831  * promiscuous flags unconditionally.
 832  */
 833 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 834                          struct slave *old_active)
 835 {
 836         struct netdev_hw_addr *ha;
 837
 838         if (!USES_PRIMARY(bond->params.mode))
 839                 /* nothing to do -  mc list is already up-to-date on
 840                  * all slaves
 841                  */
 842                 return;
 843
 844         if (old_active) {
 845                 if (bond->dev->flags & IFF_PROMISC)
 846                         dev_set_promiscuity(old_active->dev, -1);
 847
 848                 if (bond->dev->flags & IFF_ALLMULTI)
 849                         dev_set_allmulti(old_active->dev, -1);
 850
 851                 netif_addr_lock_bh(bond->dev);
 852                 netdev_for_each_mc_addr(ha, bond->dev)
 853                         dev_mc_del(old_active->dev, ha->addr);
 854                 netif_addr_unlock_bh(bond->dev);
 855         }
 856
 857         if (new_active) {
 858                 /* FIXME: Signal errors upstream. */
 859                 if (bond->dev->flags & IFF_PROMISC)
 860                         dev_set_promiscuity(new_active->dev, 1);
 861
 862                 if (bond->dev->flags & IFF_ALLMULTI)
 863                         dev_set_allmulti(new_active->dev, 1);
 864
 865                 netif_addr_lock_bh(bond->dev);
 866                 netdev_for_each_mc_addr(ha, bond->dev)
 867                         dev_mc_add(new_active->dev, ha->addr);
 868                 netif_addr_unlock_bh(bond->dev);
 869         }
 870 }
 871
 872 /*
 873  * bond_do_fail_over_mac
 874  *
 875  * Perform special MAC address swapping for fail_over_mac settings
 876  *
 877  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 878  */
 879 static void bond_do_fail_over_mac(struct bonding *bond,
 880                                   struct slave *new_active,
 881                                   struct slave *old_active)
 882         __releases(&bond->curr_slave_lock)
 883         __releases(&bond->lock)
 884         __acquires(&bond->lock)
 885         __acquires(&bond->curr_slave_lock)
 886 {
 887         u8 tmp_mac[ETH_ALEN];
 888         struct sockaddr saddr;
 889         int rv;
 890
 891         switch (bond->params.fail_over_mac) {
 892         case BOND_FOM_ACTIVE:
 893                 if (new_active) {
 894                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 895                                new_active->dev->addr_len);
 896                         write_unlock_bh(&bond->curr_slave_lock);
 897                         read_unlock(&bond->lock);
 898                         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
 899                         read_lock(&bond->lock);
 900                         write_lock_bh(&bond->curr_slave_lock);
 901                 }
 902                 break;
 903         case BOND_FOM_FOLLOW:
 904                 /*
 905                  * if new_active && old_active, swap them
 906                  * if just old_active, do nothing (going to no active slave)
 907                  * if just new_active, set new_active to bond's MAC
 908                  */
 909                 if (!new_active)
 910                         return;
 911
 912                 write_unlock_bh(&bond->curr_slave_lock);
 913                 read_unlock(&bond->lock);
 914
 915                 if (old_active) {
 916                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 917                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 918                                ETH_ALEN);
 919                         saddr.sa_family = new_active->dev->type;
 920                 } else {
 921                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 922                         saddr.sa_family = bond->dev->type;
 923                 }
 924
 925                 rv = dev_set_mac_address(new_active->dev, &saddr);
 926                 if (rv) {
 927                         pr_err("%s: Error %d setting MAC of slave %s\n",
 928                                bond->dev->name, -rv, new_active->dev->name);
 929                         goto out;
 930                 }
 931
 932                 if (!old_active)
 933                         goto out;
 934
 935                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 936                 saddr.sa_family = old_active->dev->type;
 937
 938                 rv = dev_set_mac_address(old_active->dev, &saddr);
 939                 if (rv)
 940                         pr_err("%s: Error %d setting MAC of slave %s\n",
 941                                bond->dev->name, -rv, new_active->dev->name);
 942 out:
 943                 read_lock(&bond->lock);
 944                 write_lock_bh(&bond->curr_slave_lock);
 945                 break;
 946         default:
 947                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 948                        bond->dev->name, bond->params.fail_over_mac);
 949                 break;
 950         }
 951
 952 }
 953
 954 static bool bond_should_change_active(struct bonding *bond)
 955 {
 956         struct slave *prim = bond->primary_slave;
 957         struct slave *curr = bond->curr_active_slave;
 958
 959         if (!prim || !curr || curr->link != BOND_LINK_UP)
 960                 return true;
 961         if (bond->force_primary) {
 962                 bond->force_primary = false;
 963                 return true;
 964         }
 965         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 966             (prim->speed < curr->speed ||
 967              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 968                 return false;
 969         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 970                 return false;
 971         return true;
 972 }
 973
 974 /**
 975  * find_best_interface - select the best available slave to be the active one
 976  * @bond: our bonding struct
 977  *
 978  * Warning: Caller must hold curr_slave_lock for writing.
 979  */
 980 static struct slave *bond_find_best_slave(struct bonding *bond)
 981 {
 982         struct slave *new_active, *old_active;
 983         struct slave *bestslave = NULL;
 984         int mintime = bond->params.updelay;
 985         int i;
 986
 987         new_active = bond->curr_active_slave;
 988
 989         if (!new_active) { /* there were no active slaves left */
 990                 if (bond->slave_cnt > 0)   /* found one slave */
 991                         new_active = bond->first_slave;
 992                 else
 993                         return NULL; /* still no slave, return NULL */
 994         }
 995
 996         if ((bond->primary_slave) &&
 997             bond->primary_slave->link == BOND_LINK_UP &&
 998             bond_should_change_active(bond)) {
 999                 new_active = bond->primary_slave;
1000         }
1001
1002         /* remember where to stop iterating over the slaves */
1003         old_active = new_active;
1004
1005         bond_for_each_slave_from(bond, new_active, i, old_active) {
1006                 if (new_active->link == BOND_LINK_UP) {
1007                         return new_active;
1008                 } else if (new_active->link == BOND_LINK_BACK &&
1009                            IS_UP(new_active->dev)) {
1010                         /* link up, but waiting for stabilization */
1011                         if (new_active->delay < mintime) {
1012                                 mintime = new_active->delay;
1013                                 bestslave = new_active;
1014                         }
1015                 }
1016         }
1017
1018         return bestslave;
1019 }
1020
1021 static bool bond_should_notify_peers(struct bonding *bond)
1022 {
1023         struct slave *slave = bond->curr_active_slave;
1024
1025         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1026                  bond->dev->name, slave ? slave->dev->name : "NULL");
1027
1028         if (!slave || !bond->send_peer_notif ||
1029             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1030                 return false;
1031
1032         bond->send_peer_notif--;
1033         return true;
1034 }
1035
1036 /**
1037  * change_active_interface - change the active slave into the specified one
1038  * @bond: our bonding struct
1039  * @new: the new slave to make the active one
1040  *
1041  * Set the new slave to the bond's settings and unset them on the old
1042  * curr_active_slave.
1043  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1044  *
1045  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1046  * because it is apparently the best available slave we have, even though its
1047  * updelay hasn't timed out yet.
1048  *
1049  * If new_active is not NULL, caller must hold bond->lock for read and
1050  * curr_slave_lock for write_bh.
1051  */
1052 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1053 {
1054         struct slave *old_active = bond->curr_active_slave;
1055
1056         if (old_active == new_active)
1057                 return;
1058
1059         if (new_active) {
1060                 new_active->jiffies = jiffies;
1061
1062                 if (new_active->link == BOND_LINK_BACK) {
1063                         if (USES_PRIMARY(bond->params.mode)) {
1064                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1065                                         bond->dev->name, new_active->dev->name,
1066                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1067                         }
1068
1069                         new_active->delay = 0;
1070                         new_active->link = BOND_LINK_UP;
1071
1072                         if (bond->params.mode == BOND_MODE_8023AD)
1073                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1074
1075                         if (bond_is_lb(bond))
1076                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1077                 } else {
1078                         if (USES_PRIMARY(bond->params.mode)) {
1079                                 pr_info("%s: making interface %s the new active one.\n",
1080                                         bond->dev->name, new_active->dev->name);
1081                         }
1082                 }
1083         }
1084
1085         if (USES_PRIMARY(bond->params.mode))
1086                 bond_mc_swap(bond, new_active, old_active);
1087
1088         if (bond_is_lb(bond)) {
1089                 bond_alb_handle_active_change(bond, new_active);
1090                 if (old_active)
1091                         bond_set_slave_inactive_flags(old_active);
1092                 if (new_active)
1093                         bond_set_slave_active_flags(new_active);
1094         } else {
1095                 bond->curr_active_slave = new_active;
1096         }
1097
1098         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1099                 if (old_active)
1100                         bond_set_slave_inactive_flags(old_active);
1101
1102                 if (new_active) {
1103                         bool should_notify_peers = false;
1104
1105                         bond_set_slave_active_flags(new_active);
1106
1107                         if (bond->params.fail_over_mac)
1108                                 bond_do_fail_over_mac(bond, new_active,
1109                                                       old_active);
1110
1111                         if (netif_running(bond->dev)) {
1112                                 bond->send_peer_notif =
1113                                         bond->params.num_peer_notif;
1114                                 should_notify_peers =
1115                                         bond_should_notify_peers(bond);
1116                         }
1117
1118                         write_unlock_bh(&bond->curr_slave_lock);
1119                         read_unlock(&bond->lock);
1120
1121                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1122                         if (should_notify_peers)
1123                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1124                                                          bond->dev);
1125
1126                         read_lock(&bond->lock);
1127                         write_lock_bh(&bond->curr_slave_lock);
1128                 }
1129         }
1130
1131         /* resend IGMP joins since active slave has changed or
1132          * all were sent on curr_active_slave.
1133          * resend only if bond is brought up with the affected
1134          * bonding modes and the retransmission is enabled */
1135         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1136             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1137              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1138                 bond->igmp_retrans = bond->params.resend_igmp;
1139                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1140         }
1141 }
1142
1143 /**
1144  * bond_select_active_slave - select a new active slave, if needed
1145  * @bond: our bonding struct
1146  *
1147  * This functions should be called when one of the following occurs:
1148  * - The old curr_active_slave has been released or lost its link.
1149  * - The primary_slave has got its link back.
1150  * - A slave has got its link back and there's no old curr_active_slave.
1151  *
1152  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1153  */
1154 void bond_select_active_slave(struct bonding *bond)
1155 {
1156         struct slave *best_slave;
1157         int rv;
1158
1159         best_slave = bond_find_best_slave(bond);
1160         if (best_slave != bond->curr_active_slave) {
1161                 bond_change_active_slave(bond, best_slave);
1162                 rv = bond_set_carrier(bond);
1163                 if (!rv)
1164                         return;
1165
1166                 if (netif_carrier_ok(bond->dev)) {
1167                         pr_info("%s: first active interface up!\n",
1168                                 bond->dev->name);
1169                 } else {
1170                         pr_info("%s: now running without any active interface !\n",
1171                                 bond->dev->name);
1172                 }
1173         }
1174 }
1175
1176 /*--------------------------- slave list handling ---------------------------*/
1177
1178 /*
1179  * This function attaches the slave to the end of list.
1180  *
1181  * bond->lock held for writing by caller.
1182  */
1183 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1184 {
1185         if (bond->first_slave == NULL) { /* attaching the first slave */
1186                 new_slave->next = new_slave;
1187                 new_slave->prev = new_slave;
1188                 bond->first_slave = new_slave;
1189         } else {
1190                 new_slave->next = bond->first_slave;
1191                 new_slave->prev = bond->first_slave->prev;
1192                 new_slave->next->prev = new_slave;
1193                 new_slave->prev->next = new_slave;
1194         }
1195
1196         bond->slave_cnt++;
1197 }
1198
1199 /*
1200  * This function detaches the slave from the list.
1201  * WARNING: no check is made to verify if the slave effectively
1202  * belongs to <bond>.
1203  * Nothing is freed on return, structures are just unchained.
1204  * If any slave pointer in bond was pointing to <slave>,
1205  * it should be changed by the calling function.
1206  *
1207  * bond->lock held for writing by caller.
1208  */
1209 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1210 {
1211         if (slave->next)
1212                 slave->next->prev = slave->prev;
1213
1214         if (slave->prev)
1215                 slave->prev->next = slave->next;
1216
1217         if (bond->first_slave == slave) { /* slave is the first slave */
1218                 if (bond->slave_cnt > 1) { /* there are more slave */
1219                         bond->first_slave = slave->next;
1220                 } else {
1221                         bond->first_slave = NULL; /* slave was the last one */
1222                 }
1223         }
1224
1225         slave->next = NULL;
1226         slave->prev = NULL;
1227         bond->slave_cnt--;
1228 }
1229
1230 #ifdef CONFIG_NET_POLL_CONTROLLER
1231 static inline int slave_enable_netpoll(struct slave *slave)
1232 {
1233         struct netpoll *np;
1234         int err = 0;
1235
1236         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1237         err = -ENOMEM;
1238         if (!np)
1239                 goto out;
1240
1241         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1242         if (err) {
1243                 kfree(np);
1244                 goto out;
1245         }
1246         slave->np = np;
1247 out:
1248         return err;
1249 }
1250 static inline void slave_disable_netpoll(struct slave *slave)
1251 {
1252         struct netpoll *np = slave->np;
1253
1254         if (!np)
1255                 return;
1256
1257         slave->np = NULL;
1258         __netpoll_free_async(np);
1259 }
1260 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1261 {
1262         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1263                 return false;
1264         if (!slave_dev->netdev_ops->ndo_poll_controller)
1265                 return false;
1266         return true;
1267 }
1268
1269 static void bond_poll_controller(struct net_device *bond_dev)
1270 {
1271 }
1272
1273 static void __bond_netpoll_cleanup(struct bonding *bond)
1274 {
1275         struct slave *slave;
1276         int i;
1277
1278         bond_for_each_slave(bond, slave, i)
1279                 if (IS_UP(slave->dev))
1280                         slave_disable_netpoll(slave);
1281 }
1282 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1283 {
1284         struct bonding *bond = netdev_priv(bond_dev);
1285
1286         read_lock(&bond->lock);
1287         __bond_netpoll_cleanup(bond);
1288         read_unlock(&bond->lock);
1289 }
1290
1291 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1292 {
1293         struct bonding *bond = netdev_priv(dev);
1294         struct slave *slave;
1295         int i, err = 0;
1296
1297         read_lock(&bond->lock);
1298         bond_for_each_slave(bond, slave, i) {
1299                 err = slave_enable_netpoll(slave);
1300                 if (err) {
1301                         __bond_netpoll_cleanup(bond);
1302                         break;
1303                 }
1304         }
1305         read_unlock(&bond->lock);
1306         return err;
1307 }
1308
1309 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1310 {
1311         return bond->dev->npinfo;
1312 }
1313
1314 #else
1315 static inline int slave_enable_netpoll(struct slave *slave)
1316 {
1317         return 0;
1318 }
1319 static inline void slave_disable_netpoll(struct slave *slave)
1320 {
1321 }
1322 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1323 {
1324 }
1325 #endif
1326
1327 /*---------------------------------- IOCTL ----------------------------------*/
1328
1329 static void bond_set_dev_addr(struct net_device *bond_dev,
1330                               struct net_device *slave_dev)
1331 {
1332         pr_debug("bond_dev=%p\n", bond_dev);
1333         pr_debug("slave_dev=%p\n", slave_dev);
1334         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1335         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1336         bond_dev->addr_assign_type = NET_ADDR_SET;
1337         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
1338 }
1339
1340 static netdev_features_t bond_fix_features(struct net_device *dev,
1341         netdev_features_t features)
1342 {
1343         struct slave *slave;
1344         struct bonding *bond = netdev_priv(dev);
1345         netdev_features_t mask;
1346         int i;
1347
1348         read_lock(&bond->lock);
1349
1350         if (!bond->first_slave) {
1351                 /* Disable adding VLANs to empty bond. But why? --mq */
1352                 features |= NETIF_F_VLAN_CHALLENGED;
1353                 goto out;
1354         }
1355
1356         mask = features;
1357         features &= ~NETIF_F_ONE_FOR_ALL;
1358         features |= NETIF_F_ALL_FOR_ALL;
1359
1360         bond_for_each_slave(bond, slave, i) {
1361                 features = netdev_increment_features(features,
1362                                                      slave->dev->features,
1363                                                      mask);
1364         }
1365         features = netdev_add_tso_features(features, mask);
1366
1367 out:
1368         read_unlock(&bond->lock);
1369         return features;
1370 }
1371
1372 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1373                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1374                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1375
1376 static void bond_compute_features(struct bonding *bond)
1377 {
1378         struct slave *slave;
1379         struct net_device *bond_dev = bond->dev;
1380         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1381         unsigned short max_hard_header_len = ETH_HLEN;
1382         unsigned int gso_max_size = GSO_MAX_SIZE;
1383         u16 gso_max_segs = GSO_MAX_SEGS;
1384         int i;
1385         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1386
1387         read_lock(&bond->lock);
1388
1389         if (!bond->first_slave)
1390                 goto done;
1391
1392         bond_for_each_slave(bond, slave, i) {
1393                 vlan_features = netdev_increment_features(vlan_features,
1394                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1395
1396                 dst_release_flag &= slave->dev->priv_flags;
1397                 if (slave->dev->hard_header_len > max_hard_header_len)
1398                         max_hard_header_len = slave->dev->hard_header_len;
1399
1400                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1401                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1402         }
1403
1404 done:
1405         bond_dev->vlan_features = vlan_features;
1406         bond_dev->hard_header_len = max_hard_header_len;
1407         bond_dev->gso_max_segs = gso_max_segs;
1408         netif_set_gso_max_size(bond_dev, gso_max_size);
1409
1410         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1411         bond_dev->priv_flags = flags | dst_release_flag;
1412
1413         read_unlock(&bond->lock);
1414
1415         netdev_change_features(bond_dev);
1416 }
1417
1418 static void bond_setup_by_slave(struct net_device *bond_dev,
1419                                 struct net_device *slave_dev)
1420 {
1421         struct bonding *bond = netdev_priv(bond_dev);
1422
1423         bond_dev->header_ops        = slave_dev->header_ops;
1424
1425         bond_dev->type              = slave_dev->type;
1426         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1427         bond_dev->addr_len          = slave_dev->addr_len;
1428
1429         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1430                 slave_dev->addr_len);
1431         bond->setup_by_slave = 1;
1432 }
1433
1434 /* On bonding slaves other than the currently active slave, suppress
1435  * duplicates except for alb non-mcast/bcast.
1436  */
1437 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1438                                             struct slave *slave,
1439                                             struct bonding *bond)
1440 {
1441         if (bond_is_slave_inactive(slave)) {
1442                 if (bond->params.mode == BOND_MODE_ALB &&
1443                     skb->pkt_type != PACKET_BROADCAST &&
1444                     skb->pkt_type != PACKET_MULTICAST)
1445                         return false;
1446                 return true;
1447         }
1448         return false;
1449 }
1450
1451 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1452 {
1453         struct sk_buff *skb = *pskb;
1454         struct slave *slave;
1455         struct bonding *bond;
1456         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1457                           struct slave *);
1458         int ret = RX_HANDLER_ANOTHER;
1459
1460         skb = skb_share_check(skb, GFP_ATOMIC);
1461         if (unlikely(!skb))
1462                 return RX_HANDLER_CONSUMED;
1463
1464         *pskb = skb;
1465
1466         slave = bond_slave_get_rcu(skb->dev);
1467         bond = slave->bond;
1468
1469         if (bond->params.arp_interval)
1470                 slave->dev->last_rx = jiffies;
1471
1472         recv_probe = ACCESS_ONCE(bond->recv_probe);
1473         if (recv_probe) {
1474                 ret = recv_probe(skb, bond, slave);
1475                 if (ret == RX_HANDLER_CONSUMED) {
1476                         consume_skb(skb);
1477                         return ret;
1478                 }
1479         }
1480
1481         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1482                 return RX_HANDLER_EXACT;
1483         }
1484
1485         skb->dev = bond->dev;
1486
1487         if (bond->params.mode == BOND_MODE_ALB &&
1488             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1489             skb->pkt_type == PACKET_HOST) {
1490
1491                 if (unlikely(skb_cow_head(skb,
1492                                           skb->data - skb_mac_header(skb)))) {
1493                         kfree_skb(skb);
1494                         return RX_HANDLER_CONSUMED;
1495                 }
1496                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1497         }
1498
1499         return ret;
1500 }
1501
1502 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1503                                       struct net_device *slave_dev)
1504 {
1505         int err;
1506
1507         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1508         if (err)
1509                 return err;
1510         slave_dev->flags |= IFF_SLAVE;
1511         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1512         return 0;
1513 }
1514
1515 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1516                                   struct net_device *slave_dev)
1517 {
1518         netdev_upper_dev_unlink(slave_dev, bond_dev);
1519         slave_dev->flags &= ~IFF_SLAVE;
1520         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1521 }
1522
1523 /* enslave device <slave> to bond device <master> */
1524 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1525 {
1526         struct bonding *bond = netdev_priv(bond_dev);
1527         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1528         struct slave *new_slave = NULL;
1529         struct netdev_hw_addr *ha;
1530         struct sockaddr addr;
1531         int link_reporting;
1532         int res = 0;
1533
1534         if (!bond->params.use_carrier &&
1535             slave_dev->ethtool_ops->get_link == NULL &&
1536             slave_ops->ndo_do_ioctl == NULL) {
1537                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1538                            bond_dev->name, slave_dev->name);
1539         }
1540
1541         /* already enslaved */
1542         if (slave_dev->flags & IFF_SLAVE) {
1543                 pr_debug("Error, Device was already enslaved\n");
1544                 return -EBUSY;
1545         }
1546
1547         /* vlan challenged mutual exclusion */
1548         /* no need to lock since we're protected by rtnl_lock */
1549         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1550                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1551                 if (vlan_uses_dev(bond_dev)) {
1552                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1553                                bond_dev->name, slave_dev->name, bond_dev->name);
1554                         return -EPERM;
1555                 } else {
1556                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1557                                    bond_dev->name, slave_dev->name,
1558                                    slave_dev->name, bond_dev->name);
1559                 }
1560         } else {
1561                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1562         }
1563
1564         /*
1565          * Old ifenslave binaries are no longer supported.  These can
1566          * be identified with moderate accuracy by the state of the slave:
1567          * the current ifenslave will set the interface down prior to
1568          * enslaving it; the old ifenslave will not.
1569          */
1570         if ((slave_dev->flags & IFF_UP)) {
1571                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1572                        slave_dev->name);
1573                 res = -EPERM;
1574                 goto err_undo_flags;
1575         }
1576
1577         /* set bonding device ether type by slave - bonding netdevices are
1578          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1579          * there is a need to override some of the type dependent attribs/funcs.
1580          *
1581          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1582          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1583          */
1584         if (bond->slave_cnt == 0) {
1585                 if (bond_dev->type != slave_dev->type) {
1586                         pr_debug("%s: change device type from %d to %d\n",
1587                                  bond_dev->name,
1588                                  bond_dev->type, slave_dev->type);
1589
1590                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1591                                                        bond_dev);
1592                         res = notifier_to_errno(res);
1593                         if (res) {
1594                                 pr_err("%s: refused to change device type\n",
1595                                        bond_dev->name);
1596                                 res = -EBUSY;
1597                                 goto err_undo_flags;
1598                         }
1599
1600                         /* Flush unicast and multicast addresses */
1601                         dev_uc_flush(bond_dev);
1602                         dev_mc_flush(bond_dev);
1603
1604                         if (slave_dev->type != ARPHRD_ETHER)
1605                                 bond_setup_by_slave(bond_dev, slave_dev);
1606                         else {
1607                                 ether_setup(bond_dev);
1608                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1609                         }
1610
1611                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1612                                                  bond_dev);
1613                 }
1614         } else if (bond_dev->type != slave_dev->type) {
1615                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1616                        slave_dev->name,
1617                        slave_dev->type, bond_dev->type);
1618                 res = -EINVAL;
1619                 goto err_undo_flags;
1620         }
1621
1622         if (slave_ops->ndo_set_mac_address == NULL) {
1623                 if (bond->slave_cnt == 0) {
1624                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1625                                    bond_dev->name);
1626                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1627                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1628                         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",
1629                                bond_dev->name);
1630                         res = -EOPNOTSUPP;
1631                         goto err_undo_flags;
1632                 }
1633         }
1634
1635         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1636
1637         /* If this is the first slave, then we need to set the master's hardware
1638          * address to be the same as the slave's. */
1639         if (bond->slave_cnt == 0 && bond->dev_addr_from_first)
1640                 bond_set_dev_addr(bond->dev, slave_dev);
1641
1642         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1643         if (!new_slave) {
1644                 res = -ENOMEM;
1645                 goto err_undo_flags;
1646         }
1647
1648         /*
1649          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1650          * is set via sysfs or module option if desired.
1651          */
1652         new_slave->queue_id = 0;
1653
1654         /* Save slave's original mtu and then set it to match the bond */
1655         new_slave->original_mtu = slave_dev->mtu;
1656         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1657         if (res) {
1658                 pr_debug("Error %d calling dev_set_mtu\n", res);
1659                 goto err_free;
1660         }
1661
1662         /*
1663          * Save slave's original ("permanent") mac address for modes
1664          * that need it, and for restoring it upon release, and then
1665          * set it to the master's address
1666          */
1667         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1668
1669         if (!bond->params.fail_over_mac) {
1670                 /*
1671                  * Set slave to master's mac address.  The application already
1672                  * set the master's mac address to that of the first slave
1673                  */
1674                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1675                 addr.sa_family = slave_dev->type;
1676                 res = dev_set_mac_address(slave_dev, &addr);
1677                 if (res) {
1678                         pr_debug("Error %d calling set_mac_address\n", res);
1679                         goto err_restore_mtu;
1680                 }
1681         }
1682
1683         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1684         if (res) {
1685                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1686                 goto err_restore_mac;
1687         }
1688
1689         /* open the slave since the application closed it */
1690         res = dev_open(slave_dev);
1691         if (res) {
1692                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1693                 goto err_unset_master;
1694         }
1695
1696         new_slave->bond = bond;
1697         new_slave->dev = slave_dev;
1698         slave_dev->priv_flags |= IFF_BONDING;
1699
1700         if (bond_is_lb(bond)) {
1701                 /* bond_alb_init_slave() must be called before all other stages since
1702                  * it might fail and we do not want to have to undo everything
1703                  */
1704                 res = bond_alb_init_slave(bond, new_slave);
1705                 if (res)
1706                         goto err_close;
1707         }
1708
1709         /* If the mode USES_PRIMARY, then the new slave gets the
1710          * master's promisc (and mc) settings only if it becomes the
1711          * curr_active_slave, and that is taken care of later when calling
1712          * bond_change_active()
1713          */
1714         if (!USES_PRIMARY(bond->params.mode)) {
1715                 /* set promiscuity level to new slave */
1716                 if (bond_dev->flags & IFF_PROMISC) {
1717                         res = dev_set_promiscuity(slave_dev, 1);
1718                         if (res)
1719                                 goto err_close;
1720                 }
1721
1722                 /* set allmulti level to new slave */
1723                 if (bond_dev->flags & IFF_ALLMULTI) {
1724                         res = dev_set_allmulti(slave_dev, 1);
1725                         if (res)
1726                                 goto err_close;
1727                 }
1728
1729                 netif_addr_lock_bh(bond_dev);
1730                 /* upload master's mc_list to new slave */
1731                 netdev_for_each_mc_addr(ha, bond_dev)
1732                         dev_mc_add(slave_dev, ha->addr);
1733                 netif_addr_unlock_bh(bond_dev);
1734         }
1735
1736         if (bond->params.mode == BOND_MODE_8023AD) {
1737                 /* add lacpdu mc addr to mc list */
1738                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1739
1740                 dev_mc_add(slave_dev, lacpdu_multicast);
1741         }
1742
1743         bond_add_vlans_on_slave(bond, slave_dev);
1744
1745         write_lock_bh(&bond->lock);
1746
1747         bond_attach_slave(bond, new_slave);
1748
1749         new_slave->delay = 0;
1750         new_slave->link_failure_count = 0;
1751
1752         write_unlock_bh(&bond->lock);
1753
1754         bond_compute_features(bond);
1755
1756         bond_update_speed_duplex(new_slave);
1757
1758         read_lock(&bond->lock);
1759
1760         new_slave->last_arp_rx = jiffies -
1761                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1762
1763         if (bond->params.miimon && !bond->params.use_carrier) {
1764                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1765
1766                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1767                         /*
1768                          * miimon is set but a bonded network driver
1769                          * does not support ETHTOOL/MII and
1770                          * arp_interval is not set.  Note: if
1771                          * use_carrier is enabled, we will never go
1772                          * here (because netif_carrier is always
1773                          * supported); thus, we don't need to change
1774                          * the messages for netif_carrier.
1775                          */
1776                         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",
1777                                bond_dev->name, slave_dev->name);
1778                 } else if (link_reporting == -1) {
1779                         /* unable get link status using mii/ethtool */
1780                         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",
1781                                    bond_dev->name, slave_dev->name);
1782                 }
1783         }
1784
1785         /* check for initial state */
1786         if (bond->params.miimon) {
1787                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1788                         if (bond->params.updelay) {
1789                                 new_slave->link = BOND_LINK_BACK;
1790                                 new_slave->delay = bond->params.updelay;
1791                         } else {
1792                                 new_slave->link = BOND_LINK_UP;
1793                         }
1794                 } else {
1795                         new_slave->link = BOND_LINK_DOWN;
1796                 }
1797         } else if (bond->params.arp_interval) {
1798                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1799                         BOND_LINK_UP : BOND_LINK_DOWN);
1800         } else {
1801                 new_slave->link = BOND_LINK_UP;
1802         }
1803
1804         if (new_slave->link != BOND_LINK_DOWN)
1805                 new_slave->jiffies = jiffies;
1806         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1807                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1808                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1809
1810         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1811                 /* if there is a primary slave, remember it */
1812                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1813                         bond->primary_slave = new_slave;
1814                         bond->force_primary = true;
1815                 }
1816         }
1817
1818         write_lock_bh(&bond->curr_slave_lock);
1819
1820         switch (bond->params.mode) {
1821         case BOND_MODE_ACTIVEBACKUP:
1822                 bond_set_slave_inactive_flags(new_slave);
1823                 bond_select_active_slave(bond);
1824                 break;
1825         case BOND_MODE_8023AD:
1826                 /* in 802.3ad mode, the internal mechanism
1827                  * will activate the slaves in the selected
1828                  * aggregator
1829                  */
1830                 bond_set_slave_inactive_flags(new_slave);
1831                 /* if this is the first slave */
1832                 if (bond->slave_cnt == 1) {
1833                         SLAVE_AD_INFO(new_slave).id = 1;
1834                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1835                          * can be called only after the mac address of the bond is set
1836                          */
1837                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1838                 } else {
1839                         SLAVE_AD_INFO(new_slave).id =
1840                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1841                 }
1842
1843                 bond_3ad_bind_slave(new_slave);
1844                 break;
1845         case BOND_MODE_TLB:
1846         case BOND_MODE_ALB:
1847                 bond_set_active_slave(new_slave);
1848                 bond_set_slave_inactive_flags(new_slave);
1849                 bond_select_active_slave(bond);
1850                 break;
1851         default:
1852                 pr_debug("This slave is always active in trunk mode\n");
1853
1854                 /* always active in trunk mode */
1855                 bond_set_active_slave(new_slave);
1856
1857                 /* In trunking mode there is little meaning to curr_active_slave
1858                  * anyway (it holds no special properties of the bond device),
1859                  * so we can change it without calling change_active_interface()
1860                  */
1861                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1862                         bond->curr_active_slave = new_slave;
1863
1864                 break;
1865         } /* switch(bond_mode) */
1866
1867         write_unlock_bh(&bond->curr_slave_lock);
1868
1869         bond_set_carrier(bond);
1870
1871 #ifdef CONFIG_NET_POLL_CONTROLLER
1872         slave_dev->npinfo = bond_netpoll_info(bond);
1873         if (slave_dev->npinfo) {
1874                 if (slave_enable_netpoll(new_slave)) {
1875                         read_unlock(&bond->lock);
1876                         pr_info("Error, %s: master_dev is using netpoll, "
1877                                  "but new slave device does not support netpoll.\n",
1878                                  bond_dev->name);
1879                         res = -EBUSY;
1880                         goto err_detach;
1881                 }
1882         }
1883 #endif
1884
1885         read_unlock(&bond->lock);
1886
1887         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1888         if (res)
1889                 goto err_detach;
1890
1891         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1892                                          new_slave);
1893         if (res) {
1894                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1895                 goto err_dest_symlinks;
1896         }
1897
1898         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1899                 bond_dev->name, slave_dev->name,
1900                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1901                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1902
1903         /* enslave is successful */
1904         return 0;
1905
1906 /* Undo stages on error */
1907 err_dest_symlinks:
1908         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1909
1910 err_detach:
1911         if (!USES_PRIMARY(bond->params.mode)) {
1912                 netif_addr_lock_bh(bond_dev);
1913                 bond_mc_list_flush(bond_dev, slave_dev);
1914                 netif_addr_unlock_bh(bond_dev);
1915         }
1916         bond_del_vlans_from_slave(bond, slave_dev);
1917         write_lock_bh(&bond->lock);
1918         bond_detach_slave(bond, new_slave);
1919         if (bond->primary_slave == new_slave)
1920                 bond->primary_slave = NULL;
1921         if (bond->curr_active_slave == new_slave) {
1922                 bond_change_active_slave(bond, NULL);
1923                 write_unlock_bh(&bond->lock);
1924                 read_lock(&bond->lock);
1925                 write_lock_bh(&bond->curr_slave_lock);
1926                 bond_select_active_slave(bond);
1927                 write_unlock_bh(&bond->curr_slave_lock);
1928                 read_unlock(&bond->lock);
1929         } else {
1930                 write_unlock_bh(&bond->lock);
1931         }
1932         slave_disable_netpoll(new_slave);
1933
1934 err_close:
1935         slave_dev->priv_flags &= ~IFF_BONDING;
1936         dev_close(slave_dev);
1937
1938 err_unset_master:
1939         bond_upper_dev_unlink(bond_dev, slave_dev);
1940
1941 err_restore_mac:
1942         if (!bond->params.fail_over_mac) {
1943                 /* XXX TODO - fom follow mode needs to change master's
1944                  * MAC if this slave's MAC is in use by the bond, or at
1945                  * least print a warning.
1946                  */
1947                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1948                 addr.sa_family = slave_dev->type;
1949                 dev_set_mac_address(slave_dev, &addr);
1950         }
1951
1952 err_restore_mtu:
1953         dev_set_mtu(slave_dev, new_slave->original_mtu);
1954
1955 err_free:
1956         kfree(new_slave);
1957
1958 err_undo_flags:
1959         bond_compute_features(bond);
1960         /* Enslave of first slave has failed and we need to fix master's mac */
1961         if (bond->slave_cnt == 0 &&
1962             ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1963                 eth_hw_addr_random(bond_dev);
1964
1965         return res;
1966 }
1967
1968 /*
1969  * Try to release the slave device <slave> from the bond device <master>
1970  * It is legal to access curr_active_slave without a lock because all the function
1971  * is write-locked. If "all" is true it means that the function is being called
1972  * while destroying a bond interface and all slaves are being released.
1973  *
1974  * The rules for slave state should be:
1975  *   for Active/Backup:
1976  *     Active stays on all backups go down
1977  *   for Bonded connections:
1978  *     The first up interface should be left on and all others downed.
1979  */
1980 static int __bond_release_one(struct net_device *bond_dev,
1981                               struct net_device *slave_dev,
1982                               bool all)
1983 {
1984         struct bonding *bond = netdev_priv(bond_dev);
1985         struct slave *slave, *oldcurrent;
1986         struct sockaddr addr;
1987         netdev_features_t old_features = bond_dev->features;
1988
1989         /* slave is not a slave or master is not master of this slave */
1990         if (!(slave_dev->flags & IFF_SLAVE) ||
1991             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1992                 pr_err("%s: Error: cannot release %s.\n",
1993                        bond_dev->name, slave_dev->name);
1994                 return -EINVAL;
1995         }
1996
1997         block_netpoll_tx();
1998         write_lock_bh(&bond->lock);
1999
2000         slave = bond_get_slave_by_dev(bond, slave_dev);
2001         if (!slave) {
2002                 /* not a slave of this bond */
2003                 pr_info("%s: %s not enslaved\n",
2004                         bond_dev->name, slave_dev->name);
2005                 write_unlock_bh(&bond->lock);
2006                 unblock_netpoll_tx();
2007                 return -EINVAL;
2008         }
2009
2010         write_unlock_bh(&bond->lock);
2011         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2012          * for this slave anymore.
2013          */
2014         netdev_rx_handler_unregister(slave_dev);
2015         write_lock_bh(&bond->lock);
2016
2017         if (!all && !bond->params.fail_over_mac) {
2018                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
2019                     bond->slave_cnt > 1)
2020                         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",
2021                                    bond_dev->name, slave_dev->name,
2022                                    slave->perm_hwaddr,
2023                                    bond_dev->name, slave_dev->name);
2024         }
2025
2026         /* Inform AD package of unbinding of slave. */
2027         if (bond->params.mode == BOND_MODE_8023AD) {
2028                 /* must be called before the slave is
2029                  * detached from the list
2030                  */
2031                 bond_3ad_unbind_slave(slave);
2032         }
2033
2034         pr_info("%s: releasing %s interface %s\n",
2035                 bond_dev->name,
2036                 bond_is_active_slave(slave) ? "active" : "backup",
2037                 slave_dev->name);
2038
2039         oldcurrent = bond->curr_active_slave;
2040
2041         bond->current_arp_slave = NULL;
2042
2043         /* release the slave from its bond */
2044         bond_detach_slave(bond, slave);
2045
2046         if (bond->primary_slave == slave)
2047                 bond->primary_slave = NULL;
2048
2049         if (oldcurrent == slave)
2050                 bond_change_active_slave(bond, NULL);
2051
2052         if (bond_is_lb(bond)) {
2053                 /* Must be called only after the slave has been
2054                  * detached from the list and the curr_active_slave
2055                  * has been cleared (if our_slave == old_current),
2056                  * but before a new active slave is selected.
2057                  */
2058                 write_unlock_bh(&bond->lock);
2059                 bond_alb_deinit_slave(bond, slave);
2060                 write_lock_bh(&bond->lock);
2061         }
2062
2063         if (all) {
2064                 bond->curr_active_slave = NULL;
2065         } else if (oldcurrent == slave) {
2066                 /*
2067                  * Note that we hold RTNL over this sequence, so there
2068                  * is no concern that another slave add/remove event
2069                  * will interfere.
2070                  */
2071                 write_unlock_bh(&bond->lock);
2072                 read_lock(&bond->lock);
2073                 write_lock_bh(&bond->curr_slave_lock);
2074
2075                 bond_select_active_slave(bond);
2076
2077                 write_unlock_bh(&bond->curr_slave_lock);
2078                 read_unlock(&bond->lock);
2079                 write_lock_bh(&bond->lock);
2080         }
2081
2082         if (bond->slave_cnt == 0) {
2083                 bond_set_carrier(bond);
2084                 eth_hw_addr_random(bond_dev);
2085                 bond->dev_addr_from_first = true;
2086
2087                 if (bond_vlan_used(bond)) {
2088                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2089                                    bond_dev->name, bond_dev->name);
2090                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2091                                    bond_dev->name);
2092                 }
2093         }
2094
2095         write_unlock_bh(&bond->lock);
2096         unblock_netpoll_tx();
2097
2098         if (bond->slave_cnt == 0) {
2099                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2100                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2101         }
2102
2103         bond_compute_features(bond);
2104         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2105             (old_features & NETIF_F_VLAN_CHALLENGED))
2106                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2107                         bond_dev->name, slave_dev->name, bond_dev->name);
2108
2109         /* must do this from outside any spinlocks */
2110         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2111
2112         bond_del_vlans_from_slave(bond, slave_dev);
2113
2114         /* If the mode USES_PRIMARY, then we should only remove its
2115          * promisc and mc settings if it was the curr_active_slave, but that was
2116          * already taken care of above when we detached the slave
2117          */
2118         if (!USES_PRIMARY(bond->params.mode)) {
2119                 /* unset promiscuity level from slave */
2120                 if (bond_dev->flags & IFF_PROMISC)
2121                         dev_set_promiscuity(slave_dev, -1);
2122
2123                 /* unset allmulti level from slave */
2124                 if (bond_dev->flags & IFF_ALLMULTI)
2125                         dev_set_allmulti(slave_dev, -1);
2126
2127                 /* flush master's mc_list from slave */
2128                 netif_addr_lock_bh(bond_dev);
2129                 bond_mc_list_flush(bond_dev, slave_dev);
2130                 netif_addr_unlock_bh(bond_dev);
2131         }
2132
2133         bond_upper_dev_unlink(bond_dev, slave_dev);
2134
2135         slave_disable_netpoll(slave);
2136
2137         /* close slave before restoring its mac address */
2138         dev_close(slave_dev);
2139
2140         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2141                 /* restore original ("permanent") mac address */
2142                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2143                 addr.sa_family = slave_dev->type;
2144                 dev_set_mac_address(slave_dev, &addr);
2145         }
2146
2147         dev_set_mtu(slave_dev, slave->original_mtu);
2148
2149         slave_dev->priv_flags &= ~IFF_BONDING;
2150
2151         kfree(slave);
2152
2153         return 0;  /* deletion OK */
2154 }
2155
2156 /* A wrapper used because of ndo_del_link */
2157 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2158 {
2159         return __bond_release_one(bond_dev, slave_dev, false);
2160 }
2161
2162 /*
2163 * First release a slave and then destroy the bond if no more slaves are left.
2164 * Must be under rtnl_lock when this function is called.
2165 */
2166 static int  bond_release_and_destroy(struct net_device *bond_dev,
2167                                      struct net_device *slave_dev)
2168 {
2169         struct bonding *bond = netdev_priv(bond_dev);
2170         int ret;
2171
2172         ret = bond_release(bond_dev, slave_dev);
2173         if ((ret == 0) && (bond->slave_cnt == 0)) {
2174                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2175                 pr_info("%s: destroying bond %s.\n",
2176                         bond_dev->name, bond_dev->name);
2177                 unregister_netdevice(bond_dev);
2178         }
2179         return ret;
2180 }
2181
2182 /*
2183  * This function changes the active slave to slave <slave_dev>.
2184  * It returns -EINVAL in the following cases.
2185  *  - <slave_dev> is not found in the list.
2186  *  - There is not active slave now.
2187  *  - <slave_dev> is already active.
2188  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2189  *  - <slave_dev> is not running.
2190  * In these cases, this function does nothing.
2191  * In the other cases, current_slave pointer is changed and 0 is returned.
2192  */
2193 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2194 {
2195         struct bonding *bond = netdev_priv(bond_dev);
2196         struct slave *old_active = NULL;
2197         struct slave *new_active = NULL;
2198         int res = 0;
2199
2200         if (!USES_PRIMARY(bond->params.mode))
2201                 return -EINVAL;
2202
2203         /* Verify that bond_dev is indeed the master of slave_dev */
2204         if (!(slave_dev->flags & IFF_SLAVE) ||
2205             !netdev_has_upper_dev(slave_dev, bond_dev))
2206                 return -EINVAL;
2207
2208         read_lock(&bond->lock);
2209
2210         read_lock(&bond->curr_slave_lock);
2211         old_active = bond->curr_active_slave;
2212         read_unlock(&bond->curr_slave_lock);
2213
2214         new_active = bond_get_slave_by_dev(bond, slave_dev);
2215
2216         /*
2217          * Changing to the current active: do nothing; return success.
2218          */
2219         if (new_active && (new_active == old_active)) {
2220                 read_unlock(&bond->lock);
2221                 return 0;
2222         }
2223
2224         if ((new_active) &&
2225             (old_active) &&
2226             (new_active->link == BOND_LINK_UP) &&
2227             IS_UP(new_active->dev)) {
2228                 block_netpoll_tx();
2229                 write_lock_bh(&bond->curr_slave_lock);
2230                 bond_change_active_slave(bond, new_active);
2231                 write_unlock_bh(&bond->curr_slave_lock);
2232                 unblock_netpoll_tx();
2233         } else
2234                 res = -EINVAL;
2235
2236         read_unlock(&bond->lock);
2237
2238         return res;
2239 }
2240
2241 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2242 {
2243         struct bonding *bond = netdev_priv(bond_dev);
2244
2245         info->bond_mode = bond->params.mode;
2246         info->miimon = bond->params.miimon;
2247
2248         read_lock(&bond->lock);
2249         info->num_slaves = bond->slave_cnt;
2250         read_unlock(&bond->lock);
2251
2252         return 0;
2253 }
2254
2255 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2256 {
2257         struct bonding *bond = netdev_priv(bond_dev);
2258         struct slave *slave;
2259         int i, res = -ENODEV;
2260
2261         read_lock(&bond->lock);
2262
2263         bond_for_each_slave(bond, slave, i) {
2264                 if (i == (int)info->slave_id) {
2265                         res = 0;
2266                         strcpy(info->slave_name, slave->dev->name);
2267                         info->link = slave->link;
2268                         info->state = bond_slave_state(slave);
2269                         info->link_failure_count = slave->link_failure_count;
2270                         break;
2271                 }
2272         }
2273
2274         read_unlock(&bond->lock);
2275
2276         return res;
2277 }
2278
2279 /*-------------------------------- Monitoring -------------------------------*/
2280
2281
2282 static int bond_miimon_inspect(struct bonding *bond)
2283 {
2284         struct slave *slave;
2285         int i, link_state, commit = 0;
2286         bool ignore_updelay;
2287
2288         ignore_updelay = !bond->curr_active_slave ? true : false;
2289
2290         bond_for_each_slave(bond, slave, i) {
2291                 slave->new_link = BOND_LINK_NOCHANGE;
2292
2293                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2294
2295                 switch (slave->link) {
2296                 case BOND_LINK_UP:
2297                         if (link_state)
2298                                 continue;
2299
2300                         slave->link = BOND_LINK_FAIL;
2301                         slave->delay = bond->params.downdelay;
2302                         if (slave->delay) {
2303                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2304                                         bond->dev->name,
2305                                         (bond->params.mode ==
2306                                          BOND_MODE_ACTIVEBACKUP) ?
2307                                         (bond_is_active_slave(slave) ?
2308                                          "active " : "backup ") : "",
2309                                         slave->dev->name,
2310                                         bond->params.downdelay * bond->params.miimon);
2311                         }
2312                         /*FALLTHRU*/
2313                 case BOND_LINK_FAIL:
2314                         if (link_state) {
2315                                 /*
2316                                  * recovered before downdelay expired
2317                                  */
2318                                 slave->link = BOND_LINK_UP;
2319                                 slave->jiffies = jiffies;
2320                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2321                                         bond->dev->name,
2322                                         (bond->params.downdelay - slave->delay) *
2323                                         bond->params.miimon,
2324                                         slave->dev->name);
2325                                 continue;
2326                         }
2327
2328                         if (slave->delay <= 0) {
2329                                 slave->new_link = BOND_LINK_DOWN;
2330                                 commit++;
2331                                 continue;
2332                         }
2333
2334                         slave->delay--;
2335                         break;
2336
2337                 case BOND_LINK_DOWN:
2338                         if (!link_state)
2339                                 continue;
2340
2341                         slave->link = BOND_LINK_BACK;
2342                         slave->delay = bond->params.updelay;
2343
2344                         if (slave->delay) {
2345                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2346                                         bond->dev->name, slave->dev->name,
2347                                         ignore_updelay ? 0 :
2348                                         bond->params.updelay *
2349                                         bond->params.miimon);
2350                         }
2351                         /*FALLTHRU*/
2352                 case BOND_LINK_BACK:
2353                         if (!link_state) {
2354                                 slave->link = BOND_LINK_DOWN;
2355                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2356                                         bond->dev->name,
2357                                         (bond->params.updelay - slave->delay) *
2358                                         bond->params.miimon,
2359                                         slave->dev->name);
2360
2361                                 continue;
2362                         }
2363
2364                         if (ignore_updelay)
2365                                 slave->delay = 0;
2366
2367                         if (slave->delay <= 0) {
2368                                 slave->new_link = BOND_LINK_UP;
2369                                 commit++;
2370                                 ignore_updelay = false;
2371                                 continue;
2372                         }
2373
2374                         slave->delay--;
2375                         break;
2376                 }
2377         }
2378
2379         return commit;
2380 }
2381
2382 static void bond_miimon_commit(struct bonding *bond)
2383 {
2384         struct slave *slave;
2385         int i;
2386
2387         bond_for_each_slave(bond, slave, i) {
2388                 switch (slave->new_link) {
2389                 case BOND_LINK_NOCHANGE:
2390                         continue;
2391
2392                 case BOND_LINK_UP:
2393                         slave->link = BOND_LINK_UP;
2394                         slave->jiffies = jiffies;
2395
2396                         if (bond->params.mode == BOND_MODE_8023AD) {
2397                                 /* prevent it from being the active one */
2398                                 bond_set_backup_slave(slave);
2399                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2400                                 /* make it immediately active */
2401                                 bond_set_active_slave(slave);
2402                         } else if (slave != bond->primary_slave) {
2403                                 /* prevent it from being the active one */
2404                                 bond_set_backup_slave(slave);
2405                         }
2406
2407                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2408                                 bond->dev->name, slave->dev->name,
2409                                 slave->speed, slave->duplex ? "full" : "half");
2410
2411                         /* notify ad that the link status has changed */
2412                         if (bond->params.mode == BOND_MODE_8023AD)
2413                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2414
2415                         if (bond_is_lb(bond))
2416                                 bond_alb_handle_link_change(bond, slave,
2417                                                             BOND_LINK_UP);
2418
2419                         if (!bond->curr_active_slave ||
2420                             (slave == bond->primary_slave))
2421                                 goto do_failover;
2422
2423                         continue;
2424
2425                 case BOND_LINK_DOWN:
2426                         if (slave->link_failure_count < UINT_MAX)
2427                                 slave->link_failure_count++;
2428
2429                         slave->link = BOND_LINK_DOWN;
2430
2431                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2432                             bond->params.mode == BOND_MODE_8023AD)
2433                                 bond_set_slave_inactive_flags(slave);
2434
2435                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2436                                 bond->dev->name, slave->dev->name);
2437
2438                         if (bond->params.mode == BOND_MODE_8023AD)
2439                                 bond_3ad_handle_link_change(slave,
2440                                                             BOND_LINK_DOWN);
2441
2442                         if (bond_is_lb(bond))
2443                                 bond_alb_handle_link_change(bond, slave,
2444                                                             BOND_LINK_DOWN);
2445
2446                         if (slave == bond->curr_active_slave)
2447                                 goto do_failover;
2448
2449                         continue;
2450
2451                 default:
2452                         pr_err("%s: invalid new link %d on slave %s\n",
2453                                bond->dev->name, slave->new_link,
2454                                slave->dev->name);
2455                         slave->new_link = BOND_LINK_NOCHANGE;
2456
2457                         continue;
2458                 }
2459
2460 do_failover:
2461                 ASSERT_RTNL();
2462                 block_netpoll_tx();
2463                 write_lock_bh(&bond->curr_slave_lock);
2464                 bond_select_active_slave(bond);
2465                 write_unlock_bh(&bond->curr_slave_lock);
2466                 unblock_netpoll_tx();
2467         }
2468
2469         bond_set_carrier(bond);
2470 }
2471
2472 /*
2473  * bond_mii_monitor
2474  *
2475  * Really a wrapper that splits the mii monitor into two phases: an
2476  * inspection, then (if inspection indicates something needs to be done)
2477  * an acquisition of appropriate locks followed by a commit phase to
2478  * implement whatever link state changes are indicated.
2479  */
2480 void bond_mii_monitor(struct work_struct *work)
2481 {
2482         struct bonding *bond = container_of(work, struct bonding,
2483                                             mii_work.work);
2484         bool should_notify_peers = false;
2485         unsigned long delay;
2486
2487         read_lock(&bond->lock);
2488
2489         delay = msecs_to_jiffies(bond->params.miimon);
2490
2491         if (bond->slave_cnt == 0)
2492                 goto re_arm;
2493
2494         should_notify_peers = bond_should_notify_peers(bond);
2495
2496         if (bond_miimon_inspect(bond)) {
2497                 read_unlock(&bond->lock);
2498
2499                 /* Race avoidance with bond_close cancel of workqueue */
2500                 if (!rtnl_trylock()) {
2501                         read_lock(&bond->lock);
2502                         delay = 1;
2503                         should_notify_peers = false;
2504                         goto re_arm;
2505                 }
2506
2507                 read_lock(&bond->lock);
2508
2509                 bond_miimon_commit(bond);
2510
2511                 read_unlock(&bond->lock);
2512                 rtnl_unlock();  /* might sleep, hold no other locks */
2513                 read_lock(&bond->lock);
2514         }
2515
2516 re_arm:
2517         if (bond->params.miimon)
2518                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2519
2520         read_unlock(&bond->lock);
2521
2522         if (should_notify_peers) {
2523                 if (!rtnl_trylock()) {
2524                         read_lock(&bond->lock);
2525                         bond->send_peer_notif++;
2526                         read_unlock(&bond->lock);
2527                         return;
2528                 }
2529                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2530                 rtnl_unlock();
2531         }
2532 }
2533
2534 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2535 {
2536         struct vlan_entry *vlan;
2537         struct net_device *vlan_dev;
2538
2539         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2540                 return 1;
2541
2542         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2543                 rcu_read_lock();
2544                 vlan_dev = __vlan_find_dev_deep(bond->dev, htons(ETH_P_8021Q),
2545                                                 vlan->vlan_id);
2546                 rcu_read_unlock();
2547                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2548                         return 1;
2549         }
2550
2551         return 0;
2552 }
2553
2554 /*
2555  * We go to the (large) trouble of VLAN tagging ARP frames because
2556  * switches in VLAN mode (especially if ports are configured as
2557  * "native" to a VLAN) might not pass non-tagged frames.
2558  */
2559 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2560 {
2561         struct sk_buff *skb;
2562
2563         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2564                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2565
2566         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2567                          NULL, slave_dev->dev_addr, NULL);
2568
2569         if (!skb) {
2570                 pr_err("ARP packet allocation failed\n");
2571                 return;
2572         }
2573         if (vlan_id) {
2574                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2575                 if (!skb) {
2576                         pr_err("failed to insert VLAN tag\n");
2577                         return;
2578                 }
2579         }
2580         arp_xmit(skb);
2581 }
2582
2583
2584 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2585 {
2586         int i, vlan_id;
2587         __be32 *targets = bond->params.arp_targets;
2588         struct vlan_entry *vlan;
2589         struct net_device *vlan_dev = NULL;
2590         struct rtable *rt;
2591
2592         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2593                 __be32 addr;
2594                 if (!targets[i])
2595                         break;
2596                 pr_debug("basa: target %pI4\n", &targets[i]);
2597                 if (!bond_vlan_used(bond)) {
2598                         pr_debug("basa: empty vlan: arp_send\n");
2599                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2600                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2601                                       addr, 0);
2602                         continue;
2603                 }
2604
2605                 /*
2606                  * If VLANs are configured, we do a route lookup to
2607                  * determine which VLAN interface would be used, so we
2608                  * can tag the ARP with the proper VLAN tag.
2609                  */
2610                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2611                                      RTO_ONLINK, 0);
2612                 if (IS_ERR(rt)) {
2613                         if (net_ratelimit()) {
2614                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2615                                            bond->dev->name, &targets[i]);
2616                         }
2617                         continue;
2618                 }
2619
2620                 /*
2621                  * This target is not on a VLAN
2622                  */
2623                 if (rt->dst.dev == bond->dev) {
2624                         ip_rt_put(rt);
2625                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2626                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2627                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2628                                       addr, 0);
2629                         continue;
2630                 }
2631
2632                 vlan_id = 0;
2633                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2634                         rcu_read_lock();
2635                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2636                                                         htons(ETH_P_8021Q),
2637                                                         vlan->vlan_id);
2638                         rcu_read_unlock();
2639                         if (vlan_dev == rt->dst.dev) {
2640                                 vlan_id = vlan->vlan_id;
2641                                 pr_debug("basa: vlan match on %s %d\n",
2642                                        vlan_dev->name, vlan_id);
2643                                 break;
2644                         }
2645                 }
2646
2647                 if (vlan_id && vlan_dev) {
2648                         ip_rt_put(rt);
2649                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2650                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2651                                       addr, vlan_id);
2652                         continue;
2653                 }
2654
2655                 if (net_ratelimit()) {
2656                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2657                                    bond->dev->name, &targets[i],
2658                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2659                 }
2660                 ip_rt_put(rt);
2661         }
2662 }
2663
2664 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2665 {
2666         int i;
2667         __be32 *targets = bond->params.arp_targets;
2668
2669         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2670                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2671                          &sip, &tip, i, &targets[i],
2672                          bond_has_this_ip(bond, tip));
2673                 if (sip == targets[i]) {
2674                         if (bond_has_this_ip(bond, tip))
2675                                 slave->last_arp_rx = jiffies;
2676                         return;
2677                 }
2678         }
2679 }
2680
2681 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2682                         struct slave *slave)
2683 {
2684         struct arphdr *arp = (struct arphdr *)skb->data;
2685         unsigned char *arp_ptr;
2686         __be32 sip, tip;
2687         int alen;
2688
2689         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2690                 return RX_HANDLER_ANOTHER;
2691
2692         read_lock(&bond->lock);
2693         alen = arp_hdr_len(bond->dev);
2694
2695         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2696                  bond->dev->name, skb->dev->name);
2697
2698         if (alen > skb_headlen(skb)) {
2699                 arp = kmalloc(alen, GFP_ATOMIC);
2700                 if (!arp)
2701                         goto out_unlock;
2702                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2703                         goto out_unlock;
2704         }
2705
2706         if (arp->ar_hln != bond->dev->addr_len ||
2707             skb->pkt_type == PACKET_OTHERHOST ||
2708             skb->pkt_type == PACKET_LOOPBACK ||
2709             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2710             arp->ar_pro != htons(ETH_P_IP) ||
2711             arp->ar_pln != 4)
2712                 goto out_unlock;
2713
2714         arp_ptr = (unsigned char *)(arp + 1);
2715         arp_ptr += bond->dev->addr_len;
2716         memcpy(&sip, arp_ptr, 4);
2717         arp_ptr += 4 + bond->dev->addr_len;
2718         memcpy(&tip, arp_ptr, 4);
2719
2720         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2721                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2722                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2723                  &sip, &tip);
2724
2725         /*
2726          * Backup slaves won't see the ARP reply, but do come through
2727          * here for each ARP probe (so we swap the sip/tip to validate
2728          * the probe).  In a "redundant switch, common router" type of
2729          * configuration, the ARP probe will (hopefully) travel from
2730          * the active, through one switch, the router, then the other
2731          * switch before reaching the backup.
2732          */
2733         if (bond_is_active_slave(slave))
2734                 bond_validate_arp(bond, slave, sip, tip);
2735         else
2736                 bond_validate_arp(bond, slave, tip, sip);
2737
2738 out_unlock:
2739         read_unlock(&bond->lock);
2740         if (arp != (struct arphdr *)skb->data)
2741                 kfree(arp);
2742         return RX_HANDLER_ANOTHER;
2743 }
2744
2745 /*
2746  * this function is called regularly to monitor each slave's link
2747  * ensuring that traffic is being sent and received when arp monitoring
2748  * is used in load-balancing mode. if the adapter has been dormant, then an
2749  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2750  * arp monitoring in active backup mode.
2751  */
2752 void bond_loadbalance_arp_mon(struct work_struct *work)
2753 {
2754         struct bonding *bond = container_of(work, struct bonding,
2755                                             arp_work.work);
2756         struct slave *slave, *oldcurrent;
2757         int do_failover = 0;
2758         int delta_in_ticks, extra_ticks;
2759         int i;
2760
2761         read_lock(&bond->lock);
2762
2763         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2764         extra_ticks = delta_in_ticks / 2;
2765
2766         if (bond->slave_cnt == 0)
2767                 goto re_arm;
2768
2769         read_lock(&bond->curr_slave_lock);
2770         oldcurrent = bond->curr_active_slave;
2771         read_unlock(&bond->curr_slave_lock);
2772
2773         /* see if any of the previous devices are up now (i.e. they have
2774          * xmt and rcv traffic). the curr_active_slave does not come into
2775          * the picture unless it is null. also, slave->jiffies is not needed
2776          * here because we send an arp on each slave and give a slave as
2777          * long as it needs to get the tx/rx within the delta.
2778          * TODO: what about up/down delay in arp mode? it wasn't here before
2779          *       so it can wait
2780          */
2781         bond_for_each_slave(bond, slave, i) {
2782                 unsigned long trans_start = dev_trans_start(slave->dev);
2783
2784                 if (slave->link != BOND_LINK_UP) {
2785                         if (time_in_range(jiffies,
2786                                 trans_start - delta_in_ticks,
2787                                 trans_start + delta_in_ticks + extra_ticks) &&
2788                             time_in_range(jiffies,
2789                                 slave->dev->last_rx - delta_in_ticks,
2790                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2791
2792                                 slave->link  = BOND_LINK_UP;
2793                                 bond_set_active_slave(slave);
2794
2795                                 /* primary_slave has no meaning in round-robin
2796                                  * mode. the window of a slave being up and
2797                                  * curr_active_slave being null after enslaving
2798                                  * is closed.
2799                                  */
2800                                 if (!oldcurrent) {
2801                                         pr_info("%s: link status definitely up for interface %s, ",
2802                                                 bond->dev->name,
2803                                                 slave->dev->name);
2804                                         do_failover = 1;
2805                                 } else {
2806                                         pr_info("%s: interface %s is now up\n",
2807                                                 bond->dev->name,
2808                                                 slave->dev->name);
2809                                 }
2810                         }
2811                 } else {
2812                         /* slave->link == BOND_LINK_UP */
2813
2814                         /* not all switches will respond to an arp request
2815                          * when the source ip is 0, so don't take the link down
2816                          * if we don't know our ip yet
2817                          */
2818                         if (!time_in_range(jiffies,
2819                                 trans_start - delta_in_ticks,
2820                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2821                             !time_in_range(jiffies,
2822                                 slave->dev->last_rx - delta_in_ticks,
2823                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2824
2825                                 slave->link  = BOND_LINK_DOWN;
2826                                 bond_set_backup_slave(slave);
2827
2828                                 if (slave->link_failure_count < UINT_MAX)
2829                                         slave->link_failure_count++;
2830
2831                                 pr_info("%s: interface %s is now down.\n",
2832                                         bond->dev->name,
2833                                         slave->dev->name);
2834
2835                                 if (slave == oldcurrent)
2836                                         do_failover = 1;
2837                         }
2838                 }
2839
2840                 /* note: if switch is in round-robin mode, all links
2841                  * must tx arp to ensure all links rx an arp - otherwise
2842                  * links may oscillate or not come up at all; if switch is
2843                  * in something like xor mode, there is nothing we can
2844                  * do - all replies will be rx'ed on same link causing slaves
2845                  * to be unstable during low/no traffic periods
2846                  */
2847                 if (IS_UP(slave->dev))
2848                         bond_arp_send_all(bond, slave);
2849         }
2850
2851         if (do_failover) {
2852                 block_netpoll_tx();
2853                 write_lock_bh(&bond->curr_slave_lock);
2854
2855                 bond_select_active_slave(bond);
2856
2857                 write_unlock_bh(&bond->curr_slave_lock);
2858                 unblock_netpoll_tx();
2859         }
2860
2861 re_arm:
2862         if (bond->params.arp_interval)
2863                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2864
2865         read_unlock(&bond->lock);
2866 }
2867
2868 /*
2869  * Called to inspect slaves for active-backup mode ARP monitor link state
2870  * changes.  Sets new_link in slaves to specify what action should take
2871  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2872  * to link states must be committed.
2873  *
2874  * Called with bond->lock held for read.
2875  */
2876 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2877 {
2878         struct slave *slave;
2879         int i, commit = 0;
2880         unsigned long trans_start;
2881         int extra_ticks;
2882
2883         /* All the time comparisons below need some extra time. Otherwise, on
2884          * fast networks the ARP probe/reply may arrive within the same jiffy
2885          * as it was sent.  Then, the next time the ARP monitor is run, one
2886          * arp_interval will already have passed in the comparisons.
2887          */
2888         extra_ticks = delta_in_ticks / 2;
2889
2890         bond_for_each_slave(bond, slave, i) {
2891                 slave->new_link = BOND_LINK_NOCHANGE;
2892
2893                 if (slave->link != BOND_LINK_UP) {
2894                         if (time_in_range(jiffies,
2895                                 slave_last_rx(bond, slave) - delta_in_ticks,
2896                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2897
2898                                 slave->new_link = BOND_LINK_UP;
2899                                 commit++;
2900                         }
2901
2902                         continue;
2903                 }
2904
2905                 /*
2906                  * Give slaves 2*delta after being enslaved or made
2907                  * active.  This avoids bouncing, as the last receive
2908                  * times need a full ARP monitor cycle to be updated.
2909                  */
2910                 if (time_in_range(jiffies,
2911                                   slave->jiffies - delta_in_ticks,
2912                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2913                         continue;
2914
2915                 /*
2916                  * Backup slave is down if:
2917                  * - No current_arp_slave AND
2918                  * - more than 3*delta since last receive AND
2919                  * - the bond has an IP address
2920                  *
2921                  * Note: a non-null current_arp_slave indicates
2922                  * the curr_active_slave went down and we are
2923                  * searching for a new one; under this condition
2924                  * we only take the curr_active_slave down - this
2925                  * gives each slave a chance to tx/rx traffic
2926                  * before being taken out
2927                  */
2928                 if (!bond_is_active_slave(slave) &&
2929                     !bond->current_arp_slave &&
2930                     !time_in_range(jiffies,
2931                         slave_last_rx(bond, slave) - delta_in_ticks,
2932                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2933
2934                         slave->new_link = BOND_LINK_DOWN;
2935                         commit++;
2936                 }
2937
2938                 /*
2939                  * Active slave is down if:
2940                  * - more than 2*delta since transmitting OR
2941                  * - (more than 2*delta since receive AND
2942                  *    the bond has an IP address)
2943                  */
2944                 trans_start = dev_trans_start(slave->dev);
2945                 if (bond_is_active_slave(slave) &&
2946                     (!time_in_range(jiffies,
2947                         trans_start - delta_in_ticks,
2948                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
2949                      !time_in_range(jiffies,
2950                         slave_last_rx(bond, slave) - delta_in_ticks,
2951                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2952
2953                         slave->new_link = BOND_LINK_DOWN;
2954                         commit++;
2955                 }
2956         }
2957
2958         return commit;
2959 }
2960
2961 /*
2962  * Called to commit link state changes noted by inspection step of
2963  * active-backup mode ARP monitor.
2964  *
2965  * Called with RTNL and bond->lock for read.
2966  */
2967 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2968 {
2969         struct slave *slave;
2970         int i;
2971         unsigned long trans_start;
2972
2973         bond_for_each_slave(bond, slave, i) {
2974                 switch (slave->new_link) {
2975                 case BOND_LINK_NOCHANGE:
2976                         continue;
2977
2978                 case BOND_LINK_UP:
2979                         trans_start = dev_trans_start(slave->dev);
2980                         if ((!bond->curr_active_slave &&
2981                              time_in_range(jiffies,
2982                                            trans_start - delta_in_ticks,
2983                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2984                             bond->curr_active_slave != slave) {
2985                                 slave->link = BOND_LINK_UP;
2986                                 if (bond->current_arp_slave) {
2987                                         bond_set_slave_inactive_flags(
2988                                                 bond->current_arp_slave);
2989                                         bond->current_arp_slave = NULL;
2990                                 }
2991
2992                                 pr_info("%s: link status definitely up for interface %s.\n",
2993                                         bond->dev->name, slave->dev->name);
2994
2995                                 if (!bond->curr_active_slave ||
2996                                     (slave == bond->primary_slave))
2997                                         goto do_failover;
2998
2999                         }
3000
3001                         continue;
3002
3003                 case BOND_LINK_DOWN:
3004                         if (slave->link_failure_count < UINT_MAX)
3005                                 slave->link_failure_count++;
3006
3007                         slave->link = BOND_LINK_DOWN;
3008                         bond_set_slave_inactive_flags(slave);
3009
3010                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3011                                 bond->dev->name, slave->dev->name);
3012
3013                         if (slave == bond->curr_active_slave) {
3014                                 bond->current_arp_slave = NULL;
3015                                 goto do_failover;
3016                         }
3017
3018                         continue;
3019
3020                 default:
3021                         pr_err("%s: impossible: new_link %d on slave %s\n",
3022                                bond->dev->name, slave->new_link,
3023                                slave->dev->name);
3024                         continue;
3025                 }
3026
3027 do_failover:
3028                 ASSERT_RTNL();
3029                 block_netpoll_tx();
3030                 write_lock_bh(&bond->curr_slave_lock);
3031                 bond_select_active_slave(bond);
3032                 write_unlock_bh(&bond->curr_slave_lock);
3033                 unblock_netpoll_tx();
3034         }
3035
3036         bond_set_carrier(bond);
3037 }
3038
3039 /*
3040  * Send ARP probes for active-backup mode ARP monitor.
3041  *
3042  * Called with bond->lock held for read.
3043  */
3044 static void bond_ab_arp_probe(struct bonding *bond)
3045 {
3046         struct slave *slave;
3047         int i;
3048
3049         read_lock(&bond->curr_slave_lock);
3050
3051         if (bond->current_arp_slave && bond->curr_active_slave)
3052                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3053                         bond->current_arp_slave->dev->name,
3054                         bond->curr_active_slave->dev->name);
3055
3056         if (bond->curr_active_slave) {
3057                 bond_arp_send_all(bond, bond->curr_active_slave);
3058                 read_unlock(&bond->curr_slave_lock);
3059                 return;
3060         }
3061
3062         read_unlock(&bond->curr_slave_lock);
3063
3064         /* if we don't have a curr_active_slave, search for the next available
3065          * backup slave from the current_arp_slave and make it the candidate
3066          * for becoming the curr_active_slave
3067          */
3068
3069         if (!bond->current_arp_slave) {
3070                 bond->current_arp_slave = bond->first_slave;
3071                 if (!bond->current_arp_slave)
3072                         return;
3073         }
3074
3075         bond_set_slave_inactive_flags(bond->current_arp_slave);
3076
3077         /* search for next candidate */
3078         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3079                 if (IS_UP(slave->dev)) {
3080                         slave->link = BOND_LINK_BACK;
3081                         bond_set_slave_active_flags(slave);
3082                         bond_arp_send_all(bond, slave);
3083                         slave->jiffies = jiffies;
3084                         bond->current_arp_slave = slave;
3085                         break;
3086                 }
3087
3088                 /* if the link state is up at this point, we
3089                  * mark it down - this can happen if we have
3090                  * simultaneous link failures and
3091                  * reselect_active_interface doesn't make this
3092                  * one the current slave so it is still marked
3093                  * up when it is actually down
3094                  */
3095                 if (slave->link == BOND_LINK_UP) {
3096                         slave->link = BOND_LINK_DOWN;
3097                         if (slave->link_failure_count < UINT_MAX)
3098                                 slave->link_failure_count++;
3099
3100                         bond_set_slave_inactive_flags(slave);
3101
3102                         pr_info("%s: backup interface %s is now down.\n",
3103                                 bond->dev->name, slave->dev->name);
3104                 }
3105         }
3106 }
3107
3108 void bond_activebackup_arp_mon(struct work_struct *work)
3109 {
3110         struct bonding *bond = container_of(work, struct bonding,
3111                                             arp_work.work);
3112         bool should_notify_peers = false;
3113         int delta_in_ticks;
3114
3115         read_lock(&bond->lock);
3116
3117         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3118
3119         if (bond->slave_cnt == 0)
3120                 goto re_arm;
3121
3122         should_notify_peers = bond_should_notify_peers(bond);
3123
3124         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3125                 read_unlock(&bond->lock);
3126
3127                 /* Race avoidance with bond_close flush of workqueue */
3128                 if (!rtnl_trylock()) {
3129                         read_lock(&bond->lock);
3130                         delta_in_ticks = 1;
3131                         should_notify_peers = false;
3132                         goto re_arm;
3133                 }
3134
3135                 read_lock(&bond->lock);
3136
3137                 bond_ab_arp_commit(bond, delta_in_ticks);
3138
3139                 read_unlock(&bond->lock);
3140                 rtnl_unlock();
3141                 read_lock(&bond->lock);
3142         }
3143
3144         bond_ab_arp_probe(bond);
3145
3146 re_arm:
3147         if (bond->params.arp_interval)
3148                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3149
3150         read_unlock(&bond->lock);
3151
3152         if (should_notify_peers) {
3153                 if (!rtnl_trylock()) {
3154                         read_lock(&bond->lock);
3155                         bond->send_peer_notif++;
3156                         read_unlock(&bond->lock);
3157                         return;
3158                 }
3159                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3160                 rtnl_unlock();
3161         }
3162 }
3163
3164 /*-------------------------- netdev event handling --------------------------*/
3165
3166 /*
3167  * Change device name
3168  */
3169 static int bond_event_changename(struct bonding *bond)
3170 {
3171         bond_remove_proc_entry(bond);
3172         bond_create_proc_entry(bond);
3173
3174         bond_debug_reregister(bond);
3175
3176         return NOTIFY_DONE;
3177 }
3178
3179 static int bond_master_netdev_event(unsigned long event,
3180                                     struct net_device *bond_dev)
3181 {
3182         struct bonding *event_bond = netdev_priv(bond_dev);
3183
3184         switch (event) {
3185         case NETDEV_CHANGENAME:
3186                 return bond_event_changename(event_bond);
3187         case NETDEV_UNREGISTER:
3188                 bond_remove_proc_entry(event_bond);
3189                 break;
3190         case NETDEV_REGISTER:
3191                 bond_create_proc_entry(event_bond);
3192                 break;
3193         default:
3194                 break;
3195         }
3196
3197         return NOTIFY_DONE;
3198 }
3199
3200 static int bond_slave_netdev_event(unsigned long event,
3201                                    struct net_device *slave_dev)
3202 {
3203         struct slave *slave = bond_slave_get_rtnl(slave_dev);
3204         struct bonding *bond;
3205         struct net_device *bond_dev;
3206         u32 old_speed;
3207         u8 old_duplex;
3208
3209         /* A netdev event can be generated while enslaving a device
3210          * before netdev_rx_handler_register is called in which case
3211          * slave will be NULL
3212          */
3213         if (!slave)
3214                 return NOTIFY_DONE;
3215         bond_dev = slave->bond->dev;
3216         bond = slave->bond;
3217
3218         switch (event) {
3219         case NETDEV_UNREGISTER:
3220                 if (bond->setup_by_slave)
3221                         bond_release_and_destroy(bond_dev, slave_dev);
3222                 else
3223                         bond_release(bond_dev, slave_dev);
3224                 break;
3225         case NETDEV_UP:
3226         case NETDEV_CHANGE:
3227                 old_speed = slave->speed;
3228                 old_duplex = slave->duplex;
3229
3230                 bond_update_speed_duplex(slave);
3231
3232                 if (bond->params.mode == BOND_MODE_8023AD) {
3233                         if (old_speed != slave->speed)
3234                                 bond_3ad_adapter_speed_changed(slave);
3235                         if (old_duplex != slave->duplex)
3236                                 bond_3ad_adapter_duplex_changed(slave);
3237                 }
3238                 break;
3239         case NETDEV_DOWN:
3240                 /*
3241                  * ... Or is it this?
3242                  */
3243                 break;
3244         case NETDEV_CHANGEMTU:
3245                 /*
3246                  * TODO: Should slaves be allowed to
3247                  * independently alter their MTU?  For
3248                  * an active-backup bond, slaves need
3249                  * not be the same type of device, so
3250                  * MTUs may vary.  For other modes,
3251                  * slaves arguably should have the
3252                  * same MTUs. To do this, we'd need to
3253                  * take over the slave's change_mtu
3254                  * function for the duration of their
3255                  * servitude.
3256                  */
3257                 break;
3258         case NETDEV_CHANGENAME:
3259                 /*
3260                  * TODO: handle changing the primary's name
3261                  */
3262                 break;
3263         case NETDEV_FEAT_CHANGE:
3264                 bond_compute_features(bond);
3265                 break;
3266         default:
3267                 break;
3268         }
3269
3270         return NOTIFY_DONE;
3271 }
3272
3273 /*
3274  * bond_netdev_event: handle netdev notifier chain events.
3275  *
3276  * This function receives events for the netdev chain.  The caller (an
3277  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3278  * locks for us to safely manipulate the slave devices (RTNL lock,
3279  * dev_probe_lock).
3280  */
3281 static int bond_netdev_event(struct notifier_block *this,
3282                              unsigned long event, void *ptr)
3283 {
3284         struct net_device *event_dev = (struct net_device *)ptr;
3285
3286         pr_debug("event_dev: %s, event: %lx\n",
3287                  event_dev ? event_dev->name : "None",
3288                  event);
3289
3290         if (!(event_dev->priv_flags & IFF_BONDING))
3291                 return NOTIFY_DONE;
3292
3293         if (event_dev->flags & IFF_MASTER) {
3294                 pr_debug("IFF_MASTER\n");
3295                 return bond_master_netdev_event(event, event_dev);
3296         }
3297
3298         if (event_dev->flags & IFF_SLAVE) {
3299                 pr_debug("IFF_SLAVE\n");
3300                 return bond_slave_netdev_event(event, event_dev);
3301         }
3302
3303         return NOTIFY_DONE;
3304 }
3305
3306 static struct notifier_block bond_netdev_notifier = {
3307         .notifier_call = bond_netdev_event,
3308 };
3309
3310 /*---------------------------- Hashing Policies -----------------------------*/
3311
3312 /*
3313  * Hash for the output device based upon layer 2 data
3314  */
3315 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3316 {
3317         struct ethhdr *data = (struct ethhdr *)skb->data;
3318
3319         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3320                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3321
3322         return 0;
3323 }
3324
3325 /*
3326  * Hash for the output device based upon layer 2 and layer 3 data. If
3327  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3328  */
3329 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3330 {
3331         const struct ethhdr *data;
3332         const struct iphdr *iph;
3333         const struct ipv6hdr *ipv6h;
3334         u32 v6hash;
3335         const __be32 *s, *d;
3336
3337         if (skb->protocol == htons(ETH_P_IP) &&
3338             pskb_network_may_pull(skb, sizeof(*iph))) {
3339                 iph = ip_hdr(skb);
3340                 data = (struct ethhdr *)skb->data;
3341                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3342                         (data->h_dest[5] ^ data->h_source[5])) % count;
3343         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3344                    pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3345                 ipv6h = ipv6_hdr(skb);
3346                 data = (struct ethhdr *)skb->data;
3347                 s = &ipv6h->saddr.s6_addr32[0];
3348                 d = &ipv6h->daddr.s6_addr32[0];
3349                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3350                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3351                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3352         }
3353
3354         return bond_xmit_hash_policy_l2(skb, count);
3355 }
3356
3357 /*
3358  * Hash for the output device based upon layer 3 and layer 4 data. If
3359  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3360  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3361  */
3362 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3363 {
3364         u32 layer4_xor = 0;
3365         const struct iphdr *iph;
3366         const struct ipv6hdr *ipv6h;
3367         const __be32 *s, *d;
3368         const __be16 *l4 = NULL;
3369         __be16 _l4[2];
3370         int noff = skb_network_offset(skb);
3371         int poff;
3372
3373         if (skb->protocol == htons(ETH_P_IP) &&
3374             pskb_may_pull(skb, noff + sizeof(*iph))) {
3375                 iph = ip_hdr(skb);
3376                 poff = proto_ports_offset(iph->protocol);
3377
3378                 if (!ip_is_fragment(iph) && poff >= 0) {
3379                         l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3380                                                 sizeof(_l4), &_l4);
3381                         if (l4)
3382                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3383                 }
3384                 return (layer4_xor ^
3385                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3386         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3387                    pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3388                 ipv6h = ipv6_hdr(skb);
3389                 poff = proto_ports_offset(ipv6h->nexthdr);
3390                 if (poff >= 0) {
3391                         l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3392                                                 sizeof(_l4), &_l4);
3393                         if (l4)
3394                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3395                 }
3396                 s = &ipv6h->saddr.s6_addr32[0];
3397                 d = &ipv6h->daddr.s6_addr32[0];
3398                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3399                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3400                                (layer4_xor >> 8);
3401                 return layer4_xor % count;
3402         }
3403
3404         return bond_xmit_hash_policy_l2(skb, count);
3405 }
3406
3407 /*-------------------------- Device entry points ----------------------------*/
3408
3409 static void bond_work_init_all(struct bonding *bond)
3410 {
3411         INIT_DELAYED_WORK(&bond->mcast_work,
3412                           bond_resend_igmp_join_requests_delayed);
3413         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3414         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3415         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3416                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3417         else
3418                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3419         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3420 }
3421
3422 static void bond_work_cancel_all(struct bonding *bond)
3423 {
3424         cancel_delayed_work_sync(&bond->mii_work);
3425         cancel_delayed_work_sync(&bond->arp_work);
3426         cancel_delayed_work_sync(&bond->alb_work);
3427         cancel_delayed_work_sync(&bond->ad_work);
3428         cancel_delayed_work_sync(&bond->mcast_work);
3429 }
3430
3431 static int bond_open(struct net_device *bond_dev)
3432 {
3433         struct bonding *bond = netdev_priv(bond_dev);
3434         struct slave *slave;
3435         int i;
3436
3437         /* reset slave->backup and slave->inactive */
3438         read_lock(&bond->lock);
3439         if (bond->slave_cnt > 0) {
3440                 read_lock(&bond->curr_slave_lock);
3441                 bond_for_each_slave(bond, slave, i) {
3442                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3443                                 && (slave != bond->curr_active_slave)) {
3444                                 bond_set_slave_inactive_flags(slave);
3445                         } else {
3446                                 bond_set_slave_active_flags(slave);
3447                         }
3448                 }
3449                 read_unlock(&bond->curr_slave_lock);
3450         }
3451         read_unlock(&bond->lock);
3452
3453         bond_work_init_all(bond);
3454
3455         if (bond_is_lb(bond)) {
3456                 /* bond_alb_initialize must be called before the timer
3457                  * is started.
3458                  */
3459                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3460                         return -ENOMEM;
3461                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3462         }
3463
3464         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3465                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3466
3467         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3468                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3469                 if (bond->params.arp_validate)
3470                         bond->recv_probe = bond_arp_rcv;
3471         }
3472
3473         if (bond->params.mode == BOND_MODE_8023AD) {
3474                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3475                 /* register to receive LACPDUs */
3476                 bond->recv_probe = bond_3ad_lacpdu_recv;
3477                 bond_3ad_initiate_agg_selection(bond, 1);
3478         }
3479
3480         return 0;
3481 }
3482
3483 static int bond_close(struct net_device *bond_dev)
3484 {
3485         struct bonding *bond = netdev_priv(bond_dev);
3486
3487         write_lock_bh(&bond->lock);
3488         bond->send_peer_notif = 0;
3489         write_unlock_bh(&bond->lock);
3490
3491         bond_work_cancel_all(bond);
3492         if (bond_is_lb(bond)) {
3493                 /* Must be called only after all
3494                  * slaves have been released
3495                  */
3496                 bond_alb_deinitialize(bond);
3497         }
3498         bond->recv_probe = NULL;
3499
3500         return 0;
3501 }
3502
3503 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3504                                                 struct rtnl_link_stats64 *stats)
3505 {
3506         struct bonding *bond = netdev_priv(bond_dev);
3507         struct rtnl_link_stats64 temp;
3508         struct slave *slave;
3509         int i;
3510
3511         memset(stats, 0, sizeof(*stats));
3512
3513         read_lock_bh(&bond->lock);
3514
3515         bond_for_each_slave(bond, slave, i) {
3516                 const struct rtnl_link_stats64 *sstats =
3517                         dev_get_stats(slave->dev, &temp);
3518
3519                 stats->rx_packets += sstats->rx_packets;
3520                 stats->rx_bytes += sstats->rx_bytes;
3521                 stats->rx_errors += sstats->rx_errors;
3522                 stats->rx_dropped += sstats->rx_dropped;
3523
3524                 stats->tx_packets += sstats->tx_packets;
3525                 stats->tx_bytes += sstats->tx_bytes;
3526                 stats->tx_errors += sstats->tx_errors;
3527                 stats->tx_dropped += sstats->tx_dropped;
3528
3529                 stats->multicast += sstats->multicast;
3530                 stats->collisions += sstats->collisions;
3531
3532                 stats->rx_length_errors += sstats->rx_length_errors;
3533                 stats->rx_over_errors += sstats->rx_over_errors;
3534                 stats->rx_crc_errors += sstats->rx_crc_errors;
3535                 stats->rx_frame_errors += sstats->rx_frame_errors;
3536                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3537                 stats->rx_missed_errors += sstats->rx_missed_errors;
3538
3539                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3540                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3541                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3542                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3543                 stats->tx_window_errors += sstats->tx_window_errors;
3544         }
3545
3546         read_unlock_bh(&bond->lock);
3547
3548         return stats;
3549 }
3550
3551 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3552 {
3553         struct net_device *slave_dev = NULL;
3554         struct ifbond k_binfo;
3555         struct ifbond __user *u_binfo = NULL;
3556         struct ifslave k_sinfo;
3557         struct ifslave __user *u_sinfo = NULL;
3558         struct mii_ioctl_data *mii = NULL;
3559         struct net *net;
3560         int res = 0;
3561
3562         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3563
3564         switch (cmd) {
3565         case SIOCGMIIPHY:
3566                 mii = if_mii(ifr);
3567                 if (!mii)
3568                         return -EINVAL;
3569
3570                 mii->phy_id = 0;
3571                 /* Fall Through */
3572         case SIOCGMIIREG:
3573                 /*
3574                  * We do this again just in case we were called by SIOCGMIIREG
3575                  * instead of SIOCGMIIPHY.
3576                  */
3577                 mii = if_mii(ifr);
3578                 if (!mii)
3579                         return -EINVAL;
3580
3581
3582                 if (mii->reg_num == 1) {
3583                         struct bonding *bond = netdev_priv(bond_dev);
3584                         mii->val_out = 0;
3585                         read_lock(&bond->lock);
3586                         read_lock(&bond->curr_slave_lock);
3587                         if (netif_carrier_ok(bond->dev))
3588                                 mii->val_out = BMSR_LSTATUS;
3589
3590                         read_unlock(&bond->curr_slave_lock);
3591                         read_unlock(&bond->lock);
3592                 }
3593
3594                 return 0;
3595         case BOND_INFO_QUERY_OLD:
3596         case SIOCBONDINFOQUERY:
3597                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3598
3599                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3600                         return -EFAULT;
3601
3602                 res = bond_info_query(bond_dev, &k_binfo);
3603                 if (res == 0 &&
3604                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3605                         return -EFAULT;
3606
3607                 return res;
3608         case BOND_SLAVE_INFO_QUERY_OLD:
3609         case SIOCBONDSLAVEINFOQUERY:
3610                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3611
3612                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3613                         return -EFAULT;
3614
3615                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3616                 if (res == 0 &&
3617                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3618                         return -EFAULT;
3619
3620                 return res;
3621         default:
3622                 /* Go on */
3623                 break;
3624         }
3625
3626         net = dev_net(bond_dev);
3627
3628         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3629                 return -EPERM;
3630
3631         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3632
3633         pr_debug("slave_dev=%p:\n", slave_dev);
3634
3635         if (!slave_dev)
3636                 res = -ENODEV;
3637         else {
3638                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3639                 switch (cmd) {
3640                 case BOND_ENSLAVE_OLD:
3641                 case SIOCBONDENSLAVE:
3642                         res = bond_enslave(bond_dev, slave_dev);
3643                         break;
3644                 case BOND_RELEASE_OLD:
3645                 case SIOCBONDRELEASE:
3646                         res = bond_release(bond_dev, slave_dev);
3647                         break;
3648                 case BOND_SETHWADDR_OLD:
3649                 case SIOCBONDSETHWADDR:
3650                         bond_set_dev_addr(bond_dev, slave_dev);
3651                         res = 0;
3652                         break;
3653                 case BOND_CHANGE_ACTIVE_OLD:
3654                 case SIOCBONDCHANGEACTIVE:
3655                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3656                         break;
3657                 default:
3658                         res = -EOPNOTSUPP;
3659                 }
3660
3661                 dev_put(slave_dev);
3662         }
3663
3664         return res;
3665 }
3666
3667 static bool bond_addr_in_mc_list(unsigned char *addr,
3668                                  struct netdev_hw_addr_list *list,
3669                                  int addrlen)
3670 {
3671         struct netdev_hw_addr *ha;
3672
3673         netdev_hw_addr_list_for_each(ha, list)
3674                 if (!memcmp(ha->addr, addr, addrlen))
3675                         return true;
3676
3677         return false;
3678 }
3679
3680 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3681 {
3682         struct bonding *bond = netdev_priv(bond_dev);
3683
3684         if (change & IFF_PROMISC)
3685                 bond_set_promiscuity(bond,
3686                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3687
3688         if (change & IFF_ALLMULTI)
3689                 bond_set_allmulti(bond,
3690                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3691 }
3692
3693 static void bond_set_multicast_list(struct net_device *bond_dev)
3694 {
3695         struct bonding *bond = netdev_priv(bond_dev);
3696         struct netdev_hw_addr *ha;
3697         bool found;
3698
3699         read_lock(&bond->lock);
3700
3701         /* looking for addresses to add to slaves' mc list */
3702         netdev_for_each_mc_addr(ha, bond_dev) {
3703                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3704                                              bond_dev->addr_len);
3705                 if (!found)
3706                         bond_mc_add(bond, ha->addr);
3707         }
3708
3709         /* looking for addresses to delete from slaves' list */
3710         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3711                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3712                                              bond_dev->addr_len);
3713                 if (!found)
3714                         bond_mc_del(bond, ha->addr);
3715         }
3716
3717         /* save master's multicast list */
3718         __hw_addr_flush(&bond->mc_list);
3719         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3720                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3721
3722         read_unlock(&bond->lock);
3723 }
3724
3725 static int bond_neigh_init(struct neighbour *n)
3726 {
3727         struct bonding *bond = netdev_priv(n->dev);
3728         struct slave *slave = bond->first_slave;
3729         const struct net_device_ops *slave_ops;
3730         struct neigh_parms parms;
3731         int ret;
3732
3733         if (!slave)
3734                 return 0;
3735
3736         slave_ops = slave->dev->netdev_ops;
3737
3738         if (!slave_ops->ndo_neigh_setup)
3739                 return 0;
3740
3741         parms.neigh_setup = NULL;
3742         parms.neigh_cleanup = NULL;
3743         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3744         if (ret)
3745                 return ret;
3746
3747         /*
3748          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3749          * after the last slave has been detached.  Assumes that all slaves
3750          * utilize the same neigh_cleanup (true at this writing as only user
3751          * is ipoib).
3752          */
3753         n->parms->neigh_cleanup = parms.neigh_cleanup;
3754
3755         if (!parms.neigh_setup)
3756                 return 0;
3757
3758         return parms.neigh_setup(n);
3759 }
3760
3761 /*
3762  * The bonding ndo_neigh_setup is called at init time beofre any
3763  * slave exists. So we must declare proxy setup function which will
3764  * be used at run time to resolve the actual slave neigh param setup.
3765  */
3766 static int bond_neigh_setup(struct net_device *dev,
3767                             struct neigh_parms *parms)
3768 {
3769         parms->neigh_setup   = bond_neigh_init;
3770
3771         return 0;
3772 }
3773
3774 /*
3775  * Change the MTU of all of a master's slaves to match the master
3776  */
3777 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3778 {
3779         struct bonding *bond = netdev_priv(bond_dev);
3780         struct slave *slave, *stop_at;
3781         int res = 0;
3782         int i;
3783
3784         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3785                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3786
3787         /* Can't hold bond->lock with bh disabled here since
3788          * some base drivers panic. On the other hand we can't
3789          * hold bond->lock without bh disabled because we'll
3790          * deadlock. The only solution is to rely on the fact
3791          * that we're under rtnl_lock here, and the slaves
3792          * list won't change. This doesn't solve the problem
3793          * of setting the slave's MTU while it is
3794          * transmitting, but the assumption is that the base
3795          * driver can handle that.
3796          *
3797          * TODO: figure out a way to safely iterate the slaves
3798          * list, but without holding a lock around the actual
3799          * call to the base driver.
3800          */
3801
3802         bond_for_each_slave(bond, slave, i) {
3803                 pr_debug("s %p s->p %p c_m %p\n",
3804                          slave,
3805                          slave->prev,
3806                          slave->dev->netdev_ops->ndo_change_mtu);
3807
3808                 res = dev_set_mtu(slave->dev, new_mtu);
3809
3810                 if (res) {
3811                         /* If we failed to set the slave's mtu to the new value
3812                          * we must abort the operation even in ACTIVE_BACKUP
3813                          * mode, because if we allow the backup slaves to have
3814                          * different mtu values than the active slave we'll
3815                          * need to change their mtu when doing a failover. That
3816                          * means changing their mtu from timer context, which
3817                          * is probably not a good idea.
3818                          */
3819                         pr_debug("err %d %s\n", res, slave->dev->name);
3820                         goto unwind;
3821                 }
3822         }
3823
3824         bond_dev->mtu = new_mtu;
3825
3826         return 0;
3827
3828 unwind:
3829         /* unwind from head to the slave that failed */
3830         stop_at = slave;
3831         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3832                 int tmp_res;
3833
3834                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3835                 if (tmp_res) {
3836                         pr_debug("unwind err %d dev %s\n",
3837                                  tmp_res, slave->dev->name);
3838                 }
3839         }
3840
3841         return res;
3842 }
3843
3844 /*
3845  * Change HW address
3846  *
3847  * Note that many devices must be down to change the HW address, and
3848  * downing the master releases all slaves.  We can make bonds full of
3849  * bonding devices to test this, however.
3850  */
3851 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3852 {
3853         struct bonding *bond = netdev_priv(bond_dev);
3854         struct sockaddr *sa = addr, tmp_sa;
3855         struct slave *slave, *stop_at;
3856         int res = 0;
3857         int i;
3858
3859         if (bond->params.mode == BOND_MODE_ALB)
3860                 return bond_alb_set_mac_address(bond_dev, addr);
3861
3862
3863         pr_debug("bond=%p, name=%s\n",
3864                  bond, bond_dev ? bond_dev->name : "None");
3865
3866         /*
3867          * If fail_over_mac is set to active, do nothing and return
3868          * success.  Returning an error causes ifenslave to fail.
3869          */
3870         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3871                 return 0;
3872
3873         if (!is_valid_ether_addr(sa->sa_data))
3874                 return -EADDRNOTAVAIL;
3875
3876         /* Can't hold bond->lock with bh disabled here since
3877          * some base drivers panic. On the other hand we can't
3878          * hold bond->lock without bh disabled because we'll
3879          * deadlock. The only solution is to rely on the fact
3880          * that we're under rtnl_lock here, and the slaves
3881          * list won't change. This doesn't solve the problem
3882          * of setting the slave's hw address while it is
3883          * transmitting, but the assumption is that the base
3884          * driver can handle that.
3885          *
3886          * TODO: figure out a way to safely iterate the slaves
3887          * list, but without holding a lock around the actual
3888          * call to the base driver.
3889          */
3890
3891         bond_for_each_slave(bond, slave, i) {
3892                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3893                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3894
3895                 if (slave_ops->ndo_set_mac_address == NULL) {
3896                         res = -EOPNOTSUPP;
3897                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3898                         goto unwind;
3899                 }
3900
3901                 res = dev_set_mac_address(slave->dev, addr);
3902                 if (res) {
3903                         /* TODO: consider downing the slave
3904                          * and retry ?
3905                          * User should expect communications
3906                          * breakage anyway until ARP finish
3907                          * updating, so...
3908                          */
3909                         pr_debug("err %d %s\n", res, slave->dev->name);
3910                         goto unwind;
3911                 }
3912         }
3913
3914         /* success */
3915         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3916         return 0;
3917
3918 unwind:
3919         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3920         tmp_sa.sa_family = bond_dev->type;
3921
3922         /* unwind from head to the slave that failed */
3923         stop_at = slave;
3924         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3925                 int tmp_res;
3926
3927                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3928                 if (tmp_res) {
3929                         pr_debug("unwind err %d dev %s\n",
3930                                  tmp_res, slave->dev->name);
3931                 }
3932         }
3933
3934         return res;
3935 }
3936
3937 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3938 {
3939         struct bonding *bond = netdev_priv(bond_dev);
3940         struct slave *slave, *start_at;
3941         int i, slave_no, res = 1;
3942         struct iphdr *iph = ip_hdr(skb);
3943
3944         /*
3945          * Start with the curr_active_slave that joined the bond as the
3946          * default for sending IGMP traffic.  For failover purposes one
3947          * needs to maintain some consistency for the interface that will
3948          * send the join/membership reports.  The curr_active_slave found
3949          * will send all of this type of traffic.
3950          */
3951         if ((iph->protocol == IPPROTO_IGMP) &&
3952             (skb->protocol == htons(ETH_P_IP))) {
3953
3954                 read_lock(&bond->curr_slave_lock);
3955                 slave = bond->curr_active_slave;
3956                 read_unlock(&bond->curr_slave_lock);
3957
3958                 if (!slave)
3959                         goto out;
3960         } else {
3961                 /*
3962                  * Concurrent TX may collide on rr_tx_counter; we accept
3963                  * that as being rare enough not to justify using an
3964                  * atomic op here.
3965                  */
3966                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3967
3968                 bond_for_each_slave(bond, slave, i) {
3969                         slave_no--;
3970                         if (slave_no < 0)
3971                                 break;
3972                 }
3973         }
3974
3975         start_at = slave;
3976         bond_for_each_slave_from(bond, slave, i, start_at) {
3977                 if (IS_UP(slave->dev) &&
3978                     (slave->link == BOND_LINK_UP) &&
3979                     bond_is_active_slave(slave)) {
3980                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3981                         break;
3982                 }
3983         }
3984
3985 out:
3986         if (res) {
3987                 /* no suitable interface, frame not sent */
3988                 kfree_skb(skb);
3989         }
3990
3991         return NETDEV_TX_OK;
3992 }
3993
3994
3995 /*
3996  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3997  * the bond has a usable interface.
3998  */
3999 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4000 {
4001         struct bonding *bond = netdev_priv(bond_dev);
4002         int res = 1;
4003
4004         read_lock(&bond->curr_slave_lock);
4005
4006         if (bond->curr_active_slave)
4007                 res = bond_dev_queue_xmit(bond, skb,
4008                         bond->curr_active_slave->dev);
4009
4010         read_unlock(&bond->curr_slave_lock);
4011
4012         if (res)
4013                 /* no suitable interface, frame not sent */
4014                 kfree_skb(skb);
4015
4016         return NETDEV_TX_OK;
4017 }
4018
4019 /*
4020  * In bond_xmit_xor() , we determine the output device by using a pre-
4021  * determined xmit_hash_policy(), If the selected device is not enabled,
4022  * find the next active slave.
4023  */
4024 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4025 {
4026         struct bonding *bond = netdev_priv(bond_dev);
4027         struct slave *slave, *start_at;
4028         int slave_no;
4029         int i;
4030         int res = 1;
4031
4032         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4033
4034         bond_for_each_slave(bond, slave, i) {
4035                 slave_no--;
4036                 if (slave_no < 0)
4037                         break;
4038         }
4039
4040         start_at = slave;
4041
4042         bond_for_each_slave_from(bond, slave, i, start_at) {
4043                 if (IS_UP(slave->dev) &&
4044                     (slave->link == BOND_LINK_UP) &&
4045                     bond_is_active_slave(slave)) {
4046                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4047                         break;
4048                 }
4049         }
4050
4051         if (res) {
4052                 /* no suitable interface, frame not sent */
4053                 kfree_skb(skb);
4054         }
4055
4056         return NETDEV_TX_OK;
4057 }
4058
4059 /*
4060  * in broadcast mode, we send everything to all usable interfaces.
4061  */
4062 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4063 {
4064         struct bonding *bond = netdev_priv(bond_dev);
4065         struct slave *slave, *start_at;
4066         struct net_device *tx_dev = NULL;
4067         int i;
4068         int res = 1;
4069
4070         read_lock(&bond->curr_slave_lock);
4071         start_at = bond->curr_active_slave;
4072         read_unlock(&bond->curr_slave_lock);
4073
4074         if (!start_at)
4075                 goto out;
4076
4077         bond_for_each_slave_from(bond, slave, i, start_at) {
4078                 if (IS_UP(slave->dev) &&
4079                     (slave->link == BOND_LINK_UP) &&
4080                     bond_is_active_slave(slave)) {
4081                         if (tx_dev) {
4082                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4083                                 if (!skb2) {
4084                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4085                                                bond_dev->name);
4086                                         continue;
4087                                 }
4088
4089                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4090                                 if (res) {
4091                                         kfree_skb(skb2);
4092                                         continue;
4093                                 }
4094                         }
4095                         tx_dev = slave->dev;
4096                 }
4097         }
4098
4099         if (tx_dev)
4100                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4101
4102 out:
4103         if (res)
4104                 /* no suitable interface, frame not sent */
4105                 kfree_skb(skb);
4106
4107         /* frame sent to all suitable interfaces */
4108         return NETDEV_TX_OK;
4109 }
4110
4111 /*------------------------- Device initialization ---------------------------*/
4112
4113 static void bond_set_xmit_hash_policy(struct bonding *bond)
4114 {
4115         switch (bond->params.xmit_policy) {
4116         case BOND_XMIT_POLICY_LAYER23:
4117                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4118                 break;
4119         case BOND_XMIT_POLICY_LAYER34:
4120                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4121                 break;
4122         case BOND_XMIT_POLICY_LAYER2:
4123         default:
4124                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4125                 break;
4126         }
4127 }
4128
4129 /*
4130  * Lookup the slave that corresponds to a qid
4131  */
4132 static inline int bond_slave_override(struct bonding *bond,
4133                                       struct sk_buff *skb)
4134 {
4135         int i, res = 1;
4136         struct slave *slave = NULL;
4137         struct slave *check_slave;
4138
4139         if (!skb->queue_mapping)
4140                 return 1;
4141
4142         /* Find out if any slaves have the same mapping as this skb. */
4143         bond_for_each_slave(bond, check_slave, i) {
4144                 if (check_slave->queue_id == skb->queue_mapping) {
4145                         slave = check_slave;
4146                         break;
4147                 }
4148         }
4149
4150         /* If the slave isn't UP, use default transmit policy. */
4151         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4152             (slave->link == BOND_LINK_UP)) {
4153                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4154         }
4155
4156         return res;
4157 }
4158
4159
4160 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4161 {
4162         /*
4163          * This helper function exists to help dev_pick_tx get the correct
4164          * destination queue.  Using a helper function skips a call to
4165          * skb_tx_hash and will put the skbs in the queue we expect on their
4166          * way down to the bonding driver.
4167          */
4168         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4169
4170         /*
4171          * Save the original txq to restore before passing to the driver
4172          */
4173         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4174
4175         if (unlikely(txq >= dev->real_num_tx_queues)) {
4176                 do {
4177                         txq -= dev->real_num_tx_queues;
4178                 } while (txq >= dev->real_num_tx_queues);
4179         }
4180         return txq;
4181 }
4182
4183 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4184 {
4185         struct bonding *bond = netdev_priv(dev);
4186
4187         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4188                 if (!bond_slave_override(bond, skb))
4189                         return NETDEV_TX_OK;
4190         }
4191
4192         switch (bond->params.mode) {
4193         case BOND_MODE_ROUNDROBIN:
4194                 return bond_xmit_roundrobin(skb, dev);
4195         case BOND_MODE_ACTIVEBACKUP:
4196                 return bond_xmit_activebackup(skb, dev);
4197         case BOND_MODE_XOR:
4198                 return bond_xmit_xor(skb, dev);
4199         case BOND_MODE_BROADCAST:
4200                 return bond_xmit_broadcast(skb, dev);
4201         case BOND_MODE_8023AD:
4202                 return bond_3ad_xmit_xor(skb, dev);
4203         case BOND_MODE_ALB:
4204         case BOND_MODE_TLB:
4205                 return bond_alb_xmit(skb, dev);
4206         default:
4207                 /* Should never happen, mode already checked */
4208                 pr_err("%s: Error: Unknown bonding mode %d\n",
4209                        dev->name, bond->params.mode);
4210                 WARN_ON_ONCE(1);
4211                 kfree_skb(skb);
4212                 return NETDEV_TX_OK;
4213         }
4214 }
4215
4216 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4217 {
4218         struct bonding *bond = netdev_priv(dev);
4219         netdev_tx_t ret = NETDEV_TX_OK;
4220
4221         /*
4222          * If we risk deadlock from transmitting this in the
4223          * netpoll path, tell netpoll to queue the frame for later tx
4224          */
4225         if (is_netpoll_tx_blocked(dev))
4226                 return NETDEV_TX_BUSY;
4227
4228         read_lock(&bond->lock);
4229
4230         if (bond->slave_cnt)
4231                 ret = __bond_start_xmit(skb, dev);
4232         else
4233                 kfree_skb(skb);
4234
4235         read_unlock(&bond->lock);
4236
4237         return ret;
4238 }
4239
4240 /*
4241  * set bond mode specific net device operations
4242  */
4243 void bond_set_mode_ops(struct bonding *bond, int mode)
4244 {
4245         struct net_device *bond_dev = bond->dev;
4246
4247         switch (mode) {
4248         case BOND_MODE_ROUNDROBIN:
4249                 break;
4250         case BOND_MODE_ACTIVEBACKUP:
4251                 break;
4252         case BOND_MODE_XOR:
4253                 bond_set_xmit_hash_policy(bond);
4254                 break;
4255         case BOND_MODE_BROADCAST:
4256                 break;
4257         case BOND_MODE_8023AD:
4258                 bond_set_xmit_hash_policy(bond);
4259                 break;
4260         case BOND_MODE_ALB:
4261                 /* FALLTHRU */
4262         case BOND_MODE_TLB:
4263                 break;
4264         default:
4265                 /* Should never happen, mode already checked */
4266                 pr_err("%s: Error: Unknown bonding mode %d\n",
4267                        bond_dev->name, mode);
4268                 break;
4269         }
4270 }
4271
4272 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4273                                      struct ethtool_cmd *ecmd)
4274 {
4275         struct bonding *bond = netdev_priv(bond_dev);
4276         struct slave *slave;
4277         int i;
4278         unsigned long speed = 0;
4279
4280         ecmd->duplex = DUPLEX_UNKNOWN;
4281         ecmd->port = PORT_OTHER;
4282
4283         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4284          * do not need to check mode.  Though link speed might not represent
4285          * the true receive or transmit bandwidth (not all modes are symmetric)
4286          * this is an accurate maximum.
4287          */
4288         read_lock(&bond->lock);
4289         bond_for_each_slave(bond, slave, i) {
4290                 if (SLAVE_IS_OK(slave)) {
4291                         if (slave->speed != SPEED_UNKNOWN)
4292                                 speed += slave->speed;
4293                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4294                             slave->duplex != DUPLEX_UNKNOWN)
4295                                 ecmd->duplex = slave->duplex;
4296                 }
4297         }
4298         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4299         read_unlock(&bond->lock);
4300         return 0;
4301 }
4302
4303 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4304                                      struct ethtool_drvinfo *drvinfo)
4305 {
4306         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4307         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4308         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4309                  BOND_ABI_VERSION);
4310 }
4311
4312 static const struct ethtool_ops bond_ethtool_ops = {
4313         .get_drvinfo            = bond_ethtool_get_drvinfo,
4314         .get_settings           = bond_ethtool_get_settings,
4315         .get_link               = ethtool_op_get_link,
4316 };
4317
4318 static const struct net_device_ops bond_netdev_ops = {
4319         .ndo_init               = bond_init,
4320         .ndo_uninit             = bond_uninit,
4321         .ndo_open               = bond_open,
4322         .ndo_stop               = bond_close,
4323         .ndo_start_xmit         = bond_start_xmit,
4324         .ndo_select_queue       = bond_select_queue,
4325         .ndo_get_stats64        = bond_get_stats,
4326         .ndo_do_ioctl           = bond_do_ioctl,
4327         .ndo_change_rx_flags    = bond_change_rx_flags,
4328         .ndo_set_rx_mode        = bond_set_multicast_list,
4329         .ndo_change_mtu         = bond_change_mtu,
4330         .ndo_set_mac_address    = bond_set_mac_address,
4331         .ndo_neigh_setup        = bond_neigh_setup,
4332         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4333         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4334 #ifdef CONFIG_NET_POLL_CONTROLLER
4335         .ndo_netpoll_setup      = bond_netpoll_setup,
4336         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4337         .ndo_poll_controller    = bond_poll_controller,
4338 #endif
4339         .ndo_add_slave          = bond_enslave,
4340         .ndo_del_slave          = bond_release,
4341         .ndo_fix_features       = bond_fix_features,
4342 };
4343
4344 static const struct device_type bond_type = {
4345         .name = "bond",
4346 };
4347
4348 static void bond_destructor(struct net_device *bond_dev)
4349 {
4350         struct bonding *bond = netdev_priv(bond_dev);
4351         if (bond->wq)
4352                 destroy_workqueue(bond->wq);
4353         free_netdev(bond_dev);
4354 }
4355
4356 static void bond_setup(struct net_device *bond_dev)
4357 {
4358         struct bonding *bond = netdev_priv(bond_dev);
4359
4360         /* initialize rwlocks */
4361         rwlock_init(&bond->lock);
4362         rwlock_init(&bond->curr_slave_lock);
4363
4364         bond->params = bonding_defaults;
4365
4366         /* Initialize pointers */
4367         bond->dev = bond_dev;
4368         INIT_LIST_HEAD(&bond->vlan_list);
4369
4370         /* Initialize the device entry points */
4371         ether_setup(bond_dev);
4372         bond_dev->netdev_ops = &bond_netdev_ops;
4373         bond_dev->ethtool_ops = &bond_ethtool_ops;
4374         bond_set_mode_ops(bond, bond->params.mode);
4375
4376         bond_dev->destructor = bond_destructor;
4377
4378         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4379
4380         /* Initialize the device options */
4381         bond_dev->tx_queue_len = 0;
4382         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4383         bond_dev->priv_flags |= IFF_BONDING;
4384         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4385
4386         /* At first, we block adding VLANs. That's the only way to
4387          * prevent problems that occur when adding VLANs over an
4388          * empty bond. The block will be removed once non-challenged
4389          * slaves are enslaved.
4390          */
4391         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4392
4393         /* don't acquire bond device's netif_tx_lock when
4394          * transmitting */
4395         bond_dev->features |= NETIF_F_LLTX;
4396
4397         /* By default, we declare the bond to be fully
4398          * VLAN hardware accelerated capable. Special
4399          * care is taken in the various xmit functions
4400          * when there are slaves that are not hw accel
4401          * capable
4402          */
4403
4404         bond_dev->hw_features = BOND_VLAN_FEATURES |
4405                                 NETIF_F_HW_VLAN_CTAG_TX |
4406                                 NETIF_F_HW_VLAN_CTAG_RX |
4407                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4408
4409         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4410         bond_dev->features |= bond_dev->hw_features;
4411 }
4412
4413 /*
4414 * Destroy a bonding device.
4415 * Must be under rtnl_lock when this function is called.
4416 */
4417 static void bond_uninit(struct net_device *bond_dev)
4418 {
4419         struct bonding *bond = netdev_priv(bond_dev);
4420         struct vlan_entry *vlan, *tmp;
4421
4422         bond_netpoll_cleanup(bond_dev);
4423
4424         /* Release the bonded slaves */
4425         while (bond->first_slave != NULL)
4426                 __bond_release_one(bond_dev, bond->first_slave->dev, true);
4427         pr_info("%s: released all slaves\n", bond_dev->name);
4428
4429         list_del(&bond->bond_list);
4430
4431         bond_debug_unregister(bond);
4432
4433         __hw_addr_flush(&bond->mc_list);
4434
4435         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4436                 list_del(&vlan->vlan_list);
4437                 kfree(vlan);
4438         }
4439 }
4440
4441 /*------------------------- Module initialization ---------------------------*/
4442
4443 /*
4444  * Convert string input module parms.  Accept either the
4445  * number of the mode or its string name.  A bit complicated because
4446  * some mode names are substrings of other names, and calls from sysfs
4447  * may have whitespace in the name (trailing newlines, for example).
4448  */
4449 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4450 {
4451         int modeint = -1, i, rv;
4452         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4453
4454         for (p = (char *)buf; *p; p++)
4455                 if (!(isdigit(*p) || isspace(*p)))
4456                         break;
4457
4458         if (*p)
4459                 rv = sscanf(buf, "%20s", modestr);
4460         else
4461                 rv = sscanf(buf, "%d", &modeint);
4462
4463         if (!rv)
4464                 return -1;
4465
4466         for (i = 0; tbl[i].modename; i++) {
4467                 if (modeint == tbl[i].mode)
4468                         return tbl[i].mode;
4469                 if (strcmp(modestr, tbl[i].modename) == 0)
4470                         return tbl[i].mode;
4471         }
4472
4473         return -1;
4474 }
4475
4476 static int bond_check_params(struct bond_params *params)
4477 {
4478         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4479
4480         /*
4481          * Convert string parameters.
4482          */
4483         if (mode) {
4484                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4485                 if (bond_mode == -1) {
4486                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4487                                mode == NULL ? "NULL" : mode);
4488                         return -EINVAL;
4489                 }
4490         }
4491
4492         if (xmit_hash_policy) {
4493                 if ((bond_mode != BOND_MODE_XOR) &&
4494                     (bond_mode != BOND_MODE_8023AD)) {
4495                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4496                                bond_mode_name(bond_mode));
4497                 } else {
4498                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4499                                                         xmit_hashtype_tbl);
4500                         if (xmit_hashtype == -1) {
4501                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4502                                        xmit_hash_policy == NULL ? "NULL" :
4503                                        xmit_hash_policy);
4504                                 return -EINVAL;
4505                         }
4506                 }
4507         }
4508
4509         if (lacp_rate) {
4510                 if (bond_mode != BOND_MODE_8023AD) {
4511                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4512                                 bond_mode_name(bond_mode));
4513                 } else {
4514                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4515                         if (lacp_fast == -1) {
4516                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4517                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4518                                 return -EINVAL;
4519                         }
4520                 }
4521         }
4522
4523         if (ad_select) {
4524                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4525                 if (params->ad_select == -1) {
4526                         pr_err("Error: Invalid ad_select \"%s\"\n",
4527                                ad_select == NULL ? "NULL" : ad_select);
4528                         return -EINVAL;
4529                 }
4530
4531                 if (bond_mode != BOND_MODE_8023AD) {
4532                         pr_warning("ad_select param only affects 802.3ad mode\n");
4533                 }
4534         } else {
4535                 params->ad_select = BOND_AD_STABLE;
4536         }
4537
4538         if (max_bonds < 0) {
4539                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4540                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4541                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4542         }
4543
4544         if (miimon < 0) {
4545                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4546                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4547                 miimon = BOND_LINK_MON_INTERV;
4548         }
4549
4550         if (updelay < 0) {
4551                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4552                            updelay, INT_MAX);
4553                 updelay = 0;
4554         }
4555
4556         if (downdelay < 0) {
4557                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4558                            downdelay, INT_MAX);
4559                 downdelay = 0;
4560         }
4561
4562         if ((use_carrier != 0) && (use_carrier != 1)) {
4563                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4564                            use_carrier);
4565                 use_carrier = 1;
4566         }
4567
4568         if (num_peer_notif < 0 || num_peer_notif > 255) {
4569                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4570                            num_peer_notif);
4571                 num_peer_notif = 1;
4572         }
4573
4574         /* reset values for 802.3ad */
4575         if (bond_mode == BOND_MODE_8023AD) {
4576                 if (!miimon) {
4577                         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");
4578                         pr_warning("Forcing miimon to 100msec\n");
4579                         miimon = 100;
4580                 }
4581         }
4582
4583         if (tx_queues < 1 || tx_queues > 255) {
4584                 pr_warning("Warning: tx_queues (%d) should be between "
4585                            "1 and 255, resetting to %d\n",
4586                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4587                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4588         }
4589
4590         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4591                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4592                            "not of valid value (0/1), so it was set to "
4593                            "0\n", all_slaves_active);
4594                 all_slaves_active = 0;
4595         }
4596
4597         if (resend_igmp < 0 || resend_igmp > 255) {
4598                 pr_warning("Warning: resend_igmp (%d) should be between "
4599                            "0 and 255, resetting to %d\n",
4600                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4601                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4602         }
4603
4604         /* reset values for TLB/ALB */
4605         if ((bond_mode == BOND_MODE_TLB) ||
4606             (bond_mode == BOND_MODE_ALB)) {
4607                 if (!miimon) {
4608                         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");
4609                         pr_warning("Forcing miimon to 100msec\n");
4610                         miimon = 100;
4611                 }
4612         }
4613
4614         if (bond_mode == BOND_MODE_ALB) {
4615                 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",
4616                           updelay);
4617         }
4618
4619         if (!miimon) {
4620                 if (updelay || downdelay) {
4621                         /* just warn the user the up/down delay will have
4622                          * no effect since miimon is zero...
4623                          */
4624                         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",
4625                                    updelay, downdelay);
4626                 }
4627         } else {
4628                 /* don't allow arp monitoring */
4629                 if (arp_interval) {
4630                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4631                                    miimon, arp_interval);
4632                         arp_interval = 0;
4633                 }
4634
4635                 if ((updelay % miimon) != 0) {
4636                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4637                                    updelay, miimon,
4638                                    (updelay / miimon) * miimon);
4639                 }
4640
4641                 updelay /= miimon;
4642
4643                 if ((downdelay % miimon) != 0) {
4644                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4645                                    downdelay, miimon,
4646                                    (downdelay / miimon) * miimon);
4647                 }
4648
4649                 downdelay /= miimon;
4650         }
4651
4652         if (arp_interval < 0) {
4653                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4654                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4655                 arp_interval = BOND_LINK_ARP_INTERV;
4656         }
4657
4658         for (arp_ip_count = 0, i = 0;
4659              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4660                 /* not complete check, but should be good enough to
4661                    catch mistakes */
4662                 __be32 ip = in_aton(arp_ip_target[i]);
4663                 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4664                     ip == htonl(INADDR_BROADCAST)) {
4665                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4666                                    arp_ip_target[i]);
4667                         arp_interval = 0;
4668                 } else {
4669                         arp_target[arp_ip_count++] = ip;
4670                 }
4671         }
4672
4673         if (arp_interval && !arp_ip_count) {
4674                 /* don't allow arping if no arp_ip_target given... */
4675                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4676                            arp_interval);
4677                 arp_interval = 0;
4678         }
4679
4680         if (arp_validate) {
4681                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4682                         pr_err("arp_validate only supported in active-backup mode\n");
4683                         return -EINVAL;
4684                 }
4685                 if (!arp_interval) {
4686                         pr_err("arp_validate requires arp_interval\n");
4687                         return -EINVAL;
4688                 }
4689
4690                 arp_validate_value = bond_parse_parm(arp_validate,
4691                                                      arp_validate_tbl);
4692                 if (arp_validate_value == -1) {
4693                         pr_err("Error: invalid arp_validate \"%s\"\n",
4694                                arp_validate == NULL ? "NULL" : arp_validate);
4695                         return -EINVAL;
4696                 }
4697         } else
4698                 arp_validate_value = 0;
4699
4700         if (miimon) {
4701                 pr_info("MII link monitoring set to %d ms\n", miimon);
4702         } else if (arp_interval) {
4703                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4704                         arp_interval,
4705                         arp_validate_tbl[arp_validate_value].modename,
4706                         arp_ip_count);
4707
4708                 for (i = 0; i < arp_ip_count; i++)
4709                         pr_info(" %s", arp_ip_target[i]);
4710
4711                 pr_info("\n");
4712
4713         } else if (max_bonds) {
4714                 /* miimon and arp_interval not set, we need one so things
4715                  * work as expected, see bonding.txt for details
4716                  */
4717                 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");
4718         }
4719
4720         if (primary && !USES_PRIMARY(bond_mode)) {
4721                 /* currently, using a primary only makes sense
4722                  * in active backup, TLB or ALB modes
4723                  */
4724                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4725                            primary, bond_mode_name(bond_mode));
4726                 primary = NULL;
4727         }
4728
4729         if (primary && primary_reselect) {
4730                 primary_reselect_value = bond_parse_parm(primary_reselect,
4731                                                          pri_reselect_tbl);
4732                 if (primary_reselect_value == -1) {
4733                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4734                                primary_reselect ==
4735                                         NULL ? "NULL" : primary_reselect);
4736                         return -EINVAL;
4737                 }
4738         } else {
4739                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4740         }
4741
4742         if (fail_over_mac) {
4743                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4744                                                       fail_over_mac_tbl);
4745                 if (fail_over_mac_value == -1) {
4746                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4747                                arp_validate == NULL ? "NULL" : arp_validate);
4748                         return -EINVAL;
4749                 }
4750
4751                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4752                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4753         } else {
4754                 fail_over_mac_value = BOND_FOM_NONE;
4755         }
4756
4757         /* fill params struct with the proper values */
4758         params->mode = bond_mode;
4759         params->xmit_policy = xmit_hashtype;
4760         params->miimon = miimon;
4761         params->num_peer_notif = num_peer_notif;
4762         params->arp_interval = arp_interval;
4763         params->arp_validate = arp_validate_value;
4764         params->updelay = updelay;
4765         params->downdelay = downdelay;
4766         params->use_carrier = use_carrier;
4767         params->lacp_fast = lacp_fast;
4768         params->primary[0] = 0;
4769         params->primary_reselect = primary_reselect_value;
4770         params->fail_over_mac = fail_over_mac_value;
4771         params->tx_queues = tx_queues;
4772         params->all_slaves_active = all_slaves_active;
4773         params->resend_igmp = resend_igmp;
4774         params->min_links = min_links;
4775
4776         if (primary) {
4777                 strncpy(params->primary, primary, IFNAMSIZ);
4778                 params->primary[IFNAMSIZ - 1] = 0;
4779         }
4780
4781         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4782
4783         return 0;
4784 }
4785
4786 static struct lock_class_key bonding_netdev_xmit_lock_key;
4787 static struct lock_class_key bonding_netdev_addr_lock_key;
4788 static struct lock_class_key bonding_tx_busylock_key;
4789
4790 static void bond_set_lockdep_class_one(struct net_device *dev,
4791                                        struct netdev_queue *txq,
4792                                        void *_unused)
4793 {
4794         lockdep_set_class(&txq->_xmit_lock,
4795                           &bonding_netdev_xmit_lock_key);
4796 }
4797
4798 static void bond_set_lockdep_class(struct net_device *dev)
4799 {
4800         lockdep_set_class(&dev->addr_list_lock,
4801                           &bonding_netdev_addr_lock_key);
4802         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4803         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4804 }
4805
4806 /*
4807  * Called from registration process
4808  */
4809 static int bond_init(struct net_device *bond_dev)
4810 {
4811         struct bonding *bond = netdev_priv(bond_dev);
4812         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4813         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4814
4815         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4816
4817         /*
4818          * Initialize locks that may be required during
4819          * en/deslave operations.  All of the bond_open work
4820          * (of which this is part) should really be moved to
4821          * a phase prior to dev_open
4822          */
4823         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4824         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4825
4826         bond->wq = create_singlethread_workqueue(bond_dev->name);
4827         if (!bond->wq)
4828                 return -ENOMEM;
4829
4830         bond_set_lockdep_class(bond_dev);
4831
4832         list_add_tail(&bond->bond_list, &bn->dev_list);
4833
4834         bond_prepare_sysfs_group(bond);
4835
4836         bond_debug_register(bond);
4837
4838         /* Ensure valid dev_addr */
4839         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4840             bond_dev->addr_assign_type == NET_ADDR_PERM) {
4841                 eth_hw_addr_random(bond_dev);
4842                 bond->dev_addr_from_first = true;
4843         }
4844
4845         __hw_addr_init(&bond->mc_list);
4846         return 0;
4847 }
4848
4849 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4850 {
4851         if (tb[IFLA_ADDRESS]) {
4852                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4853                         return -EINVAL;
4854                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4855                         return -EADDRNOTAVAIL;
4856         }
4857         return 0;
4858 }
4859
4860 static unsigned int bond_get_num_tx_queues(void)
4861 {
4862         return tx_queues;
4863 }
4864
4865 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4866         .kind                   = "bond",
4867         .priv_size              = sizeof(struct bonding),
4868         .setup                  = bond_setup,
4869         .validate               = bond_validate,
4870         .get_num_tx_queues      = bond_get_num_tx_queues,
4871         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4872                                                              as for TX queues */
4873 };
4874
4875 /* Create a new bond based on the specified name and bonding parameters.
4876  * If name is NULL, obtain a suitable "bond%d" name for us.
4877  * Caller must NOT hold rtnl_lock; we need to release it here before we
4878  * set up our sysfs entries.
4879  */
4880 int bond_create(struct net *net, const char *name)
4881 {
4882         struct net_device *bond_dev;
4883         int res;
4884
4885         rtnl_lock();
4886
4887         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4888                                    name ? name : "bond%d",
4889                                    bond_setup, tx_queues);
4890         if (!bond_dev) {
4891                 pr_err("%s: eek! can't alloc netdev!\n", name);
4892                 rtnl_unlock();
4893                 return -ENOMEM;
4894         }
4895
4896         dev_net_set(bond_dev, net);
4897         bond_dev->rtnl_link_ops = &bond_link_ops;
4898
4899         res = register_netdevice(bond_dev);
4900
4901         netif_carrier_off(bond_dev);
4902
4903         rtnl_unlock();
4904         if (res < 0)
4905                 bond_destructor(bond_dev);
4906         return res;
4907 }
4908
4909 static int __net_init bond_net_init(struct net *net)
4910 {
4911         struct bond_net *bn = net_generic(net, bond_net_id);
4912
4913         bn->net = net;
4914         INIT_LIST_HEAD(&bn->dev_list);
4915
4916         bond_create_proc_dir(bn);
4917         bond_create_sysfs(bn);
4918
4919         return 0;
4920 }
4921
4922 static void __net_exit bond_net_exit(struct net *net)
4923 {
4924         struct bond_net *bn = net_generic(net, bond_net_id);
4925         struct bonding *bond, *tmp_bond;
4926         LIST_HEAD(list);
4927
4928         bond_destroy_sysfs(bn);
4929         bond_destroy_proc_dir(bn);
4930
4931         /* Kill off any bonds created after unregistering bond rtnl ops */
4932         rtnl_lock();
4933         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4934                 unregister_netdevice_queue(bond->dev, &list);
4935         unregister_netdevice_many(&list);
4936         rtnl_unlock();
4937 }
4938
4939 static struct pernet_operations bond_net_ops = {
4940         .init = bond_net_init,
4941         .exit = bond_net_exit,
4942         .id   = &bond_net_id,
4943         .size = sizeof(struct bond_net),
4944 };
4945
4946 static int __init bonding_init(void)
4947 {
4948         int i;
4949         int res;
4950
4951         pr_info("%s", bond_version);
4952
4953         res = bond_check_params(&bonding_defaults);
4954         if (res)
4955                 goto out;
4956
4957         res = register_pernet_subsys(&bond_net_ops);
4958         if (res)
4959                 goto out;
4960
4961         res = rtnl_link_register(&bond_link_ops);
4962         if (res)
4963                 goto err_link;
4964
4965         bond_create_debugfs();
4966
4967         for (i = 0; i < max_bonds; i++) {
4968                 res = bond_create(&init_net, NULL);
4969                 if (res)
4970                         goto err;
4971         }
4972
4973         register_netdevice_notifier(&bond_netdev_notifier);
4974 out:
4975         return res;
4976 err:
4977         rtnl_link_unregister(&bond_link_ops);
4978 err_link:
4979         unregister_pernet_subsys(&bond_net_ops);
4980         goto out;
4981
4982 }
4983
4984 static void __exit bonding_exit(void)
4985 {
4986         unregister_netdevice_notifier(&bond_netdev_notifier);
4987
4988         bond_destroy_debugfs();
4989
4990         rtnl_link_unregister(&bond_link_ops);
4991         unregister_pernet_subsys(&bond_net_ops);
4992
4993 #ifdef CONFIG_NET_POLL_CONTROLLER
4994         /*
4995          * Make sure we don't have an imbalance on our netpoll blocking
4996          */
4997         WARN_ON(atomic_read(&netpoll_block_tx));
4998 #endif
4999 }
5000
5001 module_init(bonding_init);
5002 module_exit(bonding_exit);
5003 MODULE_LICENSE("GPL");
5004 MODULE_VERSION(DRV_VERSION);
5005 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5006 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5007 MODULE_ALIAS_RTNL_LINK("bond");