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, uint16_t vid)
 432 {
 433         struct bonding *bond = netdev_priv(bond_dev);
 434         struct slave *slave, *stop_at;
 435         int i, res;
 436
 437         bond_for_each_slave(bond, slave, i) {
 438                 res = vlan_vid_add(slave->dev, vid);
 439                 if (res)
 440                         goto unwind;
 441         }
 442
 443         res = bond_add_vlan(bond, vid);
 444         if (res) {
 445                 pr_err("%s: Error: Failed to add vlan id %d\n",
 446                        bond_dev->name, vid);
 447                 return res;
 448         }
 449
 450         return 0;
 451
 452 unwind:
 453         /* unwind from head to the slave that failed */
 454         stop_at = slave;
 455         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
 456                 vlan_vid_del(slave->dev, vid);
 457
 458         return res;
 459 }
 460
 461 /**
 462  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 463  * @bond_dev: bonding net device that got called
 464  * @vid: vlan id being removed
 465  */
 466 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 467 {
 468         struct bonding *bond = netdev_priv(bond_dev);
 469         struct slave *slave;
 470         int i, res;
 471
 472         bond_for_each_slave(bond, slave, i)
 473                 vlan_vid_del(slave->dev, vid);
 474
 475         res = bond_del_vlan(bond, vid);
 476         if (res) {
 477                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 478                        bond_dev->name, vid);
 479                 return res;
 480         }
 481
 482         return 0;
 483 }
 484
 485 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 486 {
 487         struct vlan_entry *vlan;
 488         int res;
 489
 490         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 491                 res = vlan_vid_add(slave_dev, vlan->vlan_id);
 492                 if (res)
 493                         pr_warning("%s: Failed to add vlan id %d to device %s\n",
 494                                    bond->dev->name, vlan->vlan_id,
 495                                    slave_dev->name);
 496         }
 497 }
 498
 499 static void bond_del_vlans_from_slave(struct bonding *bond,
 500                                       struct net_device *slave_dev)
 501 {
 502         struct vlan_entry *vlan;
 503
 504         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 505                 if (!vlan->vlan_id)
 506                         continue;
 507                 vlan_vid_del(slave_dev, vlan->vlan_id);
 508         }
 509 }
 510
 511 /*------------------------------- Link status -------------------------------*/
 512
 513 /*
 514  * Set the carrier state for the master according to the state of its
 515  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 516  * do special 802.3ad magic.
 517  *
 518  * Returns zero if carrier state does not change, nonzero if it does.
 519  */
 520 static int bond_set_carrier(struct bonding *bond)
 521 {
 522         struct slave *slave;
 523         int i;
 524
 525         if (bond->slave_cnt == 0)
 526                 goto down;
 527
 528         if (bond->params.mode == BOND_MODE_8023AD)
 529                 return bond_3ad_set_carrier(bond);
 530
 531         bond_for_each_slave(bond, slave, i) {
 532                 if (slave->link == BOND_LINK_UP) {
 533                         if (!netif_carrier_ok(bond->dev)) {
 534                                 netif_carrier_on(bond->dev);
 535                                 return 1;
 536                         }
 537                         return 0;
 538                 }
 539         }
 540
 541 down:
 542         if (netif_carrier_ok(bond->dev)) {
 543                 netif_carrier_off(bond->dev);
 544                 return 1;
 545         }
 546         return 0;
 547 }
 548
 549 /*
 550  * Get link speed and duplex from the slave's base driver
 551  * using ethtool. If for some reason the call fails or the
 552  * values are invalid, set speed and duplex to -1,
 553  * and return.
 554  */
 555 static void bond_update_speed_duplex(struct slave *slave)
 556 {
 557         struct net_device *slave_dev = slave->dev;
 558         struct ethtool_cmd ecmd;
 559         u32 slave_speed;
 560         int res;
 561
 562         slave->speed = SPEED_UNKNOWN;
 563         slave->duplex = DUPLEX_UNKNOWN;
 564
 565         res = __ethtool_get_settings(slave_dev, &ecmd);
 566         if (res < 0)
 567                 return;
 568
 569         slave_speed = ethtool_cmd_speed(&ecmd);
 570         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 571                 return;
 572
 573         switch (ecmd.duplex) {
 574         case DUPLEX_FULL:
 575         case DUPLEX_HALF:
 576                 break;
 577         default:
 578                 return;
 579         }
 580
 581         slave->speed = slave_speed;
 582         slave->duplex = ecmd.duplex;
 583
 584         return;
 585 }
 586
 587 /*
 588  * if <dev> supports MII link status reporting, check its link status.
 589  *
 590  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 591  * depending upon the setting of the use_carrier parameter.
 592  *
 593  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 594  * can't tell and just pretend it is), or 0, meaning that the link is
 595  * down.
 596  *
 597  * If reporting is non-zero, instead of faking link up, return -1 if
 598  * both ETHTOOL and MII ioctls fail (meaning the device does not
 599  * support them).  If use_carrier is set, return whatever it says.
 600  * It'd be nice if there was a good way to tell if a driver supports
 601  * netif_carrier, but there really isn't.
 602  */
 603 static int bond_check_dev_link(struct bonding *bond,
 604                                struct net_device *slave_dev, int reporting)
 605 {
 606         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 607         int (*ioctl)(struct net_device *, struct ifreq *, int);
 608         struct ifreq ifr;
 609         struct mii_ioctl_data *mii;
 610
 611         if (!reporting && !netif_running(slave_dev))
 612                 return 0;
 613
 614         if (bond->params.use_carrier)
 615                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 616
 617         /* Try to get link status using Ethtool first. */
 618         if (slave_dev->ethtool_ops->get_link)
 619                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 620                         BMSR_LSTATUS : 0;
 621
 622         /* Ethtool can't be used, fallback to MII ioctls. */
 623         ioctl = slave_ops->ndo_do_ioctl;
 624         if (ioctl) {
 625                 /* TODO: set pointer to correct ioctl on a per team member */
 626                 /*       bases to make this more efficient. that is, once  */
 627                 /*       we determine the correct ioctl, we will always    */
 628                 /*       call it and not the others for that team          */
 629                 /*       member.                                           */
 630
 631                 /*
 632                  * We cannot assume that SIOCGMIIPHY will also read a
 633                  * register; not all network drivers (e.g., e100)
 634                  * support that.
 635                  */
 636
 637                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 638                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 639                 mii = if_mii(&ifr);
 640                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 641                         mii->reg_num = MII_BMSR;
 642                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 643                                 return mii->val_out & BMSR_LSTATUS;
 644                 }
 645         }
 646
 647         /*
 648          * If reporting, report that either there's no dev->do_ioctl,
 649          * or both SIOCGMIIREG and get_link failed (meaning that we
 650          * cannot report link status).  If not reporting, pretend
 651          * we're ok.
 652          */
 653         return reporting ? -1 : BMSR_LSTATUS;
 654 }
 655
 656 /*----------------------------- Multicast list ------------------------------*/
 657
 658 /*
 659  * Push the promiscuity flag down to appropriate slaves
 660  */
 661 static int bond_set_promiscuity(struct bonding *bond, int inc)
 662 {
 663         int err = 0;
 664         if (USES_PRIMARY(bond->params.mode)) {
 665                 /* write lock already acquired */
 666                 if (bond->curr_active_slave) {
 667                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 668                                                   inc);
 669                 }
 670         } else {
 671                 struct slave *slave;
 672                 int i;
 673                 bond_for_each_slave(bond, slave, i) {
 674                         err = dev_set_promiscuity(slave->dev, inc);
 675                         if (err)
 676                                 return err;
 677                 }
 678         }
 679         return err;
 680 }
 681
 682 /*
 683  * Push the allmulti flag down to all slaves
 684  */
 685 static int bond_set_allmulti(struct bonding *bond, int inc)
 686 {
 687         int err = 0;
 688         if (USES_PRIMARY(bond->params.mode)) {
 689                 /* write lock already acquired */
 690                 if (bond->curr_active_slave) {
 691                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 692                                                inc);
 693                 }
 694         } else {
 695                 struct slave *slave;
 696                 int i;
 697                 bond_for_each_slave(bond, slave, i) {
 698                         err = dev_set_allmulti(slave->dev, inc);
 699                         if (err)
 700                                 return err;
 701                 }
 702         }
 703         return err;
 704 }
 705
 706 /*
 707  * Add a Multicast address to slaves
 708  * according to mode
 709  */
 710 static void bond_mc_add(struct bonding *bond, void *addr)
 711 {
 712         if (USES_PRIMARY(bond->params.mode)) {
 713                 /* write lock already acquired */
 714                 if (bond->curr_active_slave)
 715                         dev_mc_add(bond->curr_active_slave->dev, addr);
 716         } else {
 717                 struct slave *slave;
 718                 int i;
 719
 720                 bond_for_each_slave(bond, slave, i)
 721                         dev_mc_add(slave->dev, addr);
 722         }
 723 }
 724
 725 /*
 726  * Remove a multicast address from slave
 727  * according to mode
 728  */
 729 static void bond_mc_del(struct bonding *bond, void *addr)
 730 {
 731         if (USES_PRIMARY(bond->params.mode)) {
 732                 /* write lock already acquired */
 733                 if (bond->curr_active_slave)
 734                         dev_mc_del(bond->curr_active_slave->dev, addr);
 735         } else {
 736                 struct slave *slave;
 737                 int i;
 738                 bond_for_each_slave(bond, slave, i) {
 739                         dev_mc_del(slave->dev, addr);
 740                 }
 741         }
 742 }
 743
 744
 745 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 746 {
 747         struct in_device *in_dev;
 748
 749         in_dev = __in_dev_get_rcu(dev);
 750         if (in_dev)
 751                 ip_mc_rejoin_groups(in_dev);
 752 }
 753
 754 /*
 755  * Retrieve the list of registered multicast addresses for the bonding
 756  * device and retransmit an IGMP JOIN request to the current active
 757  * slave.
 758  */
 759 static void bond_resend_igmp_join_requests(struct bonding *bond)
 760 {
 761         struct net_device *bond_dev, *vlan_dev, *upper_dev;
 762         struct vlan_entry *vlan;
 763
 764         rcu_read_lock();
 765         read_lock(&bond->lock);
 766
 767         bond_dev = bond->dev;
 768
 769         /* rejoin all groups on bond device */
 770         __bond_resend_igmp_join_requests(bond_dev);
 771
 772         /*
 773          * if bond is enslaved to a bridge,
 774          * then rejoin all groups on its master
 775          */
 776         upper_dev = netdev_master_upper_dev_get_rcu(bond_dev);
 777         if (upper_dev && upper_dev->priv_flags & IFF_EBRIDGE)
 778                 __bond_resend_igmp_join_requests(upper_dev);
 779
 780         /* rejoin all groups on vlan devices */
 781         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 782                 vlan_dev = __vlan_find_dev_deep(bond_dev,
 783                                                 vlan->vlan_id);
 784                 if (vlan_dev)
 785                         __bond_resend_igmp_join_requests(vlan_dev);
 786         }
 787
 788         if (--bond->igmp_retrans > 0)
 789                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 790
 791         read_unlock(&bond->lock);
 792         rcu_read_unlock();
 793 }
 794
 795 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 796 {
 797         struct bonding *bond = container_of(work, struct bonding,
 798                                             mcast_work.work);
 799         rcu_read_lock();
 800         bond_resend_igmp_join_requests(bond);
 801         rcu_read_unlock();
 802 }
 803
 804 /*
 805  * flush all members of flush->mc_list from device dev->mc_list
 806  */
 807 static void bond_mc_list_flush(struct net_device *bond_dev,
 808                                struct net_device *slave_dev)
 809 {
 810         struct bonding *bond = netdev_priv(bond_dev);
 811         struct netdev_hw_addr *ha;
 812
 813         netdev_for_each_mc_addr(ha, bond_dev)
 814                 dev_mc_del(slave_dev, ha->addr);
 815
 816         if (bond->params.mode == BOND_MODE_8023AD) {
 817                 /* del lacpdu mc addr from mc list */
 818                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 819
 820                 dev_mc_del(slave_dev, lacpdu_multicast);
 821         }
 822 }
 823
 824 /*--------------------------- Active slave change ---------------------------*/
 825
 826 /*
 827  * Update the mc list and multicast-related flags for the new and
 828  * old active slaves (if any) according to the multicast mode, and
 829  * promiscuous flags unconditionally.
 830  */
 831 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 832                          struct slave *old_active)
 833 {
 834         struct netdev_hw_addr *ha;
 835
 836         if (!USES_PRIMARY(bond->params.mode))
 837                 /* nothing to do -  mc list is already up-to-date on
 838                  * all slaves
 839                  */
 840                 return;
 841
 842         if (old_active) {
 843                 if (bond->dev->flags & IFF_PROMISC)
 844                         dev_set_promiscuity(old_active->dev, -1);
 845
 846                 if (bond->dev->flags & IFF_ALLMULTI)
 847                         dev_set_allmulti(old_active->dev, -1);
 848
 849                 netdev_for_each_mc_addr(ha, bond->dev)
 850                         dev_mc_del(old_active->dev, ha->addr);
 851         }
 852
 853         if (new_active) {
 854                 /* FIXME: Signal errors upstream. */
 855                 if (bond->dev->flags & IFF_PROMISC)
 856                         dev_set_promiscuity(new_active->dev, 1);
 857
 858                 if (bond->dev->flags & IFF_ALLMULTI)
 859                         dev_set_allmulti(new_active->dev, 1);
 860
 861                 netdev_for_each_mc_addr(ha, bond->dev)
 862                         dev_mc_add(new_active->dev, ha->addr);
 863         }
 864 }
 865
 866 /*
 867  * bond_do_fail_over_mac
 868  *
 869  * Perform special MAC address swapping for fail_over_mac settings
 870  *
 871  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 872  */
 873 static void bond_do_fail_over_mac(struct bonding *bond,
 874                                   struct slave *new_active,
 875                                   struct slave *old_active)
 876         __releases(&bond->curr_slave_lock)
 877         __releases(&bond->lock)
 878         __acquires(&bond->lock)
 879         __acquires(&bond->curr_slave_lock)
 880 {
 881         u8 tmp_mac[ETH_ALEN];
 882         struct sockaddr saddr;
 883         int rv;
 884
 885         switch (bond->params.fail_over_mac) {
 886         case BOND_FOM_ACTIVE:
 887                 if (new_active) {
 888                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 889                                new_active->dev->addr_len);
 890                         write_unlock_bh(&bond->curr_slave_lock);
 891                         read_unlock(&bond->lock);
 892                         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
 893                         read_lock(&bond->lock);
 894                         write_lock_bh(&bond->curr_slave_lock);
 895                 }
 896                 break;
 897         case BOND_FOM_FOLLOW:
 898                 /*
 899                  * if new_active && old_active, swap them
 900                  * if just old_active, do nothing (going to no active slave)
 901                  * if just new_active, set new_active to bond's MAC
 902                  */
 903                 if (!new_active)
 904                         return;
 905
 906                 write_unlock_bh(&bond->curr_slave_lock);
 907                 read_unlock(&bond->lock);
 908
 909                 if (old_active) {
 910                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 911                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 912                                ETH_ALEN);
 913                         saddr.sa_family = new_active->dev->type;
 914                 } else {
 915                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 916                         saddr.sa_family = bond->dev->type;
 917                 }
 918
 919                 rv = dev_set_mac_address(new_active->dev, &saddr);
 920                 if (rv) {
 921                         pr_err("%s: Error %d setting MAC of slave %s\n",
 922                                bond->dev->name, -rv, new_active->dev->name);
 923                         goto out;
 924                 }
 925
 926                 if (!old_active)
 927                         goto out;
 928
 929                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 930                 saddr.sa_family = old_active->dev->type;
 931
 932                 rv = dev_set_mac_address(old_active->dev, &saddr);
 933                 if (rv)
 934                         pr_err("%s: Error %d setting MAC of slave %s\n",
 935                                bond->dev->name, -rv, new_active->dev->name);
 936 out:
 937                 read_lock(&bond->lock);
 938                 write_lock_bh(&bond->curr_slave_lock);
 939                 break;
 940         default:
 941                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 942                        bond->dev->name, bond->params.fail_over_mac);
 943                 break;
 944         }
 945
 946 }
 947
 948 static bool bond_should_change_active(struct bonding *bond)
 949 {
 950         struct slave *prim = bond->primary_slave;
 951         struct slave *curr = bond->curr_active_slave;
 952
 953         if (!prim || !curr || curr->link != BOND_LINK_UP)
 954                 return true;
 955         if (bond->force_primary) {
 956                 bond->force_primary = false;
 957                 return true;
 958         }
 959         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 960             (prim->speed < curr->speed ||
 961              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 962                 return false;
 963         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 964                 return false;
 965         return true;
 966 }
 967
 968 /**
 969  * find_best_interface - select the best available slave to be the active one
 970  * @bond: our bonding struct
 971  *
 972  * Warning: Caller must hold curr_slave_lock for writing.
 973  */
 974 static struct slave *bond_find_best_slave(struct bonding *bond)
 975 {
 976         struct slave *new_active, *old_active;
 977         struct slave *bestslave = NULL;
 978         int mintime = bond->params.updelay;
 979         int i;
 980
 981         new_active = bond->curr_active_slave;
 982
 983         if (!new_active) { /* there were no active slaves left */
 984                 if (bond->slave_cnt > 0)   /* found one slave */
 985                         new_active = bond->first_slave;
 986                 else
 987                         return NULL; /* still no slave, return NULL */
 988         }
 989
 990         if ((bond->primary_slave) &&
 991             bond->primary_slave->link == BOND_LINK_UP &&
 992             bond_should_change_active(bond)) {
 993                 new_active = bond->primary_slave;
 994         }
 995
 996         /* remember where to stop iterating over the slaves */
 997         old_active = new_active;
 998
 999         bond_for_each_slave_from(bond, new_active, i, old_active) {
1000                 if (new_active->link == BOND_LINK_UP) {
1001                         return new_active;
1002                 } else if (new_active->link == BOND_LINK_BACK &&
1003                            IS_UP(new_active->dev)) {
1004                         /* link up, but waiting for stabilization */
1005                         if (new_active->delay < mintime) {
1006                                 mintime = new_active->delay;
1007                                 bestslave = new_active;
1008                         }
1009                 }
1010         }
1011
1012         return bestslave;
1013 }
1014
1015 static bool bond_should_notify_peers(struct bonding *bond)
1016 {
1017         struct slave *slave = bond->curr_active_slave;
1018
1019         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1020                  bond->dev->name, slave ? slave->dev->name : "NULL");
1021
1022         if (!slave || !bond->send_peer_notif ||
1023             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1024                 return false;
1025
1026         bond->send_peer_notif--;
1027         return true;
1028 }
1029
1030 /**
1031  * change_active_interface - change the active slave into the specified one
1032  * @bond: our bonding struct
1033  * @new: the new slave to make the active one
1034  *
1035  * Set the new slave to the bond's settings and unset them on the old
1036  * curr_active_slave.
1037  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1038  *
1039  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1040  * because it is apparently the best available slave we have, even though its
1041  * updelay hasn't timed out yet.
1042  *
1043  * If new_active is not NULL, caller must hold bond->lock for read and
1044  * curr_slave_lock for write_bh.
1045  */
1046 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1047 {
1048         struct slave *old_active = bond->curr_active_slave;
1049
1050         if (old_active == new_active)
1051                 return;
1052
1053         if (new_active) {
1054                 new_active->jiffies = jiffies;
1055
1056                 if (new_active->link == BOND_LINK_BACK) {
1057                         if (USES_PRIMARY(bond->params.mode)) {
1058                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1059                                         bond->dev->name, new_active->dev->name,
1060                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1061                         }
1062
1063                         new_active->delay = 0;
1064                         new_active->link = BOND_LINK_UP;
1065
1066                         if (bond->params.mode == BOND_MODE_8023AD)
1067                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1068
1069                         if (bond_is_lb(bond))
1070                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1071                 } else {
1072                         if (USES_PRIMARY(bond->params.mode)) {
1073                                 pr_info("%s: making interface %s the new active one.\n",
1074                                         bond->dev->name, new_active->dev->name);
1075                         }
1076                 }
1077         }
1078
1079         if (USES_PRIMARY(bond->params.mode))
1080                 bond_mc_swap(bond, new_active, old_active);
1081
1082         if (bond_is_lb(bond)) {
1083                 bond_alb_handle_active_change(bond, new_active);
1084                 if (old_active)
1085                         bond_set_slave_inactive_flags(old_active);
1086                 if (new_active)
1087                         bond_set_slave_active_flags(new_active);
1088         } else {
1089                 bond->curr_active_slave = new_active;
1090         }
1091
1092         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1093                 if (old_active)
1094                         bond_set_slave_inactive_flags(old_active);
1095
1096                 if (new_active) {
1097                         bool should_notify_peers = false;
1098
1099                         bond_set_slave_active_flags(new_active);
1100
1101                         if (bond->params.fail_over_mac)
1102                                 bond_do_fail_over_mac(bond, new_active,
1103                                                       old_active);
1104
1105                         if (netif_running(bond->dev)) {
1106                                 bond->send_peer_notif =
1107                                         bond->params.num_peer_notif;
1108                                 should_notify_peers =
1109                                         bond_should_notify_peers(bond);
1110                         }
1111
1112                         write_unlock_bh(&bond->curr_slave_lock);
1113                         read_unlock(&bond->lock);
1114
1115                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1116                         if (should_notify_peers)
1117                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1118                                                          bond->dev);
1119
1120                         read_lock(&bond->lock);
1121                         write_lock_bh(&bond->curr_slave_lock);
1122                 }
1123         }
1124
1125         /* resend IGMP joins since active slave has changed or
1126          * all were sent on curr_active_slave.
1127          * resend only if bond is brought up with the affected
1128          * bonding modes and the retransmission is enabled */
1129         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1130             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1131              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1132                 bond->igmp_retrans = bond->params.resend_igmp;
1133                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1134         }
1135 }
1136
1137 /**
1138  * bond_select_active_slave - select a new active slave, if needed
1139  * @bond: our bonding struct
1140  *
1141  * This functions should be called when one of the following occurs:
1142  * - The old curr_active_slave has been released or lost its link.
1143  * - The primary_slave has got its link back.
1144  * - A slave has got its link back and there's no old curr_active_slave.
1145  *
1146  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1147  */
1148 void bond_select_active_slave(struct bonding *bond)
1149 {
1150         struct slave *best_slave;
1151         int rv;
1152
1153         best_slave = bond_find_best_slave(bond);
1154         if (best_slave != bond->curr_active_slave) {
1155                 bond_change_active_slave(bond, best_slave);
1156                 rv = bond_set_carrier(bond);
1157                 if (!rv)
1158                         return;
1159
1160                 if (netif_carrier_ok(bond->dev)) {
1161                         pr_info("%s: first active interface up!\n",
1162                                 bond->dev->name);
1163                 } else {
1164                         pr_info("%s: now running without any active interface !\n",
1165                                 bond->dev->name);
1166                 }
1167         }
1168 }
1169
1170 /*--------------------------- slave list handling ---------------------------*/
1171
1172 /*
1173  * This function attaches the slave to the end of list.
1174  *
1175  * bond->lock held for writing by caller.
1176  */
1177 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1178 {
1179         if (bond->first_slave == NULL) { /* attaching the first slave */
1180                 new_slave->next = new_slave;
1181                 new_slave->prev = new_slave;
1182                 bond->first_slave = new_slave;
1183         } else {
1184                 new_slave->next = bond->first_slave;
1185                 new_slave->prev = bond->first_slave->prev;
1186                 new_slave->next->prev = new_slave;
1187                 new_slave->prev->next = new_slave;
1188         }
1189
1190         bond->slave_cnt++;
1191 }
1192
1193 /*
1194  * This function detaches the slave from the list.
1195  * WARNING: no check is made to verify if the slave effectively
1196  * belongs to <bond>.
1197  * Nothing is freed on return, structures are just unchained.
1198  * If any slave pointer in bond was pointing to <slave>,
1199  * it should be changed by the calling function.
1200  *
1201  * bond->lock held for writing by caller.
1202  */
1203 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1204 {
1205         if (slave->next)
1206                 slave->next->prev = slave->prev;
1207
1208         if (slave->prev)
1209                 slave->prev->next = slave->next;
1210
1211         if (bond->first_slave == slave) { /* slave is the first slave */
1212                 if (bond->slave_cnt > 1) { /* there are more slave */
1213                         bond->first_slave = slave->next;
1214                 } else {
1215                         bond->first_slave = NULL; /* slave was the last one */
1216                 }
1217         }
1218
1219         slave->next = NULL;
1220         slave->prev = NULL;
1221         bond->slave_cnt--;
1222 }
1223
1224 #ifdef CONFIG_NET_POLL_CONTROLLER
1225 static inline int slave_enable_netpoll(struct slave *slave)
1226 {
1227         struct netpoll *np;
1228         int err = 0;
1229
1230         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1231         err = -ENOMEM;
1232         if (!np)
1233                 goto out;
1234
1235         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1236         if (err) {
1237                 kfree(np);
1238                 goto out;
1239         }
1240         slave->np = np;
1241 out:
1242         return err;
1243 }
1244 static inline void slave_disable_netpoll(struct slave *slave)
1245 {
1246         struct netpoll *np = slave->np;
1247
1248         if (!np)
1249                 return;
1250
1251         slave->np = NULL;
1252         __netpoll_free_async(np);
1253 }
1254 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1255 {
1256         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1257                 return false;
1258         if (!slave_dev->netdev_ops->ndo_poll_controller)
1259                 return false;
1260         return true;
1261 }
1262
1263 static void bond_poll_controller(struct net_device *bond_dev)
1264 {
1265 }
1266
1267 static void __bond_netpoll_cleanup(struct bonding *bond)
1268 {
1269         struct slave *slave;
1270         int i;
1271
1272         bond_for_each_slave(bond, slave, i)
1273                 if (IS_UP(slave->dev))
1274                         slave_disable_netpoll(slave);
1275 }
1276 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1277 {
1278         struct bonding *bond = netdev_priv(bond_dev);
1279
1280         read_lock(&bond->lock);
1281         __bond_netpoll_cleanup(bond);
1282         read_unlock(&bond->lock);
1283 }
1284
1285 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1286 {
1287         struct bonding *bond = netdev_priv(dev);
1288         struct slave *slave;
1289         int i, err = 0;
1290
1291         read_lock(&bond->lock);
1292         bond_for_each_slave(bond, slave, i) {
1293                 err = slave_enable_netpoll(slave);
1294                 if (err) {
1295                         __bond_netpoll_cleanup(bond);
1296                         break;
1297                 }
1298         }
1299         read_unlock(&bond->lock);
1300         return err;
1301 }
1302
1303 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1304 {
1305         return bond->dev->npinfo;
1306 }
1307
1308 #else
1309 static inline int slave_enable_netpoll(struct slave *slave)
1310 {
1311         return 0;
1312 }
1313 static inline void slave_disable_netpoll(struct slave *slave)
1314 {
1315 }
1316 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1317 {
1318 }
1319 #endif
1320
1321 /*---------------------------------- IOCTL ----------------------------------*/
1322
1323 static void bond_set_dev_addr(struct net_device *bond_dev,
1324                               struct net_device *slave_dev)
1325 {
1326         pr_debug("bond_dev=%p\n", bond_dev);
1327         pr_debug("slave_dev=%p\n", slave_dev);
1328         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1329         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1330         bond_dev->addr_assign_type = NET_ADDR_SET;
1331         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
1332 }
1333
1334 static netdev_features_t bond_fix_features(struct net_device *dev,
1335         netdev_features_t features)
1336 {
1337         struct slave *slave;
1338         struct bonding *bond = netdev_priv(dev);
1339         netdev_features_t mask;
1340         int i;
1341
1342         read_lock(&bond->lock);
1343
1344         if (!bond->first_slave) {
1345                 /* Disable adding VLANs to empty bond. But why? --mq */
1346                 features |= NETIF_F_VLAN_CHALLENGED;
1347                 goto out;
1348         }
1349
1350         mask = features;
1351         features &= ~NETIF_F_ONE_FOR_ALL;
1352         features |= NETIF_F_ALL_FOR_ALL;
1353
1354         bond_for_each_slave(bond, slave, i) {
1355                 features = netdev_increment_features(features,
1356                                                      slave->dev->features,
1357                                                      mask);
1358         }
1359
1360 out:
1361         read_unlock(&bond->lock);
1362         return features;
1363 }
1364
1365 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1366                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1367                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1368
1369 static void bond_compute_features(struct bonding *bond)
1370 {
1371         struct slave *slave;
1372         struct net_device *bond_dev = bond->dev;
1373         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1374         unsigned short max_hard_header_len = ETH_HLEN;
1375         unsigned int gso_max_size = GSO_MAX_SIZE;
1376         u16 gso_max_segs = GSO_MAX_SEGS;
1377         int i;
1378         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1379
1380         read_lock(&bond->lock);
1381
1382         if (!bond->first_slave)
1383                 goto done;
1384
1385         bond_for_each_slave(bond, slave, i) {
1386                 vlan_features = netdev_increment_features(vlan_features,
1387                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1388
1389                 dst_release_flag &= slave->dev->priv_flags;
1390                 if (slave->dev->hard_header_len > max_hard_header_len)
1391                         max_hard_header_len = slave->dev->hard_header_len;
1392
1393                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1394                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1395         }
1396
1397 done:
1398         bond_dev->vlan_features = vlan_features;
1399         bond_dev->hard_header_len = max_hard_header_len;
1400         bond_dev->gso_max_segs = gso_max_segs;
1401         netif_set_gso_max_size(bond_dev, gso_max_size);
1402
1403         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1404         bond_dev->priv_flags = flags | dst_release_flag;
1405
1406         read_unlock(&bond->lock);
1407
1408         netdev_change_features(bond_dev);
1409 }
1410
1411 static void bond_setup_by_slave(struct net_device *bond_dev,
1412                                 struct net_device *slave_dev)
1413 {
1414         struct bonding *bond = netdev_priv(bond_dev);
1415
1416         bond_dev->header_ops        = slave_dev->header_ops;
1417
1418         bond_dev->type              = slave_dev->type;
1419         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1420         bond_dev->addr_len          = slave_dev->addr_len;
1421
1422         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1423                 slave_dev->addr_len);
1424         bond->setup_by_slave = 1;
1425 }
1426
1427 /* On bonding slaves other than the currently active slave, suppress
1428  * duplicates except for alb non-mcast/bcast.
1429  */
1430 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1431                                             struct slave *slave,
1432                                             struct bonding *bond)
1433 {
1434         if (bond_is_slave_inactive(slave)) {
1435                 if (bond->params.mode == BOND_MODE_ALB &&
1436                     skb->pkt_type != PACKET_BROADCAST &&
1437                     skb->pkt_type != PACKET_MULTICAST)
1438                         return false;
1439                 return true;
1440         }
1441         return false;
1442 }
1443
1444 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1445 {
1446         struct sk_buff *skb = *pskb;
1447         struct slave *slave;
1448         struct bonding *bond;
1449         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1450                           struct slave *);
1451         int ret = RX_HANDLER_ANOTHER;
1452
1453         skb = skb_share_check(skb, GFP_ATOMIC);
1454         if (unlikely(!skb))
1455                 return RX_HANDLER_CONSUMED;
1456
1457         *pskb = skb;
1458
1459         slave = bond_slave_get_rcu(skb->dev);
1460         bond = slave->bond;
1461
1462         if (bond->params.arp_interval)
1463                 slave->dev->last_rx = jiffies;
1464
1465         recv_probe = ACCESS_ONCE(bond->recv_probe);
1466         if (recv_probe) {
1467                 ret = recv_probe(skb, bond, slave);
1468                 if (ret == RX_HANDLER_CONSUMED) {
1469                         consume_skb(skb);
1470                         return ret;
1471                 }
1472         }
1473
1474         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1475                 return RX_HANDLER_EXACT;
1476         }
1477
1478         skb->dev = bond->dev;
1479
1480         if (bond->params.mode == BOND_MODE_ALB &&
1481             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1482             skb->pkt_type == PACKET_HOST) {
1483
1484                 if (unlikely(skb_cow_head(skb,
1485                                           skb->data - skb_mac_header(skb)))) {
1486                         kfree_skb(skb);
1487                         return RX_HANDLER_CONSUMED;
1488                 }
1489                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1490         }
1491
1492         return ret;
1493 }
1494
1495 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1496                                       struct net_device *slave_dev)
1497 {
1498         int err;
1499
1500         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1501         if (err)
1502                 return err;
1503         slave_dev->flags |= IFF_SLAVE;
1504         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1505         return 0;
1506 }
1507
1508 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1509                                   struct net_device *slave_dev)
1510 {
1511         netdev_upper_dev_unlink(slave_dev, bond_dev);
1512         slave_dev->flags &= ~IFF_SLAVE;
1513         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1514 }
1515
1516 /* enslave device <slave> to bond device <master> */
1517 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1518 {
1519         struct bonding *bond = netdev_priv(bond_dev);
1520         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1521         struct slave *new_slave = NULL;
1522         struct netdev_hw_addr *ha;
1523         struct sockaddr addr;
1524         int link_reporting;
1525         int res = 0;
1526
1527         if (!bond->params.use_carrier &&
1528             slave_dev->ethtool_ops->get_link == NULL &&
1529             slave_ops->ndo_do_ioctl == NULL) {
1530                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1531                            bond_dev->name, slave_dev->name);
1532         }
1533
1534         /* already enslaved */
1535         if (slave_dev->flags & IFF_SLAVE) {
1536                 pr_debug("Error, Device was already enslaved\n");
1537                 return -EBUSY;
1538         }
1539
1540         /* vlan challenged mutual exclusion */
1541         /* no need to lock since we're protected by rtnl_lock */
1542         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1543                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1544                 if (vlan_uses_dev(bond_dev)) {
1545                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1546                                bond_dev->name, slave_dev->name, bond_dev->name);
1547                         return -EPERM;
1548                 } else {
1549                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1550                                    bond_dev->name, slave_dev->name,
1551                                    slave_dev->name, bond_dev->name);
1552                 }
1553         } else {
1554                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1555         }
1556
1557         /*
1558          * Old ifenslave binaries are no longer supported.  These can
1559          * be identified with moderate accuracy by the state of the slave:
1560          * the current ifenslave will set the interface down prior to
1561          * enslaving it; the old ifenslave will not.
1562          */
1563         if ((slave_dev->flags & IFF_UP)) {
1564                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1565                        slave_dev->name);
1566                 res = -EPERM;
1567                 goto err_undo_flags;
1568         }
1569
1570         /* set bonding device ether type by slave - bonding netdevices are
1571          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1572          * there is a need to override some of the type dependent attribs/funcs.
1573          *
1574          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1575          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1576          */
1577         if (bond->slave_cnt == 0) {
1578                 if (bond_dev->type != slave_dev->type) {
1579                         pr_debug("%s: change device type from %d to %d\n",
1580                                  bond_dev->name,
1581                                  bond_dev->type, slave_dev->type);
1582
1583                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1584                                                        bond_dev);
1585                         res = notifier_to_errno(res);
1586                         if (res) {
1587                                 pr_err("%s: refused to change device type\n",
1588                                        bond_dev->name);
1589                                 res = -EBUSY;
1590                                 goto err_undo_flags;
1591                         }
1592
1593                         /* Flush unicast and multicast addresses */
1594                         dev_uc_flush(bond_dev);
1595                         dev_mc_flush(bond_dev);
1596
1597                         if (slave_dev->type != ARPHRD_ETHER)
1598                                 bond_setup_by_slave(bond_dev, slave_dev);
1599                         else {
1600                                 ether_setup(bond_dev);
1601                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1602                         }
1603
1604                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1605                                                  bond_dev);
1606                 }
1607         } else if (bond_dev->type != slave_dev->type) {
1608                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1609                        slave_dev->name,
1610                        slave_dev->type, bond_dev->type);
1611                 res = -EINVAL;
1612                 goto err_undo_flags;
1613         }
1614
1615         if (slave_ops->ndo_set_mac_address == NULL) {
1616                 if (bond->slave_cnt == 0) {
1617                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1618                                    bond_dev->name);
1619                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1620                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1621                         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",
1622                                bond_dev->name);
1623                         res = -EOPNOTSUPP;
1624                         goto err_undo_flags;
1625                 }
1626         }
1627
1628         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1629
1630         /* If this is the first slave, then we need to set the master's hardware
1631          * address to be the same as the slave's. */
1632         if (bond->slave_cnt == 0 && bond->dev_addr_from_first)
1633                 bond_set_dev_addr(bond->dev, slave_dev);
1634
1635         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1636         if (!new_slave) {
1637                 res = -ENOMEM;
1638                 goto err_undo_flags;
1639         }
1640
1641         /*
1642          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1643          * is set via sysfs or module option if desired.
1644          */
1645         new_slave->queue_id = 0;
1646
1647         /* Save slave's original mtu and then set it to match the bond */
1648         new_slave->original_mtu = slave_dev->mtu;
1649         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1650         if (res) {
1651                 pr_debug("Error %d calling dev_set_mtu\n", res);
1652                 goto err_free;
1653         }
1654
1655         /*
1656          * Save slave's original ("permanent") mac address for modes
1657          * that need it, and for restoring it upon release, and then
1658          * set it to the master's address
1659          */
1660         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1661
1662         if (!bond->params.fail_over_mac) {
1663                 /*
1664                  * Set slave to master's mac address.  The application already
1665                  * set the master's mac address to that of the first slave
1666                  */
1667                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1668                 addr.sa_family = slave_dev->type;
1669                 res = dev_set_mac_address(slave_dev, &addr);
1670                 if (res) {
1671                         pr_debug("Error %d calling set_mac_address\n", res);
1672                         goto err_restore_mtu;
1673                 }
1674         }
1675
1676         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1677         if (res) {
1678                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1679                 goto err_restore_mac;
1680         }
1681
1682         /* open the slave since the application closed it */
1683         res = dev_open(slave_dev);
1684         if (res) {
1685                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1686                 goto err_unset_master;
1687         }
1688
1689         new_slave->bond = bond;
1690         new_slave->dev = slave_dev;
1691         slave_dev->priv_flags |= IFF_BONDING;
1692
1693         if (bond_is_lb(bond)) {
1694                 /* bond_alb_init_slave() must be called before all other stages since
1695                  * it might fail and we do not want to have to undo everything
1696                  */
1697                 res = bond_alb_init_slave(bond, new_slave);
1698                 if (res)
1699                         goto err_close;
1700         }
1701
1702         /* If the mode USES_PRIMARY, then the new slave gets the
1703          * master's promisc (and mc) settings only if it becomes the
1704          * curr_active_slave, and that is taken care of later when calling
1705          * bond_change_active()
1706          */
1707         if (!USES_PRIMARY(bond->params.mode)) {
1708                 /* set promiscuity level to new slave */
1709                 if (bond_dev->flags & IFF_PROMISC) {
1710                         res = dev_set_promiscuity(slave_dev, 1);
1711                         if (res)
1712                                 goto err_close;
1713                 }
1714
1715                 /* set allmulti level to new slave */
1716                 if (bond_dev->flags & IFF_ALLMULTI) {
1717                         res = dev_set_allmulti(slave_dev, 1);
1718                         if (res)
1719                                 goto err_close;
1720                 }
1721
1722                 netif_addr_lock_bh(bond_dev);
1723                 /* upload master's mc_list to new slave */
1724                 netdev_for_each_mc_addr(ha, bond_dev)
1725                         dev_mc_add(slave_dev, ha->addr);
1726                 netif_addr_unlock_bh(bond_dev);
1727         }
1728
1729         if (bond->params.mode == BOND_MODE_8023AD) {
1730                 /* add lacpdu mc addr to mc list */
1731                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1732
1733                 dev_mc_add(slave_dev, lacpdu_multicast);
1734         }
1735
1736         bond_add_vlans_on_slave(bond, slave_dev);
1737
1738         write_lock_bh(&bond->lock);
1739
1740         bond_attach_slave(bond, new_slave);
1741
1742         new_slave->delay = 0;
1743         new_slave->link_failure_count = 0;
1744
1745         write_unlock_bh(&bond->lock);
1746
1747         bond_compute_features(bond);
1748
1749         bond_update_speed_duplex(new_slave);
1750
1751         read_lock(&bond->lock);
1752
1753         new_slave->last_arp_rx = jiffies -
1754                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1755
1756         if (bond->params.miimon && !bond->params.use_carrier) {
1757                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1758
1759                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1760                         /*
1761                          * miimon is set but a bonded network driver
1762                          * does not support ETHTOOL/MII and
1763                          * arp_interval is not set.  Note: if
1764                          * use_carrier is enabled, we will never go
1765                          * here (because netif_carrier is always
1766                          * supported); thus, we don't need to change
1767                          * the messages for netif_carrier.
1768                          */
1769                         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",
1770                                bond_dev->name, slave_dev->name);
1771                 } else if (link_reporting == -1) {
1772                         /* unable get link status using mii/ethtool */
1773                         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",
1774                                    bond_dev->name, slave_dev->name);
1775                 }
1776         }
1777
1778         /* check for initial state */
1779         if (bond->params.miimon) {
1780                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1781                         if (bond->params.updelay) {
1782                                 new_slave->link = BOND_LINK_BACK;
1783                                 new_slave->delay = bond->params.updelay;
1784                         } else {
1785                                 new_slave->link = BOND_LINK_UP;
1786                         }
1787                 } else {
1788                         new_slave->link = BOND_LINK_DOWN;
1789                 }
1790         } else if (bond->params.arp_interval) {
1791                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1792                         BOND_LINK_UP : BOND_LINK_DOWN);
1793         } else {
1794                 new_slave->link = BOND_LINK_UP;
1795         }
1796
1797         if (new_slave->link != BOND_LINK_DOWN)
1798                 new_slave->jiffies = jiffies;
1799         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1800                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1801                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1802
1803         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1804                 /* if there is a primary slave, remember it */
1805                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1806                         bond->primary_slave = new_slave;
1807                         bond->force_primary = true;
1808                 }
1809         }
1810
1811         write_lock_bh(&bond->curr_slave_lock);
1812
1813         switch (bond->params.mode) {
1814         case BOND_MODE_ACTIVEBACKUP:
1815                 bond_set_slave_inactive_flags(new_slave);
1816                 bond_select_active_slave(bond);
1817                 break;
1818         case BOND_MODE_8023AD:
1819                 /* in 802.3ad mode, the internal mechanism
1820                  * will activate the slaves in the selected
1821                  * aggregator
1822                  */
1823                 bond_set_slave_inactive_flags(new_slave);
1824                 /* if this is the first slave */
1825                 if (bond->slave_cnt == 1) {
1826                         SLAVE_AD_INFO(new_slave).id = 1;
1827                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1828                          * can be called only after the mac address of the bond is set
1829                          */
1830                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1831                 } else {
1832                         SLAVE_AD_INFO(new_slave).id =
1833                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1834                 }
1835
1836                 bond_3ad_bind_slave(new_slave);
1837                 break;
1838         case BOND_MODE_TLB:
1839         case BOND_MODE_ALB:
1840                 bond_set_active_slave(new_slave);
1841                 bond_set_slave_inactive_flags(new_slave);
1842                 bond_select_active_slave(bond);
1843                 break;
1844         default:
1845                 pr_debug("This slave is always active in trunk mode\n");
1846
1847                 /* always active in trunk mode */
1848                 bond_set_active_slave(new_slave);
1849
1850                 /* In trunking mode there is little meaning to curr_active_slave
1851                  * anyway (it holds no special properties of the bond device),
1852                  * so we can change it without calling change_active_interface()
1853                  */
1854                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1855                         bond->curr_active_slave = new_slave;
1856
1857                 break;
1858         } /* switch(bond_mode) */
1859
1860         write_unlock_bh(&bond->curr_slave_lock);
1861
1862         bond_set_carrier(bond);
1863
1864 #ifdef CONFIG_NET_POLL_CONTROLLER
1865         slave_dev->npinfo = bond_netpoll_info(bond);
1866         if (slave_dev->npinfo) {
1867                 if (slave_enable_netpoll(new_slave)) {
1868                         read_unlock(&bond->lock);
1869                         pr_info("Error, %s: master_dev is using netpoll, "
1870                                  "but new slave device does not support netpoll.\n",
1871                                  bond_dev->name);
1872                         res = -EBUSY;
1873                         goto err_detach;
1874                 }
1875         }
1876 #endif
1877
1878         read_unlock(&bond->lock);
1879
1880         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1881         if (res)
1882                 goto err_detach;
1883
1884         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1885                                          new_slave);
1886         if (res) {
1887                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1888                 goto err_dest_symlinks;
1889         }
1890
1891         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1892                 bond_dev->name, slave_dev->name,
1893                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1894                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1895
1896         /* enslave is successful */
1897         return 0;
1898
1899 /* Undo stages on error */
1900 err_dest_symlinks:
1901         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1902
1903 err_detach:
1904         write_lock_bh(&bond->lock);
1905         bond_detach_slave(bond, new_slave);
1906         write_unlock_bh(&bond->lock);
1907
1908 err_close:
1909         dev_close(slave_dev);
1910
1911 err_unset_master:
1912         bond_upper_dev_unlink(bond_dev, slave_dev);
1913
1914 err_restore_mac:
1915         if (!bond->params.fail_over_mac) {
1916                 /* XXX TODO - fom follow mode needs to change master's
1917                  * MAC if this slave's MAC is in use by the bond, or at
1918                  * least print a warning.
1919                  */
1920                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1921                 addr.sa_family = slave_dev->type;
1922                 dev_set_mac_address(slave_dev, &addr);
1923         }
1924
1925 err_restore_mtu:
1926         dev_set_mtu(slave_dev, new_slave->original_mtu);
1927
1928 err_free:
1929         kfree(new_slave);
1930
1931 err_undo_flags:
1932         bond_compute_features(bond);
1933
1934         return res;
1935 }
1936
1937 /*
1938  * Try to release the slave device <slave> from the bond device <master>
1939  * It is legal to access curr_active_slave without a lock because all the function
1940  * is write-locked. If "all" is true it means that the function is being called
1941  * while destroying a bond interface and all slaves are being released.
1942  *
1943  * The rules for slave state should be:
1944  *   for Active/Backup:
1945  *     Active stays on all backups go down
1946  *   for Bonded connections:
1947  *     The first up interface should be left on and all others downed.
1948  */
1949 static int __bond_release_one(struct net_device *bond_dev,
1950                               struct net_device *slave_dev,
1951                               bool all)
1952 {
1953         struct bonding *bond = netdev_priv(bond_dev);
1954         struct slave *slave, *oldcurrent;
1955         struct sockaddr addr;
1956         netdev_features_t old_features = bond_dev->features;
1957
1958         /* slave is not a slave or master is not master of this slave */
1959         if (!(slave_dev->flags & IFF_SLAVE) ||
1960             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1961                 pr_err("%s: Error: cannot release %s.\n",
1962                        bond_dev->name, slave_dev->name);
1963                 return -EINVAL;
1964         }
1965
1966         block_netpoll_tx();
1967         write_lock_bh(&bond->lock);
1968
1969         slave = bond_get_slave_by_dev(bond, slave_dev);
1970         if (!slave) {
1971                 /* not a slave of this bond */
1972                 pr_info("%s: %s not enslaved\n",
1973                         bond_dev->name, slave_dev->name);
1974                 write_unlock_bh(&bond->lock);
1975                 unblock_netpoll_tx();
1976                 return -EINVAL;
1977         }
1978
1979         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1980          * for this slave anymore.
1981          */
1982         netdev_rx_handler_unregister(slave_dev);
1983         write_unlock_bh(&bond->lock);
1984         synchronize_net();
1985         write_lock_bh(&bond->lock);
1986
1987         if (!all && !bond->params.fail_over_mac) {
1988                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1989                     bond->slave_cnt > 1)
1990                         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",
1991                                    bond_dev->name, slave_dev->name,
1992                                    slave->perm_hwaddr,
1993                                    bond_dev->name, slave_dev->name);
1994         }
1995
1996         /* Inform AD package of unbinding of slave. */
1997         if (bond->params.mode == BOND_MODE_8023AD) {
1998                 /* must be called before the slave is
1999                  * detached from the list
2000                  */
2001                 bond_3ad_unbind_slave(slave);
2002         }
2003
2004         pr_info("%s: releasing %s interface %s\n",
2005                 bond_dev->name,
2006                 bond_is_active_slave(slave) ? "active" : "backup",
2007                 slave_dev->name);
2008
2009         oldcurrent = bond->curr_active_slave;
2010
2011         bond->current_arp_slave = NULL;
2012
2013         /* release the slave from its bond */
2014         bond_detach_slave(bond, slave);
2015
2016         if (bond->primary_slave == slave)
2017                 bond->primary_slave = NULL;
2018
2019         if (oldcurrent == slave)
2020                 bond_change_active_slave(bond, NULL);
2021
2022         if (bond_is_lb(bond)) {
2023                 /* Must be called only after the slave has been
2024                  * detached from the list and the curr_active_slave
2025                  * has been cleared (if our_slave == old_current),
2026                  * but before a new active slave is selected.
2027                  */
2028                 write_unlock_bh(&bond->lock);
2029                 bond_alb_deinit_slave(bond, slave);
2030                 write_lock_bh(&bond->lock);
2031         }
2032
2033         if (all) {
2034                 bond->curr_active_slave = NULL;
2035         } else if (oldcurrent == slave) {
2036                 /*
2037                  * Note that we hold RTNL over this sequence, so there
2038                  * is no concern that another slave add/remove event
2039                  * will interfere.
2040                  */
2041                 write_unlock_bh(&bond->lock);
2042                 read_lock(&bond->lock);
2043                 write_lock_bh(&bond->curr_slave_lock);
2044
2045                 bond_select_active_slave(bond);
2046
2047                 write_unlock_bh(&bond->curr_slave_lock);
2048                 read_unlock(&bond->lock);
2049                 write_lock_bh(&bond->lock);
2050         }
2051
2052         if (bond->slave_cnt == 0) {
2053                 bond_set_carrier(bond);
2054                 eth_hw_addr_random(bond_dev);
2055                 bond->dev_addr_from_first = true;
2056
2057                 if (bond_vlan_used(bond)) {
2058                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2059                                    bond_dev->name, bond_dev->name);
2060                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2061                                    bond_dev->name);
2062                 }
2063         }
2064
2065         write_unlock_bh(&bond->lock);
2066         unblock_netpoll_tx();
2067
2068         if (bond->slave_cnt == 0) {
2069                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2070                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2071         }
2072
2073         bond_compute_features(bond);
2074         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2075             (old_features & NETIF_F_VLAN_CHALLENGED))
2076                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2077                         bond_dev->name, slave_dev->name, bond_dev->name);
2078
2079         /* must do this from outside any spinlocks */
2080         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2081
2082         bond_del_vlans_from_slave(bond, slave_dev);
2083
2084         /* If the mode USES_PRIMARY, then we should only remove its
2085          * promisc and mc settings if it was the curr_active_slave, but that was
2086          * already taken care of above when we detached the slave
2087          */
2088         if (!USES_PRIMARY(bond->params.mode)) {
2089                 /* unset promiscuity level from slave */
2090                 if (bond_dev->flags & IFF_PROMISC)
2091                         dev_set_promiscuity(slave_dev, -1);
2092
2093                 /* unset allmulti level from slave */
2094                 if (bond_dev->flags & IFF_ALLMULTI)
2095                         dev_set_allmulti(slave_dev, -1);
2096
2097                 /* flush master's mc_list from slave */
2098                 netif_addr_lock_bh(bond_dev);
2099                 bond_mc_list_flush(bond_dev, slave_dev);
2100                 netif_addr_unlock_bh(bond_dev);
2101         }
2102
2103         bond_upper_dev_unlink(bond_dev, slave_dev);
2104
2105         slave_disable_netpoll(slave);
2106
2107         /* close slave before restoring its mac address */
2108         dev_close(slave_dev);
2109
2110         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2111                 /* restore original ("permanent") mac address */
2112                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2113                 addr.sa_family = slave_dev->type;
2114                 dev_set_mac_address(slave_dev, &addr);
2115         }
2116
2117         dev_set_mtu(slave_dev, slave->original_mtu);
2118
2119         slave_dev->priv_flags &= ~IFF_BONDING;
2120
2121         kfree(slave);
2122
2123         return 0;  /* deletion OK */
2124 }
2125
2126 /* A wrapper used because of ndo_del_link */
2127 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2128 {
2129         return __bond_release_one(bond_dev, slave_dev, false);
2130 }
2131
2132 /*
2133 * First release a slave and then destroy the bond if no more slaves are left.
2134 * Must be under rtnl_lock when this function is called.
2135 */
2136 static int  bond_release_and_destroy(struct net_device *bond_dev,
2137                                      struct net_device *slave_dev)
2138 {
2139         struct bonding *bond = netdev_priv(bond_dev);
2140         int ret;
2141
2142         ret = bond_release(bond_dev, slave_dev);
2143         if ((ret == 0) && (bond->slave_cnt == 0)) {
2144                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2145                 pr_info("%s: destroying bond %s.\n",
2146                         bond_dev->name, bond_dev->name);
2147                 unregister_netdevice(bond_dev);
2148         }
2149         return ret;
2150 }
2151
2152 /*
2153  * This function changes the active slave to slave <slave_dev>.
2154  * It returns -EINVAL in the following cases.
2155  *  - <slave_dev> is not found in the list.
2156  *  - There is not active slave now.
2157  *  - <slave_dev> is already active.
2158  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2159  *  - <slave_dev> is not running.
2160  * In these cases, this function does nothing.
2161  * In the other cases, current_slave pointer is changed and 0 is returned.
2162  */
2163 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2164 {
2165         struct bonding *bond = netdev_priv(bond_dev);
2166         struct slave *old_active = NULL;
2167         struct slave *new_active = NULL;
2168         int res = 0;
2169
2170         if (!USES_PRIMARY(bond->params.mode))
2171                 return -EINVAL;
2172
2173         /* Verify that bond_dev is indeed the master of slave_dev */
2174         if (!(slave_dev->flags & IFF_SLAVE) ||
2175             !netdev_has_upper_dev(slave_dev, bond_dev))
2176                 return -EINVAL;
2177
2178         read_lock(&bond->lock);
2179
2180         read_lock(&bond->curr_slave_lock);
2181         old_active = bond->curr_active_slave;
2182         read_unlock(&bond->curr_slave_lock);
2183
2184         new_active = bond_get_slave_by_dev(bond, slave_dev);
2185
2186         /*
2187          * Changing to the current active: do nothing; return success.
2188          */
2189         if (new_active && (new_active == old_active)) {
2190                 read_unlock(&bond->lock);
2191                 return 0;
2192         }
2193
2194         if ((new_active) &&
2195             (old_active) &&
2196             (new_active->link == BOND_LINK_UP) &&
2197             IS_UP(new_active->dev)) {
2198                 block_netpoll_tx();
2199                 write_lock_bh(&bond->curr_slave_lock);
2200                 bond_change_active_slave(bond, new_active);
2201                 write_unlock_bh(&bond->curr_slave_lock);
2202                 unblock_netpoll_tx();
2203         } else
2204                 res = -EINVAL;
2205
2206         read_unlock(&bond->lock);
2207
2208         return res;
2209 }
2210
2211 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2212 {
2213         struct bonding *bond = netdev_priv(bond_dev);
2214
2215         info->bond_mode = bond->params.mode;
2216         info->miimon = bond->params.miimon;
2217
2218         read_lock(&bond->lock);
2219         info->num_slaves = bond->slave_cnt;
2220         read_unlock(&bond->lock);
2221
2222         return 0;
2223 }
2224
2225 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2226 {
2227         struct bonding *bond = netdev_priv(bond_dev);
2228         struct slave *slave;
2229         int i, res = -ENODEV;
2230
2231         read_lock(&bond->lock);
2232
2233         bond_for_each_slave(bond, slave, i) {
2234                 if (i == (int)info->slave_id) {
2235                         res = 0;
2236                         strcpy(info->slave_name, slave->dev->name);
2237                         info->link = slave->link;
2238                         info->state = bond_slave_state(slave);
2239                         info->link_failure_count = slave->link_failure_count;
2240                         break;
2241                 }
2242         }
2243
2244         read_unlock(&bond->lock);
2245
2246         return res;
2247 }
2248
2249 /*-------------------------------- Monitoring -------------------------------*/
2250
2251
2252 static int bond_miimon_inspect(struct bonding *bond)
2253 {
2254         struct slave *slave;
2255         int i, link_state, commit = 0;
2256         bool ignore_updelay;
2257
2258         ignore_updelay = !bond->curr_active_slave ? true : false;
2259
2260         bond_for_each_slave(bond, slave, i) {
2261                 slave->new_link = BOND_LINK_NOCHANGE;
2262
2263                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2264
2265                 switch (slave->link) {
2266                 case BOND_LINK_UP:
2267                         if (link_state)
2268                                 continue;
2269
2270                         slave->link = BOND_LINK_FAIL;
2271                         slave->delay = bond->params.downdelay;
2272                         if (slave->delay) {
2273                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2274                                         bond->dev->name,
2275                                         (bond->params.mode ==
2276                                          BOND_MODE_ACTIVEBACKUP) ?
2277                                         (bond_is_active_slave(slave) ?
2278                                          "active " : "backup ") : "",
2279                                         slave->dev->name,
2280                                         bond->params.downdelay * bond->params.miimon);
2281                         }
2282                         /*FALLTHRU*/
2283                 case BOND_LINK_FAIL:
2284                         if (link_state) {
2285                                 /*
2286                                  * recovered before downdelay expired
2287                                  */
2288                                 slave->link = BOND_LINK_UP;
2289                                 slave->jiffies = jiffies;
2290                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2291                                         bond->dev->name,
2292                                         (bond->params.downdelay - slave->delay) *
2293                                         bond->params.miimon,
2294                                         slave->dev->name);
2295                                 continue;
2296                         }
2297
2298                         if (slave->delay <= 0) {
2299                                 slave->new_link = BOND_LINK_DOWN;
2300                                 commit++;
2301                                 continue;
2302                         }
2303
2304                         slave->delay--;
2305                         break;
2306
2307                 case BOND_LINK_DOWN:
2308                         if (!link_state)
2309                                 continue;
2310
2311                         slave->link = BOND_LINK_BACK;
2312                         slave->delay = bond->params.updelay;
2313
2314                         if (slave->delay) {
2315                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2316                                         bond->dev->name, slave->dev->name,
2317                                         ignore_updelay ? 0 :
2318                                         bond->params.updelay *
2319                                         bond->params.miimon);
2320                         }
2321                         /*FALLTHRU*/
2322                 case BOND_LINK_BACK:
2323                         if (!link_state) {
2324                                 slave->link = BOND_LINK_DOWN;
2325                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2326                                         bond->dev->name,
2327                                         (bond->params.updelay - slave->delay) *
2328                                         bond->params.miimon,
2329                                         slave->dev->name);
2330
2331                                 continue;
2332                         }
2333
2334                         if (ignore_updelay)
2335                                 slave->delay = 0;
2336
2337                         if (slave->delay <= 0) {
2338                                 slave->new_link = BOND_LINK_UP;
2339                                 commit++;
2340                                 ignore_updelay = false;
2341                                 continue;
2342                         }
2343
2344                         slave->delay--;
2345                         break;
2346                 }
2347         }
2348
2349         return commit;
2350 }
2351
2352 static void bond_miimon_commit(struct bonding *bond)
2353 {
2354         struct slave *slave;
2355         int i;
2356
2357         bond_for_each_slave(bond, slave, i) {
2358                 switch (slave->new_link) {
2359                 case BOND_LINK_NOCHANGE:
2360                         continue;
2361
2362                 case BOND_LINK_UP:
2363                         slave->link = BOND_LINK_UP;
2364                         slave->jiffies = jiffies;
2365
2366                         if (bond->params.mode == BOND_MODE_8023AD) {
2367                                 /* prevent it from being the active one */
2368                                 bond_set_backup_slave(slave);
2369                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2370                                 /* make it immediately active */
2371                                 bond_set_active_slave(slave);
2372                         } else if (slave != bond->primary_slave) {
2373                                 /* prevent it from being the active one */
2374                                 bond_set_backup_slave(slave);
2375                         }
2376
2377                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2378                                 bond->dev->name, slave->dev->name,
2379                                 slave->speed, slave->duplex ? "full" : "half");
2380
2381                         /* notify ad that the link status has changed */
2382                         if (bond->params.mode == BOND_MODE_8023AD)
2383                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2384
2385                         if (bond_is_lb(bond))
2386                                 bond_alb_handle_link_change(bond, slave,
2387                                                             BOND_LINK_UP);
2388
2389                         if (!bond->curr_active_slave ||
2390                             (slave == bond->primary_slave))
2391                                 goto do_failover;
2392
2393                         continue;
2394
2395                 case BOND_LINK_DOWN:
2396                         if (slave->link_failure_count < UINT_MAX)
2397                                 slave->link_failure_count++;
2398
2399                         slave->link = BOND_LINK_DOWN;
2400
2401                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2402                             bond->params.mode == BOND_MODE_8023AD)
2403                                 bond_set_slave_inactive_flags(slave);
2404
2405                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2406                                 bond->dev->name, slave->dev->name);
2407
2408                         if (bond->params.mode == BOND_MODE_8023AD)
2409                                 bond_3ad_handle_link_change(slave,
2410                                                             BOND_LINK_DOWN);
2411
2412                         if (bond_is_lb(bond))
2413                                 bond_alb_handle_link_change(bond, slave,
2414                                                             BOND_LINK_DOWN);
2415
2416                         if (slave == bond->curr_active_slave)
2417                                 goto do_failover;
2418
2419                         continue;
2420
2421                 default:
2422                         pr_err("%s: invalid new link %d on slave %s\n",
2423                                bond->dev->name, slave->new_link,
2424                                slave->dev->name);
2425                         slave->new_link = BOND_LINK_NOCHANGE;
2426
2427                         continue;
2428                 }
2429
2430 do_failover:
2431                 ASSERT_RTNL();
2432                 block_netpoll_tx();
2433                 write_lock_bh(&bond->curr_slave_lock);
2434                 bond_select_active_slave(bond);
2435                 write_unlock_bh(&bond->curr_slave_lock);
2436                 unblock_netpoll_tx();
2437         }
2438
2439         bond_set_carrier(bond);
2440 }
2441
2442 /*
2443  * bond_mii_monitor
2444  *
2445  * Really a wrapper that splits the mii monitor into two phases: an
2446  * inspection, then (if inspection indicates something needs to be done)
2447  * an acquisition of appropriate locks followed by a commit phase to
2448  * implement whatever link state changes are indicated.
2449  */
2450 void bond_mii_monitor(struct work_struct *work)
2451 {
2452         struct bonding *bond = container_of(work, struct bonding,
2453                                             mii_work.work);
2454         bool should_notify_peers = false;
2455         unsigned long delay;
2456
2457         read_lock(&bond->lock);
2458
2459         delay = msecs_to_jiffies(bond->params.miimon);
2460
2461         if (bond->slave_cnt == 0)
2462                 goto re_arm;
2463
2464         should_notify_peers = bond_should_notify_peers(bond);
2465
2466         if (bond_miimon_inspect(bond)) {
2467                 read_unlock(&bond->lock);
2468
2469                 /* Race avoidance with bond_close cancel of workqueue */
2470                 if (!rtnl_trylock()) {
2471                         read_lock(&bond->lock);
2472                         delay = 1;
2473                         should_notify_peers = false;
2474                         goto re_arm;
2475                 }
2476
2477                 read_lock(&bond->lock);
2478
2479                 bond_miimon_commit(bond);
2480
2481                 read_unlock(&bond->lock);
2482                 rtnl_unlock();  /* might sleep, hold no other locks */
2483                 read_lock(&bond->lock);
2484         }
2485
2486 re_arm:
2487         if (bond->params.miimon)
2488                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2489
2490         read_unlock(&bond->lock);
2491
2492         if (should_notify_peers) {
2493                 if (!rtnl_trylock()) {
2494                         read_lock(&bond->lock);
2495                         bond->send_peer_notif++;
2496                         read_unlock(&bond->lock);
2497                         return;
2498                 }
2499                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2500                 rtnl_unlock();
2501         }
2502 }
2503
2504 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2505 {
2506         struct vlan_entry *vlan;
2507         struct net_device *vlan_dev;
2508
2509         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2510                 return 1;
2511
2512         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2513                 rcu_read_lock();
2514                 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2515                 rcu_read_unlock();
2516                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2517                         return 1;
2518         }
2519
2520         return 0;
2521 }
2522
2523 /*
2524  * We go to the (large) trouble of VLAN tagging ARP frames because
2525  * switches in VLAN mode (especially if ports are configured as
2526  * "native" to a VLAN) might not pass non-tagged frames.
2527  */
2528 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2529 {
2530         struct sk_buff *skb;
2531
2532         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2533                  slave_dev->name, dest_ip, src_ip, vlan_id);
2534
2535         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2536                          NULL, slave_dev->dev_addr, NULL);
2537
2538         if (!skb) {
2539                 pr_err("ARP packet allocation failed\n");
2540                 return;
2541         }
2542         if (vlan_id) {
2543                 skb = vlan_put_tag(skb, vlan_id);
2544                 if (!skb) {
2545                         pr_err("failed to insert VLAN tag\n");
2546                         return;
2547                 }
2548         }
2549         arp_xmit(skb);
2550 }
2551
2552
2553 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2554 {
2555         int i, vlan_id;
2556         __be32 *targets = bond->params.arp_targets;
2557         struct vlan_entry *vlan;
2558         struct net_device *vlan_dev = NULL;
2559         struct rtable *rt;
2560
2561         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2562                 __be32 addr;
2563                 if (!targets[i])
2564                         break;
2565                 pr_debug("basa: target %x\n", targets[i]);
2566                 if (!bond_vlan_used(bond)) {
2567                         pr_debug("basa: empty vlan: arp_send\n");
2568                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2569                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2570                                       addr, 0);
2571                         continue;
2572                 }
2573
2574                 /*
2575                  * If VLANs are configured, we do a route lookup to
2576                  * determine which VLAN interface would be used, so we
2577                  * can tag the ARP with the proper VLAN tag.
2578                  */
2579                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2580                                      RTO_ONLINK, 0);
2581                 if (IS_ERR(rt)) {
2582                         if (net_ratelimit()) {
2583                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2584                                            bond->dev->name, &targets[i]);
2585                         }
2586                         continue;
2587                 }
2588
2589                 /*
2590                  * This target is not on a VLAN
2591                  */
2592                 if (rt->dst.dev == bond->dev) {
2593                         ip_rt_put(rt);
2594                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2595                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2596                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2597                                       addr, 0);
2598                         continue;
2599                 }
2600
2601                 vlan_id = 0;
2602                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2603                         rcu_read_lock();
2604                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2605                                                         vlan->vlan_id);
2606                         rcu_read_unlock();
2607                         if (vlan_dev == rt->dst.dev) {
2608                                 vlan_id = vlan->vlan_id;
2609                                 pr_debug("basa: vlan match on %s %d\n",
2610                                        vlan_dev->name, vlan_id);
2611                                 break;
2612                         }
2613                 }
2614
2615                 if (vlan_id && vlan_dev) {
2616                         ip_rt_put(rt);
2617                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2618                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2619                                       addr, vlan_id);
2620                         continue;
2621                 }
2622
2623                 if (net_ratelimit()) {
2624                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2625                                    bond->dev->name, &targets[i],
2626                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2627                 }
2628                 ip_rt_put(rt);
2629         }
2630 }
2631
2632 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2633 {
2634         int i;
2635         __be32 *targets = bond->params.arp_targets;
2636
2637         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2638                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2639                          &sip, &tip, i, &targets[i],
2640                          bond_has_this_ip(bond, tip));
2641                 if (sip == targets[i]) {
2642                         if (bond_has_this_ip(bond, tip))
2643                                 slave->last_arp_rx = jiffies;
2644                         return;
2645                 }
2646         }
2647 }
2648
2649 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2650                         struct slave *slave)
2651 {
2652         struct arphdr *arp = (struct arphdr *)skb->data;
2653         unsigned char *arp_ptr;
2654         __be32 sip, tip;
2655         int alen;
2656
2657         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2658                 return RX_HANDLER_ANOTHER;
2659
2660         read_lock(&bond->lock);
2661         alen = arp_hdr_len(bond->dev);
2662
2663         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2664                  bond->dev->name, skb->dev->name);
2665
2666         if (alen > skb_headlen(skb)) {
2667                 arp = kmalloc(alen, GFP_ATOMIC);
2668                 if (!arp)
2669                         goto out_unlock;
2670                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2671                         goto out_unlock;
2672         }
2673
2674         if (arp->ar_hln != bond->dev->addr_len ||
2675             skb->pkt_type == PACKET_OTHERHOST ||
2676             skb->pkt_type == PACKET_LOOPBACK ||
2677             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2678             arp->ar_pro != htons(ETH_P_IP) ||
2679             arp->ar_pln != 4)
2680                 goto out_unlock;
2681
2682         arp_ptr = (unsigned char *)(arp + 1);
2683         arp_ptr += bond->dev->addr_len;
2684         memcpy(&sip, arp_ptr, 4);
2685         arp_ptr += 4 + bond->dev->addr_len;
2686         memcpy(&tip, arp_ptr, 4);
2687
2688         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2689                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2690                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2691                  &sip, &tip);
2692
2693         /*
2694          * Backup slaves won't see the ARP reply, but do come through
2695          * here for each ARP probe (so we swap the sip/tip to validate
2696          * the probe).  In a "redundant switch, common router" type of
2697          * configuration, the ARP probe will (hopefully) travel from
2698          * the active, through one switch, the router, then the other
2699          * switch before reaching the backup.
2700          */
2701         if (bond_is_active_slave(slave))
2702                 bond_validate_arp(bond, slave, sip, tip);
2703         else
2704                 bond_validate_arp(bond, slave, tip, sip);
2705
2706 out_unlock:
2707         read_unlock(&bond->lock);
2708         if (arp != (struct arphdr *)skb->data)
2709                 kfree(arp);
2710         return RX_HANDLER_ANOTHER;
2711 }
2712
2713 /*
2714  * this function is called regularly to monitor each slave's link
2715  * ensuring that traffic is being sent and received when arp monitoring
2716  * is used in load-balancing mode. if the adapter has been dormant, then an
2717  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2718  * arp monitoring in active backup mode.
2719  */
2720 void bond_loadbalance_arp_mon(struct work_struct *work)
2721 {
2722         struct bonding *bond = container_of(work, struct bonding,
2723                                             arp_work.work);
2724         struct slave *slave, *oldcurrent;
2725         int do_failover = 0;
2726         int delta_in_ticks, extra_ticks;
2727         int i;
2728
2729         read_lock(&bond->lock);
2730
2731         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2732         extra_ticks = delta_in_ticks / 2;
2733
2734         if (bond->slave_cnt == 0)
2735                 goto re_arm;
2736
2737         read_lock(&bond->curr_slave_lock);
2738         oldcurrent = bond->curr_active_slave;
2739         read_unlock(&bond->curr_slave_lock);
2740
2741         /* see if any of the previous devices are up now (i.e. they have
2742          * xmt and rcv traffic). the curr_active_slave does not come into
2743          * the picture unless it is null. also, slave->jiffies is not needed
2744          * here because we send an arp on each slave and give a slave as
2745          * long as it needs to get the tx/rx within the delta.
2746          * TODO: what about up/down delay in arp mode? it wasn't here before
2747          *       so it can wait
2748          */
2749         bond_for_each_slave(bond, slave, i) {
2750                 unsigned long trans_start = dev_trans_start(slave->dev);
2751
2752                 if (slave->link != BOND_LINK_UP) {
2753                         if (time_in_range(jiffies,
2754                                 trans_start - delta_in_ticks,
2755                                 trans_start + delta_in_ticks + extra_ticks) &&
2756                             time_in_range(jiffies,
2757                                 slave->dev->last_rx - delta_in_ticks,
2758                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2759
2760                                 slave->link  = BOND_LINK_UP;
2761                                 bond_set_active_slave(slave);
2762
2763                                 /* primary_slave has no meaning in round-robin
2764                                  * mode. the window of a slave being up and
2765                                  * curr_active_slave being null after enslaving
2766                                  * is closed.
2767                                  */
2768                                 if (!oldcurrent) {
2769                                         pr_info("%s: link status definitely up for interface %s, ",
2770                                                 bond->dev->name,
2771                                                 slave->dev->name);
2772                                         do_failover = 1;
2773                                 } else {
2774                                         pr_info("%s: interface %s is now up\n",
2775                                                 bond->dev->name,
2776                                                 slave->dev->name);
2777                                 }
2778                         }
2779                 } else {
2780                         /* slave->link == BOND_LINK_UP */
2781
2782                         /* not all switches will respond to an arp request
2783                          * when the source ip is 0, so don't take the link down
2784                          * if we don't know our ip yet
2785                          */
2786                         if (!time_in_range(jiffies,
2787                                 trans_start - delta_in_ticks,
2788                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2789                             !time_in_range(jiffies,
2790                                 slave->dev->last_rx - delta_in_ticks,
2791                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2792
2793                                 slave->link  = BOND_LINK_DOWN;
2794                                 bond_set_backup_slave(slave);
2795
2796                                 if (slave->link_failure_count < UINT_MAX)
2797                                         slave->link_failure_count++;
2798
2799                                 pr_info("%s: interface %s is now down.\n",
2800                                         bond->dev->name,
2801                                         slave->dev->name);
2802
2803                                 if (slave == oldcurrent)
2804                                         do_failover = 1;
2805                         }
2806                 }
2807
2808                 /* note: if switch is in round-robin mode, all links
2809                  * must tx arp to ensure all links rx an arp - otherwise
2810                  * links may oscillate or not come up at all; if switch is
2811                  * in something like xor mode, there is nothing we can
2812                  * do - all replies will be rx'ed on same link causing slaves
2813                  * to be unstable during low/no traffic periods
2814                  */
2815                 if (IS_UP(slave->dev))
2816                         bond_arp_send_all(bond, slave);
2817         }
2818
2819         if (do_failover) {
2820                 block_netpoll_tx();
2821                 write_lock_bh(&bond->curr_slave_lock);
2822
2823                 bond_select_active_slave(bond);
2824
2825                 write_unlock_bh(&bond->curr_slave_lock);
2826                 unblock_netpoll_tx();
2827         }
2828
2829 re_arm:
2830         if (bond->params.arp_interval)
2831                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2832
2833         read_unlock(&bond->lock);
2834 }
2835
2836 /*
2837  * Called to inspect slaves for active-backup mode ARP monitor link state
2838  * changes.  Sets new_link in slaves to specify what action should take
2839  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2840  * to link states must be committed.
2841  *
2842  * Called with bond->lock held for read.
2843  */
2844 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2845 {
2846         struct slave *slave;
2847         int i, commit = 0;
2848         unsigned long trans_start;
2849         int extra_ticks;
2850
2851         /* All the time comparisons below need some extra time. Otherwise, on
2852          * fast networks the ARP probe/reply may arrive within the same jiffy
2853          * as it was sent.  Then, the next time the ARP monitor is run, one
2854          * arp_interval will already have passed in the comparisons.
2855          */
2856         extra_ticks = delta_in_ticks / 2;
2857
2858         bond_for_each_slave(bond, slave, i) {
2859                 slave->new_link = BOND_LINK_NOCHANGE;
2860
2861                 if (slave->link != BOND_LINK_UP) {
2862                         if (time_in_range(jiffies,
2863                                 slave_last_rx(bond, slave) - delta_in_ticks,
2864                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2865
2866                                 slave->new_link = BOND_LINK_UP;
2867                                 commit++;
2868                         }
2869
2870                         continue;
2871                 }
2872
2873                 /*
2874                  * Give slaves 2*delta after being enslaved or made
2875                  * active.  This avoids bouncing, as the last receive
2876                  * times need a full ARP monitor cycle to be updated.
2877                  */
2878                 if (time_in_range(jiffies,
2879                                   slave->jiffies - delta_in_ticks,
2880                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2881                         continue;
2882
2883                 /*
2884                  * Backup slave is down if:
2885                  * - No current_arp_slave AND
2886                  * - more than 3*delta since last receive AND
2887                  * - the bond has an IP address
2888                  *
2889                  * Note: a non-null current_arp_slave indicates
2890                  * the curr_active_slave went down and we are
2891                  * searching for a new one; under this condition
2892                  * we only take the curr_active_slave down - this
2893                  * gives each slave a chance to tx/rx traffic
2894                  * before being taken out
2895                  */
2896                 if (!bond_is_active_slave(slave) &&
2897                     !bond->current_arp_slave &&
2898                     !time_in_range(jiffies,
2899                         slave_last_rx(bond, slave) - delta_in_ticks,
2900                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2901
2902                         slave->new_link = BOND_LINK_DOWN;
2903                         commit++;
2904                 }
2905
2906                 /*
2907                  * Active slave is down if:
2908                  * - more than 2*delta since transmitting OR
2909                  * - (more than 2*delta since receive AND
2910                  *    the bond has an IP address)
2911                  */
2912                 trans_start = dev_trans_start(slave->dev);
2913                 if (bond_is_active_slave(slave) &&
2914                     (!time_in_range(jiffies,
2915                         trans_start - delta_in_ticks,
2916                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
2917                      !time_in_range(jiffies,
2918                         slave_last_rx(bond, slave) - delta_in_ticks,
2919                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2920
2921                         slave->new_link = BOND_LINK_DOWN;
2922                         commit++;
2923                 }
2924         }
2925
2926         return commit;
2927 }
2928
2929 /*
2930  * Called to commit link state changes noted by inspection step of
2931  * active-backup mode ARP monitor.
2932  *
2933  * Called with RTNL and bond->lock for read.
2934  */
2935 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2936 {
2937         struct slave *slave;
2938         int i;
2939         unsigned long trans_start;
2940
2941         bond_for_each_slave(bond, slave, i) {
2942                 switch (slave->new_link) {
2943                 case BOND_LINK_NOCHANGE:
2944                         continue;
2945
2946                 case BOND_LINK_UP:
2947                         trans_start = dev_trans_start(slave->dev);
2948                         if ((!bond->curr_active_slave &&
2949                              time_in_range(jiffies,
2950                                            trans_start - delta_in_ticks,
2951                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2952                             bond->curr_active_slave != slave) {
2953                                 slave->link = BOND_LINK_UP;
2954                                 if (bond->current_arp_slave) {
2955                                         bond_set_slave_inactive_flags(
2956                                                 bond->current_arp_slave);
2957                                         bond->current_arp_slave = NULL;
2958                                 }
2959
2960                                 pr_info("%s: link status definitely up for interface %s.\n",
2961                                         bond->dev->name, slave->dev->name);
2962
2963                                 if (!bond->curr_active_slave ||
2964                                     (slave == bond->primary_slave))
2965                                         goto do_failover;
2966
2967                         }
2968
2969                         continue;
2970
2971                 case BOND_LINK_DOWN:
2972                         if (slave->link_failure_count < UINT_MAX)
2973                                 slave->link_failure_count++;
2974
2975                         slave->link = BOND_LINK_DOWN;
2976                         bond_set_slave_inactive_flags(slave);
2977
2978                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2979                                 bond->dev->name, slave->dev->name);
2980
2981                         if (slave == bond->curr_active_slave) {
2982                                 bond->current_arp_slave = NULL;
2983                                 goto do_failover;
2984                         }
2985
2986                         continue;
2987
2988                 default:
2989                         pr_err("%s: impossible: new_link %d on slave %s\n",
2990                                bond->dev->name, slave->new_link,
2991                                slave->dev->name);
2992                         continue;
2993                 }
2994
2995 do_failover:
2996                 ASSERT_RTNL();
2997                 block_netpoll_tx();
2998                 write_lock_bh(&bond->curr_slave_lock);
2999                 bond_select_active_slave(bond);
3000                 write_unlock_bh(&bond->curr_slave_lock);
3001                 unblock_netpoll_tx();
3002         }
3003
3004         bond_set_carrier(bond);
3005 }
3006
3007 /*
3008  * Send ARP probes for active-backup mode ARP monitor.
3009  *
3010  * Called with bond->lock held for read.
3011  */
3012 static void bond_ab_arp_probe(struct bonding *bond)
3013 {
3014         struct slave *slave;
3015         int i;
3016
3017         read_lock(&bond->curr_slave_lock);
3018
3019         if (bond->current_arp_slave && bond->curr_active_slave)
3020                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3021                         bond->current_arp_slave->dev->name,
3022                         bond->curr_active_slave->dev->name);
3023
3024         if (bond->curr_active_slave) {
3025                 bond_arp_send_all(bond, bond->curr_active_slave);
3026                 read_unlock(&bond->curr_slave_lock);
3027                 return;
3028         }
3029
3030         read_unlock(&bond->curr_slave_lock);
3031
3032         /* if we don't have a curr_active_slave, search for the next available
3033          * backup slave from the current_arp_slave and make it the candidate
3034          * for becoming the curr_active_slave
3035          */
3036
3037         if (!bond->current_arp_slave) {
3038                 bond->current_arp_slave = bond->first_slave;
3039                 if (!bond->current_arp_slave)
3040                         return;
3041         }
3042
3043         bond_set_slave_inactive_flags(bond->current_arp_slave);
3044
3045         /* search for next candidate */
3046         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3047                 if (IS_UP(slave->dev)) {
3048                         slave->link = BOND_LINK_BACK;
3049                         bond_set_slave_active_flags(slave);
3050                         bond_arp_send_all(bond, slave);
3051                         slave->jiffies = jiffies;
3052                         bond->current_arp_slave = slave;
3053                         break;
3054                 }
3055
3056                 /* if the link state is up at this point, we
3057                  * mark it down - this can happen if we have
3058                  * simultaneous link failures and
3059                  * reselect_active_interface doesn't make this
3060                  * one the current slave so it is still marked
3061                  * up when it is actually down
3062                  */
3063                 if (slave->link == BOND_LINK_UP) {
3064                         slave->link = BOND_LINK_DOWN;
3065                         if (slave->link_failure_count < UINT_MAX)
3066                                 slave->link_failure_count++;
3067
3068                         bond_set_slave_inactive_flags(slave);
3069
3070                         pr_info("%s: backup interface %s is now down.\n",
3071                                 bond->dev->name, slave->dev->name);
3072                 }
3073         }
3074 }
3075
3076 void bond_activebackup_arp_mon(struct work_struct *work)
3077 {
3078         struct bonding *bond = container_of(work, struct bonding,
3079                                             arp_work.work);
3080         bool should_notify_peers = false;
3081         int delta_in_ticks;
3082
3083         read_lock(&bond->lock);
3084
3085         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3086
3087         if (bond->slave_cnt == 0)
3088                 goto re_arm;
3089
3090         should_notify_peers = bond_should_notify_peers(bond);
3091
3092         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3093                 read_unlock(&bond->lock);
3094
3095                 /* Race avoidance with bond_close flush of workqueue */
3096                 if (!rtnl_trylock()) {
3097                         read_lock(&bond->lock);
3098                         delta_in_ticks = 1;
3099                         should_notify_peers = false;
3100                         goto re_arm;
3101                 }
3102
3103                 read_lock(&bond->lock);
3104
3105                 bond_ab_arp_commit(bond, delta_in_ticks);
3106
3107                 read_unlock(&bond->lock);
3108                 rtnl_unlock();
3109                 read_lock(&bond->lock);
3110         }
3111
3112         bond_ab_arp_probe(bond);
3113
3114 re_arm:
3115         if (bond->params.arp_interval)
3116                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3117
3118         read_unlock(&bond->lock);
3119
3120         if (should_notify_peers) {
3121                 if (!rtnl_trylock()) {
3122                         read_lock(&bond->lock);
3123                         bond->send_peer_notif++;
3124                         read_unlock(&bond->lock);
3125                         return;
3126                 }
3127                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3128                 rtnl_unlock();
3129         }
3130 }
3131
3132 /*-------------------------- netdev event handling --------------------------*/
3133
3134 /*
3135  * Change device name
3136  */
3137 static int bond_event_changename(struct bonding *bond)
3138 {
3139         bond_remove_proc_entry(bond);
3140         bond_create_proc_entry(bond);
3141
3142         bond_debug_reregister(bond);
3143
3144         return NOTIFY_DONE;
3145 }
3146
3147 static int bond_master_netdev_event(unsigned long event,
3148                                     struct net_device *bond_dev)
3149 {
3150         struct bonding *event_bond = netdev_priv(bond_dev);
3151
3152         switch (event) {
3153         case NETDEV_CHANGENAME:
3154                 return bond_event_changename(event_bond);
3155         case NETDEV_UNREGISTER:
3156                 bond_remove_proc_entry(event_bond);
3157                 break;
3158         case NETDEV_REGISTER:
3159                 bond_create_proc_entry(event_bond);
3160                 break;
3161         default:
3162                 break;
3163         }
3164
3165         return NOTIFY_DONE;
3166 }
3167
3168 static int bond_slave_netdev_event(unsigned long event,
3169                                    struct net_device *slave_dev)
3170 {
3171         struct slave *slave = bond_slave_get_rtnl(slave_dev);
3172         struct bonding *bond = slave->bond;
3173         struct net_device *bond_dev = slave->bond->dev;
3174         u32 old_speed;
3175         u8 old_duplex;
3176
3177         switch (event) {
3178         case NETDEV_UNREGISTER:
3179                 if (bond->setup_by_slave)
3180                         bond_release_and_destroy(bond_dev, slave_dev);
3181                 else
3182                         bond_release(bond_dev, slave_dev);
3183                 break;
3184         case NETDEV_UP:
3185         case NETDEV_CHANGE:
3186                 old_speed = slave->speed;
3187                 old_duplex = slave->duplex;
3188
3189                 bond_update_speed_duplex(slave);
3190
3191                 if (bond->params.mode == BOND_MODE_8023AD) {
3192                         if (old_speed != slave->speed)
3193                                 bond_3ad_adapter_speed_changed(slave);
3194                         if (old_duplex != slave->duplex)
3195                                 bond_3ad_adapter_duplex_changed(slave);
3196                 }
3197                 break;
3198         case NETDEV_DOWN:
3199                 /*
3200                  * ... Or is it this?
3201                  */
3202                 break;
3203         case NETDEV_CHANGEMTU:
3204                 /*
3205                  * TODO: Should slaves be allowed to
3206                  * independently alter their MTU?  For
3207                  * an active-backup bond, slaves need
3208                  * not be the same type of device, so
3209                  * MTUs may vary.  For other modes,
3210                  * slaves arguably should have the
3211                  * same MTUs. To do this, we'd need to
3212                  * take over the slave's change_mtu
3213                  * function for the duration of their
3214                  * servitude.
3215                  */
3216                 break;
3217         case NETDEV_CHANGENAME:
3218                 /*
3219                  * TODO: handle changing the primary's name
3220                  */
3221                 break;
3222         case NETDEV_FEAT_CHANGE:
3223                 bond_compute_features(bond);
3224                 break;
3225         default:
3226                 break;
3227         }
3228
3229         return NOTIFY_DONE;
3230 }
3231
3232 /*
3233  * bond_netdev_event: handle netdev notifier chain events.
3234  *
3235  * This function receives events for the netdev chain.  The caller (an
3236  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3237  * locks for us to safely manipulate the slave devices (RTNL lock,
3238  * dev_probe_lock).
3239  */
3240 static int bond_netdev_event(struct notifier_block *this,
3241                              unsigned long event, void *ptr)
3242 {
3243         struct net_device *event_dev = (struct net_device *)ptr;
3244
3245         pr_debug("event_dev: %s, event: %lx\n",
3246                  event_dev ? event_dev->name : "None",
3247                  event);
3248
3249         if (!(event_dev->priv_flags & IFF_BONDING))
3250                 return NOTIFY_DONE;
3251
3252         if (event_dev->flags & IFF_MASTER) {
3253                 pr_debug("IFF_MASTER\n");
3254                 return bond_master_netdev_event(event, event_dev);
3255         }
3256
3257         if (event_dev->flags & IFF_SLAVE) {
3258                 pr_debug("IFF_SLAVE\n");
3259                 return bond_slave_netdev_event(event, event_dev);
3260         }
3261
3262         return NOTIFY_DONE;
3263 }
3264
3265 static struct notifier_block bond_netdev_notifier = {
3266         .notifier_call = bond_netdev_event,
3267 };
3268
3269 /*---------------------------- Hashing Policies -----------------------------*/
3270
3271 /*
3272  * Hash for the output device based upon layer 2 data
3273  */
3274 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3275 {
3276         struct ethhdr *data = (struct ethhdr *)skb->data;
3277
3278         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3279                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3280
3281         return 0;
3282 }
3283
3284 /*
3285  * Hash for the output device based upon layer 2 and layer 3 data. If
3286  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3287  */
3288 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3289 {
3290         struct ethhdr *data = (struct ethhdr *)skb->data;
3291         struct iphdr *iph;
3292         struct ipv6hdr *ipv6h;
3293         u32 v6hash;
3294         __be32 *s, *d;
3295
3296         if (skb->protocol == htons(ETH_P_IP) &&
3297             skb_network_header_len(skb) >= sizeof(*iph)) {
3298                 iph = ip_hdr(skb);
3299                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3300                         (data->h_dest[5] ^ data->h_source[5])) % count;
3301         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3302                    skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3303                 ipv6h = ipv6_hdr(skb);
3304                 s = &ipv6h->saddr.s6_addr32[0];
3305                 d = &ipv6h->daddr.s6_addr32[0];
3306                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3307                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3308                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3309         }
3310
3311         return bond_xmit_hash_policy_l2(skb, count);
3312 }
3313
3314 /*
3315  * Hash for the output device based upon layer 3 and layer 4 data. If
3316  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3317  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3318  */
3319 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3320 {
3321         u32 layer4_xor = 0;
3322         struct iphdr *iph;
3323         struct ipv6hdr *ipv6h;
3324         __be32 *s, *d;
3325         __be16 *layer4hdr;
3326
3327         if (skb->protocol == htons(ETH_P_IP) &&
3328             skb_network_header_len(skb) >= sizeof(*iph)) {
3329                 iph = ip_hdr(skb);
3330                 if (!ip_is_fragment(iph) &&
3331                     (iph->protocol == IPPROTO_TCP ||
3332                      iph->protocol == IPPROTO_UDP) &&
3333                     (skb_headlen(skb) - skb_network_offset(skb) >=
3334                      iph->ihl * sizeof(u32) + sizeof(*layer4hdr) * 2)) {
3335                         layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3336                         layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3337                 }
3338                 return (layer4_xor ^
3339                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3340         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3341                    skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3342                 ipv6h = ipv6_hdr(skb);
3343                 if ((ipv6h->nexthdr == IPPROTO_TCP ||
3344                      ipv6h->nexthdr == IPPROTO_UDP) &&
3345                     (skb_headlen(skb) - skb_network_offset(skb) >=
3346                      sizeof(*ipv6h) + sizeof(*layer4hdr) * 2)) {
3347                         layer4hdr = (__be16 *)(ipv6h + 1);
3348                         layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3349                 }
3350                 s = &ipv6h->saddr.s6_addr32[0];
3351                 d = &ipv6h->daddr.s6_addr32[0];
3352                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3353                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3354                                (layer4_xor >> 8);
3355                 return layer4_xor % count;
3356         }
3357
3358         return bond_xmit_hash_policy_l2(skb, count);
3359 }
3360
3361 /*-------------------------- Device entry points ----------------------------*/
3362
3363 static void bond_work_init_all(struct bonding *bond)
3364 {
3365         INIT_DELAYED_WORK(&bond->mcast_work,
3366                           bond_resend_igmp_join_requests_delayed);
3367         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3368         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3369         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3370                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3371         else
3372                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3373         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3374 }
3375
3376 static void bond_work_cancel_all(struct bonding *bond)
3377 {
3378         cancel_delayed_work_sync(&bond->mii_work);
3379         cancel_delayed_work_sync(&bond->arp_work);
3380         cancel_delayed_work_sync(&bond->alb_work);
3381         cancel_delayed_work_sync(&bond->ad_work);
3382         cancel_delayed_work_sync(&bond->mcast_work);
3383 }
3384
3385 static int bond_open(struct net_device *bond_dev)
3386 {
3387         struct bonding *bond = netdev_priv(bond_dev);
3388         struct slave *slave;
3389         int i;
3390
3391         /* reset slave->backup and slave->inactive */
3392         read_lock(&bond->lock);
3393         if (bond->slave_cnt > 0) {
3394                 read_lock(&bond->curr_slave_lock);
3395                 bond_for_each_slave(bond, slave, i) {
3396                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3397                                 && (slave != bond->curr_active_slave)) {
3398                                 bond_set_slave_inactive_flags(slave);
3399                         } else {
3400                                 bond_set_slave_active_flags(slave);
3401                         }
3402                 }
3403                 read_unlock(&bond->curr_slave_lock);
3404         }
3405         read_unlock(&bond->lock);
3406
3407         bond_work_init_all(bond);
3408
3409         if (bond_is_lb(bond)) {
3410                 /* bond_alb_initialize must be called before the timer
3411                  * is started.
3412                  */
3413                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3414                         return -ENOMEM;
3415                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3416         }
3417
3418         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3419                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3420
3421         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3422                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3423                 if (bond->params.arp_validate)
3424                         bond->recv_probe = bond_arp_rcv;
3425         }
3426
3427         if (bond->params.mode == BOND_MODE_8023AD) {
3428                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3429                 /* register to receive LACPDUs */
3430                 bond->recv_probe = bond_3ad_lacpdu_recv;
3431                 bond_3ad_initiate_agg_selection(bond, 1);
3432         }
3433
3434         return 0;
3435 }
3436
3437 static int bond_close(struct net_device *bond_dev)
3438 {
3439         struct bonding *bond = netdev_priv(bond_dev);
3440
3441         write_lock_bh(&bond->lock);
3442         bond->send_peer_notif = 0;
3443         write_unlock_bh(&bond->lock);
3444
3445         bond_work_cancel_all(bond);
3446         if (bond_is_lb(bond)) {
3447                 /* Must be called only after all
3448                  * slaves have been released
3449                  */
3450                 bond_alb_deinitialize(bond);
3451         }
3452         bond->recv_probe = NULL;
3453
3454         return 0;
3455 }
3456
3457 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3458                                                 struct rtnl_link_stats64 *stats)
3459 {
3460         struct bonding *bond = netdev_priv(bond_dev);
3461         struct rtnl_link_stats64 temp;
3462         struct slave *slave;
3463         int i;
3464
3465         memset(stats, 0, sizeof(*stats));
3466
3467         read_lock_bh(&bond->lock);
3468
3469         bond_for_each_slave(bond, slave, i) {
3470                 const struct rtnl_link_stats64 *sstats =
3471                         dev_get_stats(slave->dev, &temp);
3472
3473                 stats->rx_packets += sstats->rx_packets;
3474                 stats->rx_bytes += sstats->rx_bytes;
3475                 stats->rx_errors += sstats->rx_errors;
3476                 stats->rx_dropped += sstats->rx_dropped;
3477
3478                 stats->tx_packets += sstats->tx_packets;
3479                 stats->tx_bytes += sstats->tx_bytes;
3480                 stats->tx_errors += sstats->tx_errors;
3481                 stats->tx_dropped += sstats->tx_dropped;
3482
3483                 stats->multicast += sstats->multicast;
3484                 stats->collisions += sstats->collisions;
3485
3486                 stats->rx_length_errors += sstats->rx_length_errors;
3487                 stats->rx_over_errors += sstats->rx_over_errors;
3488                 stats->rx_crc_errors += sstats->rx_crc_errors;
3489                 stats->rx_frame_errors += sstats->rx_frame_errors;
3490                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3491                 stats->rx_missed_errors += sstats->rx_missed_errors;
3492
3493                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3494                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3495                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3496                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3497                 stats->tx_window_errors += sstats->tx_window_errors;
3498         }
3499
3500         read_unlock_bh(&bond->lock);
3501
3502         return stats;
3503 }
3504
3505 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3506 {
3507         struct net_device *slave_dev = NULL;
3508         struct ifbond k_binfo;
3509         struct ifbond __user *u_binfo = NULL;
3510         struct ifslave k_sinfo;
3511         struct ifslave __user *u_sinfo = NULL;
3512         struct mii_ioctl_data *mii = NULL;
3513         struct net *net;
3514         int res = 0;
3515
3516         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3517
3518         switch (cmd) {
3519         case SIOCGMIIPHY:
3520                 mii = if_mii(ifr);
3521                 if (!mii)
3522                         return -EINVAL;
3523
3524                 mii->phy_id = 0;
3525                 /* Fall Through */
3526         case SIOCGMIIREG:
3527                 /*
3528                  * We do this again just in case we were called by SIOCGMIIREG
3529                  * instead of SIOCGMIIPHY.
3530                  */
3531                 mii = if_mii(ifr);
3532                 if (!mii)
3533                         return -EINVAL;
3534
3535
3536                 if (mii->reg_num == 1) {
3537                         struct bonding *bond = netdev_priv(bond_dev);
3538                         mii->val_out = 0;
3539                         read_lock(&bond->lock);
3540                         read_lock(&bond->curr_slave_lock);
3541                         if (netif_carrier_ok(bond->dev))
3542                                 mii->val_out = BMSR_LSTATUS;
3543
3544                         read_unlock(&bond->curr_slave_lock);
3545                         read_unlock(&bond->lock);
3546                 }
3547
3548                 return 0;
3549         case BOND_INFO_QUERY_OLD:
3550         case SIOCBONDINFOQUERY:
3551                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3552
3553                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3554                         return -EFAULT;
3555
3556                 res = bond_info_query(bond_dev, &k_binfo);
3557                 if (res == 0 &&
3558                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3559                         return -EFAULT;
3560
3561                 return res;
3562         case BOND_SLAVE_INFO_QUERY_OLD:
3563         case SIOCBONDSLAVEINFOQUERY:
3564                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3565
3566                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3567                         return -EFAULT;
3568
3569                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3570                 if (res == 0 &&
3571                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3572                         return -EFAULT;
3573
3574                 return res;
3575         default:
3576                 /* Go on */
3577                 break;
3578         }
3579
3580         net = dev_net(bond_dev);
3581
3582         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3583                 return -EPERM;
3584
3585         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3586
3587         pr_debug("slave_dev=%p:\n", slave_dev);
3588
3589         if (!slave_dev)
3590                 res = -ENODEV;
3591         else {
3592                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3593                 switch (cmd) {
3594                 case BOND_ENSLAVE_OLD:
3595                 case SIOCBONDENSLAVE:
3596                         res = bond_enslave(bond_dev, slave_dev);
3597                         break;
3598                 case BOND_RELEASE_OLD:
3599                 case SIOCBONDRELEASE:
3600                         res = bond_release(bond_dev, slave_dev);
3601                         break;
3602                 case BOND_SETHWADDR_OLD:
3603                 case SIOCBONDSETHWADDR:
3604                         bond_set_dev_addr(bond_dev, slave_dev);
3605                         res = 0;
3606                         break;
3607                 case BOND_CHANGE_ACTIVE_OLD:
3608                 case SIOCBONDCHANGEACTIVE:
3609                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3610                         break;
3611                 default:
3612                         res = -EOPNOTSUPP;
3613                 }
3614
3615                 dev_put(slave_dev);
3616         }
3617
3618         return res;
3619 }
3620
3621 static bool bond_addr_in_mc_list(unsigned char *addr,
3622                                  struct netdev_hw_addr_list *list,
3623                                  int addrlen)
3624 {
3625         struct netdev_hw_addr *ha;
3626
3627         netdev_hw_addr_list_for_each(ha, list)
3628                 if (!memcmp(ha->addr, addr, addrlen))
3629                         return true;
3630
3631         return false;
3632 }
3633
3634 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3635 {
3636         struct bonding *bond = netdev_priv(bond_dev);
3637
3638         if (change & IFF_PROMISC)
3639                 bond_set_promiscuity(bond,
3640                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3641
3642         if (change & IFF_ALLMULTI)
3643                 bond_set_allmulti(bond,
3644                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3645 }
3646
3647 static void bond_set_multicast_list(struct net_device *bond_dev)
3648 {
3649         struct bonding *bond = netdev_priv(bond_dev);
3650         struct netdev_hw_addr *ha;
3651         bool found;
3652
3653         read_lock(&bond->lock);
3654
3655         /* looking for addresses to add to slaves' mc list */
3656         netdev_for_each_mc_addr(ha, bond_dev) {
3657                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3658                                              bond_dev->addr_len);
3659                 if (!found)
3660                         bond_mc_add(bond, ha->addr);
3661         }
3662
3663         /* looking for addresses to delete from slaves' list */
3664         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3665                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3666                                              bond_dev->addr_len);
3667                 if (!found)
3668                         bond_mc_del(bond, ha->addr);
3669         }
3670
3671         /* save master's multicast list */
3672         __hw_addr_flush(&bond->mc_list);
3673         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3674                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3675
3676         read_unlock(&bond->lock);
3677 }
3678
3679 static int bond_neigh_init(struct neighbour *n)
3680 {
3681         struct bonding *bond = netdev_priv(n->dev);
3682         struct slave *slave = bond->first_slave;
3683         const struct net_device_ops *slave_ops;
3684         struct neigh_parms parms;
3685         int ret;
3686
3687         if (!slave)
3688                 return 0;
3689
3690         slave_ops = slave->dev->netdev_ops;
3691
3692         if (!slave_ops->ndo_neigh_setup)
3693                 return 0;
3694
3695         parms.neigh_setup = NULL;
3696         parms.neigh_cleanup = NULL;
3697         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3698         if (ret)
3699                 return ret;
3700
3701         /*
3702          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3703          * after the last slave has been detached.  Assumes that all slaves
3704          * utilize the same neigh_cleanup (true at this writing as only user
3705          * is ipoib).
3706          */
3707         n->parms->neigh_cleanup = parms.neigh_cleanup;
3708
3709         if (!parms.neigh_setup)
3710                 return 0;
3711
3712         return parms.neigh_setup(n);
3713 }
3714
3715 /*
3716  * The bonding ndo_neigh_setup is called at init time beofre any
3717  * slave exists. So we must declare proxy setup function which will
3718  * be used at run time to resolve the actual slave neigh param setup.
3719  */
3720 static int bond_neigh_setup(struct net_device *dev,
3721                             struct neigh_parms *parms)
3722 {
3723         parms->neigh_setup   = bond_neigh_init;
3724
3725         return 0;
3726 }
3727
3728 /*
3729  * Change the MTU of all of a master's slaves to match the master
3730  */
3731 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3732 {
3733         struct bonding *bond = netdev_priv(bond_dev);
3734         struct slave *slave, *stop_at;
3735         int res = 0;
3736         int i;
3737
3738         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3739                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3740
3741         /* Can't hold bond->lock with bh disabled here since
3742          * some base drivers panic. On the other hand we can't
3743          * hold bond->lock without bh disabled because we'll
3744          * deadlock. The only solution is to rely on the fact
3745          * that we're under rtnl_lock here, and the slaves
3746          * list won't change. This doesn't solve the problem
3747          * of setting the slave's MTU while it is
3748          * transmitting, but the assumption is that the base
3749          * driver can handle that.
3750          *
3751          * TODO: figure out a way to safely iterate the slaves
3752          * list, but without holding a lock around the actual
3753          * call to the base driver.
3754          */
3755
3756         bond_for_each_slave(bond, slave, i) {
3757                 pr_debug("s %p s->p %p c_m %p\n",
3758                          slave,
3759                          slave->prev,
3760                          slave->dev->netdev_ops->ndo_change_mtu);
3761
3762                 res = dev_set_mtu(slave->dev, new_mtu);
3763
3764                 if (res) {
3765                         /* If we failed to set the slave's mtu to the new value
3766                          * we must abort the operation even in ACTIVE_BACKUP
3767                          * mode, because if we allow the backup slaves to have
3768                          * different mtu values than the active slave we'll
3769                          * need to change their mtu when doing a failover. That
3770                          * means changing their mtu from timer context, which
3771                          * is probably not a good idea.
3772                          */
3773                         pr_debug("err %d %s\n", res, slave->dev->name);
3774                         goto unwind;
3775                 }
3776         }
3777
3778         bond_dev->mtu = new_mtu;
3779
3780         return 0;
3781
3782 unwind:
3783         /* unwind from head to the slave that failed */
3784         stop_at = slave;
3785         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3786                 int tmp_res;
3787
3788                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3789                 if (tmp_res) {
3790                         pr_debug("unwind err %d dev %s\n",
3791                                  tmp_res, slave->dev->name);
3792                 }
3793         }
3794
3795         return res;
3796 }
3797
3798 /*
3799  * Change HW address
3800  *
3801  * Note that many devices must be down to change the HW address, and
3802  * downing the master releases all slaves.  We can make bonds full of
3803  * bonding devices to test this, however.
3804  */
3805 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3806 {
3807         struct bonding *bond = netdev_priv(bond_dev);
3808         struct sockaddr *sa = addr, tmp_sa;
3809         struct slave *slave, *stop_at;
3810         int res = 0;
3811         int i;
3812
3813         if (bond->params.mode == BOND_MODE_ALB)
3814                 return bond_alb_set_mac_address(bond_dev, addr);
3815
3816
3817         pr_debug("bond=%p, name=%s\n",
3818                  bond, bond_dev ? bond_dev->name : "None");
3819
3820         /*
3821          * If fail_over_mac is set to active, do nothing and return
3822          * success.  Returning an error causes ifenslave to fail.
3823          */
3824         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3825                 return 0;
3826
3827         if (!is_valid_ether_addr(sa->sa_data))
3828                 return -EADDRNOTAVAIL;
3829
3830         /* Can't hold bond->lock with bh disabled here since
3831          * some base drivers panic. On the other hand we can't
3832          * hold bond->lock without bh disabled because we'll
3833          * deadlock. The only solution is to rely on the fact
3834          * that we're under rtnl_lock here, and the slaves
3835          * list won't change. This doesn't solve the problem
3836          * of setting the slave's hw address while it is
3837          * transmitting, but the assumption is that the base
3838          * driver can handle that.
3839          *
3840          * TODO: figure out a way to safely iterate the slaves
3841          * list, but without holding a lock around the actual
3842          * call to the base driver.
3843          */
3844
3845         bond_for_each_slave(bond, slave, i) {
3846                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3847                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3848
3849                 if (slave_ops->ndo_set_mac_address == NULL) {
3850                         res = -EOPNOTSUPP;
3851                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3852                         goto unwind;
3853                 }
3854
3855                 res = dev_set_mac_address(slave->dev, addr);
3856                 if (res) {
3857                         /* TODO: consider downing the slave
3858                          * and retry ?
3859                          * User should expect communications
3860                          * breakage anyway until ARP finish
3861                          * updating, so...
3862                          */
3863                         pr_debug("err %d %s\n", res, slave->dev->name);
3864                         goto unwind;
3865                 }
3866         }
3867
3868         /* success */
3869         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3870         return 0;
3871
3872 unwind:
3873         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3874         tmp_sa.sa_family = bond_dev->type;
3875
3876         /* unwind from head to the slave that failed */
3877         stop_at = slave;
3878         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3879                 int tmp_res;
3880
3881                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3882                 if (tmp_res) {
3883                         pr_debug("unwind err %d dev %s\n",
3884                                  tmp_res, slave->dev->name);
3885                 }
3886         }
3887
3888         return res;
3889 }
3890
3891 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3892 {
3893         struct bonding *bond = netdev_priv(bond_dev);
3894         struct slave *slave, *start_at;
3895         int i, slave_no, res = 1;
3896         struct iphdr *iph = ip_hdr(skb);
3897
3898         /*
3899          * Start with the curr_active_slave that joined the bond as the
3900          * default for sending IGMP traffic.  For failover purposes one
3901          * needs to maintain some consistency for the interface that will
3902          * send the join/membership reports.  The curr_active_slave found
3903          * will send all of this type of traffic.
3904          */
3905         if ((iph->protocol == IPPROTO_IGMP) &&
3906             (skb->protocol == htons(ETH_P_IP))) {
3907
3908                 read_lock(&bond->curr_slave_lock);
3909                 slave = bond->curr_active_slave;
3910                 read_unlock(&bond->curr_slave_lock);
3911
3912                 if (!slave)
3913                         goto out;
3914         } else {
3915                 /*
3916                  * Concurrent TX may collide on rr_tx_counter; we accept
3917                  * that as being rare enough not to justify using an
3918                  * atomic op here.
3919                  */
3920                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3921
3922                 bond_for_each_slave(bond, slave, i) {
3923                         slave_no--;
3924                         if (slave_no < 0)
3925                                 break;
3926                 }
3927         }
3928
3929         start_at = slave;
3930         bond_for_each_slave_from(bond, slave, i, start_at) {
3931                 if (IS_UP(slave->dev) &&
3932                     (slave->link == BOND_LINK_UP) &&
3933                     bond_is_active_slave(slave)) {
3934                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3935                         break;
3936                 }
3937         }
3938
3939 out:
3940         if (res) {
3941                 /* no suitable interface, frame not sent */
3942                 kfree_skb(skb);
3943         }
3944
3945         return NETDEV_TX_OK;
3946 }
3947
3948
3949 /*
3950  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3951  * the bond has a usable interface.
3952  */
3953 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3954 {
3955         struct bonding *bond = netdev_priv(bond_dev);
3956         int res = 1;
3957
3958         read_lock(&bond->curr_slave_lock);
3959
3960         if (bond->curr_active_slave)
3961                 res = bond_dev_queue_xmit(bond, skb,
3962                         bond->curr_active_slave->dev);
3963
3964         read_unlock(&bond->curr_slave_lock);
3965
3966         if (res)
3967                 /* no suitable interface, frame not sent */
3968                 kfree_skb(skb);
3969
3970         return NETDEV_TX_OK;
3971 }
3972
3973 /*
3974  * In bond_xmit_xor() , we determine the output device by using a pre-
3975  * determined xmit_hash_policy(), If the selected device is not enabled,
3976  * find the next active slave.
3977  */
3978 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3979 {
3980         struct bonding *bond = netdev_priv(bond_dev);
3981         struct slave *slave, *start_at;
3982         int slave_no;
3983         int i;
3984         int res = 1;
3985
3986         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
3987
3988         bond_for_each_slave(bond, slave, i) {
3989                 slave_no--;
3990                 if (slave_no < 0)
3991                         break;
3992         }
3993
3994         start_at = slave;
3995
3996         bond_for_each_slave_from(bond, slave, i, start_at) {
3997                 if (IS_UP(slave->dev) &&
3998                     (slave->link == BOND_LINK_UP) &&
3999                     bond_is_active_slave(slave)) {
4000                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4001                         break;
4002                 }
4003         }
4004
4005         if (res) {
4006                 /* no suitable interface, frame not sent */
4007                 kfree_skb(skb);
4008         }
4009
4010         return NETDEV_TX_OK;
4011 }
4012
4013 /*
4014  * in broadcast mode, we send everything to all usable interfaces.
4015  */
4016 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4017 {
4018         struct bonding *bond = netdev_priv(bond_dev);
4019         struct slave *slave, *start_at;
4020         struct net_device *tx_dev = NULL;
4021         int i;
4022         int res = 1;
4023
4024         read_lock(&bond->curr_slave_lock);
4025         start_at = bond->curr_active_slave;
4026         read_unlock(&bond->curr_slave_lock);
4027
4028         if (!start_at)
4029                 goto out;
4030
4031         bond_for_each_slave_from(bond, slave, i, start_at) {
4032                 if (IS_UP(slave->dev) &&
4033                     (slave->link == BOND_LINK_UP) &&
4034                     bond_is_active_slave(slave)) {
4035                         if (tx_dev) {
4036                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4037                                 if (!skb2) {
4038                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4039                                                bond_dev->name);
4040                                         continue;
4041                                 }
4042
4043                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4044                                 if (res) {
4045                                         kfree_skb(skb2);
4046                                         continue;
4047                                 }
4048                         }
4049                         tx_dev = slave->dev;
4050                 }
4051         }
4052
4053         if (tx_dev)
4054                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4055
4056 out:
4057         if (res)
4058                 /* no suitable interface, frame not sent */
4059                 kfree_skb(skb);
4060
4061         /* frame sent to all suitable interfaces */
4062         return NETDEV_TX_OK;
4063 }
4064
4065 /*------------------------- Device initialization ---------------------------*/
4066
4067 static void bond_set_xmit_hash_policy(struct bonding *bond)
4068 {
4069         switch (bond->params.xmit_policy) {
4070         case BOND_XMIT_POLICY_LAYER23:
4071                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4072                 break;
4073         case BOND_XMIT_POLICY_LAYER34:
4074                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4075                 break;
4076         case BOND_XMIT_POLICY_LAYER2:
4077         default:
4078                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4079                 break;
4080         }
4081 }
4082
4083 /*
4084  * Lookup the slave that corresponds to a qid
4085  */
4086 static inline int bond_slave_override(struct bonding *bond,
4087                                       struct sk_buff *skb)
4088 {
4089         int i, res = 1;
4090         struct slave *slave = NULL;
4091         struct slave *check_slave;
4092
4093         if (!skb->queue_mapping)
4094                 return 1;
4095
4096         /* Find out if any slaves have the same mapping as this skb. */
4097         bond_for_each_slave(bond, check_slave, i) {
4098                 if (check_slave->queue_id == skb->queue_mapping) {
4099                         slave = check_slave;
4100                         break;
4101                 }
4102         }
4103
4104         /* If the slave isn't UP, use default transmit policy. */
4105         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4106             (slave->link == BOND_LINK_UP)) {
4107                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4108         }
4109
4110         return res;
4111 }
4112
4113
4114 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4115 {
4116         /*
4117          * This helper function exists to help dev_pick_tx get the correct
4118          * destination queue.  Using a helper function skips a call to
4119          * skb_tx_hash and will put the skbs in the queue we expect on their
4120          * way down to the bonding driver.
4121          */
4122         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4123
4124         /*
4125          * Save the original txq to restore before passing to the driver
4126          */
4127         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4128
4129         if (unlikely(txq >= dev->real_num_tx_queues)) {
4130                 do {
4131                         txq -= dev->real_num_tx_queues;
4132                 } while (txq >= dev->real_num_tx_queues);
4133         }
4134         return txq;
4135 }
4136
4137 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4138 {
4139         struct bonding *bond = netdev_priv(dev);
4140
4141         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4142                 if (!bond_slave_override(bond, skb))
4143                         return NETDEV_TX_OK;
4144         }
4145
4146         switch (bond->params.mode) {
4147         case BOND_MODE_ROUNDROBIN:
4148                 return bond_xmit_roundrobin(skb, dev);
4149         case BOND_MODE_ACTIVEBACKUP:
4150                 return bond_xmit_activebackup(skb, dev);
4151         case BOND_MODE_XOR:
4152                 return bond_xmit_xor(skb, dev);
4153         case BOND_MODE_BROADCAST:
4154                 return bond_xmit_broadcast(skb, dev);
4155         case BOND_MODE_8023AD:
4156                 return bond_3ad_xmit_xor(skb, dev);
4157         case BOND_MODE_ALB:
4158         case BOND_MODE_TLB:
4159                 return bond_alb_xmit(skb, dev);
4160         default:
4161                 /* Should never happen, mode already checked */
4162                 pr_err("%s: Error: Unknown bonding mode %d\n",
4163                        dev->name, bond->params.mode);
4164                 WARN_ON_ONCE(1);
4165                 kfree_skb(skb);
4166                 return NETDEV_TX_OK;
4167         }
4168 }
4169
4170 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4171 {
4172         struct bonding *bond = netdev_priv(dev);
4173         netdev_tx_t ret = NETDEV_TX_OK;
4174
4175         /*
4176          * If we risk deadlock from transmitting this in the
4177          * netpoll path, tell netpoll to queue the frame for later tx
4178          */
4179         if (is_netpoll_tx_blocked(dev))
4180                 return NETDEV_TX_BUSY;
4181
4182         read_lock(&bond->lock);
4183
4184         if (bond->slave_cnt)
4185                 ret = __bond_start_xmit(skb, dev);
4186         else
4187                 kfree_skb(skb);
4188
4189         read_unlock(&bond->lock);
4190
4191         return ret;
4192 }
4193
4194 /*
4195  * set bond mode specific net device operations
4196  */
4197 void bond_set_mode_ops(struct bonding *bond, int mode)
4198 {
4199         struct net_device *bond_dev = bond->dev;
4200
4201         switch (mode) {
4202         case BOND_MODE_ROUNDROBIN:
4203                 break;
4204         case BOND_MODE_ACTIVEBACKUP:
4205                 break;
4206         case BOND_MODE_XOR:
4207                 bond_set_xmit_hash_policy(bond);
4208                 break;
4209         case BOND_MODE_BROADCAST:
4210                 break;
4211         case BOND_MODE_8023AD:
4212                 bond_set_xmit_hash_policy(bond);
4213                 break;
4214         case BOND_MODE_ALB:
4215                 /* FALLTHRU */
4216         case BOND_MODE_TLB:
4217                 break;
4218         default:
4219                 /* Should never happen, mode already checked */
4220                 pr_err("%s: Error: Unknown bonding mode %d\n",
4221                        bond_dev->name, mode);
4222                 break;
4223         }
4224 }
4225
4226 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4227                                      struct ethtool_drvinfo *drvinfo)
4228 {
4229         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4230         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4231         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4232                  BOND_ABI_VERSION);
4233 }
4234
4235 static const struct ethtool_ops bond_ethtool_ops = {
4236         .get_drvinfo            = bond_ethtool_get_drvinfo,
4237         .get_link               = ethtool_op_get_link,
4238 };
4239
4240 static const struct net_device_ops bond_netdev_ops = {
4241         .ndo_init               = bond_init,
4242         .ndo_uninit             = bond_uninit,
4243         .ndo_open               = bond_open,
4244         .ndo_stop               = bond_close,
4245         .ndo_start_xmit         = bond_start_xmit,
4246         .ndo_select_queue       = bond_select_queue,
4247         .ndo_get_stats64        = bond_get_stats,
4248         .ndo_do_ioctl           = bond_do_ioctl,
4249         .ndo_change_rx_flags    = bond_change_rx_flags,
4250         .ndo_set_rx_mode        = bond_set_multicast_list,
4251         .ndo_change_mtu         = bond_change_mtu,
4252         .ndo_set_mac_address    = bond_set_mac_address,
4253         .ndo_neigh_setup        = bond_neigh_setup,
4254         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4255         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4256 #ifdef CONFIG_NET_POLL_CONTROLLER
4257         .ndo_netpoll_setup      = bond_netpoll_setup,
4258         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4259         .ndo_poll_controller    = bond_poll_controller,
4260 #endif
4261         .ndo_add_slave          = bond_enslave,
4262         .ndo_del_slave          = bond_release,
4263         .ndo_fix_features       = bond_fix_features,
4264 };
4265
4266 static const struct device_type bond_type = {
4267         .name = "bond",
4268 };
4269
4270 static void bond_destructor(struct net_device *bond_dev)
4271 {
4272         struct bonding *bond = netdev_priv(bond_dev);
4273         if (bond->wq)
4274                 destroy_workqueue(bond->wq);
4275         free_netdev(bond_dev);
4276 }
4277
4278 static void bond_setup(struct net_device *bond_dev)
4279 {
4280         struct bonding *bond = netdev_priv(bond_dev);
4281
4282         /* initialize rwlocks */
4283         rwlock_init(&bond->lock);
4284         rwlock_init(&bond->curr_slave_lock);
4285
4286         bond->params = bonding_defaults;
4287
4288         /* Initialize pointers */
4289         bond->dev = bond_dev;
4290         INIT_LIST_HEAD(&bond->vlan_list);
4291
4292         /* Initialize the device entry points */
4293         ether_setup(bond_dev);
4294         bond_dev->netdev_ops = &bond_netdev_ops;
4295         bond_dev->ethtool_ops = &bond_ethtool_ops;
4296         bond_set_mode_ops(bond, bond->params.mode);
4297
4298         bond_dev->destructor = bond_destructor;
4299
4300         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4301
4302         /* Initialize the device options */
4303         bond_dev->tx_queue_len = 0;
4304         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4305         bond_dev->priv_flags |= IFF_BONDING;
4306         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4307
4308         /* At first, we block adding VLANs. That's the only way to
4309          * prevent problems that occur when adding VLANs over an
4310          * empty bond. The block will be removed once non-challenged
4311          * slaves are enslaved.
4312          */
4313         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4314
4315         /* don't acquire bond device's netif_tx_lock when
4316          * transmitting */
4317         bond_dev->features |= NETIF_F_LLTX;
4318
4319         /* By default, we declare the bond to be fully
4320          * VLAN hardware accelerated capable. Special
4321          * care is taken in the various xmit functions
4322          * when there are slaves that are not hw accel
4323          * capable
4324          */
4325
4326         bond_dev->hw_features = BOND_VLAN_FEATURES |
4327                                 NETIF_F_HW_VLAN_TX |
4328                                 NETIF_F_HW_VLAN_RX |
4329                                 NETIF_F_HW_VLAN_FILTER;
4330
4331         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4332         bond_dev->features |= bond_dev->hw_features;
4333 }
4334
4335 /*
4336 * Destroy a bonding device.
4337 * Must be under rtnl_lock when this function is called.
4338 */
4339 static void bond_uninit(struct net_device *bond_dev)
4340 {
4341         struct bonding *bond = netdev_priv(bond_dev);
4342         struct vlan_entry *vlan, *tmp;
4343
4344         bond_netpoll_cleanup(bond_dev);
4345
4346         /* Release the bonded slaves */
4347         while (bond->first_slave != NULL)
4348                 __bond_release_one(bond_dev, bond->first_slave->dev, true);
4349         pr_info("%s: released all slaves\n", bond_dev->name);
4350
4351         list_del(&bond->bond_list);
4352
4353         bond_debug_unregister(bond);
4354
4355         __hw_addr_flush(&bond->mc_list);
4356
4357         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4358                 list_del(&vlan->vlan_list);
4359                 kfree(vlan);
4360         }
4361 }
4362
4363 /*------------------------- Module initialization ---------------------------*/
4364
4365 /*
4366  * Convert string input module parms.  Accept either the
4367  * number of the mode or its string name.  A bit complicated because
4368  * some mode names are substrings of other names, and calls from sysfs
4369  * may have whitespace in the name (trailing newlines, for example).
4370  */
4371 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4372 {
4373         int modeint = -1, i, rv;
4374         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4375
4376         for (p = (char *)buf; *p; p++)
4377                 if (!(isdigit(*p) || isspace(*p)))
4378                         break;
4379
4380         if (*p)
4381                 rv = sscanf(buf, "%20s", modestr);
4382         else
4383                 rv = sscanf(buf, "%d", &modeint);
4384
4385         if (!rv)
4386                 return -1;
4387
4388         for (i = 0; tbl[i].modename; i++) {
4389                 if (modeint == tbl[i].mode)
4390                         return tbl[i].mode;
4391                 if (strcmp(modestr, tbl[i].modename) == 0)
4392                         return tbl[i].mode;
4393         }
4394
4395         return -1;
4396 }
4397
4398 static int bond_check_params(struct bond_params *params)
4399 {
4400         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4401
4402         /*
4403          * Convert string parameters.
4404          */
4405         if (mode) {
4406                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4407                 if (bond_mode == -1) {
4408                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4409                                mode == NULL ? "NULL" : mode);
4410                         return -EINVAL;
4411                 }
4412         }
4413
4414         if (xmit_hash_policy) {
4415                 if ((bond_mode != BOND_MODE_XOR) &&
4416                     (bond_mode != BOND_MODE_8023AD)) {
4417                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4418                                bond_mode_name(bond_mode));
4419                 } else {
4420                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4421                                                         xmit_hashtype_tbl);
4422                         if (xmit_hashtype == -1) {
4423                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4424                                        xmit_hash_policy == NULL ? "NULL" :
4425                                        xmit_hash_policy);
4426                                 return -EINVAL;
4427                         }
4428                 }
4429         }
4430
4431         if (lacp_rate) {
4432                 if (bond_mode != BOND_MODE_8023AD) {
4433                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4434                                 bond_mode_name(bond_mode));
4435                 } else {
4436                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4437                         if (lacp_fast == -1) {
4438                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4439                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4440                                 return -EINVAL;
4441                         }
4442                 }
4443         }
4444
4445         if (ad_select) {
4446                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4447                 if (params->ad_select == -1) {
4448                         pr_err("Error: Invalid ad_select \"%s\"\n",
4449                                ad_select == NULL ? "NULL" : ad_select);
4450                         return -EINVAL;
4451                 }
4452
4453                 if (bond_mode != BOND_MODE_8023AD) {
4454                         pr_warning("ad_select param only affects 802.3ad mode\n");
4455                 }
4456         } else {
4457                 params->ad_select = BOND_AD_STABLE;
4458         }
4459
4460         if (max_bonds < 0) {
4461                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4462                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4463                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4464         }
4465
4466         if (miimon < 0) {
4467                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4468                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4469                 miimon = BOND_LINK_MON_INTERV;
4470         }
4471
4472         if (updelay < 0) {
4473                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4474                            updelay, INT_MAX);
4475                 updelay = 0;
4476         }
4477
4478         if (downdelay < 0) {
4479                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4480                            downdelay, INT_MAX);
4481                 downdelay = 0;
4482         }
4483
4484         if ((use_carrier != 0) && (use_carrier != 1)) {
4485                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4486                            use_carrier);
4487                 use_carrier = 1;
4488         }
4489
4490         if (num_peer_notif < 0 || num_peer_notif > 255) {
4491                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4492                            num_peer_notif);
4493                 num_peer_notif = 1;
4494         }
4495
4496         /* reset values for 802.3ad */
4497         if (bond_mode == BOND_MODE_8023AD) {
4498                 if (!miimon) {
4499                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4500                         pr_warning("Forcing miimon to 100msec\n");
4501                         miimon = 100;
4502                 }
4503         }
4504
4505         if (tx_queues < 1 || tx_queues > 255) {
4506                 pr_warning("Warning: tx_queues (%d) should be between "
4507                            "1 and 255, resetting to %d\n",
4508                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4509                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4510         }
4511
4512         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4513                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4514                            "not of valid value (0/1), so it was set to "
4515                            "0\n", all_slaves_active);
4516                 all_slaves_active = 0;
4517         }
4518
4519         if (resend_igmp < 0 || resend_igmp > 255) {
4520                 pr_warning("Warning: resend_igmp (%d) should be between "
4521                            "0 and 255, resetting to %d\n",
4522                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4523                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4524         }
4525
4526         /* reset values for TLB/ALB */
4527         if ((bond_mode == BOND_MODE_TLB) ||
4528             (bond_mode == BOND_MODE_ALB)) {
4529                 if (!miimon) {
4530                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4531                         pr_warning("Forcing miimon to 100msec\n");
4532                         miimon = 100;
4533                 }
4534         }
4535
4536         if (bond_mode == BOND_MODE_ALB) {
4537                 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4538                           updelay);
4539         }
4540
4541         if (!miimon) {
4542                 if (updelay || downdelay) {
4543                         /* just warn the user the up/down delay will have
4544                          * no effect since miimon is zero...
4545                          */
4546                         pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4547                                    updelay, downdelay);
4548                 }
4549         } else {
4550                 /* don't allow arp monitoring */
4551                 if (arp_interval) {
4552                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4553                                    miimon, arp_interval);
4554                         arp_interval = 0;
4555                 }
4556
4557                 if ((updelay % miimon) != 0) {
4558                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4559                                    updelay, miimon,
4560                                    (updelay / miimon) * miimon);
4561                 }
4562
4563                 updelay /= miimon;
4564
4565                 if ((downdelay % miimon) != 0) {
4566                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4567                                    downdelay, miimon,
4568                                    (downdelay / miimon) * miimon);
4569                 }
4570
4571                 downdelay /= miimon;
4572         }
4573
4574         if (arp_interval < 0) {
4575                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4576                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4577                 arp_interval = BOND_LINK_ARP_INTERV;
4578         }
4579
4580         for (arp_ip_count = 0;
4581              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4582              arp_ip_count++) {
4583                 /* not complete check, but should be good enough to
4584                    catch mistakes */
4585                 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4586                 if (!isdigit(arp_ip_target[arp_ip_count][0]) ||
4587                     ip == 0 || ip == htonl(INADDR_BROADCAST)) {
4588                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4589                                    arp_ip_target[arp_ip_count]);
4590                         arp_interval = 0;
4591                 } else {
4592                         arp_target[arp_ip_count] = ip;
4593                 }
4594         }
4595
4596         if (arp_interval && !arp_ip_count) {
4597                 /* don't allow arping if no arp_ip_target given... */
4598                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4599                            arp_interval);
4600                 arp_interval = 0;
4601         }
4602
4603         if (arp_validate) {
4604                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4605                         pr_err("arp_validate only supported in active-backup mode\n");
4606                         return -EINVAL;
4607                 }
4608                 if (!arp_interval) {
4609                         pr_err("arp_validate requires arp_interval\n");
4610                         return -EINVAL;
4611                 }
4612
4613                 arp_validate_value = bond_parse_parm(arp_validate,
4614                                                      arp_validate_tbl);
4615                 if (arp_validate_value == -1) {
4616                         pr_err("Error: invalid arp_validate \"%s\"\n",
4617                                arp_validate == NULL ? "NULL" : arp_validate);
4618                         return -EINVAL;
4619                 }
4620         } else
4621                 arp_validate_value = 0;
4622
4623         if (miimon) {
4624                 pr_info("MII link monitoring set to %d ms\n", miimon);
4625         } else if (arp_interval) {
4626                 int i;
4627
4628                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4629                         arp_interval,
4630                         arp_validate_tbl[arp_validate_value].modename,
4631                         arp_ip_count);
4632
4633                 for (i = 0; i < arp_ip_count; i++)
4634                         pr_info(" %s", arp_ip_target[i]);
4635
4636                 pr_info("\n");
4637
4638         } else if (max_bonds) {
4639                 /* miimon and arp_interval not set, we need one so things
4640                  * work as expected, see bonding.txt for details
4641                  */
4642                 pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
4643         }
4644
4645         if (primary && !USES_PRIMARY(bond_mode)) {
4646                 /* currently, using a primary only makes sense
4647                  * in active backup, TLB or ALB modes
4648                  */
4649                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4650                            primary, bond_mode_name(bond_mode));
4651                 primary = NULL;
4652         }
4653
4654         if (primary && primary_reselect) {
4655                 primary_reselect_value = bond_parse_parm(primary_reselect,
4656                                                          pri_reselect_tbl);
4657                 if (primary_reselect_value == -1) {
4658                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4659                                primary_reselect ==
4660                                         NULL ? "NULL" : primary_reselect);
4661                         return -EINVAL;
4662                 }
4663         } else {
4664                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4665         }
4666
4667         if (fail_over_mac) {
4668                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4669                                                       fail_over_mac_tbl);
4670                 if (fail_over_mac_value == -1) {
4671                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4672                                arp_validate == NULL ? "NULL" : arp_validate);
4673                         return -EINVAL;
4674                 }
4675
4676                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4677                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4678         } else {
4679                 fail_over_mac_value = BOND_FOM_NONE;
4680         }
4681
4682         /* fill params struct with the proper values */
4683         params->mode = bond_mode;
4684         params->xmit_policy = xmit_hashtype;
4685         params->miimon = miimon;
4686         params->num_peer_notif = num_peer_notif;
4687         params->arp_interval = arp_interval;
4688         params->arp_validate = arp_validate_value;
4689         params->updelay = updelay;
4690         params->downdelay = downdelay;
4691         params->use_carrier = use_carrier;
4692         params->lacp_fast = lacp_fast;
4693         params->primary[0] = 0;
4694         params->primary_reselect = primary_reselect_value;
4695         params->fail_over_mac = fail_over_mac_value;
4696         params->tx_queues = tx_queues;
4697         params->all_slaves_active = all_slaves_active;
4698         params->resend_igmp = resend_igmp;
4699         params->min_links = min_links;
4700
4701         if (primary) {
4702                 strncpy(params->primary, primary, IFNAMSIZ);
4703                 params->primary[IFNAMSIZ - 1] = 0;
4704         }
4705
4706         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4707
4708         return 0;
4709 }
4710
4711 static struct lock_class_key bonding_netdev_xmit_lock_key;
4712 static struct lock_class_key bonding_netdev_addr_lock_key;
4713 static struct lock_class_key bonding_tx_busylock_key;
4714
4715 static void bond_set_lockdep_class_one(struct net_device *dev,
4716                                        struct netdev_queue *txq,
4717                                        void *_unused)
4718 {
4719         lockdep_set_class(&txq->_xmit_lock,
4720                           &bonding_netdev_xmit_lock_key);
4721 }
4722
4723 static void bond_set_lockdep_class(struct net_device *dev)
4724 {
4725         lockdep_set_class(&dev->addr_list_lock,
4726                           &bonding_netdev_addr_lock_key);
4727         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4728         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4729 }
4730
4731 /*
4732  * Called from registration process
4733  */
4734 static int bond_init(struct net_device *bond_dev)
4735 {
4736         struct bonding *bond = netdev_priv(bond_dev);
4737         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4738         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4739
4740         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4741
4742         /*
4743          * Initialize locks that may be required during
4744          * en/deslave operations.  All of the bond_open work
4745          * (of which this is part) should really be moved to
4746          * a phase prior to dev_open
4747          */
4748         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4749         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4750
4751         bond->wq = create_singlethread_workqueue(bond_dev->name);
4752         if (!bond->wq)
4753                 return -ENOMEM;
4754
4755         bond_set_lockdep_class(bond_dev);
4756
4757         list_add_tail(&bond->bond_list, &bn->dev_list);
4758
4759         bond_prepare_sysfs_group(bond);
4760
4761         bond_debug_register(bond);
4762
4763         /* Ensure valid dev_addr */
4764         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4765             bond_dev->addr_assign_type == NET_ADDR_PERM) {
4766                 eth_hw_addr_random(bond_dev);
4767                 bond->dev_addr_from_first = true;
4768         }
4769
4770         __hw_addr_init(&bond->mc_list);
4771         return 0;
4772 }
4773
4774 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4775 {
4776         if (tb[IFLA_ADDRESS]) {
4777                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4778                         return -EINVAL;
4779                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4780                         return -EADDRNOTAVAIL;
4781         }
4782         return 0;
4783 }
4784
4785 static unsigned int bond_get_num_tx_queues(void)
4786 {
4787         return tx_queues;
4788 }
4789
4790 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4791         .kind                   = "bond",
4792         .priv_size              = sizeof(struct bonding),
4793         .setup                  = bond_setup,
4794         .validate               = bond_validate,
4795         .get_num_tx_queues      = bond_get_num_tx_queues,
4796         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4797                                                              as for TX queues */
4798 };
4799
4800 /* Create a new bond based on the specified name and bonding parameters.
4801  * If name is NULL, obtain a suitable "bond%d" name for us.
4802  * Caller must NOT hold rtnl_lock; we need to release it here before we
4803  * set up our sysfs entries.
4804  */
4805 int bond_create(struct net *net, const char *name)
4806 {
4807         struct net_device *bond_dev;
4808         int res;
4809
4810         rtnl_lock();
4811
4812         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4813                                    name ? name : "bond%d",
4814                                    bond_setup, tx_queues);
4815         if (!bond_dev) {
4816                 pr_err("%s: eek! can't alloc netdev!\n", name);
4817                 rtnl_unlock();
4818                 return -ENOMEM;
4819         }
4820
4821         dev_net_set(bond_dev, net);
4822         bond_dev->rtnl_link_ops = &bond_link_ops;
4823
4824         res = register_netdevice(bond_dev);
4825
4826         netif_carrier_off(bond_dev);
4827
4828         rtnl_unlock();
4829         if (res < 0)
4830                 bond_destructor(bond_dev);
4831         return res;
4832 }
4833
4834 static int __net_init bond_net_init(struct net *net)
4835 {
4836         struct bond_net *bn = net_generic(net, bond_net_id);
4837
4838         bn->net = net;
4839         INIT_LIST_HEAD(&bn->dev_list);
4840
4841         bond_create_proc_dir(bn);
4842         bond_create_sysfs(bn);
4843
4844         return 0;
4845 }
4846
4847 static void __net_exit bond_net_exit(struct net *net)
4848 {
4849         struct bond_net *bn = net_generic(net, bond_net_id);
4850
4851         bond_destroy_sysfs(bn);
4852         bond_destroy_proc_dir(bn);
4853 }
4854
4855 static struct pernet_operations bond_net_ops = {
4856         .init = bond_net_init,
4857         .exit = bond_net_exit,
4858         .id   = &bond_net_id,
4859         .size = sizeof(struct bond_net),
4860 };
4861
4862 static int __init bonding_init(void)
4863 {
4864         int i;
4865         int res;
4866
4867         pr_info("%s", bond_version);
4868
4869         res = bond_check_params(&bonding_defaults);
4870         if (res)
4871                 goto out;
4872
4873         res = register_pernet_subsys(&bond_net_ops);
4874         if (res)
4875                 goto out;
4876
4877         res = rtnl_link_register(&bond_link_ops);
4878         if (res)
4879                 goto err_link;
4880
4881         bond_create_debugfs();
4882
4883         for (i = 0; i < max_bonds; i++) {
4884                 res = bond_create(&init_net, NULL);
4885                 if (res)
4886                         goto err;
4887         }
4888
4889         register_netdevice_notifier(&bond_netdev_notifier);
4890 out:
4891         return res;
4892 err:
4893         rtnl_link_unregister(&bond_link_ops);
4894 err_link:
4895         unregister_pernet_subsys(&bond_net_ops);
4896         goto out;
4897
4898 }
4899
4900 static void __exit bonding_exit(void)
4901 {
4902         unregister_netdevice_notifier(&bond_netdev_notifier);
4903
4904         bond_destroy_debugfs();
4905
4906         rtnl_link_unregister(&bond_link_ops);
4907         unregister_pernet_subsys(&bond_net_ops);
4908
4909 #ifdef CONFIG_NET_POLL_CONTROLLER
4910         /*
4911          * Make sure we don't have an imbalance on our netpoll blocking
4912          */
4913         WARN_ON(atomic_read(&netpoll_block_tx));
4914 #endif
4915 }
4916
4917 module_init(bonding_init);
4918 module_exit(bonding_exit);
4919 MODULE_LICENSE("GPL");
4920 MODULE_VERSION(DRV_VERSION);
4921 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4922 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4923 MODULE_ALIAS_RTNL_LINK("bond");