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         rcu_read_lock();
 750         in_dev = __in_dev_get_rcu(dev);
 751         if (in_dev)
 752                 ip_mc_rejoin_groups(in_dev);
 753         rcu_read_unlock();
 754 }
 755
 756 /*
 757  * Retrieve the list of registered multicast addresses for the bonding
 758  * device and retransmit an IGMP JOIN request to the current active
 759  * slave.
 760  */
 761 static void bond_resend_igmp_join_requests(struct bonding *bond)
 762 {
 763         struct net_device *bond_dev, *vlan_dev, *master_dev;
 764         struct vlan_entry *vlan;
 765
 766         read_lock(&bond->lock);
 767
 768         bond_dev = bond->dev;
 769
 770         /* rejoin all groups on bond device */
 771         __bond_resend_igmp_join_requests(bond_dev);
 772
 773         /*
 774          * if bond is enslaved to a bridge,
 775          * then rejoin all groups on its master
 776          */
 777         master_dev = bond_dev->master;
 778         if (master_dev)
 779                 if ((master_dev->priv_flags & IFF_EBRIDGE)
 780                         && (bond_dev->priv_flags & IFF_BRIDGE_PORT))
 781                         __bond_resend_igmp_join_requests(master_dev);
 782
 783         /* rejoin all groups on vlan devices */
 784         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 785                 rcu_read_lock();
 786                 vlan_dev = __vlan_find_dev_deep(bond_dev,
 787                                                 vlan->vlan_id);
 788                 rcu_read_unlock();
 789                 if (vlan_dev)
 790                         __bond_resend_igmp_join_requests(vlan_dev);
 791         }
 792
 793         if (--bond->igmp_retrans > 0)
 794                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 795
 796         read_unlock(&bond->lock);
 797 }
 798
 799 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 800 {
 801         struct bonding *bond = container_of(work, struct bonding,
 802                                             mcast_work.work);
 803         bond_resend_igmp_join_requests(bond);
 804 }
 805
 806 /*
 807  * flush all members of flush->mc_list from device dev->mc_list
 808  */
 809 static void bond_mc_list_flush(struct net_device *bond_dev,
 810                                struct net_device *slave_dev)
 811 {
 812         struct bonding *bond = netdev_priv(bond_dev);
 813         struct netdev_hw_addr *ha;
 814
 815         netdev_for_each_mc_addr(ha, bond_dev)
 816                 dev_mc_del(slave_dev, ha->addr);
 817
 818         if (bond->params.mode == BOND_MODE_8023AD) {
 819                 /* del lacpdu mc addr from mc list */
 820                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 821
 822                 dev_mc_del(slave_dev, lacpdu_multicast);
 823         }
 824 }
 825
 826 /*--------------------------- Active slave change ---------------------------*/
 827
 828 /*
 829  * Update the mc list and multicast-related flags for the new and
 830  * old active slaves (if any) according to the multicast mode, and
 831  * promiscuous flags unconditionally.
 832  */
 833 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 834                          struct slave *old_active)
 835 {
 836         struct netdev_hw_addr *ha;
 837
 838         if (!USES_PRIMARY(bond->params.mode))
 839                 /* nothing to do -  mc list is already up-to-date on
 840                  * all slaves
 841                  */
 842                 return;
 843
 844         if (old_active) {
 845                 if (bond->dev->flags & IFF_PROMISC)
 846                         dev_set_promiscuity(old_active->dev, -1);
 847
 848                 if (bond->dev->flags & IFF_ALLMULTI)
 849                         dev_set_allmulti(old_active->dev, -1);
 850
 851                 netdev_for_each_mc_addr(ha, bond->dev)
 852                         dev_mc_del(old_active->dev, ha->addr);
 853         }
 854
 855         if (new_active) {
 856                 /* FIXME: Signal errors upstream. */
 857                 if (bond->dev->flags & IFF_PROMISC)
 858                         dev_set_promiscuity(new_active->dev, 1);
 859
 860                 if (bond->dev->flags & IFF_ALLMULTI)
 861                         dev_set_allmulti(new_active->dev, 1);
 862
 863                 netdev_for_each_mc_addr(ha, bond->dev)
 864                         dev_mc_add(new_active->dev, ha->addr);
 865         }
 866 }
 867
 868 /*
 869  * bond_do_fail_over_mac
 870  *
 871  * Perform special MAC address swapping for fail_over_mac settings
 872  *
 873  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 874  */
 875 static void bond_do_fail_over_mac(struct bonding *bond,
 876                                   struct slave *new_active,
 877                                   struct slave *old_active)
 878         __releases(&bond->curr_slave_lock)
 879         __releases(&bond->lock)
 880         __acquires(&bond->lock)
 881         __acquires(&bond->curr_slave_lock)
 882 {
 883         u8 tmp_mac[ETH_ALEN];
 884         struct sockaddr saddr;
 885         int rv;
 886
 887         switch (bond->params.fail_over_mac) {
 888         case BOND_FOM_ACTIVE:
 889                 if (new_active) {
 890                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 891                                new_active->dev->addr_len);
 892                         write_unlock_bh(&bond->curr_slave_lock);
 893                         read_unlock(&bond->lock);
 894                         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
 895                         read_lock(&bond->lock);
 896                         write_lock_bh(&bond->curr_slave_lock);
 897                 }
 898                 break;
 899         case BOND_FOM_FOLLOW:
 900                 /*
 901                  * if new_active && old_active, swap them
 902                  * if just old_active, do nothing (going to no active slave)
 903                  * if just new_active, set new_active to bond's MAC
 904                  */
 905                 if (!new_active)
 906                         return;
 907
 908                 write_unlock_bh(&bond->curr_slave_lock);
 909                 read_unlock(&bond->lock);
 910
 911                 if (old_active) {
 912                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 913                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 914                                ETH_ALEN);
 915                         saddr.sa_family = new_active->dev->type;
 916                 } else {
 917                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 918                         saddr.sa_family = bond->dev->type;
 919                 }
 920
 921                 rv = dev_set_mac_address(new_active->dev, &saddr);
 922                 if (rv) {
 923                         pr_err("%s: Error %d setting MAC of slave %s\n",
 924                                bond->dev->name, -rv, new_active->dev->name);
 925                         goto out;
 926                 }
 927
 928                 if (!old_active)
 929                         goto out;
 930
 931                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 932                 saddr.sa_family = old_active->dev->type;
 933
 934                 rv = dev_set_mac_address(old_active->dev, &saddr);
 935                 if (rv)
 936                         pr_err("%s: Error %d setting MAC of slave %s\n",
 937                                bond->dev->name, -rv, new_active->dev->name);
 938 out:
 939                 read_lock(&bond->lock);
 940                 write_lock_bh(&bond->curr_slave_lock);
 941                 break;
 942         default:
 943                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 944                        bond->dev->name, bond->params.fail_over_mac);
 945                 break;
 946         }
 947
 948 }
 949
 950 static bool bond_should_change_active(struct bonding *bond)
 951 {
 952         struct slave *prim = bond->primary_slave;
 953         struct slave *curr = bond->curr_active_slave;
 954
 955         if (!prim || !curr || curr->link != BOND_LINK_UP)
 956                 return true;
 957         if (bond->force_primary) {
 958                 bond->force_primary = false;
 959                 return true;
 960         }
 961         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 962             (prim->speed < curr->speed ||
 963              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 964                 return false;
 965         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 966                 return false;
 967         return true;
 968 }
 969
 970 /**
 971  * find_best_interface - select the best available slave to be the active one
 972  * @bond: our bonding struct
 973  *
 974  * Warning: Caller must hold curr_slave_lock for writing.
 975  */
 976 static struct slave *bond_find_best_slave(struct bonding *bond)
 977 {
 978         struct slave *new_active, *old_active;
 979         struct slave *bestslave = NULL;
 980         int mintime = bond->params.updelay;
 981         int i;
 982
 983         new_active = bond->curr_active_slave;
 984
 985         if (!new_active) { /* there were no active slaves left */
 986                 if (bond->slave_cnt > 0)   /* found one slave */
 987                         new_active = bond->first_slave;
 988                 else
 989                         return NULL; /* still no slave, return NULL */
 990         }
 991
 992         if ((bond->primary_slave) &&
 993             bond->primary_slave->link == BOND_LINK_UP &&
 994             bond_should_change_active(bond)) {
 995                 new_active = bond->primary_slave;
 996         }
 997
 998         /* remember where to stop iterating over the slaves */
 999         old_active = new_active;
1000
1001         bond_for_each_slave_from(bond, new_active, i, old_active) {
1002                 if (new_active->link == BOND_LINK_UP) {
1003                         return new_active;
1004                 } else if (new_active->link == BOND_LINK_BACK &&
1005                            IS_UP(new_active->dev)) {
1006                         /* link up, but waiting for stabilization */
1007                         if (new_active->delay < mintime) {
1008                                 mintime = new_active->delay;
1009                                 bestslave = new_active;
1010                         }
1011                 }
1012         }
1013
1014         return bestslave;
1015 }
1016
1017 static bool bond_should_notify_peers(struct bonding *bond)
1018 {
1019         struct slave *slave = bond->curr_active_slave;
1020
1021         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1022                  bond->dev->name, slave ? slave->dev->name : "NULL");
1023
1024         if (!slave || !bond->send_peer_notif ||
1025             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1026                 return false;
1027
1028         bond->send_peer_notif--;
1029         return true;
1030 }
1031
1032 /**
1033  * change_active_interface - change the active slave into the specified one
1034  * @bond: our bonding struct
1035  * @new: the new slave to make the active one
1036  *
1037  * Set the new slave to the bond's settings and unset them on the old
1038  * curr_active_slave.
1039  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1040  *
1041  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1042  * because it is apparently the best available slave we have, even though its
1043  * updelay hasn't timed out yet.
1044  *
1045  * If new_active is not NULL, caller must hold bond->lock for read and
1046  * curr_slave_lock for write_bh.
1047  */
1048 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1049 {
1050         struct slave *old_active = bond->curr_active_slave;
1051
1052         if (old_active == new_active)
1053                 return;
1054
1055         if (new_active) {
1056                 new_active->jiffies = jiffies;
1057
1058                 if (new_active->link == BOND_LINK_BACK) {
1059                         if (USES_PRIMARY(bond->params.mode)) {
1060                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1061                                         bond->dev->name, new_active->dev->name,
1062                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1063                         }
1064
1065                         new_active->delay = 0;
1066                         new_active->link = BOND_LINK_UP;
1067
1068                         if (bond->params.mode == BOND_MODE_8023AD)
1069                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1070
1071                         if (bond_is_lb(bond))
1072                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1073                 } else {
1074                         if (USES_PRIMARY(bond->params.mode)) {
1075                                 pr_info("%s: making interface %s the new active one.\n",
1076                                         bond->dev->name, new_active->dev->name);
1077                         }
1078                 }
1079         }
1080
1081         if (USES_PRIMARY(bond->params.mode))
1082                 bond_mc_swap(bond, new_active, old_active);
1083
1084         if (bond_is_lb(bond)) {
1085                 bond_alb_handle_active_change(bond, new_active);
1086                 if (old_active)
1087                         bond_set_slave_inactive_flags(old_active);
1088                 if (new_active)
1089                         bond_set_slave_active_flags(new_active);
1090         } else {
1091                 bond->curr_active_slave = new_active;
1092         }
1093
1094         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1095                 if (old_active)
1096                         bond_set_slave_inactive_flags(old_active);
1097
1098                 if (new_active) {
1099                         bool should_notify_peers = false;
1100
1101                         bond_set_slave_active_flags(new_active);
1102
1103                         if (bond->params.fail_over_mac)
1104                                 bond_do_fail_over_mac(bond, new_active,
1105                                                       old_active);
1106
1107                         if (netif_running(bond->dev)) {
1108                                 bond->send_peer_notif =
1109                                         bond->params.num_peer_notif;
1110                                 should_notify_peers =
1111                                         bond_should_notify_peers(bond);
1112                         }
1113
1114                         write_unlock_bh(&bond->curr_slave_lock);
1115                         read_unlock(&bond->lock);
1116
1117                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1118                         if (should_notify_peers)
1119                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1120                                                          bond->dev);
1121
1122                         read_lock(&bond->lock);
1123                         write_lock_bh(&bond->curr_slave_lock);
1124                 }
1125         }
1126
1127         /* resend IGMP joins since active slave has changed or
1128          * all were sent on curr_active_slave.
1129          * resend only if bond is brought up with the affected
1130          * bonding modes and the retransmission is enabled */
1131         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1132             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1133              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1134                 bond->igmp_retrans = bond->params.resend_igmp;
1135                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1136         }
1137 }
1138
1139 /**
1140  * bond_select_active_slave - select a new active slave, if needed
1141  * @bond: our bonding struct
1142  *
1143  * This functions should be called when one of the following occurs:
1144  * - The old curr_active_slave has been released or lost its link.
1145  * - The primary_slave has got its link back.
1146  * - A slave has got its link back and there's no old curr_active_slave.
1147  *
1148  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1149  */
1150 void bond_select_active_slave(struct bonding *bond)
1151 {
1152         struct slave *best_slave;
1153         int rv;
1154
1155         best_slave = bond_find_best_slave(bond);
1156         if (best_slave != bond->curr_active_slave) {
1157                 bond_change_active_slave(bond, best_slave);
1158                 rv = bond_set_carrier(bond);
1159                 if (!rv)
1160                         return;
1161
1162                 if (netif_carrier_ok(bond->dev)) {
1163                         pr_info("%s: first active interface up!\n",
1164                                 bond->dev->name);
1165                 } else {
1166                         pr_info("%s: now running without any active interface !\n",
1167                                 bond->dev->name);
1168                 }
1169         }
1170 }
1171
1172 /*--------------------------- slave list handling ---------------------------*/
1173
1174 /*
1175  * This function attaches the slave to the end of list.
1176  *
1177  * bond->lock held for writing by caller.
1178  */
1179 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1180 {
1181         if (bond->first_slave == NULL) { /* attaching the first slave */
1182                 new_slave->next = new_slave;
1183                 new_slave->prev = new_slave;
1184                 bond->first_slave = new_slave;
1185         } else {
1186                 new_slave->next = bond->first_slave;
1187                 new_slave->prev = bond->first_slave->prev;
1188                 new_slave->next->prev = new_slave;
1189                 new_slave->prev->next = new_slave;
1190         }
1191
1192         bond->slave_cnt++;
1193 }
1194
1195 /*
1196  * This function detaches the slave from the list.
1197  * WARNING: no check is made to verify if the slave effectively
1198  * belongs to <bond>.
1199  * Nothing is freed on return, structures are just unchained.
1200  * If any slave pointer in bond was pointing to <slave>,
1201  * it should be changed by the calling function.
1202  *
1203  * bond->lock held for writing by caller.
1204  */
1205 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1206 {
1207         if (slave->next)
1208                 slave->next->prev = slave->prev;
1209
1210         if (slave->prev)
1211                 slave->prev->next = slave->next;
1212
1213         if (bond->first_slave == slave) { /* slave is the first slave */
1214                 if (bond->slave_cnt > 1) { /* there are more slave */
1215                         bond->first_slave = slave->next;
1216                 } else {
1217                         bond->first_slave = NULL; /* slave was the last one */
1218                 }
1219         }
1220
1221         slave->next = NULL;
1222         slave->prev = NULL;
1223         bond->slave_cnt--;
1224 }
1225
1226 #ifdef CONFIG_NET_POLL_CONTROLLER
1227 static inline int slave_enable_netpoll(struct slave *slave)
1228 {
1229         struct netpoll *np;
1230         int err = 0;
1231
1232         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1233         err = -ENOMEM;
1234         if (!np)
1235                 goto out;
1236
1237         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1238         if (err) {
1239                 kfree(np);
1240                 goto out;
1241         }
1242         slave->np = np;
1243 out:
1244         return err;
1245 }
1246 static inline void slave_disable_netpoll(struct slave *slave)
1247 {
1248         struct netpoll *np = slave->np;
1249
1250         if (!np)
1251                 return;
1252
1253         slave->np = NULL;
1254         __netpoll_free_rcu(np);
1255 }
1256 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1257 {
1258         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1259                 return false;
1260         if (!slave_dev->netdev_ops->ndo_poll_controller)
1261                 return false;
1262         return true;
1263 }
1264
1265 static void bond_poll_controller(struct net_device *bond_dev)
1266 {
1267 }
1268
1269 static void __bond_netpoll_cleanup(struct bonding *bond)
1270 {
1271         struct slave *slave;
1272         int i;
1273
1274         bond_for_each_slave(bond, slave, i)
1275                 if (IS_UP(slave->dev))
1276                         slave_disable_netpoll(slave);
1277 }
1278 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1279 {
1280         struct bonding *bond = netdev_priv(bond_dev);
1281
1282         read_lock(&bond->lock);
1283         __bond_netpoll_cleanup(bond);
1284         read_unlock(&bond->lock);
1285 }
1286
1287 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1288 {
1289         struct bonding *bond = netdev_priv(dev);
1290         struct slave *slave;
1291         int i, err = 0;
1292
1293         read_lock(&bond->lock);
1294         bond_for_each_slave(bond, slave, i) {
1295                 err = slave_enable_netpoll(slave);
1296                 if (err) {
1297                         __bond_netpoll_cleanup(bond);
1298                         break;
1299                 }
1300         }
1301         read_unlock(&bond->lock);
1302         return err;
1303 }
1304
1305 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1306 {
1307         return bond->dev->npinfo;
1308 }
1309
1310 #else
1311 static inline int slave_enable_netpoll(struct slave *slave)
1312 {
1313         return 0;
1314 }
1315 static inline void slave_disable_netpoll(struct slave *slave)
1316 {
1317 }
1318 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1319 {
1320 }
1321 #endif
1322
1323 /*---------------------------------- IOCTL ----------------------------------*/
1324
1325 static int bond_sethwaddr(struct net_device *bond_dev,
1326                           struct net_device *slave_dev)
1327 {
1328         pr_debug("bond_dev=%p\n", bond_dev);
1329         pr_debug("slave_dev=%p\n", slave_dev);
1330         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1331         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1332         return 0;
1333 }
1334
1335 static netdev_features_t bond_fix_features(struct net_device *dev,
1336         netdev_features_t features)
1337 {
1338         struct slave *slave;
1339         struct bonding *bond = netdev_priv(dev);
1340         netdev_features_t mask;
1341         int i;
1342
1343         read_lock(&bond->lock);
1344
1345         if (!bond->first_slave) {
1346                 /* Disable adding VLANs to empty bond. But why? --mq */
1347                 features |= NETIF_F_VLAN_CHALLENGED;
1348                 goto out;
1349         }
1350
1351         mask = features;
1352         features &= ~NETIF_F_ONE_FOR_ALL;
1353         features |= NETIF_F_ALL_FOR_ALL;
1354
1355         bond_for_each_slave(bond, slave, i) {
1356                 features = netdev_increment_features(features,
1357                                                      slave->dev->features,
1358                                                      mask);
1359         }
1360
1361 out:
1362         read_unlock(&bond->lock);
1363         return features;
1364 }
1365
1366 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1367                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1368                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1369
1370 static void bond_compute_features(struct bonding *bond)
1371 {
1372         struct slave *slave;
1373         struct net_device *bond_dev = bond->dev;
1374         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1375         unsigned short max_hard_header_len = ETH_HLEN;
1376         unsigned int gso_max_size = GSO_MAX_SIZE;
1377         u16 gso_max_segs = GSO_MAX_SEGS;
1378         int i;
1379         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1380
1381         read_lock(&bond->lock);
1382
1383         if (!bond->first_slave)
1384                 goto done;
1385
1386         bond_for_each_slave(bond, slave, i) {
1387                 vlan_features = netdev_increment_features(vlan_features,
1388                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1389
1390                 dst_release_flag &= slave->dev->priv_flags;
1391                 if (slave->dev->hard_header_len > max_hard_header_len)
1392                         max_hard_header_len = slave->dev->hard_header_len;
1393
1394                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1395                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1396         }
1397
1398 done:
1399         bond_dev->vlan_features = vlan_features;
1400         bond_dev->hard_header_len = max_hard_header_len;
1401         bond_dev->gso_max_segs = gso_max_segs;
1402         netif_set_gso_max_size(bond_dev, gso_max_size);
1403
1404         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1405         bond_dev->priv_flags = flags | dst_release_flag;
1406
1407         read_unlock(&bond->lock);
1408
1409         netdev_change_features(bond_dev);
1410 }
1411
1412 static void bond_setup_by_slave(struct net_device *bond_dev,
1413                                 struct net_device *slave_dev)
1414 {
1415         struct bonding *bond = netdev_priv(bond_dev);
1416
1417         bond_dev->header_ops        = slave_dev->header_ops;
1418
1419         bond_dev->type              = slave_dev->type;
1420         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1421         bond_dev->addr_len          = slave_dev->addr_len;
1422
1423         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1424                 slave_dev->addr_len);
1425         bond->setup_by_slave = 1;
1426 }
1427
1428 /* On bonding slaves other than the currently active slave, suppress
1429  * duplicates except for alb non-mcast/bcast.
1430  */
1431 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1432                                             struct slave *slave,
1433                                             struct bonding *bond)
1434 {
1435         if (bond_is_slave_inactive(slave)) {
1436                 if (bond->params.mode == BOND_MODE_ALB &&
1437                     skb->pkt_type != PACKET_BROADCAST &&
1438                     skb->pkt_type != PACKET_MULTICAST)
1439                         return false;
1440                 return true;
1441         }
1442         return false;
1443 }
1444
1445 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1446 {
1447         struct sk_buff *skb = *pskb;
1448         struct slave *slave;
1449         struct bonding *bond;
1450         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1451                           struct slave *);
1452         int ret = RX_HANDLER_ANOTHER;
1453
1454         skb = skb_share_check(skb, GFP_ATOMIC);
1455         if (unlikely(!skb))
1456                 return RX_HANDLER_CONSUMED;
1457
1458         *pskb = skb;
1459
1460         slave = bond_slave_get_rcu(skb->dev);
1461         bond = slave->bond;
1462
1463         if (bond->params.arp_interval)
1464                 slave->dev->last_rx = jiffies;
1465
1466         recv_probe = ACCESS_ONCE(bond->recv_probe);
1467         if (recv_probe) {
1468                 ret = recv_probe(skb, bond, slave);
1469                 if (ret == RX_HANDLER_CONSUMED) {
1470                         consume_skb(skb);
1471                         return ret;
1472                 }
1473         }
1474
1475         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1476                 return RX_HANDLER_EXACT;
1477         }
1478
1479         skb->dev = bond->dev;
1480
1481         if (bond->params.mode == BOND_MODE_ALB &&
1482             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1483             skb->pkt_type == PACKET_HOST) {
1484
1485                 if (unlikely(skb_cow_head(skb,
1486                                           skb->data - skb_mac_header(skb)))) {
1487                         kfree_skb(skb);
1488                         return RX_HANDLER_CONSUMED;
1489                 }
1490                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1491         }
1492
1493         return ret;
1494 }
1495
1496 /* enslave device <slave> to bond device <master> */
1497 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1498 {
1499         struct bonding *bond = netdev_priv(bond_dev);
1500         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1501         struct slave *new_slave = NULL;
1502         struct netdev_hw_addr *ha;
1503         struct sockaddr addr;
1504         int link_reporting;
1505         int res = 0;
1506
1507         if (!bond->params.use_carrier &&
1508             slave_dev->ethtool_ops->get_link == NULL &&
1509             slave_ops->ndo_do_ioctl == NULL) {
1510                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1511                            bond_dev->name, slave_dev->name);
1512         }
1513
1514         /* already enslaved */
1515         if (slave_dev->flags & IFF_SLAVE) {
1516                 pr_debug("Error, Device was already enslaved\n");
1517                 return -EBUSY;
1518         }
1519
1520         /* vlan challenged mutual exclusion */
1521         /* no need to lock since we're protected by rtnl_lock */
1522         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1523                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1524                 if (vlan_uses_dev(bond_dev)) {
1525                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1526                                bond_dev->name, slave_dev->name, bond_dev->name);
1527                         return -EPERM;
1528                 } else {
1529                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1530                                    bond_dev->name, slave_dev->name,
1531                                    slave_dev->name, bond_dev->name);
1532                 }
1533         } else {
1534                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1535         }
1536
1537         /*
1538          * Old ifenslave binaries are no longer supported.  These can
1539          * be identified with moderate accuracy by the state of the slave:
1540          * the current ifenslave will set the interface down prior to
1541          * enslaving it; the old ifenslave will not.
1542          */
1543         if ((slave_dev->flags & IFF_UP)) {
1544                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1545                        slave_dev->name);
1546                 res = -EPERM;
1547                 goto err_undo_flags;
1548         }
1549
1550         /* set bonding device ether type by slave - bonding netdevices are
1551          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1552          * there is a need to override some of the type dependent attribs/funcs.
1553          *
1554          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1555          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1556          */
1557         if (bond->slave_cnt == 0) {
1558                 if (bond_dev->type != slave_dev->type) {
1559                         pr_debug("%s: change device type from %d to %d\n",
1560                                  bond_dev->name,
1561                                  bond_dev->type, slave_dev->type);
1562
1563                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1564                                                        bond_dev);
1565                         res = notifier_to_errno(res);
1566                         if (res) {
1567                                 pr_err("%s: refused to change device type\n",
1568                                        bond_dev->name);
1569                                 res = -EBUSY;
1570                                 goto err_undo_flags;
1571                         }
1572
1573                         /* Flush unicast and multicast addresses */
1574                         dev_uc_flush(bond_dev);
1575                         dev_mc_flush(bond_dev);
1576
1577                         if (slave_dev->type != ARPHRD_ETHER)
1578                                 bond_setup_by_slave(bond_dev, slave_dev);
1579                         else {
1580                                 ether_setup(bond_dev);
1581                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1582                         }
1583
1584                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1585                                                  bond_dev);
1586                 }
1587         } else if (bond_dev->type != slave_dev->type) {
1588                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1589                        slave_dev->name,
1590                        slave_dev->type, bond_dev->type);
1591                 res = -EINVAL;
1592                 goto err_undo_flags;
1593         }
1594
1595         if (slave_ops->ndo_set_mac_address == NULL) {
1596                 if (bond->slave_cnt == 0) {
1597                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1598                                    bond_dev->name);
1599                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1600                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1601                         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",
1602                                bond_dev->name);
1603                         res = -EOPNOTSUPP;
1604                         goto err_undo_flags;
1605                 }
1606         }
1607
1608         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1609
1610         /* If this is the first slave, then we need to set the master's hardware
1611          * address to be the same as the slave's. */
1612         if (is_zero_ether_addr(bond->dev->dev_addr))
1613                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1614                        slave_dev->addr_len);
1615
1616
1617         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1618         if (!new_slave) {
1619                 res = -ENOMEM;
1620                 goto err_undo_flags;
1621         }
1622
1623         /*
1624          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1625          * is set via sysfs or module option if desired.
1626          */
1627         new_slave->queue_id = 0;
1628
1629         /* Save slave's original mtu and then set it to match the bond */
1630         new_slave->original_mtu = slave_dev->mtu;
1631         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1632         if (res) {
1633                 pr_debug("Error %d calling dev_set_mtu\n", res);
1634                 goto err_free;
1635         }
1636
1637         /*
1638          * Save slave's original ("permanent") mac address for modes
1639          * that need it, and for restoring it upon release, and then
1640          * set it to the master's address
1641          */
1642         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1643
1644         if (!bond->params.fail_over_mac) {
1645                 /*
1646                  * Set slave to master's mac address.  The application already
1647                  * set the master's mac address to that of the first slave
1648                  */
1649                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1650                 addr.sa_family = slave_dev->type;
1651                 res = dev_set_mac_address(slave_dev, &addr);
1652                 if (res) {
1653                         pr_debug("Error %d calling set_mac_address\n", res);
1654                         goto err_restore_mtu;
1655                 }
1656         }
1657
1658         res = netdev_set_bond_master(slave_dev, bond_dev);
1659         if (res) {
1660                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1661                 goto err_restore_mac;
1662         }
1663
1664         /* open the slave since the application closed it */
1665         res = dev_open(slave_dev);
1666         if (res) {
1667                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1668                 goto err_unset_master;
1669         }
1670
1671         new_slave->bond = bond;
1672         new_slave->dev = slave_dev;
1673         slave_dev->priv_flags |= IFF_BONDING;
1674
1675         if (bond_is_lb(bond)) {
1676                 /* bond_alb_init_slave() must be called before all other stages since
1677                  * it might fail and we do not want to have to undo everything
1678                  */
1679                 res = bond_alb_init_slave(bond, new_slave);
1680                 if (res)
1681                         goto err_close;
1682         }
1683
1684         /* If the mode USES_PRIMARY, then the new slave gets the
1685          * master's promisc (and mc) settings only if it becomes the
1686          * curr_active_slave, and that is taken care of later when calling
1687          * bond_change_active()
1688          */
1689         if (!USES_PRIMARY(bond->params.mode)) {
1690                 /* set promiscuity level to new slave */
1691                 if (bond_dev->flags & IFF_PROMISC) {
1692                         res = dev_set_promiscuity(slave_dev, 1);
1693                         if (res)
1694                                 goto err_close;
1695                 }
1696
1697                 /* set allmulti level to new slave */
1698                 if (bond_dev->flags & IFF_ALLMULTI) {
1699                         res = dev_set_allmulti(slave_dev, 1);
1700                         if (res)
1701                                 goto err_close;
1702                 }
1703
1704                 netif_addr_lock_bh(bond_dev);
1705                 /* upload master's mc_list to new slave */
1706                 netdev_for_each_mc_addr(ha, bond_dev)
1707                         dev_mc_add(slave_dev, ha->addr);
1708                 netif_addr_unlock_bh(bond_dev);
1709         }
1710
1711         if (bond->params.mode == BOND_MODE_8023AD) {
1712                 /* add lacpdu mc addr to mc list */
1713                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1714
1715                 dev_mc_add(slave_dev, lacpdu_multicast);
1716         }
1717
1718         bond_add_vlans_on_slave(bond, slave_dev);
1719
1720         write_lock_bh(&bond->lock);
1721
1722         bond_attach_slave(bond, new_slave);
1723
1724         new_slave->delay = 0;
1725         new_slave->link_failure_count = 0;
1726
1727         write_unlock_bh(&bond->lock);
1728
1729         bond_compute_features(bond);
1730
1731         read_lock(&bond->lock);
1732
1733         new_slave->last_arp_rx = jiffies -
1734                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1735
1736         if (bond->params.miimon && !bond->params.use_carrier) {
1737                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1738
1739                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1740                         /*
1741                          * miimon is set but a bonded network driver
1742                          * does not support ETHTOOL/MII and
1743                          * arp_interval is not set.  Note: if
1744                          * use_carrier is enabled, we will never go
1745                          * here (because netif_carrier is always
1746                          * supported); thus, we don't need to change
1747                          * the messages for netif_carrier.
1748                          */
1749                         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",
1750                                bond_dev->name, slave_dev->name);
1751                 } else if (link_reporting == -1) {
1752                         /* unable get link status using mii/ethtool */
1753                         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",
1754                                    bond_dev->name, slave_dev->name);
1755                 }
1756         }
1757
1758         /* check for initial state */
1759         if (bond->params.miimon) {
1760                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1761                         if (bond->params.updelay) {
1762                                 new_slave->link = BOND_LINK_BACK;
1763                                 new_slave->delay = bond->params.updelay;
1764                         } else {
1765                                 new_slave->link = BOND_LINK_UP;
1766                         }
1767                 } else {
1768                         new_slave->link = BOND_LINK_DOWN;
1769                 }
1770         } else if (bond->params.arp_interval) {
1771                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1772                         BOND_LINK_UP : BOND_LINK_DOWN);
1773         } else {
1774                 new_slave->link = BOND_LINK_UP;
1775         }
1776
1777         if (new_slave->link != BOND_LINK_DOWN)
1778                 new_slave->jiffies = jiffies;
1779         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1780                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1781                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1782
1783         bond_update_speed_duplex(new_slave);
1784
1785         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1786                 /* if there is a primary slave, remember it */
1787                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1788                         bond->primary_slave = new_slave;
1789                         bond->force_primary = true;
1790                 }
1791         }
1792
1793         write_lock_bh(&bond->curr_slave_lock);
1794
1795         switch (bond->params.mode) {
1796         case BOND_MODE_ACTIVEBACKUP:
1797                 bond_set_slave_inactive_flags(new_slave);
1798                 bond_select_active_slave(bond);
1799                 break;
1800         case BOND_MODE_8023AD:
1801                 /* in 802.3ad mode, the internal mechanism
1802                  * will activate the slaves in the selected
1803                  * aggregator
1804                  */
1805                 bond_set_slave_inactive_flags(new_slave);
1806                 /* if this is the first slave */
1807                 if (bond->slave_cnt == 1) {
1808                         SLAVE_AD_INFO(new_slave).id = 1;
1809                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1810                          * can be called only after the mac address of the bond is set
1811                          */
1812                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1813                 } else {
1814                         SLAVE_AD_INFO(new_slave).id =
1815                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1816                 }
1817
1818                 bond_3ad_bind_slave(new_slave);
1819                 break;
1820         case BOND_MODE_TLB:
1821         case BOND_MODE_ALB:
1822                 bond_set_active_slave(new_slave);
1823                 bond_set_slave_inactive_flags(new_slave);
1824                 bond_select_active_slave(bond);
1825                 break;
1826         default:
1827                 pr_debug("This slave is always active in trunk mode\n");
1828
1829                 /* always active in trunk mode */
1830                 bond_set_active_slave(new_slave);
1831
1832                 /* In trunking mode there is little meaning to curr_active_slave
1833                  * anyway (it holds no special properties of the bond device),
1834                  * so we can change it without calling change_active_interface()
1835                  */
1836                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1837                         bond->curr_active_slave = new_slave;
1838
1839                 break;
1840         } /* switch(bond_mode) */
1841
1842         write_unlock_bh(&bond->curr_slave_lock);
1843
1844         bond_set_carrier(bond);
1845
1846 #ifdef CONFIG_NET_POLL_CONTROLLER
1847         slave_dev->npinfo = bond_netpoll_info(bond);
1848         if (slave_dev->npinfo) {
1849                 if (slave_enable_netpoll(new_slave)) {
1850                         read_unlock(&bond->lock);
1851                         pr_info("Error, %s: master_dev is using netpoll, "
1852                                  "but new slave device does not support netpoll.\n",
1853                                  bond_dev->name);
1854                         res = -EBUSY;
1855                         goto err_detach;
1856                 }
1857         }
1858 #endif
1859
1860         read_unlock(&bond->lock);
1861
1862         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1863         if (res)
1864                 goto err_detach;
1865
1866         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1867                                          new_slave);
1868         if (res) {
1869                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1870                 goto err_dest_symlinks;
1871         }
1872
1873         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1874                 bond_dev->name, slave_dev->name,
1875                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1876                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1877
1878         /* enslave is successful */
1879         return 0;
1880
1881 /* Undo stages on error */
1882 err_dest_symlinks:
1883         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1884
1885 err_detach:
1886         write_lock_bh(&bond->lock);
1887         bond_detach_slave(bond, new_slave);
1888         write_unlock_bh(&bond->lock);
1889
1890 err_close:
1891         dev_close(slave_dev);
1892
1893 err_unset_master:
1894         netdev_set_bond_master(slave_dev, NULL);
1895
1896 err_restore_mac:
1897         if (!bond->params.fail_over_mac) {
1898                 /* XXX TODO - fom follow mode needs to change master's
1899                  * MAC if this slave's MAC is in use by the bond, or at
1900                  * least print a warning.
1901                  */
1902                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1903                 addr.sa_family = slave_dev->type;
1904                 dev_set_mac_address(slave_dev, &addr);
1905         }
1906
1907 err_restore_mtu:
1908         dev_set_mtu(slave_dev, new_slave->original_mtu);
1909
1910 err_free:
1911         kfree(new_slave);
1912
1913 err_undo_flags:
1914         bond_compute_features(bond);
1915
1916         return res;
1917 }
1918
1919 /*
1920  * Try to release the slave device <slave> from the bond device <master>
1921  * It is legal to access curr_active_slave without a lock because all the function
1922  * is write-locked.
1923  *
1924  * The rules for slave state should be:
1925  *   for Active/Backup:
1926  *     Active stays on all backups go down
1927  *   for Bonded connections:
1928  *     The first up interface should be left on and all others downed.
1929  */
1930 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1931 {
1932         struct bonding *bond = netdev_priv(bond_dev);
1933         struct slave *slave, *oldcurrent;
1934         struct sockaddr addr;
1935         netdev_features_t old_features = bond_dev->features;
1936
1937         /* slave is not a slave or master is not master of this slave */
1938         if (!(slave_dev->flags & IFF_SLAVE) ||
1939             (slave_dev->master != bond_dev)) {
1940                 pr_err("%s: Error: cannot release %s.\n",
1941                        bond_dev->name, slave_dev->name);
1942                 return -EINVAL;
1943         }
1944
1945         block_netpoll_tx();
1946         call_netdevice_notifiers(NETDEV_RELEASE, bond_dev);
1947         write_lock_bh(&bond->lock);
1948
1949         slave = bond_get_slave_by_dev(bond, slave_dev);
1950         if (!slave) {
1951                 /* not a slave of this bond */
1952                 pr_info("%s: %s not enslaved\n",
1953                         bond_dev->name, slave_dev->name);
1954                 write_unlock_bh(&bond->lock);
1955                 unblock_netpoll_tx();
1956                 return -EINVAL;
1957         }
1958
1959         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1960          * for this slave anymore.
1961          */
1962         netdev_rx_handler_unregister(slave_dev);
1963         write_unlock_bh(&bond->lock);
1964         synchronize_net();
1965         write_lock_bh(&bond->lock);
1966
1967         if (!bond->params.fail_over_mac) {
1968                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1969                     bond->slave_cnt > 1)
1970                         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",
1971                                    bond_dev->name, slave_dev->name,
1972                                    slave->perm_hwaddr,
1973                                    bond_dev->name, slave_dev->name);
1974         }
1975
1976         /* Inform AD package of unbinding of slave. */
1977         if (bond->params.mode == BOND_MODE_8023AD) {
1978                 /* must be called before the slave is
1979                  * detached from the list
1980                  */
1981                 bond_3ad_unbind_slave(slave);
1982         }
1983
1984         pr_info("%s: releasing %s interface %s\n",
1985                 bond_dev->name,
1986                 bond_is_active_slave(slave) ? "active" : "backup",
1987                 slave_dev->name);
1988
1989         oldcurrent = bond->curr_active_slave;
1990
1991         bond->current_arp_slave = NULL;
1992
1993         /* release the slave from its bond */
1994         bond_detach_slave(bond, slave);
1995
1996         if (bond->primary_slave == slave)
1997                 bond->primary_slave = NULL;
1998
1999         if (oldcurrent == slave)
2000                 bond_change_active_slave(bond, NULL);
2001
2002         if (bond_is_lb(bond)) {
2003                 /* Must be called only after the slave has been
2004                  * detached from the list and the curr_active_slave
2005                  * has been cleared (if our_slave == old_current),
2006                  * but before a new active slave is selected.
2007                  */
2008                 write_unlock_bh(&bond->lock);
2009                 bond_alb_deinit_slave(bond, slave);
2010                 write_lock_bh(&bond->lock);
2011         }
2012
2013         if (oldcurrent == slave) {
2014                 /*
2015                  * Note that we hold RTNL over this sequence, so there
2016                  * is no concern that another slave add/remove event
2017                  * will interfere.
2018                  */
2019                 write_unlock_bh(&bond->lock);
2020                 read_lock(&bond->lock);
2021                 write_lock_bh(&bond->curr_slave_lock);
2022
2023                 bond_select_active_slave(bond);
2024
2025                 write_unlock_bh(&bond->curr_slave_lock);
2026                 read_unlock(&bond->lock);
2027                 write_lock_bh(&bond->lock);
2028         }
2029
2030         if (bond->slave_cnt == 0) {
2031                 bond_set_carrier(bond);
2032
2033                 /* if the last slave was removed, zero the mac address
2034                  * of the master so it will be set by the application
2035                  * to the mac address of the first slave
2036                  */
2037                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2038
2039                 if (bond_vlan_used(bond)) {
2040                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2041                                    bond_dev->name, bond_dev->name);
2042                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2043                                    bond_dev->name);
2044                 }
2045         }
2046
2047         write_unlock_bh(&bond->lock);
2048         unblock_netpoll_tx();
2049
2050         if (bond->slave_cnt == 0)
2051                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2052
2053         bond_compute_features(bond);
2054         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2055             (old_features & NETIF_F_VLAN_CHALLENGED))
2056                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2057                         bond_dev->name, slave_dev->name, bond_dev->name);
2058
2059         /* must do this from outside any spinlocks */
2060         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2061
2062         bond_del_vlans_from_slave(bond, slave_dev);
2063
2064         /* If the mode USES_PRIMARY, then we should only remove its
2065          * promisc and mc settings if it was the curr_active_slave, but that was
2066          * already taken care of above when we detached the slave
2067          */
2068         if (!USES_PRIMARY(bond->params.mode)) {
2069                 /* unset promiscuity level from slave */
2070                 if (bond_dev->flags & IFF_PROMISC)
2071                         dev_set_promiscuity(slave_dev, -1);
2072
2073                 /* unset allmulti level from slave */
2074                 if (bond_dev->flags & IFF_ALLMULTI)
2075                         dev_set_allmulti(slave_dev, -1);
2076
2077                 /* flush master's mc_list from slave */
2078                 netif_addr_lock_bh(bond_dev);
2079                 bond_mc_list_flush(bond_dev, slave_dev);
2080                 netif_addr_unlock_bh(bond_dev);
2081         }
2082
2083         netdev_set_bond_master(slave_dev, NULL);
2084
2085         slave_disable_netpoll(slave);
2086
2087         /* close slave before restoring its mac address */
2088         dev_close(slave_dev);
2089
2090         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2091                 /* restore original ("permanent") mac address */
2092                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2093                 addr.sa_family = slave_dev->type;
2094                 dev_set_mac_address(slave_dev, &addr);
2095         }
2096
2097         dev_set_mtu(slave_dev, slave->original_mtu);
2098
2099         slave_dev->priv_flags &= ~IFF_BONDING;
2100
2101         kfree(slave);
2102
2103         return 0;  /* deletion OK */
2104 }
2105
2106 /*
2107 * First release a slave and then destroy the bond if no more slaves are left.
2108 * Must be under rtnl_lock when this function is called.
2109 */
2110 static int  bond_release_and_destroy(struct net_device *bond_dev,
2111                                      struct net_device *slave_dev)
2112 {
2113         struct bonding *bond = netdev_priv(bond_dev);
2114         int ret;
2115
2116         ret = bond_release(bond_dev, slave_dev);
2117         if ((ret == 0) && (bond->slave_cnt == 0)) {
2118                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2119                 pr_info("%s: destroying bond %s.\n",
2120                         bond_dev->name, bond_dev->name);
2121                 unregister_netdevice(bond_dev);
2122         }
2123         return ret;
2124 }
2125
2126 /*
2127  * This function releases all slaves.
2128  */
2129 static int bond_release_all(struct net_device *bond_dev)
2130 {
2131         struct bonding *bond = netdev_priv(bond_dev);
2132         struct slave *slave;
2133         struct net_device *slave_dev;
2134         struct sockaddr addr;
2135
2136         write_lock_bh(&bond->lock);
2137
2138         netif_carrier_off(bond_dev);
2139
2140         if (bond->slave_cnt == 0)
2141                 goto out;
2142
2143         bond->current_arp_slave = NULL;
2144         bond->primary_slave = NULL;
2145         bond_change_active_slave(bond, NULL);
2146
2147         while ((slave = bond->first_slave) != NULL) {
2148                 /* Inform AD package of unbinding of slave
2149                  * before slave is detached from the list.
2150                  */
2151                 if (bond->params.mode == BOND_MODE_8023AD)
2152                         bond_3ad_unbind_slave(slave);
2153
2154                 slave_dev = slave->dev;
2155                 bond_detach_slave(bond, slave);
2156
2157                 /* now that the slave is detached, unlock and perform
2158                  * all the undo steps that should not be called from
2159                  * within a lock.
2160                  */
2161                 write_unlock_bh(&bond->lock);
2162
2163                 /* unregister rx_handler early so bond_handle_frame wouldn't
2164                  * be called for this slave anymore.
2165                  */
2166                 netdev_rx_handler_unregister(slave_dev);
2167                 synchronize_net();
2168
2169                 if (bond_is_lb(bond)) {
2170                         /* must be called only after the slave
2171                          * has been detached from the list
2172                          */
2173                         bond_alb_deinit_slave(bond, slave);
2174                 }
2175
2176                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2177                 bond_del_vlans_from_slave(bond, slave_dev);
2178
2179                 /* If the mode USES_PRIMARY, then we should only remove its
2180                  * promisc and mc settings if it was the curr_active_slave, but that was
2181                  * already taken care of above when we detached the slave
2182                  */
2183                 if (!USES_PRIMARY(bond->params.mode)) {
2184                         /* unset promiscuity level from slave */
2185                         if (bond_dev->flags & IFF_PROMISC)
2186                                 dev_set_promiscuity(slave_dev, -1);
2187
2188                         /* unset allmulti level from slave */
2189                         if (bond_dev->flags & IFF_ALLMULTI)
2190                                 dev_set_allmulti(slave_dev, -1);
2191
2192                         /* flush master's mc_list from slave */
2193                         netif_addr_lock_bh(bond_dev);
2194                         bond_mc_list_flush(bond_dev, slave_dev);
2195                         netif_addr_unlock_bh(bond_dev);
2196                 }
2197
2198                 netdev_set_bond_master(slave_dev, NULL);
2199
2200                 slave_disable_netpoll(slave);
2201
2202                 /* close slave before restoring its mac address */
2203                 dev_close(slave_dev);
2204
2205                 if (!bond->params.fail_over_mac) {
2206                         /* restore original ("permanent") mac address*/
2207                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2208                         addr.sa_family = slave_dev->type;
2209                         dev_set_mac_address(slave_dev, &addr);
2210                 }
2211
2212                 kfree(slave);
2213
2214                 /* re-acquire the lock before getting the next slave */
2215                 write_lock_bh(&bond->lock);
2216         }
2217
2218         /* zero the mac address of the master so it will be
2219          * set by the application to the mac address of the
2220          * first slave
2221          */
2222         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2223
2224         if (bond_vlan_used(bond)) {
2225                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2226                            bond_dev->name, bond_dev->name);
2227                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2228                            bond_dev->name);
2229         }
2230
2231         pr_info("%s: released all slaves\n", bond_dev->name);
2232
2233 out:
2234         write_unlock_bh(&bond->lock);
2235
2236         bond_compute_features(bond);
2237
2238         return 0;
2239 }
2240
2241 /*
2242  * This function changes the active slave to slave <slave_dev>.
2243  * It returns -EINVAL in the following cases.
2244  *  - <slave_dev> is not found in the list.
2245  *  - There is not active slave now.
2246  *  - <slave_dev> is already active.
2247  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2248  *  - <slave_dev> is not running.
2249  * In these cases, this function does nothing.
2250  * In the other cases, current_slave pointer is changed and 0 is returned.
2251  */
2252 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2253 {
2254         struct bonding *bond = netdev_priv(bond_dev);
2255         struct slave *old_active = NULL;
2256         struct slave *new_active = NULL;
2257         int res = 0;
2258
2259         if (!USES_PRIMARY(bond->params.mode))
2260                 return -EINVAL;
2261
2262         /* Verify that master_dev is indeed the master of slave_dev */
2263         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2264                 return -EINVAL;
2265
2266         read_lock(&bond->lock);
2267
2268         read_lock(&bond->curr_slave_lock);
2269         old_active = bond->curr_active_slave;
2270         read_unlock(&bond->curr_slave_lock);
2271
2272         new_active = bond_get_slave_by_dev(bond, slave_dev);
2273
2274         /*
2275          * Changing to the current active: do nothing; return success.
2276          */
2277         if (new_active && (new_active == old_active)) {
2278                 read_unlock(&bond->lock);
2279                 return 0;
2280         }
2281
2282         if ((new_active) &&
2283             (old_active) &&
2284             (new_active->link == BOND_LINK_UP) &&
2285             IS_UP(new_active->dev)) {
2286                 block_netpoll_tx();
2287                 write_lock_bh(&bond->curr_slave_lock);
2288                 bond_change_active_slave(bond, new_active);
2289                 write_unlock_bh(&bond->curr_slave_lock);
2290                 unblock_netpoll_tx();
2291         } else
2292                 res = -EINVAL;
2293
2294         read_unlock(&bond->lock);
2295
2296         return res;
2297 }
2298
2299 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2300 {
2301         struct bonding *bond = netdev_priv(bond_dev);
2302
2303         info->bond_mode = bond->params.mode;
2304         info->miimon = bond->params.miimon;
2305
2306         read_lock(&bond->lock);
2307         info->num_slaves = bond->slave_cnt;
2308         read_unlock(&bond->lock);
2309
2310         return 0;
2311 }
2312
2313 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2314 {
2315         struct bonding *bond = netdev_priv(bond_dev);
2316         struct slave *slave;
2317         int i, res = -ENODEV;
2318
2319         read_lock(&bond->lock);
2320
2321         bond_for_each_slave(bond, slave, i) {
2322                 if (i == (int)info->slave_id) {
2323                         res = 0;
2324                         strcpy(info->slave_name, slave->dev->name);
2325                         info->link = slave->link;
2326                         info->state = bond_slave_state(slave);
2327                         info->link_failure_count = slave->link_failure_count;
2328                         break;
2329                 }
2330         }
2331
2332         read_unlock(&bond->lock);
2333
2334         return res;
2335 }
2336
2337 /*-------------------------------- Monitoring -------------------------------*/
2338
2339
2340 static int bond_miimon_inspect(struct bonding *bond)
2341 {
2342         struct slave *slave;
2343         int i, link_state, commit = 0;
2344         bool ignore_updelay;
2345
2346         ignore_updelay = !bond->curr_active_slave ? true : false;
2347
2348         bond_for_each_slave(bond, slave, i) {
2349                 slave->new_link = BOND_LINK_NOCHANGE;
2350
2351                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2352
2353                 switch (slave->link) {
2354                 case BOND_LINK_UP:
2355                         if (link_state)
2356                                 continue;
2357
2358                         slave->link = BOND_LINK_FAIL;
2359                         slave->delay = bond->params.downdelay;
2360                         if (slave->delay) {
2361                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2362                                         bond->dev->name,
2363                                         (bond->params.mode ==
2364                                          BOND_MODE_ACTIVEBACKUP) ?
2365                                         (bond_is_active_slave(slave) ?
2366                                          "active " : "backup ") : "",
2367                                         slave->dev->name,
2368                                         bond->params.downdelay * bond->params.miimon);
2369                         }
2370                         /*FALLTHRU*/
2371                 case BOND_LINK_FAIL:
2372                         if (link_state) {
2373                                 /*
2374                                  * recovered before downdelay expired
2375                                  */
2376                                 slave->link = BOND_LINK_UP;
2377                                 slave->jiffies = jiffies;
2378                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2379                                         bond->dev->name,
2380                                         (bond->params.downdelay - slave->delay) *
2381                                         bond->params.miimon,
2382                                         slave->dev->name);
2383                                 continue;
2384                         }
2385
2386                         if (slave->delay <= 0) {
2387                                 slave->new_link = BOND_LINK_DOWN;
2388                                 commit++;
2389                                 continue;
2390                         }
2391
2392                         slave->delay--;
2393                         break;
2394
2395                 case BOND_LINK_DOWN:
2396                         if (!link_state)
2397                                 continue;
2398
2399                         slave->link = BOND_LINK_BACK;
2400                         slave->delay = bond->params.updelay;
2401
2402                         if (slave->delay) {
2403                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2404                                         bond->dev->name, slave->dev->name,
2405                                         ignore_updelay ? 0 :
2406                                         bond->params.updelay *
2407                                         bond->params.miimon);
2408                         }
2409                         /*FALLTHRU*/
2410                 case BOND_LINK_BACK:
2411                         if (!link_state) {
2412                                 slave->link = BOND_LINK_DOWN;
2413                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2414                                         bond->dev->name,
2415                                         (bond->params.updelay - slave->delay) *
2416                                         bond->params.miimon,
2417                                         slave->dev->name);
2418
2419                                 continue;
2420                         }
2421
2422                         if (ignore_updelay)
2423                                 slave->delay = 0;
2424
2425                         if (slave->delay <= 0) {
2426                                 slave->new_link = BOND_LINK_UP;
2427                                 commit++;
2428                                 ignore_updelay = false;
2429                                 continue;
2430                         }
2431
2432                         slave->delay--;
2433                         break;
2434                 }
2435         }
2436
2437         return commit;
2438 }
2439
2440 static void bond_miimon_commit(struct bonding *bond)
2441 {
2442         struct slave *slave;
2443         int i;
2444
2445         bond_for_each_slave(bond, slave, i) {
2446                 switch (slave->new_link) {
2447                 case BOND_LINK_NOCHANGE:
2448                         continue;
2449
2450                 case BOND_LINK_UP:
2451                         slave->link = BOND_LINK_UP;
2452                         slave->jiffies = jiffies;
2453
2454                         if (bond->params.mode == BOND_MODE_8023AD) {
2455                                 /* prevent it from being the active one */
2456                                 bond_set_backup_slave(slave);
2457                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2458                                 /* make it immediately active */
2459                                 bond_set_active_slave(slave);
2460                         } else if (slave != bond->primary_slave) {
2461                                 /* prevent it from being the active one */
2462                                 bond_set_backup_slave(slave);
2463                         }
2464
2465                         bond_update_speed_duplex(slave);
2466
2467                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2468                                 bond->dev->name, slave->dev->name,
2469                                 slave->speed, slave->duplex ? "full" : "half");
2470
2471                         /* notify ad that the link status has changed */
2472                         if (bond->params.mode == BOND_MODE_8023AD)
2473                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2474
2475                         if (bond_is_lb(bond))
2476                                 bond_alb_handle_link_change(bond, slave,
2477                                                             BOND_LINK_UP);
2478
2479                         if (!bond->curr_active_slave ||
2480                             (slave == bond->primary_slave))
2481                                 goto do_failover;
2482
2483                         continue;
2484
2485                 case BOND_LINK_DOWN:
2486                         if (slave->link_failure_count < UINT_MAX)
2487                                 slave->link_failure_count++;
2488
2489                         slave->link = BOND_LINK_DOWN;
2490
2491                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2492                             bond->params.mode == BOND_MODE_8023AD)
2493                                 bond_set_slave_inactive_flags(slave);
2494
2495                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2496                                 bond->dev->name, slave->dev->name);
2497
2498                         if (bond->params.mode == BOND_MODE_8023AD)
2499                                 bond_3ad_handle_link_change(slave,
2500                                                             BOND_LINK_DOWN);
2501
2502                         if (bond_is_lb(bond))
2503                                 bond_alb_handle_link_change(bond, slave,
2504                                                             BOND_LINK_DOWN);
2505
2506                         if (slave == bond->curr_active_slave)
2507                                 goto do_failover;
2508
2509                         continue;
2510
2511                 default:
2512                         pr_err("%s: invalid new link %d on slave %s\n",
2513                                bond->dev->name, slave->new_link,
2514                                slave->dev->name);
2515                         slave->new_link = BOND_LINK_NOCHANGE;
2516
2517                         continue;
2518                 }
2519
2520 do_failover:
2521                 ASSERT_RTNL();
2522                 block_netpoll_tx();
2523                 write_lock_bh(&bond->curr_slave_lock);
2524                 bond_select_active_slave(bond);
2525                 write_unlock_bh(&bond->curr_slave_lock);
2526                 unblock_netpoll_tx();
2527         }
2528
2529         bond_set_carrier(bond);
2530 }
2531
2532 /*
2533  * bond_mii_monitor
2534  *
2535  * Really a wrapper that splits the mii monitor into two phases: an
2536  * inspection, then (if inspection indicates something needs to be done)
2537  * an acquisition of appropriate locks followed by a commit phase to
2538  * implement whatever link state changes are indicated.
2539  */
2540 void bond_mii_monitor(struct work_struct *work)
2541 {
2542         struct bonding *bond = container_of(work, struct bonding,
2543                                             mii_work.work);
2544         bool should_notify_peers = false;
2545         unsigned long delay;
2546
2547         read_lock(&bond->lock);
2548
2549         delay = msecs_to_jiffies(bond->params.miimon);
2550
2551         if (bond->slave_cnt == 0)
2552                 goto re_arm;
2553
2554         should_notify_peers = bond_should_notify_peers(bond);
2555
2556         if (bond_miimon_inspect(bond)) {
2557                 read_unlock(&bond->lock);
2558
2559                 /* Race avoidance with bond_close cancel of workqueue */
2560                 if (!rtnl_trylock()) {
2561                         read_lock(&bond->lock);
2562                         delay = 1;
2563                         should_notify_peers = false;
2564                         goto re_arm;
2565                 }
2566
2567                 read_lock(&bond->lock);
2568
2569                 bond_miimon_commit(bond);
2570
2571                 read_unlock(&bond->lock);
2572                 rtnl_unlock();  /* might sleep, hold no other locks */
2573                 read_lock(&bond->lock);
2574         }
2575
2576 re_arm:
2577         if (bond->params.miimon)
2578                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2579
2580         read_unlock(&bond->lock);
2581
2582         if (should_notify_peers) {
2583                 if (!rtnl_trylock()) {
2584                         read_lock(&bond->lock);
2585                         bond->send_peer_notif++;
2586                         read_unlock(&bond->lock);
2587                         return;
2588                 }
2589                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2590                 rtnl_unlock();
2591         }
2592 }
2593
2594 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2595 {
2596         struct vlan_entry *vlan;
2597         struct net_device *vlan_dev;
2598
2599         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2600                 return 1;
2601
2602         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2603                 rcu_read_lock();
2604                 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2605                 rcu_read_unlock();
2606                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2607                         return 1;
2608         }
2609
2610         return 0;
2611 }
2612
2613 /*
2614  * We go to the (large) trouble of VLAN tagging ARP frames because
2615  * switches in VLAN mode (especially if ports are configured as
2616  * "native" to a VLAN) might not pass non-tagged frames.
2617  */
2618 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2619 {
2620         struct sk_buff *skb;
2621
2622         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2623                  slave_dev->name, dest_ip, src_ip, vlan_id);
2624
2625         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2626                          NULL, slave_dev->dev_addr, NULL);
2627
2628         if (!skb) {
2629                 pr_err("ARP packet allocation failed\n");
2630                 return;
2631         }
2632         if (vlan_id) {
2633                 skb = vlan_put_tag(skb, vlan_id);
2634                 if (!skb) {
2635                         pr_err("failed to insert VLAN tag\n");
2636                         return;
2637                 }
2638         }
2639         arp_xmit(skb);
2640 }
2641
2642
2643 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2644 {
2645         int i, vlan_id;
2646         __be32 *targets = bond->params.arp_targets;
2647         struct vlan_entry *vlan;
2648         struct net_device *vlan_dev = NULL;
2649         struct rtable *rt;
2650
2651         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2652                 __be32 addr;
2653                 if (!targets[i])
2654                         break;
2655                 pr_debug("basa: target %x\n", targets[i]);
2656                 if (!bond_vlan_used(bond)) {
2657                         pr_debug("basa: empty vlan: arp_send\n");
2658                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2659                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2660                                       addr, 0);
2661                         continue;
2662                 }
2663
2664                 /*
2665                  * If VLANs are configured, we do a route lookup to
2666                  * determine which VLAN interface would be used, so we
2667                  * can tag the ARP with the proper VLAN tag.
2668                  */
2669                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2670                                      RTO_ONLINK, 0);
2671                 if (IS_ERR(rt)) {
2672                         if (net_ratelimit()) {
2673                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2674                                            bond->dev->name, &targets[i]);
2675                         }
2676                         continue;
2677                 }
2678
2679                 /*
2680                  * This target is not on a VLAN
2681                  */
2682                 if (rt->dst.dev == bond->dev) {
2683                         ip_rt_put(rt);
2684                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2685                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2686                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2687                                       addr, 0);
2688                         continue;
2689                 }
2690
2691                 vlan_id = 0;
2692                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2693                         rcu_read_lock();
2694                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2695                                                         vlan->vlan_id);
2696                         rcu_read_unlock();
2697                         if (vlan_dev == rt->dst.dev) {
2698                                 vlan_id = vlan->vlan_id;
2699                                 pr_debug("basa: vlan match on %s %d\n",
2700                                        vlan_dev->name, vlan_id);
2701                                 break;
2702                         }
2703                 }
2704
2705                 if (vlan_id && vlan_dev) {
2706                         ip_rt_put(rt);
2707                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2708                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2709                                       addr, vlan_id);
2710                         continue;
2711                 }
2712
2713                 if (net_ratelimit()) {
2714                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2715                                    bond->dev->name, &targets[i],
2716                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2717                 }
2718                 ip_rt_put(rt);
2719         }
2720 }
2721
2722 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2723 {
2724         int i;
2725         __be32 *targets = bond->params.arp_targets;
2726
2727         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2728                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2729                          &sip, &tip, i, &targets[i],
2730                          bond_has_this_ip(bond, tip));
2731                 if (sip == targets[i]) {
2732                         if (bond_has_this_ip(bond, tip))
2733                                 slave->last_arp_rx = jiffies;
2734                         return;
2735                 }
2736         }
2737 }
2738
2739 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2740                         struct slave *slave)
2741 {
2742         struct arphdr *arp = (struct arphdr *)skb->data;
2743         unsigned char *arp_ptr;
2744         __be32 sip, tip;
2745         int alen;
2746
2747         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2748                 return RX_HANDLER_ANOTHER;
2749
2750         read_lock(&bond->lock);
2751         alen = arp_hdr_len(bond->dev);
2752
2753         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2754                  bond->dev->name, skb->dev->name);
2755
2756         if (alen > skb_headlen(skb)) {
2757                 arp = kmalloc(alen, GFP_ATOMIC);
2758                 if (!arp)
2759                         goto out_unlock;
2760                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2761                         goto out_unlock;
2762         }
2763
2764         if (arp->ar_hln != bond->dev->addr_len ||
2765             skb->pkt_type == PACKET_OTHERHOST ||
2766             skb->pkt_type == PACKET_LOOPBACK ||
2767             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2768             arp->ar_pro != htons(ETH_P_IP) ||
2769             arp->ar_pln != 4)
2770                 goto out_unlock;
2771
2772         arp_ptr = (unsigned char *)(arp + 1);
2773         arp_ptr += bond->dev->addr_len;
2774         memcpy(&sip, arp_ptr, 4);
2775         arp_ptr += 4 + bond->dev->addr_len;
2776         memcpy(&tip, arp_ptr, 4);
2777
2778         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2779                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2780                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2781                  &sip, &tip);
2782
2783         /*
2784          * Backup slaves won't see the ARP reply, but do come through
2785          * here for each ARP probe (so we swap the sip/tip to validate
2786          * the probe).  In a "redundant switch, common router" type of
2787          * configuration, the ARP probe will (hopefully) travel from
2788          * the active, through one switch, the router, then the other
2789          * switch before reaching the backup.
2790          */
2791         if (bond_is_active_slave(slave))
2792                 bond_validate_arp(bond, slave, sip, tip);
2793         else
2794                 bond_validate_arp(bond, slave, tip, sip);
2795
2796 out_unlock:
2797         read_unlock(&bond->lock);
2798         if (arp != (struct arphdr *)skb->data)
2799                 kfree(arp);
2800         return RX_HANDLER_ANOTHER;
2801 }
2802
2803 /*
2804  * this function is called regularly to monitor each slave's link
2805  * ensuring that traffic is being sent and received when arp monitoring
2806  * is used in load-balancing mode. if the adapter has been dormant, then an
2807  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2808  * arp monitoring in active backup mode.
2809  */
2810 void bond_loadbalance_arp_mon(struct work_struct *work)
2811 {
2812         struct bonding *bond = container_of(work, struct bonding,
2813                                             arp_work.work);
2814         struct slave *slave, *oldcurrent;
2815         int do_failover = 0;
2816         int delta_in_ticks, extra_ticks;
2817         int i;
2818
2819         read_lock(&bond->lock);
2820
2821         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2822         extra_ticks = delta_in_ticks / 2;
2823
2824         if (bond->slave_cnt == 0)
2825                 goto re_arm;
2826
2827         read_lock(&bond->curr_slave_lock);
2828         oldcurrent = bond->curr_active_slave;
2829         read_unlock(&bond->curr_slave_lock);
2830
2831         /* see if any of the previous devices are up now (i.e. they have
2832          * xmt and rcv traffic). the curr_active_slave does not come into
2833          * the picture unless it is null. also, slave->jiffies is not needed
2834          * here because we send an arp on each slave and give a slave as
2835          * long as it needs to get the tx/rx within the delta.
2836          * TODO: what about up/down delay in arp mode? it wasn't here before
2837          *       so it can wait
2838          */
2839         bond_for_each_slave(bond, slave, i) {
2840                 unsigned long trans_start = dev_trans_start(slave->dev);
2841
2842                 if (slave->link != BOND_LINK_UP) {
2843                         if (time_in_range(jiffies,
2844                                 trans_start - delta_in_ticks,
2845                                 trans_start + delta_in_ticks + extra_ticks) &&
2846                             time_in_range(jiffies,
2847                                 slave->dev->last_rx - delta_in_ticks,
2848                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2849
2850                                 slave->link  = BOND_LINK_UP;
2851                                 bond_set_active_slave(slave);
2852
2853                                 /* primary_slave has no meaning in round-robin
2854                                  * mode. the window of a slave being up and
2855                                  * curr_active_slave being null after enslaving
2856                                  * is closed.
2857                                  */
2858                                 if (!oldcurrent) {
2859                                         pr_info("%s: link status definitely up for interface %s, ",
2860                                                 bond->dev->name,
2861                                                 slave->dev->name);
2862                                         do_failover = 1;
2863                                 } else {
2864                                         pr_info("%s: interface %s is now up\n",
2865                                                 bond->dev->name,
2866                                                 slave->dev->name);
2867                                 }
2868                         }
2869                 } else {
2870                         /* slave->link == BOND_LINK_UP */
2871
2872                         /* not all switches will respond to an arp request
2873                          * when the source ip is 0, so don't take the link down
2874                          * if we don't know our ip yet
2875                          */
2876                         if (!time_in_range(jiffies,
2877                                 trans_start - delta_in_ticks,
2878                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2879                             !time_in_range(jiffies,
2880                                 slave->dev->last_rx - delta_in_ticks,
2881                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2882
2883                                 slave->link  = BOND_LINK_DOWN;
2884                                 bond_set_backup_slave(slave);
2885
2886                                 if (slave->link_failure_count < UINT_MAX)
2887                                         slave->link_failure_count++;
2888
2889                                 pr_info("%s: interface %s is now down.\n",
2890                                         bond->dev->name,
2891                                         slave->dev->name);
2892
2893                                 if (slave == oldcurrent)
2894                                         do_failover = 1;
2895                         }
2896                 }
2897
2898                 /* note: if switch is in round-robin mode, all links
2899                  * must tx arp to ensure all links rx an arp - otherwise
2900                  * links may oscillate or not come up at all; if switch is
2901                  * in something like xor mode, there is nothing we can
2902                  * do - all replies will be rx'ed on same link causing slaves
2903                  * to be unstable during low/no traffic periods
2904                  */
2905                 if (IS_UP(slave->dev))
2906                         bond_arp_send_all(bond, slave);
2907         }
2908
2909         if (do_failover) {
2910                 block_netpoll_tx();
2911                 write_lock_bh(&bond->curr_slave_lock);
2912
2913                 bond_select_active_slave(bond);
2914
2915                 write_unlock_bh(&bond->curr_slave_lock);
2916                 unblock_netpoll_tx();
2917         }
2918
2919 re_arm:
2920         if (bond->params.arp_interval)
2921                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2922
2923         read_unlock(&bond->lock);
2924 }
2925
2926 /*
2927  * Called to inspect slaves for active-backup mode ARP monitor link state
2928  * changes.  Sets new_link in slaves to specify what action should take
2929  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2930  * to link states must be committed.
2931  *
2932  * Called with bond->lock held for read.
2933  */
2934 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2935 {
2936         struct slave *slave;
2937         int i, commit = 0;
2938         unsigned long trans_start;
2939         int extra_ticks;
2940
2941         /* All the time comparisons below need some extra time. Otherwise, on
2942          * fast networks the ARP probe/reply may arrive within the same jiffy
2943          * as it was sent.  Then, the next time the ARP monitor is run, one
2944          * arp_interval will already have passed in the comparisons.
2945          */
2946         extra_ticks = delta_in_ticks / 2;
2947
2948         bond_for_each_slave(bond, slave, i) {
2949                 slave->new_link = BOND_LINK_NOCHANGE;
2950
2951                 if (slave->link != BOND_LINK_UP) {
2952                         if (time_in_range(jiffies,
2953                                 slave_last_rx(bond, slave) - delta_in_ticks,
2954                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2955
2956                                 slave->new_link = BOND_LINK_UP;
2957                                 commit++;
2958                         }
2959
2960                         continue;
2961                 }
2962
2963                 /*
2964                  * Give slaves 2*delta after being enslaved or made
2965                  * active.  This avoids bouncing, as the last receive
2966                  * times need a full ARP monitor cycle to be updated.
2967                  */
2968                 if (time_in_range(jiffies,
2969                                   slave->jiffies - delta_in_ticks,
2970                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2971                         continue;
2972
2973                 /*
2974                  * Backup slave is down if:
2975                  * - No current_arp_slave AND
2976                  * - more than 3*delta since last receive AND
2977                  * - the bond has an IP address
2978                  *
2979                  * Note: a non-null current_arp_slave indicates
2980                  * the curr_active_slave went down and we are
2981                  * searching for a new one; under this condition
2982                  * we only take the curr_active_slave down - this
2983                  * gives each slave a chance to tx/rx traffic
2984                  * before being taken out
2985                  */
2986                 if (!bond_is_active_slave(slave) &&
2987                     !bond->current_arp_slave &&
2988                     !time_in_range(jiffies,
2989                         slave_last_rx(bond, slave) - delta_in_ticks,
2990                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2991
2992                         slave->new_link = BOND_LINK_DOWN;
2993                         commit++;
2994                 }
2995
2996                 /*
2997                  * Active slave is down if:
2998                  * - more than 2*delta since transmitting OR
2999                  * - (more than 2*delta since receive AND
3000                  *    the bond has an IP address)
3001                  */
3002                 trans_start = dev_trans_start(slave->dev);
3003                 if (bond_is_active_slave(slave) &&
3004                     (!time_in_range(jiffies,
3005                         trans_start - delta_in_ticks,
3006                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
3007                      !time_in_range(jiffies,
3008                         slave_last_rx(bond, slave) - delta_in_ticks,
3009                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
3010
3011                         slave->new_link = BOND_LINK_DOWN;
3012                         commit++;
3013                 }
3014         }
3015
3016         return commit;
3017 }
3018
3019 /*
3020  * Called to commit link state changes noted by inspection step of
3021  * active-backup mode ARP monitor.
3022  *
3023  * Called with RTNL and bond->lock for read.
3024  */
3025 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3026 {
3027         struct slave *slave;
3028         int i;
3029         unsigned long trans_start;
3030
3031         bond_for_each_slave(bond, slave, i) {
3032                 switch (slave->new_link) {
3033                 case BOND_LINK_NOCHANGE:
3034                         continue;
3035
3036                 case BOND_LINK_UP:
3037                         trans_start = dev_trans_start(slave->dev);
3038                         if ((!bond->curr_active_slave &&
3039                              time_in_range(jiffies,
3040                                            trans_start - delta_in_ticks,
3041                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
3042                             bond->curr_active_slave != slave) {
3043                                 slave->link = BOND_LINK_UP;
3044                                 if (bond->current_arp_slave) {
3045                                         bond_set_slave_inactive_flags(
3046                                                 bond->current_arp_slave);
3047                                         bond->current_arp_slave = NULL;
3048                                 }
3049
3050                                 pr_info("%s: link status definitely up for interface %s.\n",
3051                                         bond->dev->name, slave->dev->name);
3052
3053                                 if (!bond->curr_active_slave ||
3054                                     (slave == bond->primary_slave))
3055                                         goto do_failover;
3056
3057                         }
3058
3059                         continue;
3060
3061                 case BOND_LINK_DOWN:
3062                         if (slave->link_failure_count < UINT_MAX)
3063                                 slave->link_failure_count++;
3064
3065                         slave->link = BOND_LINK_DOWN;
3066                         bond_set_slave_inactive_flags(slave);
3067
3068                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3069                                 bond->dev->name, slave->dev->name);
3070
3071                         if (slave == bond->curr_active_slave) {
3072                                 bond->current_arp_slave = NULL;
3073                                 goto do_failover;
3074                         }
3075
3076                         continue;
3077
3078                 default:
3079                         pr_err("%s: impossible: new_link %d on slave %s\n",
3080                                bond->dev->name, slave->new_link,
3081                                slave->dev->name);
3082                         continue;
3083                 }
3084
3085 do_failover:
3086                 ASSERT_RTNL();
3087                 block_netpoll_tx();
3088                 write_lock_bh(&bond->curr_slave_lock);
3089                 bond_select_active_slave(bond);
3090                 write_unlock_bh(&bond->curr_slave_lock);
3091                 unblock_netpoll_tx();
3092         }
3093
3094         bond_set_carrier(bond);
3095 }
3096
3097 /*
3098  * Send ARP probes for active-backup mode ARP monitor.
3099  *
3100  * Called with bond->lock held for read.
3101  */
3102 static void bond_ab_arp_probe(struct bonding *bond)
3103 {
3104         struct slave *slave;
3105         int i;
3106
3107         read_lock(&bond->curr_slave_lock);
3108
3109         if (bond->current_arp_slave && bond->curr_active_slave)
3110                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3111                         bond->current_arp_slave->dev->name,
3112                         bond->curr_active_slave->dev->name);
3113
3114         if (bond->curr_active_slave) {
3115                 bond_arp_send_all(bond, bond->curr_active_slave);
3116                 read_unlock(&bond->curr_slave_lock);
3117                 return;
3118         }
3119
3120         read_unlock(&bond->curr_slave_lock);
3121
3122         /* if we don't have a curr_active_slave, search for the next available
3123          * backup slave from the current_arp_slave and make it the candidate
3124          * for becoming the curr_active_slave
3125          */
3126
3127         if (!bond->current_arp_slave) {
3128                 bond->current_arp_slave = bond->first_slave;
3129                 if (!bond->current_arp_slave)
3130                         return;
3131         }
3132
3133         bond_set_slave_inactive_flags(bond->current_arp_slave);
3134
3135         /* search for next candidate */
3136         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3137                 if (IS_UP(slave->dev)) {
3138                         slave->link = BOND_LINK_BACK;
3139                         bond_set_slave_active_flags(slave);
3140                         bond_arp_send_all(bond, slave);
3141                         slave->jiffies = jiffies;
3142                         bond->current_arp_slave = slave;
3143                         break;
3144                 }
3145
3146                 /* if the link state is up at this point, we
3147                  * mark it down - this can happen if we have
3148                  * simultaneous link failures and
3149                  * reselect_active_interface doesn't make this
3150                  * one the current slave so it is still marked
3151                  * up when it is actually down
3152                  */
3153                 if (slave->link == BOND_LINK_UP) {
3154                         slave->link = BOND_LINK_DOWN;
3155                         if (slave->link_failure_count < UINT_MAX)
3156                                 slave->link_failure_count++;
3157
3158                         bond_set_slave_inactive_flags(slave);
3159
3160                         pr_info("%s: backup interface %s is now down.\n",
3161                                 bond->dev->name, slave->dev->name);
3162                 }
3163         }
3164 }
3165
3166 void bond_activebackup_arp_mon(struct work_struct *work)
3167 {
3168         struct bonding *bond = container_of(work, struct bonding,
3169                                             arp_work.work);
3170         bool should_notify_peers = false;
3171         int delta_in_ticks;
3172
3173         read_lock(&bond->lock);
3174
3175         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3176
3177         if (bond->slave_cnt == 0)
3178                 goto re_arm;
3179
3180         should_notify_peers = bond_should_notify_peers(bond);
3181
3182         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3183                 read_unlock(&bond->lock);
3184
3185                 /* Race avoidance with bond_close flush of workqueue */
3186                 if (!rtnl_trylock()) {
3187                         read_lock(&bond->lock);
3188                         delta_in_ticks = 1;
3189                         should_notify_peers = false;
3190                         goto re_arm;
3191                 }
3192
3193                 read_lock(&bond->lock);
3194
3195                 bond_ab_arp_commit(bond, delta_in_ticks);
3196
3197                 read_unlock(&bond->lock);
3198                 rtnl_unlock();
3199                 read_lock(&bond->lock);
3200         }
3201
3202         bond_ab_arp_probe(bond);
3203
3204 re_arm:
3205         if (bond->params.arp_interval)
3206                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3207
3208         read_unlock(&bond->lock);
3209
3210         if (should_notify_peers) {
3211                 if (!rtnl_trylock()) {
3212                         read_lock(&bond->lock);
3213                         bond->send_peer_notif++;
3214                         read_unlock(&bond->lock);
3215                         return;
3216                 }
3217                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3218                 rtnl_unlock();
3219         }
3220 }
3221
3222 /*-------------------------- netdev event handling --------------------------*/
3223
3224 /*
3225  * Change device name
3226  */
3227 static int bond_event_changename(struct bonding *bond)
3228 {
3229         bond_remove_proc_entry(bond);
3230         bond_create_proc_entry(bond);
3231
3232         bond_debug_reregister(bond);
3233
3234         return NOTIFY_DONE;
3235 }
3236
3237 static int bond_master_netdev_event(unsigned long event,
3238                                     struct net_device *bond_dev)
3239 {
3240         struct bonding *event_bond = netdev_priv(bond_dev);
3241
3242         switch (event) {
3243         case NETDEV_CHANGENAME:
3244                 return bond_event_changename(event_bond);
3245         case NETDEV_UNREGISTER:
3246                 bond_remove_proc_entry(event_bond);
3247                 break;
3248         case NETDEV_REGISTER:
3249                 bond_create_proc_entry(event_bond);
3250                 break;
3251         default:
3252                 break;
3253         }
3254
3255         return NOTIFY_DONE;
3256 }
3257
3258 static int bond_slave_netdev_event(unsigned long event,
3259                                    struct net_device *slave_dev)
3260 {
3261         struct net_device *bond_dev = slave_dev->master;
3262         struct bonding *bond = netdev_priv(bond_dev);
3263         struct slave *slave = NULL;
3264
3265         switch (event) {
3266         case NETDEV_UNREGISTER:
3267                 if (bond_dev) {
3268                         if (bond->setup_by_slave)
3269                                 bond_release_and_destroy(bond_dev, slave_dev);
3270                         else
3271                                 bond_release(bond_dev, slave_dev);
3272                 }
3273                 break;
3274         case NETDEV_UP:
3275         case NETDEV_CHANGE:
3276                 slave = bond_get_slave_by_dev(bond, slave_dev);
3277                 if (slave) {
3278                         u32 old_speed = slave->speed;
3279                         u8  old_duplex = slave->duplex;
3280
3281                         bond_update_speed_duplex(slave);
3282
3283                         if (bond->params.mode == BOND_MODE_8023AD) {
3284                                 if (old_speed != slave->speed)
3285                                         bond_3ad_adapter_speed_changed(slave);
3286                                 if (old_duplex != slave->duplex)
3287                                         bond_3ad_adapter_duplex_changed(slave);
3288                         }
3289                 }
3290
3291                 break;
3292         case NETDEV_DOWN:
3293                 /*
3294                  * ... Or is it this?
3295                  */
3296                 break;
3297         case NETDEV_CHANGEMTU:
3298                 /*
3299                  * TODO: Should slaves be allowed to
3300                  * independently alter their MTU?  For
3301                  * an active-backup bond, slaves need
3302                  * not be the same type of device, so
3303                  * MTUs may vary.  For other modes,
3304                  * slaves arguably should have the
3305                  * same MTUs. To do this, we'd need to
3306                  * take over the slave's change_mtu
3307                  * function for the duration of their
3308                  * servitude.
3309                  */
3310                 break;
3311         case NETDEV_CHANGENAME:
3312                 /*
3313                  * TODO: handle changing the primary's name
3314                  */
3315                 break;
3316         case NETDEV_FEAT_CHANGE:
3317                 bond_compute_features(bond);
3318                 break;
3319         default:
3320                 break;
3321         }
3322
3323         return NOTIFY_DONE;
3324 }
3325
3326 /*
3327  * bond_netdev_event: handle netdev notifier chain events.
3328  *
3329  * This function receives events for the netdev chain.  The caller (an
3330  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3331  * locks for us to safely manipulate the slave devices (RTNL lock,
3332  * dev_probe_lock).
3333  */
3334 static int bond_netdev_event(struct notifier_block *this,
3335                              unsigned long event, void *ptr)
3336 {
3337         struct net_device *event_dev = (struct net_device *)ptr;
3338
3339         pr_debug("event_dev: %s, event: %lx\n",
3340                  event_dev ? event_dev->name : "None",
3341                  event);
3342
3343         if (!(event_dev->priv_flags & IFF_BONDING))
3344                 return NOTIFY_DONE;
3345
3346         if (event_dev->flags & IFF_MASTER) {
3347                 pr_debug("IFF_MASTER\n");
3348                 return bond_master_netdev_event(event, event_dev);
3349         }
3350
3351         if (event_dev->flags & IFF_SLAVE) {
3352                 pr_debug("IFF_SLAVE\n");
3353                 return bond_slave_netdev_event(event, event_dev);
3354         }
3355
3356         return NOTIFY_DONE;
3357 }
3358
3359 static struct notifier_block bond_netdev_notifier = {
3360         .notifier_call = bond_netdev_event,
3361 };
3362
3363 /*---------------------------- Hashing Policies -----------------------------*/
3364
3365 /*
3366  * Hash for the output device based upon layer 2 data
3367  */
3368 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3369 {
3370         struct ethhdr *data = (struct ethhdr *)skb->data;
3371
3372         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3373                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3374
3375         return 0;
3376 }
3377
3378 /*
3379  * Hash for the output device based upon layer 2 and layer 3 data. If
3380  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3381  */
3382 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3383 {
3384         struct ethhdr *data = (struct ethhdr *)skb->data;
3385         struct iphdr *iph;
3386         struct ipv6hdr *ipv6h;
3387         u32 v6hash;
3388         __be32 *s, *d;
3389
3390         if (skb->protocol == htons(ETH_P_IP) &&
3391             skb_network_header_len(skb) >= sizeof(*iph)) {
3392                 iph = ip_hdr(skb);
3393                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3394                         (data->h_dest[5] ^ data->h_source[5])) % count;
3395         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3396                    skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3397                 ipv6h = ipv6_hdr(skb);
3398                 s = &ipv6h->saddr.s6_addr32[0];
3399                 d = &ipv6h->daddr.s6_addr32[0];
3400                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3401                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3402                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3403         }
3404
3405         return bond_xmit_hash_policy_l2(skb, count);
3406 }
3407
3408 /*
3409  * Hash for the output device based upon layer 3 and layer 4 data. If
3410  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3411  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3412  */
3413 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3414 {
3415         u32 layer4_xor = 0;
3416         struct iphdr *iph;
3417         struct ipv6hdr *ipv6h;
3418         __be32 *s, *d;
3419         __be16 *layer4hdr;
3420
3421         if (skb->protocol == htons(ETH_P_IP) &&
3422             skb_network_header_len(skb) >= sizeof(*iph)) {
3423                 iph = ip_hdr(skb);
3424                 if (!ip_is_fragment(iph) &&
3425                     (iph->protocol == IPPROTO_TCP ||
3426                      iph->protocol == IPPROTO_UDP) &&
3427                     (skb_headlen(skb) - skb_network_offset(skb) >=
3428                      iph->ihl * sizeof(u32) + sizeof(*layer4hdr) * 2)) {
3429                         layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3430                         layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3431                 }
3432                 return (layer4_xor ^
3433                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3434         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3435                    skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3436                 ipv6h = ipv6_hdr(skb);
3437                 if ((ipv6h->nexthdr == IPPROTO_TCP ||
3438                      ipv6h->nexthdr == IPPROTO_UDP) &&
3439                     (skb_headlen(skb) - skb_network_offset(skb) >=
3440                      sizeof(*ipv6h) + sizeof(*layer4hdr) * 2)) {
3441                         layer4hdr = (__be16 *)(ipv6h + 1);
3442                         layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3443                 }
3444                 s = &ipv6h->saddr.s6_addr32[0];
3445                 d = &ipv6h->daddr.s6_addr32[0];
3446                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3447                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3448                                (layer4_xor >> 8);
3449                 return layer4_xor % count;
3450         }
3451
3452         return bond_xmit_hash_policy_l2(skb, count);
3453 }
3454
3455 /*-------------------------- Device entry points ----------------------------*/
3456
3457 static void bond_work_init_all(struct bonding *bond)
3458 {
3459         INIT_DELAYED_WORK(&bond->mcast_work,
3460                           bond_resend_igmp_join_requests_delayed);
3461         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3462         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3463         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3464                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3465         else
3466                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3467         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3468 }
3469
3470 static void bond_work_cancel_all(struct bonding *bond)
3471 {
3472         cancel_delayed_work_sync(&bond->mii_work);
3473         cancel_delayed_work_sync(&bond->arp_work);
3474         cancel_delayed_work_sync(&bond->alb_work);
3475         cancel_delayed_work_sync(&bond->ad_work);
3476         cancel_delayed_work_sync(&bond->mcast_work);
3477 }
3478
3479 static int bond_open(struct net_device *bond_dev)
3480 {
3481         struct bonding *bond = netdev_priv(bond_dev);
3482         struct slave *slave;
3483         int i;
3484
3485         /* reset slave->backup and slave->inactive */
3486         read_lock(&bond->lock);
3487         if (bond->slave_cnt > 0) {
3488                 read_lock(&bond->curr_slave_lock);
3489                 bond_for_each_slave(bond, slave, i) {
3490                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3491                                 && (slave != bond->curr_active_slave)) {
3492                                 bond_set_slave_inactive_flags(slave);
3493                         } else {
3494                                 bond_set_slave_active_flags(slave);
3495                         }
3496                 }
3497                 read_unlock(&bond->curr_slave_lock);
3498         }
3499         read_unlock(&bond->lock);
3500
3501         bond_work_init_all(bond);
3502
3503         if (bond_is_lb(bond)) {
3504                 /* bond_alb_initialize must be called before the timer
3505                  * is started.
3506                  */
3507                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3508                         return -ENOMEM;
3509                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3510         }
3511
3512         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3513                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3514
3515         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3516                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3517                 if (bond->params.arp_validate)
3518                         bond->recv_probe = bond_arp_rcv;
3519         }
3520
3521         if (bond->params.mode == BOND_MODE_8023AD) {
3522                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3523                 /* register to receive LACPDUs */
3524                 bond->recv_probe = bond_3ad_lacpdu_recv;
3525                 bond_3ad_initiate_agg_selection(bond, 1);
3526         }
3527
3528         return 0;
3529 }
3530
3531 static int bond_close(struct net_device *bond_dev)
3532 {
3533         struct bonding *bond = netdev_priv(bond_dev);
3534
3535         write_lock_bh(&bond->lock);
3536         bond->send_peer_notif = 0;
3537         write_unlock_bh(&bond->lock);
3538
3539         bond_work_cancel_all(bond);
3540         if (bond_is_lb(bond)) {
3541                 /* Must be called only after all
3542                  * slaves have been released
3543                  */
3544                 bond_alb_deinitialize(bond);
3545         }
3546         bond->recv_probe = NULL;
3547
3548         return 0;
3549 }
3550
3551 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3552                                                 struct rtnl_link_stats64 *stats)
3553 {
3554         struct bonding *bond = netdev_priv(bond_dev);
3555         struct rtnl_link_stats64 temp;
3556         struct slave *slave;
3557         int i;
3558
3559         memset(stats, 0, sizeof(*stats));
3560
3561         read_lock_bh(&bond->lock);
3562
3563         bond_for_each_slave(bond, slave, i) {
3564                 const struct rtnl_link_stats64 *sstats =
3565                         dev_get_stats(slave->dev, &temp);
3566
3567                 stats->rx_packets += sstats->rx_packets;
3568                 stats->rx_bytes += sstats->rx_bytes;
3569                 stats->rx_errors += sstats->rx_errors;
3570                 stats->rx_dropped += sstats->rx_dropped;
3571
3572                 stats->tx_packets += sstats->tx_packets;
3573                 stats->tx_bytes += sstats->tx_bytes;
3574                 stats->tx_errors += sstats->tx_errors;
3575                 stats->tx_dropped += sstats->tx_dropped;
3576
3577                 stats->multicast += sstats->multicast;
3578                 stats->collisions += sstats->collisions;
3579
3580                 stats->rx_length_errors += sstats->rx_length_errors;
3581                 stats->rx_over_errors += sstats->rx_over_errors;
3582                 stats->rx_crc_errors += sstats->rx_crc_errors;
3583                 stats->rx_frame_errors += sstats->rx_frame_errors;
3584                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3585                 stats->rx_missed_errors += sstats->rx_missed_errors;
3586
3587                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3588                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3589                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3590                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3591                 stats->tx_window_errors += sstats->tx_window_errors;
3592         }
3593
3594         read_unlock_bh(&bond->lock);
3595
3596         return stats;
3597 }
3598
3599 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3600 {
3601         struct net_device *slave_dev = NULL;
3602         struct ifbond k_binfo;
3603         struct ifbond __user *u_binfo = NULL;
3604         struct ifslave k_sinfo;
3605         struct ifslave __user *u_sinfo = NULL;
3606         struct mii_ioctl_data *mii = NULL;
3607         int res = 0;
3608
3609         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3610
3611         switch (cmd) {
3612         case SIOCGMIIPHY:
3613                 mii = if_mii(ifr);
3614                 if (!mii)
3615                         return -EINVAL;
3616
3617                 mii->phy_id = 0;
3618                 /* Fall Through */
3619         case SIOCGMIIREG:
3620                 /*
3621                  * We do this again just in case we were called by SIOCGMIIREG
3622                  * instead of SIOCGMIIPHY.
3623                  */
3624                 mii = if_mii(ifr);
3625                 if (!mii)
3626                         return -EINVAL;
3627
3628
3629                 if (mii->reg_num == 1) {
3630                         struct bonding *bond = netdev_priv(bond_dev);
3631                         mii->val_out = 0;
3632                         read_lock(&bond->lock);
3633                         read_lock(&bond->curr_slave_lock);
3634                         if (netif_carrier_ok(bond->dev))
3635                                 mii->val_out = BMSR_LSTATUS;
3636
3637                         read_unlock(&bond->curr_slave_lock);
3638                         read_unlock(&bond->lock);
3639                 }
3640
3641                 return 0;
3642         case BOND_INFO_QUERY_OLD:
3643         case SIOCBONDINFOQUERY:
3644                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3645
3646                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3647                         return -EFAULT;
3648
3649                 res = bond_info_query(bond_dev, &k_binfo);
3650                 if (res == 0 &&
3651                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3652                         return -EFAULT;
3653
3654                 return res;
3655         case BOND_SLAVE_INFO_QUERY_OLD:
3656         case SIOCBONDSLAVEINFOQUERY:
3657                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3658
3659                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3660                         return -EFAULT;
3661
3662                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3663                 if (res == 0 &&
3664                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3665                         return -EFAULT;
3666
3667                 return res;
3668         default:
3669                 /* Go on */
3670                 break;
3671         }
3672
3673         if (!capable(CAP_NET_ADMIN))
3674                 return -EPERM;
3675
3676         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3677
3678         pr_debug("slave_dev=%p:\n", slave_dev);
3679
3680         if (!slave_dev)
3681                 res = -ENODEV;
3682         else {
3683                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3684                 switch (cmd) {
3685                 case BOND_ENSLAVE_OLD:
3686                 case SIOCBONDENSLAVE:
3687                         res = bond_enslave(bond_dev, slave_dev);
3688                         break;
3689                 case BOND_RELEASE_OLD:
3690                 case SIOCBONDRELEASE:
3691                         res = bond_release(bond_dev, slave_dev);
3692                         break;
3693                 case BOND_SETHWADDR_OLD:
3694                 case SIOCBONDSETHWADDR:
3695                         res = bond_sethwaddr(bond_dev, slave_dev);
3696                         break;
3697                 case BOND_CHANGE_ACTIVE_OLD:
3698                 case SIOCBONDCHANGEACTIVE:
3699                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3700                         break;
3701                 default:
3702                         res = -EOPNOTSUPP;
3703                 }
3704
3705                 dev_put(slave_dev);
3706         }
3707
3708         return res;
3709 }
3710
3711 static bool bond_addr_in_mc_list(unsigned char *addr,
3712                                  struct netdev_hw_addr_list *list,
3713                                  int addrlen)
3714 {
3715         struct netdev_hw_addr *ha;
3716
3717         netdev_hw_addr_list_for_each(ha, list)
3718                 if (!memcmp(ha->addr, addr, addrlen))
3719                         return true;
3720
3721         return false;
3722 }
3723
3724 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3725 {
3726         struct bonding *bond = netdev_priv(bond_dev);
3727
3728         if (change & IFF_PROMISC)
3729                 bond_set_promiscuity(bond,
3730                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3731
3732         if (change & IFF_ALLMULTI)
3733                 bond_set_allmulti(bond,
3734                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3735 }
3736
3737 static void bond_set_multicast_list(struct net_device *bond_dev)
3738 {
3739         struct bonding *bond = netdev_priv(bond_dev);
3740         struct netdev_hw_addr *ha;
3741         bool found;
3742
3743         read_lock(&bond->lock);
3744
3745         /* looking for addresses to add to slaves' mc list */
3746         netdev_for_each_mc_addr(ha, bond_dev) {
3747                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3748                                              bond_dev->addr_len);
3749                 if (!found)
3750                         bond_mc_add(bond, ha->addr);
3751         }
3752
3753         /* looking for addresses to delete from slaves' list */
3754         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3755                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3756                                              bond_dev->addr_len);
3757                 if (!found)
3758                         bond_mc_del(bond, ha->addr);
3759         }
3760
3761         /* save master's multicast list */
3762         __hw_addr_flush(&bond->mc_list);
3763         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3764                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3765
3766         read_unlock(&bond->lock);
3767 }
3768
3769 static int bond_neigh_init(struct neighbour *n)
3770 {
3771         struct bonding *bond = netdev_priv(n->dev);
3772         struct slave *slave = bond->first_slave;
3773         const struct net_device_ops *slave_ops;
3774         struct neigh_parms parms;
3775         int ret;
3776
3777         if (!slave)
3778                 return 0;
3779
3780         slave_ops = slave->dev->netdev_ops;
3781
3782         if (!slave_ops->ndo_neigh_setup)
3783                 return 0;
3784
3785         parms.neigh_setup = NULL;
3786         parms.neigh_cleanup = NULL;
3787         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3788         if (ret)
3789                 return ret;
3790
3791         /*
3792          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3793          * after the last slave has been detached.  Assumes that all slaves
3794          * utilize the same neigh_cleanup (true at this writing as only user
3795          * is ipoib).
3796          */
3797         n->parms->neigh_cleanup = parms.neigh_cleanup;
3798
3799         if (!parms.neigh_setup)
3800                 return 0;
3801
3802         return parms.neigh_setup(n);
3803 }
3804
3805 /*
3806  * The bonding ndo_neigh_setup is called at init time beofre any
3807  * slave exists. So we must declare proxy setup function which will
3808  * be used at run time to resolve the actual slave neigh param setup.
3809  */
3810 static int bond_neigh_setup(struct net_device *dev,
3811                             struct neigh_parms *parms)
3812 {
3813         parms->neigh_setup   = bond_neigh_init;
3814
3815         return 0;
3816 }
3817
3818 /*
3819  * Change the MTU of all of a master's slaves to match the master
3820  */
3821 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3822 {
3823         struct bonding *bond = netdev_priv(bond_dev);
3824         struct slave *slave, *stop_at;
3825         int res = 0;
3826         int i;
3827
3828         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3829                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3830
3831         /* Can't hold bond->lock with bh disabled here since
3832          * some base drivers panic. On the other hand we can't
3833          * hold bond->lock without bh disabled because we'll
3834          * deadlock. The only solution is to rely on the fact
3835          * that we're under rtnl_lock here, and the slaves
3836          * list won't change. This doesn't solve the problem
3837          * of setting the slave's MTU while it is
3838          * transmitting, but the assumption is that the base
3839          * driver can handle that.
3840          *
3841          * TODO: figure out a way to safely iterate the slaves
3842          * list, but without holding a lock around the actual
3843          * call to the base driver.
3844          */
3845
3846         bond_for_each_slave(bond, slave, i) {
3847                 pr_debug("s %p s->p %p c_m %p\n",
3848                          slave,
3849                          slave->prev,
3850                          slave->dev->netdev_ops->ndo_change_mtu);
3851
3852                 res = dev_set_mtu(slave->dev, new_mtu);
3853
3854                 if (res) {
3855                         /* If we failed to set the slave's mtu to the new value
3856                          * we must abort the operation even in ACTIVE_BACKUP
3857                          * mode, because if we allow the backup slaves to have
3858                          * different mtu values than the active slave we'll
3859                          * need to change their mtu when doing a failover. That
3860                          * means changing their mtu from timer context, which
3861                          * is probably not a good idea.
3862                          */
3863                         pr_debug("err %d %s\n", res, slave->dev->name);
3864                         goto unwind;
3865                 }
3866         }
3867
3868         bond_dev->mtu = new_mtu;
3869
3870         return 0;
3871
3872 unwind:
3873         /* unwind from head to the slave that failed */
3874         stop_at = slave;
3875         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3876                 int tmp_res;
3877
3878                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3879                 if (tmp_res) {
3880                         pr_debug("unwind err %d dev %s\n",
3881                                  tmp_res, slave->dev->name);
3882                 }
3883         }
3884
3885         return res;
3886 }
3887
3888 /*
3889  * Change HW address
3890  *
3891  * Note that many devices must be down to change the HW address, and
3892  * downing the master releases all slaves.  We can make bonds full of
3893  * bonding devices to test this, however.
3894  */
3895 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3896 {
3897         struct bonding *bond = netdev_priv(bond_dev);
3898         struct sockaddr *sa = addr, tmp_sa;
3899         struct slave *slave, *stop_at;
3900         int res = 0;
3901         int i;
3902
3903         if (bond->params.mode == BOND_MODE_ALB)
3904                 return bond_alb_set_mac_address(bond_dev, addr);
3905
3906
3907         pr_debug("bond=%p, name=%s\n",
3908                  bond, bond_dev ? bond_dev->name : "None");
3909
3910         /*
3911          * If fail_over_mac is set to active, do nothing and return
3912          * success.  Returning an error causes ifenslave to fail.
3913          */
3914         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3915                 return 0;
3916
3917         if (!is_valid_ether_addr(sa->sa_data))
3918                 return -EADDRNOTAVAIL;
3919
3920         /* Can't hold bond->lock with bh disabled here since
3921          * some base drivers panic. On the other hand we can't
3922          * hold bond->lock without bh disabled because we'll
3923          * deadlock. The only solution is to rely on the fact
3924          * that we're under rtnl_lock here, and the slaves
3925          * list won't change. This doesn't solve the problem
3926          * of setting the slave's hw address while it is
3927          * transmitting, but the assumption is that the base
3928          * driver can handle that.
3929          *
3930          * TODO: figure out a way to safely iterate the slaves
3931          * list, but without holding a lock around the actual
3932          * call to the base driver.
3933          */
3934
3935         bond_for_each_slave(bond, slave, i) {
3936                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3937                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3938
3939                 if (slave_ops->ndo_set_mac_address == NULL) {
3940                         res = -EOPNOTSUPP;
3941                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3942                         goto unwind;
3943                 }
3944
3945                 res = dev_set_mac_address(slave->dev, addr);
3946                 if (res) {
3947                         /* TODO: consider downing the slave
3948                          * and retry ?
3949                          * User should expect communications
3950                          * breakage anyway until ARP finish
3951                          * updating, so...
3952                          */
3953                         pr_debug("err %d %s\n", res, slave->dev->name);
3954                         goto unwind;
3955                 }
3956         }
3957
3958         /* success */
3959         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3960         return 0;
3961
3962 unwind:
3963         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3964         tmp_sa.sa_family = bond_dev->type;
3965
3966         /* unwind from head to the slave that failed */
3967         stop_at = slave;
3968         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3969                 int tmp_res;
3970
3971                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3972                 if (tmp_res) {
3973                         pr_debug("unwind err %d dev %s\n",
3974                                  tmp_res, slave->dev->name);
3975                 }
3976         }
3977
3978         return res;
3979 }
3980
3981 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3982 {
3983         struct bonding *bond = netdev_priv(bond_dev);
3984         struct slave *slave, *start_at;
3985         int i, slave_no, res = 1;
3986         struct iphdr *iph = ip_hdr(skb);
3987
3988         /*
3989          * Start with the curr_active_slave that joined the bond as the
3990          * default for sending IGMP traffic.  For failover purposes one
3991          * needs to maintain some consistency for the interface that will
3992          * send the join/membership reports.  The curr_active_slave found
3993          * will send all of this type of traffic.
3994          */
3995         if ((iph->protocol == IPPROTO_IGMP) &&
3996             (skb->protocol == htons(ETH_P_IP))) {
3997
3998                 read_lock(&bond->curr_slave_lock);
3999                 slave = bond->curr_active_slave;
4000                 read_unlock(&bond->curr_slave_lock);
4001
4002                 if (!slave)
4003                         goto out;
4004         } else {
4005                 /*
4006                  * Concurrent TX may collide on rr_tx_counter; we accept
4007                  * that as being rare enough not to justify using an
4008                  * atomic op here.
4009                  */
4010                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4011
4012                 bond_for_each_slave(bond, slave, i) {
4013                         slave_no--;
4014                         if (slave_no < 0)
4015                                 break;
4016                 }
4017         }
4018
4019         start_at = slave;
4020         bond_for_each_slave_from(bond, slave, i, start_at) {
4021                 if (IS_UP(slave->dev) &&
4022                     (slave->link == BOND_LINK_UP) &&
4023                     bond_is_active_slave(slave)) {
4024                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4025                         break;
4026                 }
4027         }
4028
4029 out:
4030         if (res) {
4031                 /* no suitable interface, frame not sent */
4032                 kfree_skb(skb);
4033         }
4034
4035         return NETDEV_TX_OK;
4036 }
4037
4038
4039 /*
4040  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4041  * the bond has a usable interface.
4042  */
4043 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4044 {
4045         struct bonding *bond = netdev_priv(bond_dev);
4046         int res = 1;
4047
4048         read_lock(&bond->curr_slave_lock);
4049
4050         if (bond->curr_active_slave)
4051                 res = bond_dev_queue_xmit(bond, skb,
4052                         bond->curr_active_slave->dev);
4053
4054         read_unlock(&bond->curr_slave_lock);
4055
4056         if (res)
4057                 /* no suitable interface, frame not sent */
4058                 kfree_skb(skb);
4059
4060         return NETDEV_TX_OK;
4061 }
4062
4063 /*
4064  * In bond_xmit_xor() , we determine the output device by using a pre-
4065  * determined xmit_hash_policy(), If the selected device is not enabled,
4066  * find the next active slave.
4067  */
4068 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4069 {
4070         struct bonding *bond = netdev_priv(bond_dev);
4071         struct slave *slave, *start_at;
4072         int slave_no;
4073         int i;
4074         int res = 1;
4075
4076         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4077
4078         bond_for_each_slave(bond, slave, i) {
4079                 slave_no--;
4080                 if (slave_no < 0)
4081                         break;
4082         }
4083
4084         start_at = slave;
4085
4086         bond_for_each_slave_from(bond, slave, i, start_at) {
4087                 if (IS_UP(slave->dev) &&
4088                     (slave->link == BOND_LINK_UP) &&
4089                     bond_is_active_slave(slave)) {
4090                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4091                         break;
4092                 }
4093         }
4094
4095         if (res) {
4096                 /* no suitable interface, frame not sent */
4097                 kfree_skb(skb);
4098         }
4099
4100         return NETDEV_TX_OK;
4101 }
4102
4103 /*
4104  * in broadcast mode, we send everything to all usable interfaces.
4105  */
4106 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4107 {
4108         struct bonding *bond = netdev_priv(bond_dev);
4109         struct slave *slave, *start_at;
4110         struct net_device *tx_dev = NULL;
4111         int i;
4112         int res = 1;
4113
4114         read_lock(&bond->curr_slave_lock);
4115         start_at = bond->curr_active_slave;
4116         read_unlock(&bond->curr_slave_lock);
4117
4118         if (!start_at)
4119                 goto out;
4120
4121         bond_for_each_slave_from(bond, slave, i, start_at) {
4122                 if (IS_UP(slave->dev) &&
4123                     (slave->link == BOND_LINK_UP) &&
4124                     bond_is_active_slave(slave)) {
4125                         if (tx_dev) {
4126                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4127                                 if (!skb2) {
4128                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4129                                                bond_dev->name);
4130                                         continue;
4131                                 }
4132
4133                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4134                                 if (res) {
4135                                         kfree_skb(skb2);
4136                                         continue;
4137                                 }
4138                         }
4139                         tx_dev = slave->dev;
4140                 }
4141         }
4142
4143         if (tx_dev)
4144                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4145
4146 out:
4147         if (res)
4148                 /* no suitable interface, frame not sent */
4149                 kfree_skb(skb);
4150
4151         /* frame sent to all suitable interfaces */
4152         return NETDEV_TX_OK;
4153 }
4154
4155 /*------------------------- Device initialization ---------------------------*/
4156
4157 static void bond_set_xmit_hash_policy(struct bonding *bond)
4158 {
4159         switch (bond->params.xmit_policy) {
4160         case BOND_XMIT_POLICY_LAYER23:
4161                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4162                 break;
4163         case BOND_XMIT_POLICY_LAYER34:
4164                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4165                 break;
4166         case BOND_XMIT_POLICY_LAYER2:
4167         default:
4168                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4169                 break;
4170         }
4171 }
4172
4173 /*
4174  * Lookup the slave that corresponds to a qid
4175  */
4176 static inline int bond_slave_override(struct bonding *bond,
4177                                       struct sk_buff *skb)
4178 {
4179         int i, res = 1;
4180         struct slave *slave = NULL;
4181         struct slave *check_slave;
4182
4183         if (!skb->queue_mapping)
4184                 return 1;
4185
4186         /* Find out if any slaves have the same mapping as this skb. */
4187         bond_for_each_slave(bond, check_slave, i) {
4188                 if (check_slave->queue_id == skb->queue_mapping) {
4189                         slave = check_slave;
4190                         break;
4191                 }
4192         }
4193
4194         /* If the slave isn't UP, use default transmit policy. */
4195         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4196             (slave->link == BOND_LINK_UP)) {
4197                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4198         }
4199
4200         return res;
4201 }
4202
4203
4204 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4205 {
4206         /*
4207          * This helper function exists to help dev_pick_tx get the correct
4208          * destination queue.  Using a helper function skips a call to
4209          * skb_tx_hash and will put the skbs in the queue we expect on their
4210          * way down to the bonding driver.
4211          */
4212         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4213
4214         /*
4215          * Save the original txq to restore before passing to the driver
4216          */
4217         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4218
4219         if (unlikely(txq >= dev->real_num_tx_queues)) {
4220                 do {
4221                         txq -= dev->real_num_tx_queues;
4222                 } while (txq >= dev->real_num_tx_queues);
4223         }
4224         return txq;
4225 }
4226
4227 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4228 {
4229         struct bonding *bond = netdev_priv(dev);
4230
4231         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4232                 if (!bond_slave_override(bond, skb))
4233                         return NETDEV_TX_OK;
4234         }
4235
4236         switch (bond->params.mode) {
4237         case BOND_MODE_ROUNDROBIN:
4238                 return bond_xmit_roundrobin(skb, dev);
4239         case BOND_MODE_ACTIVEBACKUP:
4240                 return bond_xmit_activebackup(skb, dev);
4241         case BOND_MODE_XOR:
4242                 return bond_xmit_xor(skb, dev);
4243         case BOND_MODE_BROADCAST:
4244                 return bond_xmit_broadcast(skb, dev);
4245         case BOND_MODE_8023AD:
4246                 return bond_3ad_xmit_xor(skb, dev);
4247         case BOND_MODE_ALB:
4248         case BOND_MODE_TLB:
4249                 return bond_alb_xmit(skb, dev);
4250         default:
4251                 /* Should never happen, mode already checked */
4252                 pr_err("%s: Error: Unknown bonding mode %d\n",
4253                        dev->name, bond->params.mode);
4254                 WARN_ON_ONCE(1);
4255                 kfree_skb(skb);
4256                 return NETDEV_TX_OK;
4257         }
4258 }
4259
4260 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4261 {
4262         struct bonding *bond = netdev_priv(dev);
4263         netdev_tx_t ret = NETDEV_TX_OK;
4264
4265         /*
4266          * If we risk deadlock from transmitting this in the
4267          * netpoll path, tell netpoll to queue the frame for later tx
4268          */
4269         if (is_netpoll_tx_blocked(dev))
4270                 return NETDEV_TX_BUSY;
4271
4272         read_lock(&bond->lock);
4273
4274         if (bond->slave_cnt)
4275                 ret = __bond_start_xmit(skb, dev);
4276         else
4277                 kfree_skb(skb);
4278
4279         read_unlock(&bond->lock);
4280
4281         return ret;
4282 }
4283
4284 /*
4285  * set bond mode specific net device operations
4286  */
4287 void bond_set_mode_ops(struct bonding *bond, int mode)
4288 {
4289         struct net_device *bond_dev = bond->dev;
4290
4291         switch (mode) {
4292         case BOND_MODE_ROUNDROBIN:
4293                 break;
4294         case BOND_MODE_ACTIVEBACKUP:
4295                 break;
4296         case BOND_MODE_XOR:
4297                 bond_set_xmit_hash_policy(bond);
4298                 break;
4299         case BOND_MODE_BROADCAST:
4300                 break;
4301         case BOND_MODE_8023AD:
4302                 bond_set_xmit_hash_policy(bond);
4303                 break;
4304         case BOND_MODE_ALB:
4305                 /* FALLTHRU */
4306         case BOND_MODE_TLB:
4307                 break;
4308         default:
4309                 /* Should never happen, mode already checked */
4310                 pr_err("%s: Error: Unknown bonding mode %d\n",
4311                        bond_dev->name, mode);
4312                 break;
4313         }
4314 }
4315
4316 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4317                                     struct ethtool_drvinfo *drvinfo)
4318 {
4319         strncpy(drvinfo->driver, DRV_NAME, 32);
4320         strncpy(drvinfo->version, DRV_VERSION, 32);
4321         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4322 }
4323
4324 static const struct ethtool_ops bond_ethtool_ops = {
4325         .get_drvinfo            = bond_ethtool_get_drvinfo,
4326         .get_link               = ethtool_op_get_link,
4327 };
4328
4329 static const struct net_device_ops bond_netdev_ops = {
4330         .ndo_init               = bond_init,
4331         .ndo_uninit             = bond_uninit,
4332         .ndo_open               = bond_open,
4333         .ndo_stop               = bond_close,
4334         .ndo_start_xmit         = bond_start_xmit,
4335         .ndo_select_queue       = bond_select_queue,
4336         .ndo_get_stats64        = bond_get_stats,
4337         .ndo_do_ioctl           = bond_do_ioctl,
4338         .ndo_change_rx_flags    = bond_change_rx_flags,
4339         .ndo_set_rx_mode        = bond_set_multicast_list,
4340         .ndo_change_mtu         = bond_change_mtu,
4341         .ndo_set_mac_address    = bond_set_mac_address,
4342         .ndo_neigh_setup        = bond_neigh_setup,
4343         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4344         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4345 #ifdef CONFIG_NET_POLL_CONTROLLER
4346         .ndo_netpoll_setup      = bond_netpoll_setup,
4347         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4348         .ndo_poll_controller    = bond_poll_controller,
4349 #endif
4350         .ndo_add_slave          = bond_enslave,
4351         .ndo_del_slave          = bond_release,
4352         .ndo_fix_features       = bond_fix_features,
4353 };
4354
4355 static void bond_destructor(struct net_device *bond_dev)
4356 {
4357         struct bonding *bond = netdev_priv(bond_dev);
4358         if (bond->wq)
4359                 destroy_workqueue(bond->wq);
4360         free_netdev(bond_dev);
4361 }
4362
4363 static void bond_setup(struct net_device *bond_dev)
4364 {
4365         struct bonding *bond = netdev_priv(bond_dev);
4366
4367         /* initialize rwlocks */
4368         rwlock_init(&bond->lock);
4369         rwlock_init(&bond->curr_slave_lock);
4370
4371         bond->params = bonding_defaults;
4372
4373         /* Initialize pointers */
4374         bond->dev = bond_dev;
4375         INIT_LIST_HEAD(&bond->vlan_list);
4376
4377         /* Initialize the device entry points */
4378         ether_setup(bond_dev);
4379         bond_dev->netdev_ops = &bond_netdev_ops;
4380         bond_dev->ethtool_ops = &bond_ethtool_ops;
4381         bond_set_mode_ops(bond, bond->params.mode);
4382
4383         bond_dev->destructor = bond_destructor;
4384
4385         /* Initialize the device options */
4386         bond_dev->tx_queue_len = 0;
4387         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4388         bond_dev->priv_flags |= IFF_BONDING;
4389         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4390
4391         /* At first, we block adding VLANs. That's the only way to
4392          * prevent problems that occur when adding VLANs over an
4393          * empty bond. The block will be removed once non-challenged
4394          * slaves are enslaved.
4395          */
4396         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4397
4398         /* don't acquire bond device's netif_tx_lock when
4399          * transmitting */
4400         bond_dev->features |= NETIF_F_LLTX;
4401
4402         /* By default, we declare the bond to be fully
4403          * VLAN hardware accelerated capable. Special
4404          * care is taken in the various xmit functions
4405          * when there are slaves that are not hw accel
4406          * capable
4407          */
4408
4409         bond_dev->hw_features = BOND_VLAN_FEATURES |
4410                                 NETIF_F_HW_VLAN_TX |
4411                                 NETIF_F_HW_VLAN_RX |
4412                                 NETIF_F_HW_VLAN_FILTER;
4413
4414         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4415         bond_dev->features |= bond_dev->hw_features;
4416 }
4417
4418 /*
4419 * Destroy a bonding device.
4420 * Must be under rtnl_lock when this function is called.
4421 */
4422 static void bond_uninit(struct net_device *bond_dev)
4423 {
4424         struct bonding *bond = netdev_priv(bond_dev);
4425         struct vlan_entry *vlan, *tmp;
4426
4427         bond_netpoll_cleanup(bond_dev);
4428
4429         /* Release the bonded slaves */
4430         bond_release_all(bond_dev);
4431
4432         list_del(&bond->bond_list);
4433
4434         bond_debug_unregister(bond);
4435
4436         __hw_addr_flush(&bond->mc_list);
4437
4438         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4439                 list_del(&vlan->vlan_list);
4440                 kfree(vlan);
4441         }
4442 }
4443
4444 /*------------------------- Module initialization ---------------------------*/
4445
4446 /*
4447  * Convert string input module parms.  Accept either the
4448  * number of the mode or its string name.  A bit complicated because
4449  * some mode names are substrings of other names, and calls from sysfs
4450  * may have whitespace in the name (trailing newlines, for example).
4451  */
4452 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4453 {
4454         int modeint = -1, i, rv;
4455         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4456
4457         for (p = (char *)buf; *p; p++)
4458                 if (!(isdigit(*p) || isspace(*p)))
4459                         break;
4460
4461         if (*p)
4462                 rv = sscanf(buf, "%20s", modestr);
4463         else
4464                 rv = sscanf(buf, "%d", &modeint);
4465
4466         if (!rv)
4467                 return -1;
4468
4469         for (i = 0; tbl[i].modename; i++) {
4470                 if (modeint == tbl[i].mode)
4471                         return tbl[i].mode;
4472                 if (strcmp(modestr, tbl[i].modename) == 0)
4473                         return tbl[i].mode;
4474         }
4475
4476         return -1;
4477 }
4478
4479 static int bond_check_params(struct bond_params *params)
4480 {
4481         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4482
4483         /*
4484          * Convert string parameters.
4485          */
4486         if (mode) {
4487                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4488                 if (bond_mode == -1) {
4489                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4490                                mode == NULL ? "NULL" : mode);
4491                         return -EINVAL;
4492                 }
4493         }
4494
4495         if (xmit_hash_policy) {
4496                 if ((bond_mode != BOND_MODE_XOR) &&
4497                     (bond_mode != BOND_MODE_8023AD)) {
4498                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4499                                bond_mode_name(bond_mode));
4500                 } else {
4501                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4502                                                         xmit_hashtype_tbl);
4503                         if (xmit_hashtype == -1) {
4504                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4505                                        xmit_hash_policy == NULL ? "NULL" :
4506                                        xmit_hash_policy);
4507                                 return -EINVAL;
4508                         }
4509                 }
4510         }
4511
4512         if (lacp_rate) {
4513                 if (bond_mode != BOND_MODE_8023AD) {
4514                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4515                                 bond_mode_name(bond_mode));
4516                 } else {
4517                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4518                         if (lacp_fast == -1) {
4519                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4520                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4521                                 return -EINVAL;
4522                         }
4523                 }
4524         }
4525
4526         if (ad_select) {
4527                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4528                 if (params->ad_select == -1) {
4529                         pr_err("Error: Invalid ad_select \"%s\"\n",
4530                                ad_select == NULL ? "NULL" : ad_select);
4531                         return -EINVAL;
4532                 }
4533
4534                 if (bond_mode != BOND_MODE_8023AD) {
4535                         pr_warning("ad_select param only affects 802.3ad mode\n");
4536                 }
4537         } else {
4538                 params->ad_select = BOND_AD_STABLE;
4539         }
4540
4541         if (max_bonds < 0) {
4542                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4543                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4544                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4545         }
4546
4547         if (miimon < 0) {
4548                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4549                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4550                 miimon = BOND_LINK_MON_INTERV;
4551         }
4552
4553         if (updelay < 0) {
4554                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4555                            updelay, INT_MAX);
4556                 updelay = 0;
4557         }
4558
4559         if (downdelay < 0) {
4560                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4561                            downdelay, INT_MAX);
4562                 downdelay = 0;
4563         }
4564
4565         if ((use_carrier != 0) && (use_carrier != 1)) {
4566                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4567                            use_carrier);
4568                 use_carrier = 1;
4569         }
4570
4571         if (num_peer_notif < 0 || num_peer_notif > 255) {
4572                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4573                            num_peer_notif);
4574                 num_peer_notif = 1;
4575         }
4576
4577         /* reset values for 802.3ad */
4578         if (bond_mode == BOND_MODE_8023AD) {
4579                 if (!miimon) {
4580                         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");
4581                         pr_warning("Forcing miimon to 100msec\n");
4582                         miimon = 100;
4583                 }
4584         }
4585
4586         if (tx_queues < 1 || tx_queues > 255) {
4587                 pr_warning("Warning: tx_queues (%d) should be between "
4588                            "1 and 255, resetting to %d\n",
4589                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4590                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4591         }
4592
4593         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4594                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4595                            "not of valid value (0/1), so it was set to "
4596                            "0\n", all_slaves_active);
4597                 all_slaves_active = 0;
4598         }
4599
4600         if (resend_igmp < 0 || resend_igmp > 255) {
4601                 pr_warning("Warning: resend_igmp (%d) should be between "
4602                            "0 and 255, resetting to %d\n",
4603                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4604                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4605         }
4606
4607         /* reset values for TLB/ALB */
4608         if ((bond_mode == BOND_MODE_TLB) ||
4609             (bond_mode == BOND_MODE_ALB)) {
4610                 if (!miimon) {
4611                         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");
4612                         pr_warning("Forcing miimon to 100msec\n");
4613                         miimon = 100;
4614                 }
4615         }
4616
4617         if (bond_mode == BOND_MODE_ALB) {
4618                 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",
4619                           updelay);
4620         }
4621
4622         if (!miimon) {
4623                 if (updelay || downdelay) {
4624                         /* just warn the user the up/down delay will have
4625                          * no effect since miimon is zero...
4626                          */
4627                         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",
4628                                    updelay, downdelay);
4629                 }
4630         } else {
4631                 /* don't allow arp monitoring */
4632                 if (arp_interval) {
4633                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4634                                    miimon, arp_interval);
4635                         arp_interval = 0;
4636                 }
4637
4638                 if ((updelay % miimon) != 0) {
4639                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4640                                    updelay, miimon,
4641                                    (updelay / miimon) * miimon);
4642                 }
4643
4644                 updelay /= miimon;
4645
4646                 if ((downdelay % miimon) != 0) {
4647                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4648                                    downdelay, miimon,
4649                                    (downdelay / miimon) * miimon);
4650                 }
4651
4652                 downdelay /= miimon;
4653         }
4654
4655         if (arp_interval < 0) {
4656                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4657                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4658                 arp_interval = BOND_LINK_ARP_INTERV;
4659         }
4660
4661         for (arp_ip_count = 0;
4662              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4663              arp_ip_count++) {
4664                 /* not complete check, but should be good enough to
4665                    catch mistakes */
4666                 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4667                 if (!isdigit(arp_ip_target[arp_ip_count][0]) ||
4668                     ip == 0 || ip == htonl(INADDR_BROADCAST)) {
4669                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4670                                    arp_ip_target[arp_ip_count]);
4671                         arp_interval = 0;
4672                 } else {
4673                         arp_target[arp_ip_count] = ip;
4674                 }
4675         }
4676
4677         if (arp_interval && !arp_ip_count) {
4678                 /* don't allow arping if no arp_ip_target given... */
4679                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4680                            arp_interval);
4681                 arp_interval = 0;
4682         }
4683
4684         if (arp_validate) {
4685                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4686                         pr_err("arp_validate only supported in active-backup mode\n");
4687                         return -EINVAL;
4688                 }
4689                 if (!arp_interval) {
4690                         pr_err("arp_validate requires arp_interval\n");
4691                         return -EINVAL;
4692                 }
4693
4694                 arp_validate_value = bond_parse_parm(arp_validate,
4695                                                      arp_validate_tbl);
4696                 if (arp_validate_value == -1) {
4697                         pr_err("Error: invalid arp_validate \"%s\"\n",
4698                                arp_validate == NULL ? "NULL" : arp_validate);
4699                         return -EINVAL;
4700                 }
4701         } else
4702                 arp_validate_value = 0;
4703
4704         if (miimon) {
4705                 pr_info("MII link monitoring set to %d ms\n", miimon);
4706         } else if (arp_interval) {
4707                 int i;
4708
4709                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4710                         arp_interval,
4711                         arp_validate_tbl[arp_validate_value].modename,
4712                         arp_ip_count);
4713
4714                 for (i = 0; i < arp_ip_count; i++)
4715                         pr_info(" %s", arp_ip_target[i]);
4716
4717                 pr_info("\n");
4718
4719         } else if (max_bonds) {
4720                 /* miimon and arp_interval not set, we need one so things
4721                  * work as expected, see bonding.txt for details
4722                  */
4723                 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");
4724         }
4725
4726         if (primary && !USES_PRIMARY(bond_mode)) {
4727                 /* currently, using a primary only makes sense
4728                  * in active backup, TLB or ALB modes
4729                  */
4730                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4731                            primary, bond_mode_name(bond_mode));
4732                 primary = NULL;
4733         }
4734
4735         if (primary && primary_reselect) {
4736                 primary_reselect_value = bond_parse_parm(primary_reselect,
4737                                                          pri_reselect_tbl);
4738                 if (primary_reselect_value == -1) {
4739                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4740                                primary_reselect ==
4741                                         NULL ? "NULL" : primary_reselect);
4742                         return -EINVAL;
4743                 }
4744         } else {
4745                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4746         }
4747
4748         if (fail_over_mac) {
4749                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4750                                                       fail_over_mac_tbl);
4751                 if (fail_over_mac_value == -1) {
4752                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4753                                arp_validate == NULL ? "NULL" : arp_validate);
4754                         return -EINVAL;
4755                 }
4756
4757                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4758                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4759         } else {
4760                 fail_over_mac_value = BOND_FOM_NONE;
4761         }
4762
4763         /* fill params struct with the proper values */
4764         params->mode = bond_mode;
4765         params->xmit_policy = xmit_hashtype;
4766         params->miimon = miimon;
4767         params->num_peer_notif = num_peer_notif;
4768         params->arp_interval = arp_interval;
4769         params->arp_validate = arp_validate_value;
4770         params->updelay = updelay;
4771         params->downdelay = downdelay;
4772         params->use_carrier = use_carrier;
4773         params->lacp_fast = lacp_fast;
4774         params->primary[0] = 0;
4775         params->primary_reselect = primary_reselect_value;
4776         params->fail_over_mac = fail_over_mac_value;
4777         params->tx_queues = tx_queues;
4778         params->all_slaves_active = all_slaves_active;
4779         params->resend_igmp = resend_igmp;
4780         params->min_links = min_links;
4781
4782         if (primary) {
4783                 strncpy(params->primary, primary, IFNAMSIZ);
4784                 params->primary[IFNAMSIZ - 1] = 0;
4785         }
4786
4787         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4788
4789         return 0;
4790 }
4791
4792 static struct lock_class_key bonding_netdev_xmit_lock_key;
4793 static struct lock_class_key bonding_netdev_addr_lock_key;
4794 static struct lock_class_key bonding_tx_busylock_key;
4795
4796 static void bond_set_lockdep_class_one(struct net_device *dev,
4797                                        struct netdev_queue *txq,
4798                                        void *_unused)
4799 {
4800         lockdep_set_class(&txq->_xmit_lock,
4801                           &bonding_netdev_xmit_lock_key);
4802 }
4803
4804 static void bond_set_lockdep_class(struct net_device *dev)
4805 {
4806         lockdep_set_class(&dev->addr_list_lock,
4807                           &bonding_netdev_addr_lock_key);
4808         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4809         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4810 }
4811
4812 /*
4813  * Called from registration process
4814  */
4815 static int bond_init(struct net_device *bond_dev)
4816 {
4817         struct bonding *bond = netdev_priv(bond_dev);
4818         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4819         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4820
4821         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4822
4823         /*
4824          * Initialize locks that may be required during
4825          * en/deslave operations.  All of the bond_open work
4826          * (of which this is part) should really be moved to
4827          * a phase prior to dev_open
4828          */
4829         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4830         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4831
4832         bond->wq = create_singlethread_workqueue(bond_dev->name);
4833         if (!bond->wq)
4834                 return -ENOMEM;
4835
4836         bond_set_lockdep_class(bond_dev);
4837
4838         list_add_tail(&bond->bond_list, &bn->dev_list);
4839
4840         bond_prepare_sysfs_group(bond);
4841
4842         bond_debug_register(bond);
4843
4844         __hw_addr_init(&bond->mc_list);
4845         return 0;
4846 }
4847
4848 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4849 {
4850         if (tb[IFLA_ADDRESS]) {
4851                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4852                         return -EINVAL;
4853                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4854                         return -EADDRNOTAVAIL;
4855         }
4856         return 0;
4857 }
4858
4859 static unsigned int bond_get_num_tx_queues(void)
4860 {
4861         return tx_queues;
4862 }
4863
4864 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4865         .kind                   = "bond",
4866         .priv_size              = sizeof(struct bonding),
4867         .setup                  = bond_setup,
4868         .validate               = bond_validate,
4869         .get_num_tx_queues      = bond_get_num_tx_queues,
4870         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4871                                                              as for TX queues */
4872 };
4873
4874 /* Create a new bond based on the specified name and bonding parameters.
4875  * If name is NULL, obtain a suitable "bond%d" name for us.
4876  * Caller must NOT hold rtnl_lock; we need to release it here before we
4877  * set up our sysfs entries.
4878  */
4879 int bond_create(struct net *net, const char *name)
4880 {
4881         struct net_device *bond_dev;
4882         int res;
4883
4884         rtnl_lock();
4885
4886         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4887                                    name ? name : "bond%d",
4888                                    bond_setup, tx_queues);
4889         if (!bond_dev) {
4890                 pr_err("%s: eek! can't alloc netdev!\n", name);
4891                 rtnl_unlock();
4892                 return -ENOMEM;
4893         }
4894
4895         dev_net_set(bond_dev, net);
4896         bond_dev->rtnl_link_ops = &bond_link_ops;
4897
4898         res = register_netdevice(bond_dev);
4899
4900         netif_carrier_off(bond_dev);
4901
4902         rtnl_unlock();
4903         if (res < 0)
4904                 bond_destructor(bond_dev);
4905         return res;
4906 }
4907
4908 static int __net_init bond_net_init(struct net *net)
4909 {
4910         struct bond_net *bn = net_generic(net, bond_net_id);
4911
4912         bn->net = net;
4913         INIT_LIST_HEAD(&bn->dev_list);
4914
4915         bond_create_proc_dir(bn);
4916         bond_create_sysfs(bn);
4917
4918         return 0;
4919 }
4920
4921 static void __net_exit bond_net_exit(struct net *net)
4922 {
4923         struct bond_net *bn = net_generic(net, bond_net_id);
4924
4925         bond_destroy_sysfs(bn);
4926         bond_destroy_proc_dir(bn);
4927 }
4928
4929 static struct pernet_operations bond_net_ops = {
4930         .init = bond_net_init,
4931         .exit = bond_net_exit,
4932         .id   = &bond_net_id,
4933         .size = sizeof(struct bond_net),
4934 };
4935
4936 static int __init bonding_init(void)
4937 {
4938         int i;
4939         int res;
4940
4941         pr_info("%s", bond_version);
4942
4943         res = bond_check_params(&bonding_defaults);
4944         if (res)
4945                 goto out;
4946
4947         res = register_pernet_subsys(&bond_net_ops);
4948         if (res)
4949                 goto out;
4950
4951         res = rtnl_link_register(&bond_link_ops);
4952         if (res)
4953                 goto err_link;
4954
4955         bond_create_debugfs();
4956
4957         for (i = 0; i < max_bonds; i++) {
4958                 res = bond_create(&init_net, NULL);
4959                 if (res)
4960                         goto err;
4961         }
4962
4963         register_netdevice_notifier(&bond_netdev_notifier);
4964 out:
4965         return res;
4966 err:
4967         rtnl_link_unregister(&bond_link_ops);
4968 err_link:
4969         unregister_pernet_subsys(&bond_net_ops);
4970         goto out;
4971
4972 }
4973
4974 static void __exit bonding_exit(void)
4975 {
4976         unregister_netdevice_notifier(&bond_netdev_notifier);
4977
4978         bond_destroy_debugfs();
4979
4980         rtnl_link_unregister(&bond_link_ops);
4981         unregister_pernet_subsys(&bond_net_ops);
4982
4983 #ifdef CONFIG_NET_POLL_CONTROLLER
4984         /*
4985          * Make sure we don't have an imbalance on our netpoll blocking
4986          */
4987         WARN_ON(atomic_read(&netpoll_block_tx));
4988 #endif
4989 }
4990
4991 module_init(bonding_init);
4992 module_exit(bonding_exit);
4993 MODULE_LICENSE("GPL");
4994 MODULE_VERSION(DRV_VERSION);
4995 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4996 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4997 MODULE_ALIAS_RTNL_LINK("bond");