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

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon       = BOND_LINK_MON_INTERV;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *arp_all_targets;
 108 static char *fail_over_mac;
 109 static int all_slaves_active;
 110 static struct bond_params bonding_defaults;
 111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 112
 113 module_param(max_bonds, int, 0);
 114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 115 module_param(tx_queues, int, 0);
 116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 117 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 118 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 119                                "failover event (alias of num_unsol_na)");
 120 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 122                                "failover event (alias of num_grat_arp)");
 123 module_param(miimon, int, 0);
 124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 125 module_param(updelay, int, 0);
 126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 127 module_param(downdelay, int, 0);
 128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 129                             "in milliseconds");
 130 module_param(use_carrier, int, 0);
 131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 132                               "0 for off, 1 for on (default)");
 133 module_param(mode, charp, 0);
 134 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 135                        "1 for active-backup, 2 for balance-xor, "
 136                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 137                        "6 for balance-alb");
 138 module_param(primary, charp, 0);
 139 MODULE_PARM_DESC(primary, "Primary network device to use");
 140 module_param(primary_reselect, charp, 0);
 141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 142                                    "once it comes up; "
 143                                    "0 for always (default), "
 144                                    "1 for only if speed of primary is "
 145                                    "better, "
 146                                    "2 for only on active slave "
 147                                    "failure");
 148 module_param(lacp_rate, charp, 0);
 149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 150                             "0 for slow, 1 for fast");
 151 module_param(ad_select, charp, 0);
 152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 153                             "0 for stable (default), 1 for bandwidth, "
 154                             "2 for count");
 155 module_param(min_links, int, 0);
 156 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 157
 158 module_param(xmit_hash_policy, charp, 0);
 159 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 160                                    "0 for layer 2 (default), 1 for layer 3+4, "
 161                                    "2 for layer 2+3");
 162 module_param(arp_interval, int, 0);
 163 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 164 module_param_array(arp_ip_target, charp, NULL, 0);
 165 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 166 module_param(arp_validate, charp, 0);
 167 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 168                                "0 for none (default), 1 for active, "
 169                                "2 for backup, 3 for all");
 170 module_param(arp_all_targets, charp, 0);
 171 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 172 module_param(fail_over_mac, charp, 0);
 173 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 174                                 "the same MAC; 0 for none (default), "
 175                                 "1 for active, 2 for follow");
 176 module_param(all_slaves_active, int, 0);
 177 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 178                                      "by setting active flag for all slaves; "
 179                                      "0 for never (default), 1 for always.");
 180 module_param(resend_igmp, int, 0);
 181 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 182                               "link failure");
 183
 184 /*----------------------------- Global variables ----------------------------*/
 185
 186 #ifdef CONFIG_NET_POLL_CONTROLLER
 187 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 188 #endif
 189
 190 int bond_net_id __read_mostly;
 191
 192 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 193 static int arp_ip_count;
 194 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 195 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 196 static int lacp_fast;
 197
 198 const struct bond_parm_tbl bond_lacp_tbl[] = {
 199 {       "slow",         AD_LACP_SLOW},
 200 {       "fast",         AD_LACP_FAST},
 201 {       NULL,           -1},
 202 };
 203
 204 const struct bond_parm_tbl bond_mode_tbl[] = {
 205 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 206 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 207 {       "balance-xor",          BOND_MODE_XOR},
 208 {       "broadcast",            BOND_MODE_BROADCAST},
 209 {       "802.3ad",              BOND_MODE_8023AD},
 210 {       "balance-tlb",          BOND_MODE_TLB},
 211 {       "balance-alb",          BOND_MODE_ALB},
 212 {       NULL,                   -1},
 213 };
 214
 215 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 216 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 217 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 218 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 219 {       NULL,                   -1},
 220 };
 221
 222 const struct bond_parm_tbl arp_all_targets_tbl[] = {
 223 {       "any",                  BOND_ARP_TARGETS_ANY},
 224 {       "all",                  BOND_ARP_TARGETS_ALL},
 225 {       NULL,                   -1},
 226 };
 227
 228 const struct bond_parm_tbl arp_validate_tbl[] = {
 229 {       "none",                 BOND_ARP_VALIDATE_NONE},
 230 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 231 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 232 {       "all",                  BOND_ARP_VALIDATE_ALL},
 233 {       NULL,                   -1},
 234 };
 235
 236 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 237 {       "none",                 BOND_FOM_NONE},
 238 {       "active",               BOND_FOM_ACTIVE},
 239 {       "follow",               BOND_FOM_FOLLOW},
 240 {       NULL,                   -1},
 241 };
 242
 243 const struct bond_parm_tbl pri_reselect_tbl[] = {
 244 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 245 {       "better",               BOND_PRI_RESELECT_BETTER},
 246 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 247 {       NULL,                   -1},
 248 };
 249
 250 struct bond_parm_tbl ad_select_tbl[] = {
 251 {       "stable",       BOND_AD_STABLE},
 252 {       "bandwidth",    BOND_AD_BANDWIDTH},
 253 {       "count",        BOND_AD_COUNT},
 254 {       NULL,           -1},
 255 };
 256
 257 /*-------------------------- Forward declarations ---------------------------*/
 258
 259 static int bond_init(struct net_device *bond_dev);
 260 static void bond_uninit(struct net_device *bond_dev);
 261
 262 /*---------------------------- General routines -----------------------------*/
 263
 264 const char *bond_mode_name(int mode)
 265 {
 266         static const char *names[] = {
 267                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 268                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 269                 [BOND_MODE_XOR] = "load balancing (xor)",
 270                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 271                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 272                 [BOND_MODE_TLB] = "transmit load balancing",
 273                 [BOND_MODE_ALB] = "adaptive load balancing",
 274         };
 275
 276         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 277                 return "unknown";
 278
 279         return names[mode];
 280 }
 281
 282 /*---------------------------------- VLAN -----------------------------------*/
 283
 284 /**
 285  * bond_add_vlan - add a new vlan id on bond
 286  * @bond: bond that got the notification
 287  * @vlan_id: the vlan id to add
 288  *
 289  * Returns -ENOMEM if allocation failed.
 290  */
 291 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 292 {
 293         struct vlan_entry *vlan;
 294
 295         pr_debug("bond: %s, vlan id %d\n",
 296                  (bond ? bond->dev->name : "None"), vlan_id);
 297
 298         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 299         if (!vlan)
 300                 return -ENOMEM;
 301
 302         INIT_LIST_HEAD(&vlan->vlan_list);
 303         vlan->vlan_id = vlan_id;
 304
 305         write_lock_bh(&bond->lock);
 306
 307         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 308
 309         write_unlock_bh(&bond->lock);
 310
 311         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 312
 313         return 0;
 314 }
 315
 316 /**
 317  * bond_del_vlan - delete a vlan id from bond
 318  * @bond: bond that got the notification
 319  * @vlan_id: the vlan id to delete
 320  *
 321  * returns -ENODEV if @vlan_id was not found in @bond.
 322  */
 323 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 324 {
 325         struct vlan_entry *vlan;
 326         int res = -ENODEV;
 327
 328         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 329
 330         block_netpoll_tx();
 331         write_lock_bh(&bond->lock);
 332
 333         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 334                 if (vlan->vlan_id == vlan_id) {
 335                         list_del(&vlan->vlan_list);
 336
 337                         if (bond_is_lb(bond))
 338                                 bond_alb_clear_vlan(bond, vlan_id);
 339
 340                         pr_debug("removed VLAN ID %d from bond %s\n",
 341                                  vlan_id, bond->dev->name);
 342
 343                         kfree(vlan);
 344
 345                         res = 0;
 346                         goto out;
 347                 }
 348         }
 349
 350         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 351                  vlan_id, bond->dev->name);
 352
 353 out:
 354         write_unlock_bh(&bond->lock);
 355         unblock_netpoll_tx();
 356         return res;
 357 }
 358
 359 /**
 360  * bond_next_vlan - safely skip to the next item in the vlans list.
 361  * @bond: the bond we're working on
 362  * @curr: item we're advancing from
 363  *
 364  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 365  * or @curr->next otherwise (even if it is @curr itself again).
 366  *
 367  * Caller must hold bond->lock
 368  */
 369 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 370 {
 371         struct vlan_entry *next, *last;
 372
 373         if (list_empty(&bond->vlan_list))
 374                 return NULL;
 375
 376         if (!curr) {
 377                 next = list_entry(bond->vlan_list.next,
 378                                   struct vlan_entry, vlan_list);
 379         } else {
 380                 last = list_entry(bond->vlan_list.prev,
 381                                   struct vlan_entry, vlan_list);
 382                 if (last == curr) {
 383                         next = list_entry(bond->vlan_list.next,
 384                                           struct vlan_entry, vlan_list);
 385                 } else {
 386                         next = list_entry(curr->vlan_list.next,
 387                                           struct vlan_entry, vlan_list);
 388                 }
 389         }
 390
 391         return next;
 392 }
 393
 394 /**
 395  * bond_dev_queue_xmit - Prepare skb for xmit.
 396  *
 397  * @bond: bond device that got this skb for tx.
 398  * @skb: hw accel VLAN tagged skb to transmit
 399  * @slave_dev: slave that is supposed to xmit this skbuff
 400  */
 401 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 402                         struct net_device *slave_dev)
 403 {
 404         skb->dev = slave_dev;
 405
 406         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 407                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 408         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 409
 410         if (unlikely(netpoll_tx_running(bond->dev)))
 411                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 412         else
 413                 dev_queue_xmit(skb);
 414
 415         return 0;
 416 }
 417
 418 /*
 419  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 420  * We don't protect the slave list iteration with a lock because:
 421  * a. This operation is performed in IOCTL context,
 422  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 423  * c. Holding a lock with BH disabled while directly calling a base driver
 424  *    entry point is generally a BAD idea.
 425  *
 426  * The design of synchronization/protection for this operation in the 8021q
 427  * module is good for one or more VLAN devices over a single physical device
 428  * and cannot be extended for a teaming solution like bonding, so there is a
 429  * potential race condition here where a net device from the vlan group might
 430  * be referenced (either by a base driver or the 8021q code) while it is being
 431  * removed from the system. However, it turns out we're not making matters
 432  * worse, and if it works for regular VLAN usage it will work here too.
 433 */
 434
 435 /**
 436  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 437  * @bond_dev: bonding net device that got called
 438  * @vid: vlan id being added
 439  */
 440 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 441                                 __be16 proto, u16 vid)
 442 {
 443         struct bonding *bond = netdev_priv(bond_dev);
 444         struct slave *slave;
 445         int res;
 446
 447         bond_for_each_slave(bond, slave) {
 448                 res = vlan_vid_add(slave->dev, proto, vid);
 449                 if (res)
 450                         goto unwind;
 451         }
 452
 453         res = bond_add_vlan(bond, vid);
 454         if (res) {
 455                 pr_err("%s: Error: Failed to add vlan id %d\n",
 456                        bond_dev->name, vid);
 457                 return res;
 458         }
 459
 460         return 0;
 461
 462 unwind:
 463         /* unwind from head to the slave that failed */
 464         bond_for_each_slave_continue_reverse(bond, slave)
 465                 vlan_vid_del(slave->dev, proto, vid);
 466
 467         return res;
 468 }
 469
 470 /**
 471  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 472  * @bond_dev: bonding net device that got called
 473  * @vid: vlan id being removed
 474  */
 475 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 476                                  __be16 proto, u16 vid)
 477 {
 478         struct bonding *bond = netdev_priv(bond_dev);
 479         struct slave *slave;
 480         int res;
 481
 482         bond_for_each_slave(bond, slave)
 483                 vlan_vid_del(slave->dev, proto, vid);
 484
 485         res = bond_del_vlan(bond, vid);
 486         if (res) {
 487                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 488                        bond_dev->name, vid);
 489                 return res;
 490         }
 491
 492         return 0;
 493 }
 494
 495 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 496 {
 497         struct vlan_entry *vlan;
 498         int res;
 499
 500         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 501                 res = vlan_vid_add(slave_dev, htons(ETH_P_8021Q),
 502                                    vlan->vlan_id);
 503                 if (res)
 504                         pr_warning("%s: Failed to add vlan id %d to device %s\n",
 505                                    bond->dev->name, vlan->vlan_id,
 506                                    slave_dev->name);
 507         }
 508 }
 509
 510 static void bond_del_vlans_from_slave(struct bonding *bond,
 511                                       struct net_device *slave_dev)
 512 {
 513         struct vlan_entry *vlan;
 514
 515         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 516                 if (!vlan->vlan_id)
 517                         continue;
 518                 vlan_vid_del(slave_dev, htons(ETH_P_8021Q), vlan->vlan_id);
 519         }
 520 }
 521
 522 /*------------------------------- Link status -------------------------------*/
 523
 524 /*
 525  * Set the carrier state for the master according to the state of its
 526  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 527  * do special 802.3ad magic.
 528  *
 529  * Returns zero if carrier state does not change, nonzero if it does.
 530  */
 531 static int bond_set_carrier(struct bonding *bond)
 532 {
 533         struct slave *slave;
 534
 535         if (list_empty(&bond->slave_list))
 536                 goto down;
 537
 538         if (bond->params.mode == BOND_MODE_8023AD)
 539                 return bond_3ad_set_carrier(bond);
 540
 541         bond_for_each_slave(bond, slave) {
 542                 if (slave->link == BOND_LINK_UP) {
 543                         if (!netif_carrier_ok(bond->dev)) {
 544                                 netif_carrier_on(bond->dev);
 545                                 return 1;
 546                         }
 547                         return 0;
 548                 }
 549         }
 550
 551 down:
 552         if (netif_carrier_ok(bond->dev)) {
 553                 netif_carrier_off(bond->dev);
 554                 return 1;
 555         }
 556         return 0;
 557 }
 558
 559 /*
 560  * Get link speed and duplex from the slave's base driver
 561  * using ethtool. If for some reason the call fails or the
 562  * values are invalid, set speed and duplex to -1,
 563  * and return.
 564  */
 565 static void bond_update_speed_duplex(struct slave *slave)
 566 {
 567         struct net_device *slave_dev = slave->dev;
 568         struct ethtool_cmd ecmd;
 569         u32 slave_speed;
 570         int res;
 571
 572         slave->speed = SPEED_UNKNOWN;
 573         slave->duplex = DUPLEX_UNKNOWN;
 574
 575         res = __ethtool_get_settings(slave_dev, &ecmd);
 576         if (res < 0)
 577                 return;
 578
 579         slave_speed = ethtool_cmd_speed(&ecmd);
 580         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 581                 return;
 582
 583         switch (ecmd.duplex) {
 584         case DUPLEX_FULL:
 585         case DUPLEX_HALF:
 586                 break;
 587         default:
 588                 return;
 589         }
 590
 591         slave->speed = slave_speed;
 592         slave->duplex = ecmd.duplex;
 593
 594         return;
 595 }
 596
 597 /*
 598  * if <dev> supports MII link status reporting, check its link status.
 599  *
 600  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 601  * depending upon the setting of the use_carrier parameter.
 602  *
 603  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 604  * can't tell and just pretend it is), or 0, meaning that the link is
 605  * down.
 606  *
 607  * If reporting is non-zero, instead of faking link up, return -1 if
 608  * both ETHTOOL and MII ioctls fail (meaning the device does not
 609  * support them).  If use_carrier is set, return whatever it says.
 610  * It'd be nice if there was a good way to tell if a driver supports
 611  * netif_carrier, but there really isn't.
 612  */
 613 static int bond_check_dev_link(struct bonding *bond,
 614                                struct net_device *slave_dev, int reporting)
 615 {
 616         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 617         int (*ioctl)(struct net_device *, struct ifreq *, int);
 618         struct ifreq ifr;
 619         struct mii_ioctl_data *mii;
 620
 621         if (!reporting && !netif_running(slave_dev))
 622                 return 0;
 623
 624         if (bond->params.use_carrier)
 625                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 626
 627         /* Try to get link status using Ethtool first. */
 628         if (slave_dev->ethtool_ops->get_link)
 629                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 630                         BMSR_LSTATUS : 0;
 631
 632         /* Ethtool can't be used, fallback to MII ioctls. */
 633         ioctl = slave_ops->ndo_do_ioctl;
 634         if (ioctl) {
 635                 /* TODO: set pointer to correct ioctl on a per team member */
 636                 /*       bases to make this more efficient. that is, once  */
 637                 /*       we determine the correct ioctl, we will always    */
 638                 /*       call it and not the others for that team          */
 639                 /*       member.                                           */
 640
 641                 /*
 642                  * We cannot assume that SIOCGMIIPHY will also read a
 643                  * register; not all network drivers (e.g., e100)
 644                  * support that.
 645                  */
 646
 647                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 648                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 649                 mii = if_mii(&ifr);
 650                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 651                         mii->reg_num = MII_BMSR;
 652                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 653                                 return mii->val_out & BMSR_LSTATUS;
 654                 }
 655         }
 656
 657         /*
 658          * If reporting, report that either there's no dev->do_ioctl,
 659          * or both SIOCGMIIREG and get_link failed (meaning that we
 660          * cannot report link status).  If not reporting, pretend
 661          * we're ok.
 662          */
 663         return reporting ? -1 : BMSR_LSTATUS;
 664 }
 665
 666 /*----------------------------- Multicast list ------------------------------*/
 667
 668 /*
 669  * Push the promiscuity flag down to appropriate slaves
 670  */
 671 static int bond_set_promiscuity(struct bonding *bond, int inc)
 672 {
 673         int err = 0;
 674         if (USES_PRIMARY(bond->params.mode)) {
 675                 /* write lock already acquired */
 676                 if (bond->curr_active_slave) {
 677                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 678                                                   inc);
 679                 }
 680         } else {
 681                 struct slave *slave;
 682
 683                 bond_for_each_slave(bond, slave) {
 684                         err = dev_set_promiscuity(slave->dev, inc);
 685                         if (err)
 686                                 return err;
 687                 }
 688         }
 689         return err;
 690 }
 691
 692 /*
 693  * Push the allmulti flag down to all slaves
 694  */
 695 static int bond_set_allmulti(struct bonding *bond, int inc)
 696 {
 697         int err = 0;
 698         if (USES_PRIMARY(bond->params.mode)) {
 699                 /* write lock already acquired */
 700                 if (bond->curr_active_slave) {
 701                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 702                                                inc);
 703                 }
 704         } else {
 705                 struct slave *slave;
 706
 707                 bond_for_each_slave(bond, slave) {
 708                         err = dev_set_allmulti(slave->dev, inc);
 709                         if (err)
 710                                 return err;
 711                 }
 712         }
 713         return err;
 714 }
 715
 716 /*
 717  * Retrieve the list of registered multicast addresses for the bonding
 718  * device and retransmit an IGMP JOIN request to the current active
 719  * slave.
 720  */
 721 static void bond_resend_igmp_join_requests(struct bonding *bond)
 722 {
 723         if (!rtnl_trylock()) {
 724                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 725                 return;
 726         }
 727         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 728         rtnl_unlock();
 729
 730         /* We use curr_slave_lock to protect against concurrent access to
 731          * igmp_retrans from multiple running instances of this function and
 732          * bond_change_active_slave
 733          */
 734         write_lock_bh(&bond->curr_slave_lock);
 735         if (bond->igmp_retrans > 1) {
 736                 bond->igmp_retrans--;
 737                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 738         }
 739         write_unlock_bh(&bond->curr_slave_lock);
 740 }
 741
 742 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 743 {
 744         struct bonding *bond = container_of(work, struct bonding,
 745                                             mcast_work.work);
 746
 747         bond_resend_igmp_join_requests(bond);
 748 }
 749
 750 /* Flush bond's hardware addresses from slave
 751  */
 752 static void bond_hw_addr_flush(struct net_device *bond_dev,
 753                                struct net_device *slave_dev)
 754 {
 755         struct bonding *bond = netdev_priv(bond_dev);
 756
 757         dev_uc_unsync(slave_dev, bond_dev);
 758         dev_mc_unsync(slave_dev, bond_dev);
 759
 760         if (bond->params.mode == BOND_MODE_8023AD) {
 761                 /* del lacpdu mc addr from mc list */
 762                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 763
 764                 dev_mc_del(slave_dev, lacpdu_multicast);
 765         }
 766 }
 767
 768 /*--------------------------- Active slave change ---------------------------*/
 769
 770 /* Update the hardware address list and promisc/allmulti for the new and
 771  * old active slaves (if any).  Modes that are !USES_PRIMARY keep all
 772  * slaves up date at all times; only the USES_PRIMARY modes need to call
 773  * this function to swap these settings during a failover.
 774  */
 775 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 776                               struct slave *old_active)
 777 {
 778         if (old_active) {
 779                 if (bond->dev->flags & IFF_PROMISC)
 780                         dev_set_promiscuity(old_active->dev, -1);
 781
 782                 if (bond->dev->flags & IFF_ALLMULTI)
 783                         dev_set_allmulti(old_active->dev, -1);
 784
 785                 bond_hw_addr_flush(bond->dev, old_active->dev);
 786         }
 787
 788         if (new_active) {
 789                 /* FIXME: Signal errors upstream. */
 790                 if (bond->dev->flags & IFF_PROMISC)
 791                         dev_set_promiscuity(new_active->dev, 1);
 792
 793                 if (bond->dev->flags & IFF_ALLMULTI)
 794                         dev_set_allmulti(new_active->dev, 1);
 795
 796                 netif_addr_lock_bh(bond->dev);
 797                 dev_uc_sync(new_active->dev, bond->dev);
 798                 dev_mc_sync(new_active->dev, bond->dev);
 799                 netif_addr_unlock_bh(bond->dev);
 800         }
 801 }
 802
 803 /**
 804  * bond_set_dev_addr - clone slave's address to bond
 805  * @bond_dev: bond net device
 806  * @slave_dev: slave net device
 807  *
 808  * Should be called with RTNL held.
 809  */
 810 static void bond_set_dev_addr(struct net_device *bond_dev,
 811                               struct net_device *slave_dev)
 812 {
 813         pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 814                  bond_dev, slave_dev, slave_dev->addr_len);
 815         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 816         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 817         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 818 }
 819
 820 /*
 821  * bond_do_fail_over_mac
 822  *
 823  * Perform special MAC address swapping for fail_over_mac settings
 824  *
 825  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 826  */
 827 static void bond_do_fail_over_mac(struct bonding *bond,
 828                                   struct slave *new_active,
 829                                   struct slave *old_active)
 830         __releases(&bond->curr_slave_lock)
 831         __releases(&bond->lock)
 832         __acquires(&bond->lock)
 833         __acquires(&bond->curr_slave_lock)
 834 {
 835         u8 tmp_mac[ETH_ALEN];
 836         struct sockaddr saddr;
 837         int rv;
 838
 839         switch (bond->params.fail_over_mac) {
 840         case BOND_FOM_ACTIVE:
 841                 if (new_active) {
 842                         write_unlock_bh(&bond->curr_slave_lock);
 843                         read_unlock(&bond->lock);
 844                         bond_set_dev_addr(bond->dev, new_active->dev);
 845                         read_lock(&bond->lock);
 846                         write_lock_bh(&bond->curr_slave_lock);
 847                 }
 848                 break;
 849         case BOND_FOM_FOLLOW:
 850                 /*
 851                  * if new_active && old_active, swap them
 852                  * if just old_active, do nothing (going to no active slave)
 853                  * if just new_active, set new_active to bond's MAC
 854                  */
 855                 if (!new_active)
 856                         return;
 857
 858                 write_unlock_bh(&bond->curr_slave_lock);
 859                 read_unlock(&bond->lock);
 860
 861                 if (old_active) {
 862                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 863                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 864                                ETH_ALEN);
 865                         saddr.sa_family = new_active->dev->type;
 866                 } else {
 867                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 868                         saddr.sa_family = bond->dev->type;
 869                 }
 870
 871                 rv = dev_set_mac_address(new_active->dev, &saddr);
 872                 if (rv) {
 873                         pr_err("%s: Error %d setting MAC of slave %s\n",
 874                                bond->dev->name, -rv, new_active->dev->name);
 875                         goto out;
 876                 }
 877
 878                 if (!old_active)
 879                         goto out;
 880
 881                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 882                 saddr.sa_family = old_active->dev->type;
 883
 884                 rv = dev_set_mac_address(old_active->dev, &saddr);
 885                 if (rv)
 886                         pr_err("%s: Error %d setting MAC of slave %s\n",
 887                                bond->dev->name, -rv, new_active->dev->name);
 888 out:
 889                 read_lock(&bond->lock);
 890                 write_lock_bh(&bond->curr_slave_lock);
 891                 break;
 892         default:
 893                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 894                        bond->dev->name, bond->params.fail_over_mac);
 895                 break;
 896         }
 897
 898 }
 899
 900 static bool bond_should_change_active(struct bonding *bond)
 901 {
 902         struct slave *prim = bond->primary_slave;
 903         struct slave *curr = bond->curr_active_slave;
 904
 905         if (!prim || !curr || curr->link != BOND_LINK_UP)
 906                 return true;
 907         if (bond->force_primary) {
 908                 bond->force_primary = false;
 909                 return true;
 910         }
 911         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 912             (prim->speed < curr->speed ||
 913              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 914                 return false;
 915         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 916                 return false;
 917         return true;
 918 }
 919
 920 /**
 921  * find_best_interface - select the best available slave to be the active one
 922  * @bond: our bonding struct
 923  *
 924  * Warning: Caller must hold curr_slave_lock for writing.
 925  */
 926 static struct slave *bond_find_best_slave(struct bonding *bond)
 927 {
 928         struct slave *new_active, *old_active;
 929         struct slave *bestslave = NULL;
 930         int mintime = bond->params.updelay;
 931         int i;
 932
 933         new_active = bond->curr_active_slave;
 934
 935         if (!new_active) { /* there were no active slaves left */
 936                 new_active = bond_first_slave(bond);
 937                 if (!new_active)
 938                         return NULL; /* still no slave, return NULL */
 939         }
 940
 941         if ((bond->primary_slave) &&
 942             bond->primary_slave->link == BOND_LINK_UP &&
 943             bond_should_change_active(bond)) {
 944                 new_active = bond->primary_slave;
 945         }
 946
 947         /* remember where to stop iterating over the slaves */
 948         old_active = new_active;
 949
 950         bond_for_each_slave_from(bond, new_active, i, old_active) {
 951                 if (new_active->link == BOND_LINK_UP) {
 952                         return new_active;
 953                 } else if (new_active->link == BOND_LINK_BACK &&
 954                            IS_UP(new_active->dev)) {
 955                         /* link up, but waiting for stabilization */
 956                         if (new_active->delay < mintime) {
 957                                 mintime = new_active->delay;
 958                                 bestslave = new_active;
 959                         }
 960                 }
 961         }
 962
 963         return bestslave;
 964 }
 965
 966 static bool bond_should_notify_peers(struct bonding *bond)
 967 {
 968         struct slave *slave = bond->curr_active_slave;
 969
 970         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
 971                  bond->dev->name, slave ? slave->dev->name : "NULL");
 972
 973         if (!slave || !bond->send_peer_notif ||
 974             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 975                 return false;
 976
 977         bond->send_peer_notif--;
 978         return true;
 979 }
 980
 981 /**
 982  * change_active_interface - change the active slave into the specified one
 983  * @bond: our bonding struct
 984  * @new: the new slave to make the active one
 985  *
 986  * Set the new slave to the bond's settings and unset them on the old
 987  * curr_active_slave.
 988  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 989  *
 990  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 991  * because it is apparently the best available slave we have, even though its
 992  * updelay hasn't timed out yet.
 993  *
 994  * If new_active is not NULL, caller must hold bond->lock for read and
 995  * curr_slave_lock for write_bh.
 996  */
 997 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 998 {
 999         struct slave *old_active = bond->curr_active_slave;
1000
1001         if (old_active == new_active)
1002                 return;
1003
1004         if (new_active) {
1005                 new_active->jiffies = jiffies;
1006
1007                 if (new_active->link == BOND_LINK_BACK) {
1008                         if (USES_PRIMARY(bond->params.mode)) {
1009                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1010                                         bond->dev->name, new_active->dev->name,
1011                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1012                         }
1013
1014                         new_active->delay = 0;
1015                         new_active->link = BOND_LINK_UP;
1016
1017                         if (bond->params.mode == BOND_MODE_8023AD)
1018                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1019
1020                         if (bond_is_lb(bond))
1021                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1022                 } else {
1023                         if (USES_PRIMARY(bond->params.mode)) {
1024                                 pr_info("%s: making interface %s the new active one.\n",
1025                                         bond->dev->name, new_active->dev->name);
1026                         }
1027                 }
1028         }
1029
1030         if (USES_PRIMARY(bond->params.mode))
1031                 bond_hw_addr_swap(bond, new_active, old_active);
1032
1033         if (bond_is_lb(bond)) {
1034                 bond_alb_handle_active_change(bond, new_active);
1035                 if (old_active)
1036                         bond_set_slave_inactive_flags(old_active);
1037                 if (new_active)
1038                         bond_set_slave_active_flags(new_active);
1039         } else {
1040                 bond->curr_active_slave = new_active;
1041         }
1042
1043         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1044                 if (old_active)
1045                         bond_set_slave_inactive_flags(old_active);
1046
1047                 if (new_active) {
1048                         bool should_notify_peers = false;
1049
1050                         bond_set_slave_active_flags(new_active);
1051
1052                         if (bond->params.fail_over_mac)
1053                                 bond_do_fail_over_mac(bond, new_active,
1054                                                       old_active);
1055
1056                         if (netif_running(bond->dev)) {
1057                                 bond->send_peer_notif =
1058                                         bond->params.num_peer_notif;
1059                                 should_notify_peers =
1060                                         bond_should_notify_peers(bond);
1061                         }
1062
1063                         write_unlock_bh(&bond->curr_slave_lock);
1064                         read_unlock(&bond->lock);
1065
1066                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1067                         if (should_notify_peers)
1068                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1069                                                          bond->dev);
1070
1071                         read_lock(&bond->lock);
1072                         write_lock_bh(&bond->curr_slave_lock);
1073                 }
1074         }
1075
1076         /* resend IGMP joins since active slave has changed or
1077          * all were sent on curr_active_slave.
1078          * resend only if bond is brought up with the affected
1079          * bonding modes and the retransmission is enabled */
1080         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1081             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1082              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1083                 bond->igmp_retrans = bond->params.resend_igmp;
1084                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1085         }
1086 }
1087
1088 /**
1089  * bond_select_active_slave - select a new active slave, if needed
1090  * @bond: our bonding struct
1091  *
1092  * This functions should be called when one of the following occurs:
1093  * - The old curr_active_slave has been released or lost its link.
1094  * - The primary_slave has got its link back.
1095  * - A slave has got its link back and there's no old curr_active_slave.
1096  *
1097  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1098  */
1099 void bond_select_active_slave(struct bonding *bond)
1100 {
1101         struct slave *best_slave;
1102         int rv;
1103
1104         best_slave = bond_find_best_slave(bond);
1105         if (best_slave != bond->curr_active_slave) {
1106                 bond_change_active_slave(bond, best_slave);
1107                 rv = bond_set_carrier(bond);
1108                 if (!rv)
1109                         return;
1110
1111                 if (netif_carrier_ok(bond->dev)) {
1112                         pr_info("%s: first active interface up!\n",
1113                                 bond->dev->name);
1114                 } else {
1115                         pr_info("%s: now running without any active interface !\n",
1116                                 bond->dev->name);
1117                 }
1118         }
1119 }
1120
1121 /*--------------------------- slave list handling ---------------------------*/
1122
1123 /*
1124  * This function attaches the slave to the end of list.
1125  *
1126  * bond->lock held for writing by caller.
1127  */
1128 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1129 {
1130         list_add_tail(&new_slave->list, &bond->slave_list);
1131         bond->slave_cnt++;
1132 }
1133
1134 /*
1135  * This function detaches the slave from the list.
1136  * WARNING: no check is made to verify if the slave effectively
1137  * belongs to <bond>.
1138  * Nothing is freed on return, structures are just unchained.
1139  * If any slave pointer in bond was pointing to <slave>,
1140  * it should be changed by the calling function.
1141  *
1142  * bond->lock held for writing by caller.
1143  */
1144 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1145 {
1146         list_del(&slave->list);
1147         bond->slave_cnt--;
1148 }
1149
1150 #ifdef CONFIG_NET_POLL_CONTROLLER
1151 static inline int slave_enable_netpoll(struct slave *slave)
1152 {
1153         struct netpoll *np;
1154         int err = 0;
1155
1156         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1157         err = -ENOMEM;
1158         if (!np)
1159                 goto out;
1160
1161         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1162         if (err) {
1163                 kfree(np);
1164                 goto out;
1165         }
1166         slave->np = np;
1167 out:
1168         return err;
1169 }
1170 static inline void slave_disable_netpoll(struct slave *slave)
1171 {
1172         struct netpoll *np = slave->np;
1173
1174         if (!np)
1175                 return;
1176
1177         slave->np = NULL;
1178         __netpoll_free_async(np);
1179 }
1180 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1181 {
1182         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1183                 return false;
1184         if (!slave_dev->netdev_ops->ndo_poll_controller)
1185                 return false;
1186         return true;
1187 }
1188
1189 static void bond_poll_controller(struct net_device *bond_dev)
1190 {
1191 }
1192
1193 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1194 {
1195         struct bonding *bond = netdev_priv(bond_dev);
1196         struct slave *slave;
1197
1198         bond_for_each_slave(bond, slave)
1199                 if (IS_UP(slave->dev))
1200                         slave_disable_netpoll(slave);
1201 }
1202
1203 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1204 {
1205         struct bonding *bond = netdev_priv(dev);
1206         struct slave *slave;
1207         int err = 0;
1208
1209         bond_for_each_slave(bond, slave) {
1210                 err = slave_enable_netpoll(slave);
1211                 if (err) {
1212                         bond_netpoll_cleanup(dev);
1213                         break;
1214                 }
1215         }
1216         return err;
1217 }
1218 #else
1219 static inline int slave_enable_netpoll(struct slave *slave)
1220 {
1221         return 0;
1222 }
1223 static inline void slave_disable_netpoll(struct slave *slave)
1224 {
1225 }
1226 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1227 {
1228 }
1229 #endif
1230
1231 /*---------------------------------- IOCTL ----------------------------------*/
1232
1233 static netdev_features_t bond_fix_features(struct net_device *dev,
1234         netdev_features_t features)
1235 {
1236         struct slave *slave;
1237         struct bonding *bond = netdev_priv(dev);
1238         netdev_features_t mask;
1239
1240         read_lock(&bond->lock);
1241
1242         if (list_empty(&bond->slave_list)) {
1243                 /* Disable adding VLANs to empty bond. But why? --mq */
1244                 features |= NETIF_F_VLAN_CHALLENGED;
1245                 goto out;
1246         }
1247
1248         mask = features;
1249         features &= ~NETIF_F_ONE_FOR_ALL;
1250         features |= NETIF_F_ALL_FOR_ALL;
1251
1252         bond_for_each_slave(bond, slave) {
1253                 features = netdev_increment_features(features,
1254                                                      slave->dev->features,
1255                                                      mask);
1256         }
1257         features = netdev_add_tso_features(features, mask);
1258
1259 out:
1260         read_unlock(&bond->lock);
1261         return features;
1262 }
1263
1264 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1265                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1266                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1267
1268 static void bond_compute_features(struct bonding *bond)
1269 {
1270         struct slave *slave;
1271         struct net_device *bond_dev = bond->dev;
1272         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1273         unsigned short max_hard_header_len = ETH_HLEN;
1274         unsigned int gso_max_size = GSO_MAX_SIZE;
1275         u16 gso_max_segs = GSO_MAX_SEGS;
1276         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1277
1278         read_lock(&bond->lock);
1279
1280         if (list_empty(&bond->slave_list))
1281                 goto done;
1282
1283         bond_for_each_slave(bond, slave) {
1284                 vlan_features = netdev_increment_features(vlan_features,
1285                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1286
1287                 dst_release_flag &= slave->dev->priv_flags;
1288                 if (slave->dev->hard_header_len > max_hard_header_len)
1289                         max_hard_header_len = slave->dev->hard_header_len;
1290
1291                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1292                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1293         }
1294
1295 done:
1296         bond_dev->vlan_features = vlan_features;
1297         bond_dev->hard_header_len = max_hard_header_len;
1298         bond_dev->gso_max_segs = gso_max_segs;
1299         netif_set_gso_max_size(bond_dev, gso_max_size);
1300
1301         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1302         bond_dev->priv_flags = flags | dst_release_flag;
1303
1304         read_unlock(&bond->lock);
1305
1306         netdev_change_features(bond_dev);
1307 }
1308
1309 static void bond_setup_by_slave(struct net_device *bond_dev,
1310                                 struct net_device *slave_dev)
1311 {
1312         bond_dev->header_ops        = slave_dev->header_ops;
1313
1314         bond_dev->type              = slave_dev->type;
1315         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1316         bond_dev->addr_len          = slave_dev->addr_len;
1317
1318         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1319                 slave_dev->addr_len);
1320 }
1321
1322 /* On bonding slaves other than the currently active slave, suppress
1323  * duplicates except for alb non-mcast/bcast.
1324  */
1325 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1326                                             struct slave *slave,
1327                                             struct bonding *bond)
1328 {
1329         if (bond_is_slave_inactive(slave)) {
1330                 if (bond->params.mode == BOND_MODE_ALB &&
1331                     skb->pkt_type != PACKET_BROADCAST &&
1332                     skb->pkt_type != PACKET_MULTICAST)
1333                         return false;
1334                 return true;
1335         }
1336         return false;
1337 }
1338
1339 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1340 {
1341         struct sk_buff *skb = *pskb;
1342         struct slave *slave;
1343         struct bonding *bond;
1344         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1345                           struct slave *);
1346         int ret = RX_HANDLER_ANOTHER;
1347
1348         skb = skb_share_check(skb, GFP_ATOMIC);
1349         if (unlikely(!skb))
1350                 return RX_HANDLER_CONSUMED;
1351
1352         *pskb = skb;
1353
1354         slave = bond_slave_get_rcu(skb->dev);
1355         bond = slave->bond;
1356
1357         if (bond->params.arp_interval)
1358                 slave->dev->last_rx = jiffies;
1359
1360         recv_probe = ACCESS_ONCE(bond->recv_probe);
1361         if (recv_probe) {
1362                 ret = recv_probe(skb, bond, slave);
1363                 if (ret == RX_HANDLER_CONSUMED) {
1364                         consume_skb(skb);
1365                         return ret;
1366                 }
1367         }
1368
1369         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1370                 return RX_HANDLER_EXACT;
1371         }
1372
1373         skb->dev = bond->dev;
1374
1375         if (bond->params.mode == BOND_MODE_ALB &&
1376             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1377             skb->pkt_type == PACKET_HOST) {
1378
1379                 if (unlikely(skb_cow_head(skb,
1380                                           skb->data - skb_mac_header(skb)))) {
1381                         kfree_skb(skb);
1382                         return RX_HANDLER_CONSUMED;
1383                 }
1384                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1385         }
1386
1387         return ret;
1388 }
1389
1390 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1391                                       struct net_device *slave_dev)
1392 {
1393         int err;
1394
1395         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1396         if (err)
1397                 return err;
1398         slave_dev->flags |= IFF_SLAVE;
1399         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1400         return 0;
1401 }
1402
1403 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1404                                   struct net_device *slave_dev)
1405 {
1406         netdev_upper_dev_unlink(slave_dev, bond_dev);
1407         slave_dev->flags &= ~IFF_SLAVE;
1408         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1409 }
1410
1411 /* enslave device <slave> to bond device <master> */
1412 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1413 {
1414         struct bonding *bond = netdev_priv(bond_dev);
1415         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1416         struct slave *new_slave = NULL;
1417         struct sockaddr addr;
1418         int link_reporting;
1419         int res = 0, i;
1420
1421         if (!bond->params.use_carrier &&
1422             slave_dev->ethtool_ops->get_link == NULL &&
1423             slave_ops->ndo_do_ioctl == NULL) {
1424                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1425                            bond_dev->name, slave_dev->name);
1426         }
1427
1428         /* already enslaved */
1429         if (slave_dev->flags & IFF_SLAVE) {
1430                 pr_debug("Error, Device was already enslaved\n");
1431                 return -EBUSY;
1432         }
1433
1434         /* vlan challenged mutual exclusion */
1435         /* no need to lock since we're protected by rtnl_lock */
1436         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1437                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1438                 if (vlan_uses_dev(bond_dev)) {
1439                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1440                                bond_dev->name, slave_dev->name, bond_dev->name);
1441                         return -EPERM;
1442                 } else {
1443                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1444                                    bond_dev->name, slave_dev->name,
1445                                    slave_dev->name, bond_dev->name);
1446                 }
1447         } else {
1448                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1449         }
1450
1451         /*
1452          * Old ifenslave binaries are no longer supported.  These can
1453          * be identified with moderate accuracy by the state of the slave:
1454          * the current ifenslave will set the interface down prior to
1455          * enslaving it; the old ifenslave will not.
1456          */
1457         if ((slave_dev->flags & IFF_UP)) {
1458                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1459                        slave_dev->name);
1460                 res = -EPERM;
1461                 goto err_undo_flags;
1462         }
1463
1464         /* set bonding device ether type by slave - bonding netdevices are
1465          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1466          * there is a need to override some of the type dependent attribs/funcs.
1467          *
1468          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1469          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1470          */
1471         if (list_empty(&bond->slave_list)) {
1472                 if (bond_dev->type != slave_dev->type) {
1473                         pr_debug("%s: change device type from %d to %d\n",
1474                                  bond_dev->name,
1475                                  bond_dev->type, slave_dev->type);
1476
1477                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1478                                                        bond_dev);
1479                         res = notifier_to_errno(res);
1480                         if (res) {
1481                                 pr_err("%s: refused to change device type\n",
1482                                        bond_dev->name);
1483                                 res = -EBUSY;
1484                                 goto err_undo_flags;
1485                         }
1486
1487                         /* Flush unicast and multicast addresses */
1488                         dev_uc_flush(bond_dev);
1489                         dev_mc_flush(bond_dev);
1490
1491                         if (slave_dev->type != ARPHRD_ETHER)
1492                                 bond_setup_by_slave(bond_dev, slave_dev);
1493                         else {
1494                                 ether_setup(bond_dev);
1495                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1496                         }
1497
1498                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1499                                                  bond_dev);
1500                 }
1501         } else if (bond_dev->type != slave_dev->type) {
1502                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1503                        slave_dev->name,
1504                        slave_dev->type, bond_dev->type);
1505                 res = -EINVAL;
1506                 goto err_undo_flags;
1507         }
1508
1509         if (slave_ops->ndo_set_mac_address == NULL) {
1510                 if (list_empty(&bond->slave_list)) {
1511                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1512                                    bond_dev->name);
1513                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1514                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1515                         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",
1516                                bond_dev->name);
1517                         res = -EOPNOTSUPP;
1518                         goto err_undo_flags;
1519                 }
1520         }
1521
1522         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1523
1524         /* If this is the first slave, then we need to set the master's hardware
1525          * address to be the same as the slave's. */
1526         if (list_empty(&bond->slave_list) &&
1527             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1528                 bond_set_dev_addr(bond->dev, slave_dev);
1529
1530         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1531         if (!new_slave) {
1532                 res = -ENOMEM;
1533                 goto err_undo_flags;
1534         }
1535         INIT_LIST_HEAD(&new_slave->list);
1536         /*
1537          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1538          * is set via sysfs or module option if desired.
1539          */
1540         new_slave->queue_id = 0;
1541
1542         /* Save slave's original mtu and then set it to match the bond */
1543         new_slave->original_mtu = slave_dev->mtu;
1544         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1545         if (res) {
1546                 pr_debug("Error %d calling dev_set_mtu\n", res);
1547                 goto err_free;
1548         }
1549
1550         /*
1551          * Save slave's original ("permanent") mac address for modes
1552          * that need it, and for restoring it upon release, and then
1553          * set it to the master's address
1554          */
1555         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1556
1557         if (!bond->params.fail_over_mac) {
1558                 /*
1559                  * Set slave to master's mac address.  The application already
1560                  * set the master's mac address to that of the first slave
1561                  */
1562                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1563                 addr.sa_family = slave_dev->type;
1564                 res = dev_set_mac_address(slave_dev, &addr);
1565                 if (res) {
1566                         pr_debug("Error %d calling set_mac_address\n", res);
1567                         goto err_restore_mtu;
1568                 }
1569         }
1570
1571         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1572         if (res) {
1573                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1574                 goto err_restore_mac;
1575         }
1576
1577         /* open the slave since the application closed it */
1578         res = dev_open(slave_dev);
1579         if (res) {
1580                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1581                 goto err_unset_master;
1582         }
1583
1584         new_slave->bond = bond;
1585         new_slave->dev = slave_dev;
1586         slave_dev->priv_flags |= IFF_BONDING;
1587
1588         if (bond_is_lb(bond)) {
1589                 /* bond_alb_init_slave() must be called before all other stages since
1590                  * it might fail and we do not want to have to undo everything
1591                  */
1592                 res = bond_alb_init_slave(bond, new_slave);
1593                 if (res)
1594                         goto err_close;
1595         }
1596
1597         /* If the mode USES_PRIMARY, then the following is handled by
1598          * bond_change_active_slave().
1599          */
1600         if (!USES_PRIMARY(bond->params.mode)) {
1601                 /* set promiscuity level to new slave */
1602                 if (bond_dev->flags & IFF_PROMISC) {
1603                         res = dev_set_promiscuity(slave_dev, 1);
1604                         if (res)
1605                                 goto err_close;
1606                 }
1607
1608                 /* set allmulti level to new slave */
1609                 if (bond_dev->flags & IFF_ALLMULTI) {
1610                         res = dev_set_allmulti(slave_dev, 1);
1611                         if (res)
1612                                 goto err_close;
1613                 }
1614
1615                 netif_addr_lock_bh(bond_dev);
1616
1617                 dev_mc_sync_multiple(slave_dev, bond_dev);
1618                 dev_uc_sync_multiple(slave_dev, bond_dev);
1619
1620                 netif_addr_unlock_bh(bond_dev);
1621         }
1622
1623         if (bond->params.mode == BOND_MODE_8023AD) {
1624                 /* add lacpdu mc addr to mc list */
1625                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1626
1627                 dev_mc_add(slave_dev, lacpdu_multicast);
1628         }
1629
1630         bond_add_vlans_on_slave(bond, slave_dev);
1631
1632         write_lock_bh(&bond->lock);
1633
1634         bond_attach_slave(bond, new_slave);
1635
1636         new_slave->delay = 0;
1637         new_slave->link_failure_count = 0;
1638
1639         write_unlock_bh(&bond->lock);
1640
1641         bond_compute_features(bond);
1642
1643         bond_update_speed_duplex(new_slave);
1644
1645         read_lock(&bond->lock);
1646
1647         new_slave->last_arp_rx = jiffies -
1648                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1649         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1650                 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1651
1652         if (bond->params.miimon && !bond->params.use_carrier) {
1653                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1654
1655                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1656                         /*
1657                          * miimon is set but a bonded network driver
1658                          * does not support ETHTOOL/MII and
1659                          * arp_interval is not set.  Note: if
1660                          * use_carrier is enabled, we will never go
1661                          * here (because netif_carrier is always
1662                          * supported); thus, we don't need to change
1663                          * the messages for netif_carrier.
1664                          */
1665                         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",
1666                                bond_dev->name, slave_dev->name);
1667                 } else if (link_reporting == -1) {
1668                         /* unable get link status using mii/ethtool */
1669                         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",
1670                                    bond_dev->name, slave_dev->name);
1671                 }
1672         }
1673
1674         /* check for initial state */
1675         if (bond->params.miimon) {
1676                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1677                         if (bond->params.updelay) {
1678                                 new_slave->link = BOND_LINK_BACK;
1679                                 new_slave->delay = bond->params.updelay;
1680                         } else {
1681                                 new_slave->link = BOND_LINK_UP;
1682                         }
1683                 } else {
1684                         new_slave->link = BOND_LINK_DOWN;
1685                 }
1686         } else if (bond->params.arp_interval) {
1687                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1688                         BOND_LINK_UP : BOND_LINK_DOWN);
1689         } else {
1690                 new_slave->link = BOND_LINK_UP;
1691         }
1692
1693         if (new_slave->link != BOND_LINK_DOWN)
1694                 new_slave->jiffies = jiffies;
1695         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1696                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1697                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1698
1699         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1700                 /* if there is a primary slave, remember it */
1701                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1702                         bond->primary_slave = new_slave;
1703                         bond->force_primary = true;
1704                 }
1705         }
1706
1707         write_lock_bh(&bond->curr_slave_lock);
1708
1709         switch (bond->params.mode) {
1710         case BOND_MODE_ACTIVEBACKUP:
1711                 bond_set_slave_inactive_flags(new_slave);
1712                 bond_select_active_slave(bond);
1713                 break;
1714         case BOND_MODE_8023AD:
1715                 /* in 802.3ad mode, the internal mechanism
1716                  * will activate the slaves in the selected
1717                  * aggregator
1718                  */
1719                 bond_set_slave_inactive_flags(new_slave);
1720                 /* if this is the first slave */
1721                 if (bond_first_slave(bond) == new_slave) {
1722                         SLAVE_AD_INFO(new_slave).id = 1;
1723                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1724                          * can be called only after the mac address of the bond is set
1725                          */
1726                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1727                 } else {
1728                         struct slave *prev_slave;
1729
1730                         prev_slave = bond_prev_slave(bond, new_slave);
1731                         SLAVE_AD_INFO(new_slave).id =
1732                                 SLAVE_AD_INFO(prev_slave).id + 1;
1733                 }
1734
1735                 bond_3ad_bind_slave(new_slave);
1736                 break;
1737         case BOND_MODE_TLB:
1738         case BOND_MODE_ALB:
1739                 bond_set_active_slave(new_slave);
1740                 bond_set_slave_inactive_flags(new_slave);
1741                 bond_select_active_slave(bond);
1742                 break;
1743         default:
1744                 pr_debug("This slave is always active in trunk mode\n");
1745
1746                 /* always active in trunk mode */
1747                 bond_set_active_slave(new_slave);
1748
1749                 /* In trunking mode there is little meaning to curr_active_slave
1750                  * anyway (it holds no special properties of the bond device),
1751                  * so we can change it without calling change_active_interface()
1752                  */
1753                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1754                         bond->curr_active_slave = new_slave;
1755
1756                 break;
1757         } /* switch(bond_mode) */
1758
1759         write_unlock_bh(&bond->curr_slave_lock);
1760
1761         bond_set_carrier(bond);
1762
1763 #ifdef CONFIG_NET_POLL_CONTROLLER
1764         slave_dev->npinfo = bond->dev->npinfo;
1765         if (slave_dev->npinfo) {
1766                 if (slave_enable_netpoll(new_slave)) {
1767                         read_unlock(&bond->lock);
1768                         pr_info("Error, %s: master_dev is using netpoll, "
1769                                  "but new slave device does not support netpoll.\n",
1770                                  bond_dev->name);
1771                         res = -EBUSY;
1772                         goto err_detach;
1773                 }
1774         }
1775 #endif
1776
1777         read_unlock(&bond->lock);
1778
1779         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1780         if (res)
1781                 goto err_detach;
1782
1783         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1784                                          new_slave);
1785         if (res) {
1786                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1787                 goto err_dest_symlinks;
1788         }
1789
1790         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1791                 bond_dev->name, slave_dev->name,
1792                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1793                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1794
1795         /* enslave is successful */
1796         return 0;
1797
1798 /* Undo stages on error */
1799 err_dest_symlinks:
1800         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1801
1802 err_detach:
1803         if (!USES_PRIMARY(bond->params.mode))
1804                 bond_hw_addr_flush(bond_dev, slave_dev);
1805
1806         bond_del_vlans_from_slave(bond, slave_dev);
1807         write_lock_bh(&bond->lock);
1808         bond_detach_slave(bond, new_slave);
1809         if (bond->primary_slave == new_slave)
1810                 bond->primary_slave = NULL;
1811         if (bond->curr_active_slave == new_slave) {
1812                 bond_change_active_slave(bond, NULL);
1813                 write_unlock_bh(&bond->lock);
1814                 read_lock(&bond->lock);
1815                 write_lock_bh(&bond->curr_slave_lock);
1816                 bond_select_active_slave(bond);
1817                 write_unlock_bh(&bond->curr_slave_lock);
1818                 read_unlock(&bond->lock);
1819         } else {
1820                 write_unlock_bh(&bond->lock);
1821         }
1822         slave_disable_netpoll(new_slave);
1823
1824 err_close:
1825         slave_dev->priv_flags &= ~IFF_BONDING;
1826         dev_close(slave_dev);
1827
1828 err_unset_master:
1829         bond_upper_dev_unlink(bond_dev, slave_dev);
1830
1831 err_restore_mac:
1832         if (!bond->params.fail_over_mac) {
1833                 /* XXX TODO - fom follow mode needs to change master's
1834                  * MAC if this slave's MAC is in use by the bond, or at
1835                  * least print a warning.
1836                  */
1837                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1838                 addr.sa_family = slave_dev->type;
1839                 dev_set_mac_address(slave_dev, &addr);
1840         }
1841
1842 err_restore_mtu:
1843         dev_set_mtu(slave_dev, new_slave->original_mtu);
1844
1845 err_free:
1846         kfree(new_slave);
1847
1848 err_undo_flags:
1849         bond_compute_features(bond);
1850         /* Enslave of first slave has failed and we need to fix master's mac */
1851         if (list_empty(&bond->slave_list) &&
1852             ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1853                 eth_hw_addr_random(bond_dev);
1854
1855         return res;
1856 }
1857
1858 /*
1859  * Try to release the slave device <slave> from the bond device <master>
1860  * It is legal to access curr_active_slave without a lock because all the function
1861  * is write-locked. If "all" is true it means that the function is being called
1862  * while destroying a bond interface and all slaves are being released.
1863  *
1864  * The rules for slave state should be:
1865  *   for Active/Backup:
1866  *     Active stays on all backups go down
1867  *   for Bonded connections:
1868  *     The first up interface should be left on and all others downed.
1869  */
1870 static int __bond_release_one(struct net_device *bond_dev,
1871                               struct net_device *slave_dev,
1872                               bool all)
1873 {
1874         struct bonding *bond = netdev_priv(bond_dev);
1875         struct slave *slave, *oldcurrent;
1876         struct sockaddr addr;
1877         netdev_features_t old_features = bond_dev->features;
1878
1879         /* slave is not a slave or master is not master of this slave */
1880         if (!(slave_dev->flags & IFF_SLAVE) ||
1881             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1882                 pr_err("%s: Error: cannot release %s.\n",
1883                        bond_dev->name, slave_dev->name);
1884                 return -EINVAL;
1885         }
1886
1887         block_netpoll_tx();
1888         write_lock_bh(&bond->lock);
1889
1890         slave = bond_get_slave_by_dev(bond, slave_dev);
1891         if (!slave) {
1892                 /* not a slave of this bond */
1893                 pr_info("%s: %s not enslaved\n",
1894                         bond_dev->name, slave_dev->name);
1895                 write_unlock_bh(&bond->lock);
1896                 unblock_netpoll_tx();
1897                 return -EINVAL;
1898         }
1899
1900         write_unlock_bh(&bond->lock);
1901         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1902          * for this slave anymore.
1903          */
1904         netdev_rx_handler_unregister(slave_dev);
1905         write_lock_bh(&bond->lock);
1906
1907         /* Inform AD package of unbinding of slave. */
1908         if (bond->params.mode == BOND_MODE_8023AD) {
1909                 /* must be called before the slave is
1910                  * detached from the list
1911                  */
1912                 bond_3ad_unbind_slave(slave);
1913         }
1914
1915         pr_info("%s: releasing %s interface %s\n",
1916                 bond_dev->name,
1917                 bond_is_active_slave(slave) ? "active" : "backup",
1918                 slave_dev->name);
1919
1920         oldcurrent = bond->curr_active_slave;
1921
1922         bond->current_arp_slave = NULL;
1923
1924         /* release the slave from its bond */
1925         bond_detach_slave(bond, slave);
1926
1927         if (!all && !bond->params.fail_over_mac) {
1928                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1929                     !list_empty(&bond->slave_list))
1930                         pr_warn("%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",
1931                                    bond_dev->name, slave_dev->name,
1932                                    slave->perm_hwaddr,
1933                                    bond_dev->name, slave_dev->name);
1934         }
1935
1936         if (bond->primary_slave == slave)
1937                 bond->primary_slave = NULL;
1938
1939         if (oldcurrent == slave)
1940                 bond_change_active_slave(bond, NULL);
1941
1942         if (bond_is_lb(bond)) {
1943                 /* Must be called only after the slave has been
1944                  * detached from the list and the curr_active_slave
1945                  * has been cleared (if our_slave == old_current),
1946                  * but before a new active slave is selected.
1947                  */
1948                 write_unlock_bh(&bond->lock);
1949                 bond_alb_deinit_slave(bond, slave);
1950                 write_lock_bh(&bond->lock);
1951         }
1952
1953         if (all) {
1954                 bond->curr_active_slave = NULL;
1955         } else if (oldcurrent == slave) {
1956                 /*
1957                  * Note that we hold RTNL over this sequence, so there
1958                  * is no concern that another slave add/remove event
1959                  * will interfere.
1960                  */
1961                 write_unlock_bh(&bond->lock);
1962                 read_lock(&bond->lock);
1963                 write_lock_bh(&bond->curr_slave_lock);
1964
1965                 bond_select_active_slave(bond);
1966
1967                 write_unlock_bh(&bond->curr_slave_lock);
1968                 read_unlock(&bond->lock);
1969                 write_lock_bh(&bond->lock);
1970         }
1971
1972         if (list_empty(&bond->slave_list)) {
1973                 bond_set_carrier(bond);
1974                 eth_hw_addr_random(bond_dev);
1975
1976                 if (bond_vlan_used(bond)) {
1977                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1978                                    bond_dev->name, bond_dev->name);
1979                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1980                                    bond_dev->name);
1981                 }
1982         }
1983
1984         write_unlock_bh(&bond->lock);
1985         unblock_netpoll_tx();
1986
1987         if (list_empty(&bond->slave_list)) {
1988                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1989                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1990         }
1991
1992         bond_compute_features(bond);
1993         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1994             (old_features & NETIF_F_VLAN_CHALLENGED))
1995                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1996                         bond_dev->name, slave_dev->name, bond_dev->name);
1997
1998         /* must do this from outside any spinlocks */
1999         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2000
2001         bond_del_vlans_from_slave(bond, slave_dev);
2002
2003         /* If the mode USES_PRIMARY, then this cases was handled above by
2004          * bond_change_active_slave(..., NULL)
2005          */
2006         if (!USES_PRIMARY(bond->params.mode)) {
2007                 /* unset promiscuity level from slave */
2008                 if (bond_dev->flags & IFF_PROMISC)
2009                         dev_set_promiscuity(slave_dev, -1);
2010
2011                 /* unset allmulti level from slave */
2012                 if (bond_dev->flags & IFF_ALLMULTI)
2013                         dev_set_allmulti(slave_dev, -1);
2014
2015                 bond_hw_addr_flush(bond_dev, slave_dev);
2016         }
2017
2018         bond_upper_dev_unlink(bond_dev, slave_dev);
2019
2020         slave_disable_netpoll(slave);
2021
2022         /* close slave before restoring its mac address */
2023         dev_close(slave_dev);
2024
2025         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2026                 /* restore original ("permanent") mac address */
2027                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2028                 addr.sa_family = slave_dev->type;
2029                 dev_set_mac_address(slave_dev, &addr);
2030         }
2031
2032         dev_set_mtu(slave_dev, slave->original_mtu);
2033
2034         slave_dev->priv_flags &= ~IFF_BONDING;
2035
2036         kfree(slave);
2037
2038         return 0;  /* deletion OK */
2039 }
2040
2041 /* A wrapper used because of ndo_del_link */
2042 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2043 {
2044         return __bond_release_one(bond_dev, slave_dev, false);
2045 }
2046
2047 /*
2048 * First release a slave and then destroy the bond if no more slaves are left.
2049 * Must be under rtnl_lock when this function is called.
2050 */
2051 static int  bond_release_and_destroy(struct net_device *bond_dev,
2052                                      struct net_device *slave_dev)
2053 {
2054         struct bonding *bond = netdev_priv(bond_dev);
2055         int ret;
2056
2057         ret = bond_release(bond_dev, slave_dev);
2058         if (ret == 0 && list_empty(&bond->slave_list)) {
2059                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2060                 pr_info("%s: destroying bond %s.\n",
2061                         bond_dev->name, bond_dev->name);
2062                 unregister_netdevice(bond_dev);
2063         }
2064         return ret;
2065 }
2066
2067 /*
2068  * This function changes the active slave to slave <slave_dev>.
2069  * It returns -EINVAL in the following cases.
2070  *  - <slave_dev> is not found in the list.
2071  *  - There is not active slave now.
2072  *  - <slave_dev> is already active.
2073  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2074  *  - <slave_dev> is not running.
2075  * In these cases, this function does nothing.
2076  * In the other cases, current_slave pointer is changed and 0 is returned.
2077  */
2078 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2079 {
2080         struct bonding *bond = netdev_priv(bond_dev);
2081         struct slave *old_active = NULL;
2082         struct slave *new_active = NULL;
2083         int res = 0;
2084
2085         if (!USES_PRIMARY(bond->params.mode))
2086                 return -EINVAL;
2087
2088         /* Verify that bond_dev is indeed the master of slave_dev */
2089         if (!(slave_dev->flags & IFF_SLAVE) ||
2090             !netdev_has_upper_dev(slave_dev, bond_dev))
2091                 return -EINVAL;
2092
2093         read_lock(&bond->lock);
2094
2095         old_active = bond->curr_active_slave;
2096         new_active = bond_get_slave_by_dev(bond, slave_dev);
2097         /*
2098          * Changing to the current active: do nothing; return success.
2099          */
2100         if (new_active && new_active == old_active) {
2101                 read_unlock(&bond->lock);
2102                 return 0;
2103         }
2104
2105         if (new_active &&
2106             old_active &&
2107             new_active->link == BOND_LINK_UP &&
2108             IS_UP(new_active->dev)) {
2109                 block_netpoll_tx();
2110                 write_lock_bh(&bond->curr_slave_lock);
2111                 bond_change_active_slave(bond, new_active);
2112                 write_unlock_bh(&bond->curr_slave_lock);
2113                 unblock_netpoll_tx();
2114         } else
2115                 res = -EINVAL;
2116
2117         read_unlock(&bond->lock);
2118
2119         return res;
2120 }
2121
2122 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2123 {
2124         struct bonding *bond = netdev_priv(bond_dev);
2125
2126         info->bond_mode = bond->params.mode;
2127         info->miimon = bond->params.miimon;
2128
2129         read_lock(&bond->lock);
2130         info->num_slaves = bond->slave_cnt;
2131         read_unlock(&bond->lock);
2132
2133         return 0;
2134 }
2135
2136 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2137 {
2138         struct bonding *bond = netdev_priv(bond_dev);
2139         int i = 0, res = -ENODEV;
2140         struct slave *slave;
2141
2142         read_lock(&bond->lock);
2143         bond_for_each_slave(bond, slave) {
2144                 if (i++ == (int)info->slave_id) {
2145                         res = 0;
2146                         strcpy(info->slave_name, slave->dev->name);
2147                         info->link = slave->link;
2148                         info->state = bond_slave_state(slave);
2149                         info->link_failure_count = slave->link_failure_count;
2150                         break;
2151                 }
2152         }
2153         read_unlock(&bond->lock);
2154
2155         return res;
2156 }
2157
2158 /*-------------------------------- Monitoring -------------------------------*/
2159
2160
2161 static int bond_miimon_inspect(struct bonding *bond)
2162 {
2163         int link_state, commit = 0;
2164         struct slave *slave;
2165         bool ignore_updelay;
2166
2167         ignore_updelay = !bond->curr_active_slave ? true : false;
2168
2169         bond_for_each_slave(bond, slave) {
2170                 slave->new_link = BOND_LINK_NOCHANGE;
2171
2172                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2173
2174                 switch (slave->link) {
2175                 case BOND_LINK_UP:
2176                         if (link_state)
2177                                 continue;
2178
2179                         slave->link = BOND_LINK_FAIL;
2180                         slave->delay = bond->params.downdelay;
2181                         if (slave->delay) {
2182                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2183                                         bond->dev->name,
2184                                         (bond->params.mode ==
2185                                          BOND_MODE_ACTIVEBACKUP) ?
2186                                         (bond_is_active_slave(slave) ?
2187                                          "active " : "backup ") : "",
2188                                         slave->dev->name,
2189                                         bond->params.downdelay * bond->params.miimon);
2190                         }
2191                         /*FALLTHRU*/
2192                 case BOND_LINK_FAIL:
2193                         if (link_state) {
2194                                 /*
2195                                  * recovered before downdelay expired
2196                                  */
2197                                 slave->link = BOND_LINK_UP;
2198                                 slave->jiffies = jiffies;
2199                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2200                                         bond->dev->name,
2201                                         (bond->params.downdelay - slave->delay) *
2202                                         bond->params.miimon,
2203                                         slave->dev->name);
2204                                 continue;
2205                         }
2206
2207                         if (slave->delay <= 0) {
2208                                 slave->new_link = BOND_LINK_DOWN;
2209                                 commit++;
2210                                 continue;
2211                         }
2212
2213                         slave->delay--;
2214                         break;
2215
2216                 case BOND_LINK_DOWN:
2217                         if (!link_state)
2218                                 continue;
2219
2220                         slave->link = BOND_LINK_BACK;
2221                         slave->delay = bond->params.updelay;
2222
2223                         if (slave->delay) {
2224                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2225                                         bond->dev->name, slave->dev->name,
2226                                         ignore_updelay ? 0 :
2227                                         bond->params.updelay *
2228                                         bond->params.miimon);
2229                         }
2230                         /*FALLTHRU*/
2231                 case BOND_LINK_BACK:
2232                         if (!link_state) {
2233                                 slave->link = BOND_LINK_DOWN;
2234                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2235                                         bond->dev->name,
2236                                         (bond->params.updelay - slave->delay) *
2237                                         bond->params.miimon,
2238                                         slave->dev->name);
2239
2240                                 continue;
2241                         }
2242
2243                         if (ignore_updelay)
2244                                 slave->delay = 0;
2245
2246                         if (slave->delay <= 0) {
2247                                 slave->new_link = BOND_LINK_UP;
2248                                 commit++;
2249                                 ignore_updelay = false;
2250                                 continue;
2251                         }
2252
2253                         slave->delay--;
2254                         break;
2255                 }
2256         }
2257
2258         return commit;
2259 }
2260
2261 static void bond_miimon_commit(struct bonding *bond)
2262 {
2263         struct slave *slave;
2264
2265         bond_for_each_slave(bond, slave) {
2266                 switch (slave->new_link) {
2267                 case BOND_LINK_NOCHANGE:
2268                         continue;
2269
2270                 case BOND_LINK_UP:
2271                         slave->link = BOND_LINK_UP;
2272                         slave->jiffies = jiffies;
2273
2274                         if (bond->params.mode == BOND_MODE_8023AD) {
2275                                 /* prevent it from being the active one */
2276                                 bond_set_backup_slave(slave);
2277                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2278                                 /* make it immediately active */
2279                                 bond_set_active_slave(slave);
2280                         } else if (slave != bond->primary_slave) {
2281                                 /* prevent it from being the active one */
2282                                 bond_set_backup_slave(slave);
2283                         }
2284
2285                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2286                                 bond->dev->name, slave->dev->name,
2287                                 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2288                                 slave->duplex ? "full" : "half");
2289
2290                         /* notify ad that the link status has changed */
2291                         if (bond->params.mode == BOND_MODE_8023AD)
2292                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2293
2294                         if (bond_is_lb(bond))
2295                                 bond_alb_handle_link_change(bond, slave,
2296                                                             BOND_LINK_UP);
2297
2298                         if (!bond->curr_active_slave ||
2299                             (slave == bond->primary_slave))
2300                                 goto do_failover;
2301
2302                         continue;
2303
2304                 case BOND_LINK_DOWN:
2305                         if (slave->link_failure_count < UINT_MAX)
2306                                 slave->link_failure_count++;
2307
2308                         slave->link = BOND_LINK_DOWN;
2309
2310                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2311                             bond->params.mode == BOND_MODE_8023AD)
2312                                 bond_set_slave_inactive_flags(slave);
2313
2314                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2315                                 bond->dev->name, slave->dev->name);
2316
2317                         if (bond->params.mode == BOND_MODE_8023AD)
2318                                 bond_3ad_handle_link_change(slave,
2319                                                             BOND_LINK_DOWN);
2320
2321                         if (bond_is_lb(bond))
2322                                 bond_alb_handle_link_change(bond, slave,
2323                                                             BOND_LINK_DOWN);
2324
2325                         if (slave == bond->curr_active_slave)
2326                                 goto do_failover;
2327
2328                         continue;
2329
2330                 default:
2331                         pr_err("%s: invalid new link %d on slave %s\n",
2332                                bond->dev->name, slave->new_link,
2333                                slave->dev->name);
2334                         slave->new_link = BOND_LINK_NOCHANGE;
2335
2336                         continue;
2337                 }
2338
2339 do_failover:
2340                 ASSERT_RTNL();
2341                 block_netpoll_tx();
2342                 write_lock_bh(&bond->curr_slave_lock);
2343                 bond_select_active_slave(bond);
2344                 write_unlock_bh(&bond->curr_slave_lock);
2345                 unblock_netpoll_tx();
2346         }
2347
2348         bond_set_carrier(bond);
2349 }
2350
2351 /*
2352  * bond_mii_monitor
2353  *
2354  * Really a wrapper that splits the mii monitor into two phases: an
2355  * inspection, then (if inspection indicates something needs to be done)
2356  * an acquisition of appropriate locks followed by a commit phase to
2357  * implement whatever link state changes are indicated.
2358  */
2359 void bond_mii_monitor(struct work_struct *work)
2360 {
2361         struct bonding *bond = container_of(work, struct bonding,
2362                                             mii_work.work);
2363         bool should_notify_peers = false;
2364         unsigned long delay;
2365
2366         read_lock(&bond->lock);
2367
2368         delay = msecs_to_jiffies(bond->params.miimon);
2369
2370         if (list_empty(&bond->slave_list))
2371                 goto re_arm;
2372
2373         should_notify_peers = bond_should_notify_peers(bond);
2374
2375         if (bond_miimon_inspect(bond)) {
2376                 read_unlock(&bond->lock);
2377
2378                 /* Race avoidance with bond_close cancel of workqueue */
2379                 if (!rtnl_trylock()) {
2380                         read_lock(&bond->lock);
2381                         delay = 1;
2382                         should_notify_peers = false;
2383                         goto re_arm;
2384                 }
2385
2386                 read_lock(&bond->lock);
2387
2388                 bond_miimon_commit(bond);
2389
2390                 read_unlock(&bond->lock);
2391                 rtnl_unlock();  /* might sleep, hold no other locks */
2392                 read_lock(&bond->lock);
2393         }
2394
2395 re_arm:
2396         if (bond->params.miimon)
2397                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2398
2399         read_unlock(&bond->lock);
2400
2401         if (should_notify_peers) {
2402                 if (!rtnl_trylock()) {
2403                         read_lock(&bond->lock);
2404                         bond->send_peer_notif++;
2405                         read_unlock(&bond->lock);
2406                         return;
2407                 }
2408                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2409                 rtnl_unlock();
2410         }
2411 }
2412
2413 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2414 {
2415         struct vlan_entry *vlan;
2416         struct net_device *vlan_dev;
2417
2418         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2419                 return 1;
2420
2421         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2422                 rcu_read_lock();
2423                 vlan_dev = __vlan_find_dev_deep(bond->dev, htons(ETH_P_8021Q),
2424                                                 vlan->vlan_id);
2425                 rcu_read_unlock();
2426                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2427                         return 1;
2428         }
2429
2430         return 0;
2431 }
2432
2433 /*
2434  * We go to the (large) trouble of VLAN tagging ARP frames because
2435  * switches in VLAN mode (especially if ports are configured as
2436  * "native" to a VLAN) might not pass non-tagged frames.
2437  */
2438 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2439 {
2440         struct sk_buff *skb;
2441
2442         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2443                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2444
2445         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2446                          NULL, slave_dev->dev_addr, NULL);
2447
2448         if (!skb) {
2449                 pr_err("ARP packet allocation failed\n");
2450                 return;
2451         }
2452         if (vlan_id) {
2453                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2454                 if (!skb) {
2455                         pr_err("failed to insert VLAN tag\n");
2456                         return;
2457                 }
2458         }
2459         arp_xmit(skb);
2460 }
2461
2462
2463 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2464 {
2465         int i, vlan_id;
2466         __be32 *targets = bond->params.arp_targets;
2467         struct vlan_entry *vlan;
2468         struct net_device *vlan_dev = NULL;
2469         struct rtable *rt;
2470
2471         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2472                 __be32 addr;
2473                 if (!targets[i])
2474                         break;
2475                 pr_debug("basa: target %pI4\n", &targets[i]);
2476                 if (!bond_vlan_used(bond)) {
2477                         pr_debug("basa: empty vlan: arp_send\n");
2478                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2479                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2480                                       addr, 0);
2481                         continue;
2482                 }
2483
2484                 /*
2485                  * If VLANs are configured, we do a route lookup to
2486                  * determine which VLAN interface would be used, so we
2487                  * can tag the ARP with the proper VLAN tag.
2488                  */
2489                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2490                                      RTO_ONLINK, 0);
2491                 if (IS_ERR(rt)) {
2492                         if (net_ratelimit()) {
2493                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2494                                            bond->dev->name, &targets[i]);
2495                         }
2496                         continue;
2497                 }
2498
2499                 /*
2500                  * This target is not on a VLAN
2501                  */
2502                 if (rt->dst.dev == bond->dev) {
2503                         ip_rt_put(rt);
2504                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2505                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2506                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2507                                       addr, 0);
2508                         continue;
2509                 }
2510
2511                 vlan_id = 0;
2512                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2513                         rcu_read_lock();
2514                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2515                                                         htons(ETH_P_8021Q),
2516                                                         vlan->vlan_id);
2517                         rcu_read_unlock();
2518                         if (vlan_dev == rt->dst.dev) {
2519                                 vlan_id = vlan->vlan_id;
2520                                 pr_debug("basa: vlan match on %s %d\n",
2521                                        vlan_dev->name, vlan_id);
2522                                 break;
2523                         }
2524                 }
2525
2526                 if (vlan_id && vlan_dev) {
2527                         ip_rt_put(rt);
2528                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2529                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2530                                       addr, vlan_id);
2531                         continue;
2532                 }
2533
2534                 if (net_ratelimit()) {
2535                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2536                                    bond->dev->name, &targets[i],
2537                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2538                 }
2539                 ip_rt_put(rt);
2540         }
2541 }
2542
2543 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2544 {
2545         int i;
2546
2547         if (!sip || !bond_has_this_ip(bond, tip)) {
2548                 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2549                 return;
2550         }
2551
2552         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2553         if (i == -1) {
2554                 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2555                 return;
2556         }
2557         slave->last_arp_rx = jiffies;
2558         slave->target_last_arp_rx[i] = jiffies;
2559 }
2560
2561 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2562                         struct slave *slave)
2563 {
2564         struct arphdr *arp = (struct arphdr *)skb->data;
2565         unsigned char *arp_ptr;
2566         __be32 sip, tip;
2567         int alen;
2568
2569         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2570                 return RX_HANDLER_ANOTHER;
2571
2572         read_lock(&bond->lock);
2573
2574         if (!slave_do_arp_validate(bond, slave))
2575                 goto out_unlock;
2576
2577         alen = arp_hdr_len(bond->dev);
2578
2579         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2580                  bond->dev->name, skb->dev->name);
2581
2582         if (alen > skb_headlen(skb)) {
2583                 arp = kmalloc(alen, GFP_ATOMIC);
2584                 if (!arp)
2585                         goto out_unlock;
2586                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2587                         goto out_unlock;
2588         }
2589
2590         if (arp->ar_hln != bond->dev->addr_len ||
2591             skb->pkt_type == PACKET_OTHERHOST ||
2592             skb->pkt_type == PACKET_LOOPBACK ||
2593             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2594             arp->ar_pro != htons(ETH_P_IP) ||
2595             arp->ar_pln != 4)
2596                 goto out_unlock;
2597
2598         arp_ptr = (unsigned char *)(arp + 1);
2599         arp_ptr += bond->dev->addr_len;
2600         memcpy(&sip, arp_ptr, 4);
2601         arp_ptr += 4 + bond->dev->addr_len;
2602         memcpy(&tip, arp_ptr, 4);
2603
2604         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2605                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2606                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2607                  &sip, &tip);
2608
2609         /*
2610          * Backup slaves won't see the ARP reply, but do come through
2611          * here for each ARP probe (so we swap the sip/tip to validate
2612          * the probe).  In a "redundant switch, common router" type of
2613          * configuration, the ARP probe will (hopefully) travel from
2614          * the active, through one switch, the router, then the other
2615          * switch before reaching the backup.
2616          *
2617          * We 'trust' the arp requests if there is an active slave and
2618          * it received valid arp reply(s) after it became active. This
2619          * is done to avoid endless looping when we can't reach the
2620          * arp_ip_target and fool ourselves with our own arp requests.
2621          */
2622         if (bond_is_active_slave(slave))
2623                 bond_validate_arp(bond, slave, sip, tip);
2624         else if (bond->curr_active_slave &&
2625                  time_after(slave_last_rx(bond, bond->curr_active_slave),
2626                             bond->curr_active_slave->jiffies))
2627                 bond_validate_arp(bond, slave, tip, sip);
2628
2629 out_unlock:
2630         read_unlock(&bond->lock);
2631         if (arp != (struct arphdr *)skb->data)
2632                 kfree(arp);
2633         return RX_HANDLER_ANOTHER;
2634 }
2635
2636 /*
2637  * this function is called regularly to monitor each slave's link
2638  * ensuring that traffic is being sent and received when arp monitoring
2639  * is used in load-balancing mode. if the adapter has been dormant, then an
2640  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2641  * arp monitoring in active backup mode.
2642  */
2643 void bond_loadbalance_arp_mon(struct work_struct *work)
2644 {
2645         struct bonding *bond = container_of(work, struct bonding,
2646                                             arp_work.work);
2647         struct slave *slave, *oldcurrent;
2648         int do_failover = 0;
2649         int delta_in_ticks, extra_ticks;
2650
2651         read_lock(&bond->lock);
2652
2653         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2654         extra_ticks = delta_in_ticks / 2;
2655
2656         if (list_empty(&bond->slave_list))
2657                 goto re_arm;
2658
2659         oldcurrent = bond->curr_active_slave;
2660         /* see if any of the previous devices are up now (i.e. they have
2661          * xmt and rcv traffic). the curr_active_slave does not come into
2662          * the picture unless it is null. also, slave->jiffies is not needed
2663          * here because we send an arp on each slave and give a slave as
2664          * long as it needs to get the tx/rx within the delta.
2665          * TODO: what about up/down delay in arp mode? it wasn't here before
2666          *       so it can wait
2667          */
2668         bond_for_each_slave(bond, slave) {
2669                 unsigned long trans_start = dev_trans_start(slave->dev);
2670
2671                 if (slave->link != BOND_LINK_UP) {
2672                         if (time_in_range(jiffies,
2673                                 trans_start - delta_in_ticks,
2674                                 trans_start + delta_in_ticks + extra_ticks) &&
2675                             time_in_range(jiffies,
2676                                 slave->dev->last_rx - delta_in_ticks,
2677                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2678
2679                                 slave->link  = BOND_LINK_UP;
2680                                 bond_set_active_slave(slave);
2681
2682                                 /* primary_slave has no meaning in round-robin
2683                                  * mode. the window of a slave being up and
2684                                  * curr_active_slave being null after enslaving
2685                                  * is closed.
2686                                  */
2687                                 if (!oldcurrent) {
2688                                         pr_info("%s: link status definitely up for interface %s, ",
2689                                                 bond->dev->name,
2690                                                 slave->dev->name);
2691                                         do_failover = 1;
2692                                 } else {
2693                                         pr_info("%s: interface %s is now up\n",
2694                                                 bond->dev->name,
2695                                                 slave->dev->name);
2696                                 }
2697                         }
2698                 } else {
2699                         /* slave->link == BOND_LINK_UP */
2700
2701                         /* not all switches will respond to an arp request
2702                          * when the source ip is 0, so don't take the link down
2703                          * if we don't know our ip yet
2704                          */
2705                         if (!time_in_range(jiffies,
2706                                 trans_start - delta_in_ticks,
2707                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2708                             !time_in_range(jiffies,
2709                                 slave->dev->last_rx - delta_in_ticks,
2710                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2711
2712                                 slave->link  = BOND_LINK_DOWN;
2713                                 bond_set_backup_slave(slave);
2714
2715                                 if (slave->link_failure_count < UINT_MAX)
2716                                         slave->link_failure_count++;
2717
2718                                 pr_info("%s: interface %s is now down.\n",
2719                                         bond->dev->name,
2720                                         slave->dev->name);
2721
2722                                 if (slave == oldcurrent)
2723                                         do_failover = 1;
2724                         }
2725                 }
2726
2727                 /* note: if switch is in round-robin mode, all links
2728                  * must tx arp to ensure all links rx an arp - otherwise
2729                  * links may oscillate or not come up at all; if switch is
2730                  * in something like xor mode, there is nothing we can
2731                  * do - all replies will be rx'ed on same link causing slaves
2732                  * to be unstable during low/no traffic periods
2733                  */
2734                 if (IS_UP(slave->dev))
2735                         bond_arp_send_all(bond, slave);
2736         }
2737
2738         if (do_failover) {
2739                 block_netpoll_tx();
2740                 write_lock_bh(&bond->curr_slave_lock);
2741
2742                 bond_select_active_slave(bond);
2743
2744                 write_unlock_bh(&bond->curr_slave_lock);
2745                 unblock_netpoll_tx();
2746         }
2747
2748 re_arm:
2749         if (bond->params.arp_interval)
2750                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2751
2752         read_unlock(&bond->lock);
2753 }
2754
2755 /*
2756  * Called to inspect slaves for active-backup mode ARP monitor link state
2757  * changes.  Sets new_link in slaves to specify what action should take
2758  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2759  * to link states must be committed.
2760  *
2761  * Called with bond->lock held for read.
2762  */
2763 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2764 {
2765         unsigned long trans_start;
2766         struct slave *slave;
2767         int extra_ticks;
2768         int commit = 0;
2769
2770         /* All the time comparisons below need some extra time. Otherwise, on
2771          * fast networks the ARP probe/reply may arrive within the same jiffy
2772          * as it was sent.  Then, the next time the ARP monitor is run, one
2773          * arp_interval will already have passed in the comparisons.
2774          */
2775         extra_ticks = delta_in_ticks / 2;
2776
2777         bond_for_each_slave(bond, slave) {
2778                 slave->new_link = BOND_LINK_NOCHANGE;
2779
2780                 if (slave->link != BOND_LINK_UP) {
2781                         if (time_in_range(jiffies,
2782                                 slave_last_rx(bond, slave) - delta_in_ticks,
2783                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2784
2785                                 slave->new_link = BOND_LINK_UP;
2786                                 commit++;
2787                         }
2788
2789                         continue;
2790                 }
2791
2792                 /*
2793                  * Give slaves 2*delta after being enslaved or made
2794                  * active.  This avoids bouncing, as the last receive
2795                  * times need a full ARP monitor cycle to be updated.
2796                  */
2797                 if (time_in_range(jiffies,
2798                                   slave->jiffies - delta_in_ticks,
2799                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2800                         continue;
2801
2802                 /*
2803                  * Backup slave is down if:
2804                  * - No current_arp_slave AND
2805                  * - more than 3*delta since last receive AND
2806                  * - the bond has an IP address
2807                  *
2808                  * Note: a non-null current_arp_slave indicates
2809                  * the curr_active_slave went down and we are
2810                  * searching for a new one; under this condition
2811                  * we only take the curr_active_slave down - this
2812                  * gives each slave a chance to tx/rx traffic
2813                  * before being taken out
2814                  */
2815                 if (!bond_is_active_slave(slave) &&
2816                     !bond->current_arp_slave &&
2817                     !time_in_range(jiffies,
2818                         slave_last_rx(bond, slave) - delta_in_ticks,
2819                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2820
2821                         slave->new_link = BOND_LINK_DOWN;
2822                         commit++;
2823                 }
2824
2825                 /*
2826                  * Active slave is down if:
2827                  * - more than 2*delta since transmitting OR
2828                  * - (more than 2*delta since receive AND
2829                  *    the bond has an IP address)
2830                  */
2831                 trans_start = dev_trans_start(slave->dev);
2832                 if (bond_is_active_slave(slave) &&
2833                     (!time_in_range(jiffies,
2834                         trans_start - delta_in_ticks,
2835                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
2836                      !time_in_range(jiffies,
2837                         slave_last_rx(bond, slave) - delta_in_ticks,
2838                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2839
2840                         slave->new_link = BOND_LINK_DOWN;
2841                         commit++;
2842                 }
2843         }
2844
2845         return commit;
2846 }
2847
2848 /*
2849  * Called to commit link state changes noted by inspection step of
2850  * active-backup mode ARP monitor.
2851  *
2852  * Called with RTNL and bond->lock for read.
2853  */
2854 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2855 {
2856         unsigned long trans_start;
2857         struct slave *slave;
2858
2859         bond_for_each_slave(bond, slave) {
2860                 switch (slave->new_link) {
2861                 case BOND_LINK_NOCHANGE:
2862                         continue;
2863
2864                 case BOND_LINK_UP:
2865                         trans_start = dev_trans_start(slave->dev);
2866                         if ((!bond->curr_active_slave &&
2867                              time_in_range(jiffies,
2868                                            trans_start - delta_in_ticks,
2869                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2870                             bond->curr_active_slave != slave) {
2871                                 slave->link = BOND_LINK_UP;
2872                                 if (bond->current_arp_slave) {
2873                                         bond_set_slave_inactive_flags(
2874                                                 bond->current_arp_slave);
2875                                         bond->current_arp_slave = NULL;
2876                                 }
2877
2878                                 pr_info("%s: link status definitely up for interface %s.\n",
2879                                         bond->dev->name, slave->dev->name);
2880
2881                                 if (!bond->curr_active_slave ||
2882                                     (slave == bond->primary_slave))
2883                                         goto do_failover;
2884
2885                         }
2886
2887                         continue;
2888
2889                 case BOND_LINK_DOWN:
2890                         if (slave->link_failure_count < UINT_MAX)
2891                                 slave->link_failure_count++;
2892
2893                         slave->link = BOND_LINK_DOWN;
2894                         bond_set_slave_inactive_flags(slave);
2895
2896                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2897                                 bond->dev->name, slave->dev->name);
2898
2899                         if (slave == bond->curr_active_slave) {
2900                                 bond->current_arp_slave = NULL;
2901                                 goto do_failover;
2902                         }
2903
2904                         continue;
2905
2906                 default:
2907                         pr_err("%s: impossible: new_link %d on slave %s\n",
2908                                bond->dev->name, slave->new_link,
2909                                slave->dev->name);
2910                         continue;
2911                 }
2912
2913 do_failover:
2914                 ASSERT_RTNL();
2915                 block_netpoll_tx();
2916                 write_lock_bh(&bond->curr_slave_lock);
2917                 bond_select_active_slave(bond);
2918                 write_unlock_bh(&bond->curr_slave_lock);
2919                 unblock_netpoll_tx();
2920         }
2921
2922         bond_set_carrier(bond);
2923 }
2924
2925 /*
2926  * Send ARP probes for active-backup mode ARP monitor.
2927  *
2928  * Called with bond->lock held for read.
2929  */
2930 static void bond_ab_arp_probe(struct bonding *bond)
2931 {
2932         struct slave *slave, *next_slave;
2933         int i;
2934
2935         read_lock(&bond->curr_slave_lock);
2936
2937         if (bond->current_arp_slave && bond->curr_active_slave)
2938                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2939                         bond->current_arp_slave->dev->name,
2940                         bond->curr_active_slave->dev->name);
2941
2942         if (bond->curr_active_slave) {
2943                 bond_arp_send_all(bond, bond->curr_active_slave);
2944                 read_unlock(&bond->curr_slave_lock);
2945                 return;
2946         }
2947
2948         read_unlock(&bond->curr_slave_lock);
2949
2950         /* if we don't have a curr_active_slave, search for the next available
2951          * backup slave from the current_arp_slave and make it the candidate
2952          * for becoming the curr_active_slave
2953          */
2954
2955         if (!bond->current_arp_slave) {
2956                 bond->current_arp_slave = bond_first_slave(bond);
2957                 if (!bond->current_arp_slave)
2958                         return;
2959         }
2960
2961         bond_set_slave_inactive_flags(bond->current_arp_slave);
2962
2963         /* search for next candidate */
2964         next_slave = bond_next_slave(bond, bond->current_arp_slave);
2965         bond_for_each_slave_from(bond, slave, i, next_slave) {
2966                 if (IS_UP(slave->dev)) {
2967                         slave->link = BOND_LINK_BACK;
2968                         bond_set_slave_active_flags(slave);
2969                         bond_arp_send_all(bond, slave);
2970                         slave->jiffies = jiffies;
2971                         bond->current_arp_slave = slave;
2972                         break;
2973                 }
2974
2975                 /* if the link state is up at this point, we
2976                  * mark it down - this can happen if we have
2977                  * simultaneous link failures and
2978                  * reselect_active_interface doesn't make this
2979                  * one the current slave so it is still marked
2980                  * up when it is actually down
2981                  */
2982                 if (slave->link == BOND_LINK_UP) {
2983                         slave->link = BOND_LINK_DOWN;
2984                         if (slave->link_failure_count < UINT_MAX)
2985                                 slave->link_failure_count++;
2986
2987                         bond_set_slave_inactive_flags(slave);
2988
2989                         pr_info("%s: backup interface %s is now down.\n",
2990                                 bond->dev->name, slave->dev->name);
2991                 }
2992         }
2993 }
2994
2995 void bond_activebackup_arp_mon(struct work_struct *work)
2996 {
2997         struct bonding *bond = container_of(work, struct bonding,
2998                                             arp_work.work);
2999         bool should_notify_peers = false;
3000         int delta_in_ticks;
3001
3002         read_lock(&bond->lock);
3003
3004         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3005
3006         if (list_empty(&bond->slave_list))
3007                 goto re_arm;
3008
3009         should_notify_peers = bond_should_notify_peers(bond);
3010
3011         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3012                 read_unlock(&bond->lock);
3013
3014                 /* Race avoidance with bond_close flush of workqueue */
3015                 if (!rtnl_trylock()) {
3016                         read_lock(&bond->lock);
3017                         delta_in_ticks = 1;
3018                         should_notify_peers = false;
3019                         goto re_arm;
3020                 }
3021
3022                 read_lock(&bond->lock);
3023
3024                 bond_ab_arp_commit(bond, delta_in_ticks);
3025
3026                 read_unlock(&bond->lock);
3027                 rtnl_unlock();
3028                 read_lock(&bond->lock);
3029         }
3030
3031         bond_ab_arp_probe(bond);
3032
3033 re_arm:
3034         if (bond->params.arp_interval)
3035                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3036
3037         read_unlock(&bond->lock);
3038
3039         if (should_notify_peers) {
3040                 if (!rtnl_trylock()) {
3041                         read_lock(&bond->lock);
3042                         bond->send_peer_notif++;
3043                         read_unlock(&bond->lock);
3044                         return;
3045                 }
3046                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3047                 rtnl_unlock();
3048         }
3049 }
3050
3051 /*-------------------------- netdev event handling --------------------------*/
3052
3053 /*
3054  * Change device name
3055  */
3056 static int bond_event_changename(struct bonding *bond)
3057 {
3058         bond_remove_proc_entry(bond);
3059         bond_create_proc_entry(bond);
3060
3061         bond_debug_reregister(bond);
3062
3063         return NOTIFY_DONE;
3064 }
3065
3066 static int bond_master_netdev_event(unsigned long event,
3067                                     struct net_device *bond_dev)
3068 {
3069         struct bonding *event_bond = netdev_priv(bond_dev);
3070
3071         switch (event) {
3072         case NETDEV_CHANGENAME:
3073                 return bond_event_changename(event_bond);
3074         case NETDEV_UNREGISTER:
3075                 bond_remove_proc_entry(event_bond);
3076                 break;
3077         case NETDEV_REGISTER:
3078                 bond_create_proc_entry(event_bond);
3079                 break;
3080         default:
3081                 break;
3082         }
3083
3084         return NOTIFY_DONE;
3085 }
3086
3087 static int bond_slave_netdev_event(unsigned long event,
3088                                    struct net_device *slave_dev)
3089 {
3090         struct slave *slave = bond_slave_get_rtnl(slave_dev);
3091         struct bonding *bond;
3092         struct net_device *bond_dev;
3093         u32 old_speed;
3094         u8 old_duplex;
3095
3096         /* A netdev event can be generated while enslaving a device
3097          * before netdev_rx_handler_register is called in which case
3098          * slave will be NULL
3099          */
3100         if (!slave)
3101                 return NOTIFY_DONE;
3102         bond_dev = slave->bond->dev;
3103         bond = slave->bond;
3104
3105         switch (event) {
3106         case NETDEV_UNREGISTER:
3107                 if (bond_dev->type != ARPHRD_ETHER)
3108                         bond_release_and_destroy(bond_dev, slave_dev);
3109                 else
3110                         bond_release(bond_dev, slave_dev);
3111                 break;
3112         case NETDEV_UP:
3113         case NETDEV_CHANGE:
3114                 old_speed = slave->speed;
3115                 old_duplex = slave->duplex;
3116
3117                 bond_update_speed_duplex(slave);
3118
3119                 if (bond->params.mode == BOND_MODE_8023AD) {
3120                         if (old_speed != slave->speed)
3121                                 bond_3ad_adapter_speed_changed(slave);
3122                         if (old_duplex != slave->duplex)
3123                                 bond_3ad_adapter_duplex_changed(slave);
3124                 }
3125                 break;
3126         case NETDEV_DOWN:
3127                 /*
3128                  * ... Or is it this?
3129                  */
3130                 break;
3131         case NETDEV_CHANGEMTU:
3132                 /*
3133                  * TODO: Should slaves be allowed to
3134                  * independently alter their MTU?  For
3135                  * an active-backup bond, slaves need
3136                  * not be the same type of device, so
3137                  * MTUs may vary.  For other modes,
3138                  * slaves arguably should have the
3139                  * same MTUs. To do this, we'd need to
3140                  * take over the slave's change_mtu
3141                  * function for the duration of their
3142                  * servitude.
3143                  */
3144                 break;
3145         case NETDEV_CHANGENAME:
3146                 /*
3147                  * TODO: handle changing the primary's name
3148                  */
3149                 break;
3150         case NETDEV_FEAT_CHANGE:
3151                 bond_compute_features(bond);
3152                 break;
3153         case NETDEV_RESEND_IGMP:
3154                 /* Propagate to master device */
3155                 call_netdevice_notifiers(event, slave->bond->dev);
3156                 break;
3157         default:
3158                 break;
3159         }
3160
3161         return NOTIFY_DONE;
3162 }
3163
3164 /*
3165  * bond_netdev_event: handle netdev notifier chain events.
3166  *
3167  * This function receives events for the netdev chain.  The caller (an
3168  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3169  * locks for us to safely manipulate the slave devices (RTNL lock,
3170  * dev_probe_lock).
3171  */
3172 static int bond_netdev_event(struct notifier_block *this,
3173                              unsigned long event, void *ptr)
3174 {
3175         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3176
3177         pr_debug("event_dev: %s, event: %lx\n",
3178                  event_dev ? event_dev->name : "None",
3179                  event);
3180
3181         if (!(event_dev->priv_flags & IFF_BONDING))
3182                 return NOTIFY_DONE;
3183
3184         if (event_dev->flags & IFF_MASTER) {
3185                 pr_debug("IFF_MASTER\n");
3186                 return bond_master_netdev_event(event, event_dev);
3187         }
3188
3189         if (event_dev->flags & IFF_SLAVE) {
3190                 pr_debug("IFF_SLAVE\n");
3191                 return bond_slave_netdev_event(event, event_dev);
3192         }
3193
3194         return NOTIFY_DONE;
3195 }
3196
3197 static struct notifier_block bond_netdev_notifier = {
3198         .notifier_call = bond_netdev_event,
3199 };
3200
3201 /*---------------------------- Hashing Policies -----------------------------*/
3202
3203 /*
3204  * Hash for the output device based upon layer 2 data
3205  */
3206 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3207 {
3208         struct ethhdr *data = (struct ethhdr *)skb->data;
3209
3210         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3211                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3212
3213         return 0;
3214 }
3215
3216 /*
3217  * Hash for the output device based upon layer 2 and layer 3 data. If
3218  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3219  */
3220 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3221 {
3222         const struct ethhdr *data;
3223         const struct iphdr *iph;
3224         const struct ipv6hdr *ipv6h;
3225         u32 v6hash;
3226         const __be32 *s, *d;
3227
3228         if (skb->protocol == htons(ETH_P_IP) &&
3229             pskb_network_may_pull(skb, sizeof(*iph))) {
3230                 iph = ip_hdr(skb);
3231                 data = (struct ethhdr *)skb->data;
3232                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3233                         (data->h_dest[5] ^ data->h_source[5])) % count;
3234         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3235                    pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3236                 ipv6h = ipv6_hdr(skb);
3237                 data = (struct ethhdr *)skb->data;
3238                 s = &ipv6h->saddr.s6_addr32[0];
3239                 d = &ipv6h->daddr.s6_addr32[0];
3240                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3241                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3242                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3243         }
3244
3245         return bond_xmit_hash_policy_l2(skb, count);
3246 }
3247
3248 /*
3249  * Hash for the output device based upon layer 3 and layer 4 data. If
3250  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3251  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3252  */
3253 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3254 {
3255         u32 layer4_xor = 0;
3256         const struct iphdr *iph;
3257         const struct ipv6hdr *ipv6h;
3258         const __be32 *s, *d;
3259         const __be16 *l4 = NULL;
3260         __be16 _l4[2];
3261         int noff = skb_network_offset(skb);
3262         int poff;
3263
3264         if (skb->protocol == htons(ETH_P_IP) &&
3265             pskb_may_pull(skb, noff + sizeof(*iph))) {
3266                 iph = ip_hdr(skb);
3267                 poff = proto_ports_offset(iph->protocol);
3268
3269                 if (!ip_is_fragment(iph) && poff >= 0) {
3270                         l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3271                                                 sizeof(_l4), &_l4);
3272                         if (l4)
3273                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3274                 }
3275                 return (layer4_xor ^
3276                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3277         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3278                    pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3279                 ipv6h = ipv6_hdr(skb);
3280                 poff = proto_ports_offset(ipv6h->nexthdr);
3281                 if (poff >= 0) {
3282                         l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3283                                                 sizeof(_l4), &_l4);
3284                         if (l4)
3285                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3286                 }
3287                 s = &ipv6h->saddr.s6_addr32[0];
3288                 d = &ipv6h->daddr.s6_addr32[0];
3289                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3290                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3291                                (layer4_xor >> 8);
3292                 return layer4_xor % count;
3293         }
3294
3295         return bond_xmit_hash_policy_l2(skb, count);
3296 }
3297
3298 /*-------------------------- Device entry points ----------------------------*/
3299
3300 static void bond_work_init_all(struct bonding *bond)
3301 {
3302         INIT_DELAYED_WORK(&bond->mcast_work,
3303                           bond_resend_igmp_join_requests_delayed);
3304         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3305         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3306         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3307                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3308         else
3309                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3310         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3311 }
3312
3313 static void bond_work_cancel_all(struct bonding *bond)
3314 {
3315         cancel_delayed_work_sync(&bond->mii_work);
3316         cancel_delayed_work_sync(&bond->arp_work);
3317         cancel_delayed_work_sync(&bond->alb_work);
3318         cancel_delayed_work_sync(&bond->ad_work);
3319         cancel_delayed_work_sync(&bond->mcast_work);
3320 }
3321
3322 static int bond_open(struct net_device *bond_dev)
3323 {
3324         struct bonding *bond = netdev_priv(bond_dev);
3325         struct slave *slave;
3326
3327         /* reset slave->backup and slave->inactive */
3328         read_lock(&bond->lock);
3329         if (!list_empty(&bond->slave_list)) {
3330                 read_lock(&bond->curr_slave_lock);
3331                 bond_for_each_slave(bond, slave) {
3332                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3333                                 && (slave != bond->curr_active_slave)) {
3334                                 bond_set_slave_inactive_flags(slave);
3335                         } else {
3336                                 bond_set_slave_active_flags(slave);
3337                         }
3338                 }
3339                 read_unlock(&bond->curr_slave_lock);
3340         }
3341         read_unlock(&bond->lock);
3342
3343         bond_work_init_all(bond);
3344
3345         if (bond_is_lb(bond)) {
3346                 /* bond_alb_initialize must be called before the timer
3347                  * is started.
3348                  */
3349                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3350                         return -ENOMEM;
3351                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3352         }
3353
3354         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3355                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3356
3357         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3358                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3359                 if (bond->params.arp_validate)
3360                         bond->recv_probe = bond_arp_rcv;
3361         }
3362
3363         if (bond->params.mode == BOND_MODE_8023AD) {
3364                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3365                 /* register to receive LACPDUs */
3366                 bond->recv_probe = bond_3ad_lacpdu_recv;
3367                 bond_3ad_initiate_agg_selection(bond, 1);
3368         }
3369
3370         return 0;
3371 }
3372
3373 static int bond_close(struct net_device *bond_dev)
3374 {
3375         struct bonding *bond = netdev_priv(bond_dev);
3376
3377         write_lock_bh(&bond->lock);
3378         bond->send_peer_notif = 0;
3379         write_unlock_bh(&bond->lock);
3380
3381         bond_work_cancel_all(bond);
3382         if (bond_is_lb(bond)) {
3383                 /* Must be called only after all
3384                  * slaves have been released
3385                  */
3386                 bond_alb_deinitialize(bond);
3387         }
3388         bond->recv_probe = NULL;
3389
3390         return 0;
3391 }
3392
3393 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3394                                                 struct rtnl_link_stats64 *stats)
3395 {
3396         struct bonding *bond = netdev_priv(bond_dev);
3397         struct rtnl_link_stats64 temp;
3398         struct slave *slave;
3399
3400         memset(stats, 0, sizeof(*stats));
3401
3402         read_lock_bh(&bond->lock);
3403         bond_for_each_slave(bond, slave) {
3404                 const struct rtnl_link_stats64 *sstats =
3405                         dev_get_stats(slave->dev, &temp);
3406
3407                 stats->rx_packets += sstats->rx_packets;
3408                 stats->rx_bytes += sstats->rx_bytes;
3409                 stats->rx_errors += sstats->rx_errors;
3410                 stats->rx_dropped += sstats->rx_dropped;
3411
3412                 stats->tx_packets += sstats->tx_packets;
3413                 stats->tx_bytes += sstats->tx_bytes;
3414                 stats->tx_errors += sstats->tx_errors;
3415                 stats->tx_dropped += sstats->tx_dropped;
3416
3417                 stats->multicast += sstats->multicast;
3418                 stats->collisions += sstats->collisions;
3419
3420                 stats->rx_length_errors += sstats->rx_length_errors;
3421                 stats->rx_over_errors += sstats->rx_over_errors;
3422                 stats->rx_crc_errors += sstats->rx_crc_errors;
3423                 stats->rx_frame_errors += sstats->rx_frame_errors;
3424                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3425                 stats->rx_missed_errors += sstats->rx_missed_errors;
3426
3427                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3428                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3429                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3430                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3431                 stats->tx_window_errors += sstats->tx_window_errors;
3432         }
3433         read_unlock_bh(&bond->lock);
3434
3435         return stats;
3436 }
3437
3438 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3439 {
3440         struct net_device *slave_dev = NULL;
3441         struct ifbond k_binfo;
3442         struct ifbond __user *u_binfo = NULL;
3443         struct ifslave k_sinfo;
3444         struct ifslave __user *u_sinfo = NULL;
3445         struct mii_ioctl_data *mii = NULL;
3446         struct net *net;
3447         int res = 0;
3448
3449         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3450
3451         switch (cmd) {
3452         case SIOCGMIIPHY:
3453                 mii = if_mii(ifr);
3454                 if (!mii)
3455                         return -EINVAL;
3456
3457                 mii->phy_id = 0;
3458                 /* Fall Through */
3459         case SIOCGMIIREG:
3460                 /*
3461                  * We do this again just in case we were called by SIOCGMIIREG
3462                  * instead of SIOCGMIIPHY.
3463                  */
3464                 mii = if_mii(ifr);
3465                 if (!mii)
3466                         return -EINVAL;
3467
3468
3469                 if (mii->reg_num == 1) {
3470                         struct bonding *bond = netdev_priv(bond_dev);
3471                         mii->val_out = 0;
3472                         read_lock(&bond->lock);
3473                         read_lock(&bond->curr_slave_lock);
3474                         if (netif_carrier_ok(bond->dev))
3475                                 mii->val_out = BMSR_LSTATUS;
3476
3477                         read_unlock(&bond->curr_slave_lock);
3478                         read_unlock(&bond->lock);
3479                 }
3480
3481                 return 0;
3482         case BOND_INFO_QUERY_OLD:
3483         case SIOCBONDINFOQUERY:
3484                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3485
3486                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3487                         return -EFAULT;
3488
3489                 res = bond_info_query(bond_dev, &k_binfo);
3490                 if (res == 0 &&
3491                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3492                         return -EFAULT;
3493
3494                 return res;
3495         case BOND_SLAVE_INFO_QUERY_OLD:
3496         case SIOCBONDSLAVEINFOQUERY:
3497                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3498
3499                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3500                         return -EFAULT;
3501
3502                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3503                 if (res == 0 &&
3504                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3505                         return -EFAULT;
3506
3507                 return res;
3508         default:
3509                 /* Go on */
3510                 break;
3511         }
3512
3513         net = dev_net(bond_dev);
3514
3515         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3516                 return -EPERM;
3517
3518         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3519
3520         pr_debug("slave_dev=%p:\n", slave_dev);
3521
3522         if (!slave_dev)
3523                 res = -ENODEV;
3524         else {
3525                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3526                 switch (cmd) {
3527                 case BOND_ENSLAVE_OLD:
3528                 case SIOCBONDENSLAVE:
3529                         res = bond_enslave(bond_dev, slave_dev);
3530                         break;
3531                 case BOND_RELEASE_OLD:
3532                 case SIOCBONDRELEASE:
3533                         res = bond_release(bond_dev, slave_dev);
3534                         break;
3535                 case BOND_SETHWADDR_OLD:
3536                 case SIOCBONDSETHWADDR:
3537                         bond_set_dev_addr(bond_dev, slave_dev);
3538                         res = 0;
3539                         break;
3540                 case BOND_CHANGE_ACTIVE_OLD:
3541                 case SIOCBONDCHANGEACTIVE:
3542                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3543                         break;
3544                 default:
3545                         res = -EOPNOTSUPP;
3546                 }
3547
3548                 dev_put(slave_dev);
3549         }
3550
3551         return res;
3552 }
3553
3554 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3555 {
3556         struct bonding *bond = netdev_priv(bond_dev);
3557
3558         if (change & IFF_PROMISC)
3559                 bond_set_promiscuity(bond,
3560                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3561
3562         if (change & IFF_ALLMULTI)
3563                 bond_set_allmulti(bond,
3564                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3565 }
3566
3567 static void bond_set_rx_mode(struct net_device *bond_dev)
3568 {
3569         struct bonding *bond = netdev_priv(bond_dev);
3570         struct slave *slave;
3571
3572         read_lock(&bond->lock);
3573
3574         if (USES_PRIMARY(bond->params.mode)) {
3575                 read_lock(&bond->curr_slave_lock);
3576                 slave = bond->curr_active_slave;
3577                 if (slave) {
3578                         dev_uc_sync(slave->dev, bond_dev);
3579                         dev_mc_sync(slave->dev, bond_dev);
3580                 }
3581                 read_unlock(&bond->curr_slave_lock);
3582         } else {
3583                 bond_for_each_slave(bond, slave) {
3584                         dev_uc_sync_multiple(slave->dev, bond_dev);
3585                         dev_mc_sync_multiple(slave->dev, bond_dev);
3586                 }
3587         }
3588
3589         read_unlock(&bond->lock);
3590 }
3591
3592 static int bond_neigh_init(struct neighbour *n)
3593 {
3594         struct bonding *bond = netdev_priv(n->dev);
3595         const struct net_device_ops *slave_ops;
3596         struct neigh_parms parms;
3597         struct slave *slave;
3598         int ret;
3599
3600         slave = bond_first_slave(bond);
3601         if (!slave)
3602                 return 0;
3603         slave_ops = slave->dev->netdev_ops;
3604         if (!slave_ops->ndo_neigh_setup)
3605                 return 0;
3606
3607         parms.neigh_setup = NULL;
3608         parms.neigh_cleanup = NULL;
3609         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3610         if (ret)
3611                 return ret;
3612
3613         /*
3614          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3615          * after the last slave has been detached.  Assumes that all slaves
3616          * utilize the same neigh_cleanup (true at this writing as only user
3617          * is ipoib).
3618          */
3619         n->parms->neigh_cleanup = parms.neigh_cleanup;
3620
3621         if (!parms.neigh_setup)
3622                 return 0;
3623
3624         return parms.neigh_setup(n);
3625 }
3626
3627 /*
3628  * The bonding ndo_neigh_setup is called at init time beofre any
3629  * slave exists. So we must declare proxy setup function which will
3630  * be used at run time to resolve the actual slave neigh param setup.
3631  */
3632 static int bond_neigh_setup(struct net_device *dev,
3633                             struct neigh_parms *parms)
3634 {
3635         parms->neigh_setup   = bond_neigh_init;
3636
3637         return 0;
3638 }
3639
3640 /*
3641  * Change the MTU of all of a master's slaves to match the master
3642  */
3643 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3644 {
3645         struct bonding *bond = netdev_priv(bond_dev);
3646         struct slave *slave;
3647         int res = 0;
3648
3649         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3650                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3651
3652         /* Can't hold bond->lock with bh disabled here since
3653          * some base drivers panic. On the other hand we can't
3654          * hold bond->lock without bh disabled because we'll
3655          * deadlock. The only solution is to rely on the fact
3656          * that we're under rtnl_lock here, and the slaves
3657          * list won't change. This doesn't solve the problem
3658          * of setting the slave's MTU while it is
3659          * transmitting, but the assumption is that the base
3660          * driver can handle that.
3661          *
3662          * TODO: figure out a way to safely iterate the slaves
3663          * list, but without holding a lock around the actual
3664          * call to the base driver.
3665          */
3666
3667         bond_for_each_slave(bond, slave) {
3668                 pr_debug("s %p s->p %p c_m %p\n",
3669                          slave,
3670                          bond_prev_slave(bond, slave),
3671                          slave->dev->netdev_ops->ndo_change_mtu);
3672
3673                 res = dev_set_mtu(slave->dev, new_mtu);
3674
3675                 if (res) {
3676                         /* If we failed to set the slave's mtu to the new value
3677                          * we must abort the operation even in ACTIVE_BACKUP
3678                          * mode, because if we allow the backup slaves to have
3679                          * different mtu values than the active slave we'll
3680                          * need to change their mtu when doing a failover. That
3681                          * means changing their mtu from timer context, which
3682                          * is probably not a good idea.
3683                          */
3684                         pr_debug("err %d %s\n", res, slave->dev->name);
3685                         goto unwind;
3686                 }
3687         }
3688
3689         bond_dev->mtu = new_mtu;
3690
3691         return 0;
3692
3693 unwind:
3694         /* unwind from head to the slave that failed */
3695         bond_for_each_slave_continue_reverse(bond, slave) {
3696                 int tmp_res;
3697
3698                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3699                 if (tmp_res) {
3700                         pr_debug("unwind err %d dev %s\n",
3701                                  tmp_res, slave->dev->name);
3702                 }
3703         }
3704
3705         return res;
3706 }
3707
3708 /*
3709  * Change HW address
3710  *
3711  * Note that many devices must be down to change the HW address, and
3712  * downing the master releases all slaves.  We can make bonds full of
3713  * bonding devices to test this, however.
3714  */
3715 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3716 {
3717         struct bonding *bond = netdev_priv(bond_dev);
3718         struct sockaddr *sa = addr, tmp_sa;
3719         struct slave *slave;
3720         int res = 0;
3721
3722         if (bond->params.mode == BOND_MODE_ALB)
3723                 return bond_alb_set_mac_address(bond_dev, addr);
3724
3725
3726         pr_debug("bond=%p, name=%s\n",
3727                  bond, bond_dev ? bond_dev->name : "None");
3728
3729         /* If fail_over_mac is enabled, do nothing and return success.
3730          * Returning an error causes ifenslave to fail.
3731          */
3732         if (bond->params.fail_over_mac)
3733                 return 0;
3734
3735         if (!is_valid_ether_addr(sa->sa_data))
3736                 return -EADDRNOTAVAIL;
3737
3738         /* Can't hold bond->lock with bh disabled here since
3739          * some base drivers panic. On the other hand we can't
3740          * hold bond->lock without bh disabled because we'll
3741          * deadlock. The only solution is to rely on the fact
3742          * that we're under rtnl_lock here, and the slaves
3743          * list won't change. This doesn't solve the problem
3744          * of setting the slave's hw address while it is
3745          * transmitting, but the assumption is that the base
3746          * driver can handle that.
3747          *
3748          * TODO: figure out a way to safely iterate the slaves
3749          * list, but without holding a lock around the actual
3750          * call to the base driver.
3751          */
3752
3753         bond_for_each_slave(bond, slave) {
3754                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3755                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3756
3757                 if (slave_ops->ndo_set_mac_address == NULL) {
3758                         res = -EOPNOTSUPP;
3759                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3760                         goto unwind;
3761                 }
3762
3763                 res = dev_set_mac_address(slave->dev, addr);
3764                 if (res) {
3765                         /* TODO: consider downing the slave
3766                          * and retry ?
3767                          * User should expect communications
3768                          * breakage anyway until ARP finish
3769                          * updating, so...
3770                          */
3771                         pr_debug("err %d %s\n", res, slave->dev->name);
3772                         goto unwind;
3773                 }
3774         }
3775
3776         /* success */
3777         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3778         return 0;
3779
3780 unwind:
3781         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3782         tmp_sa.sa_family = bond_dev->type;
3783
3784         /* unwind from head to the slave that failed */
3785         bond_for_each_slave_continue_reverse(bond, slave) {
3786                 int tmp_res;
3787
3788                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3789                 if (tmp_res) {
3790                         pr_debug("unwind err %d dev %s\n",
3791                                  tmp_res, slave->dev->name);
3792                 }
3793         }
3794
3795         return res;
3796 }
3797
3798 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3799 {
3800         struct bonding *bond = netdev_priv(bond_dev);
3801         struct slave *slave, *start_at;
3802         int i, slave_no, res = 1;
3803         struct iphdr *iph = ip_hdr(skb);
3804
3805         /*
3806          * Start with the curr_active_slave that joined the bond as the
3807          * default for sending IGMP traffic.  For failover purposes one
3808          * needs to maintain some consistency for the interface that will
3809          * send the join/membership reports.  The curr_active_slave found
3810          * will send all of this type of traffic.
3811          */
3812         if ((iph->protocol == IPPROTO_IGMP) &&
3813             (skb->protocol == htons(ETH_P_IP))) {
3814                 slave = bond->curr_active_slave;
3815                 if (!slave)
3816                         goto out;
3817         } else {
3818                 /*
3819                  * Concurrent TX may collide on rr_tx_counter; we accept
3820                  * that as being rare enough not to justify using an
3821                  * atomic op here.
3822                  */
3823                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3824
3825                 bond_for_each_slave(bond, slave) {
3826                         slave_no--;
3827                         if (slave_no < 0)
3828                                 break;
3829                 }
3830         }
3831
3832         start_at = slave;
3833         bond_for_each_slave_from(bond, slave, i, start_at) {
3834                 if (IS_UP(slave->dev) &&
3835                     (slave->link == BOND_LINK_UP) &&
3836                     bond_is_active_slave(slave)) {
3837                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3838                         break;
3839                 }
3840         }
3841
3842 out:
3843         if (res) {
3844                 /* no suitable interface, frame not sent */
3845                 kfree_skb(skb);
3846         }
3847
3848         return NETDEV_TX_OK;
3849 }
3850
3851
3852 /*
3853  * in active-backup mode, we know that bond->curr_active_slave is always valid if
3854  * the bond has a usable interface.
3855  */
3856 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3857 {
3858         struct bonding *bond = netdev_priv(bond_dev);
3859         struct slave *slave;
3860         int res = 1;
3861
3862         slave = bond->curr_active_slave;
3863         if (slave)
3864                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3865
3866         if (res)
3867                 /* no suitable interface, frame not sent */
3868                 kfree_skb(skb);
3869
3870         return NETDEV_TX_OK;
3871 }
3872
3873 /*
3874  * In bond_xmit_xor() , we determine the output device by using a pre-
3875  * determined xmit_hash_policy(), If the selected device is not enabled,
3876  * find the next active slave.
3877  */
3878 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3879 {
3880         struct bonding *bond = netdev_priv(bond_dev);
3881         struct slave *slave, *start_at;
3882         int slave_no;
3883         int i;
3884         int res = 1;
3885
3886         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
3887
3888         bond_for_each_slave(bond, slave) {
3889                 slave_no--;
3890                 if (slave_no < 0)
3891                         break;
3892         }
3893
3894         start_at = slave;
3895
3896         bond_for_each_slave_from(bond, slave, i, start_at) {
3897                 if (IS_UP(slave->dev) &&
3898                     (slave->link == BOND_LINK_UP) &&
3899                     bond_is_active_slave(slave)) {
3900                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3901                         break;
3902                 }
3903         }
3904
3905         if (res) {
3906                 /* no suitable interface, frame not sent */
3907                 kfree_skb(skb);
3908         }
3909
3910         return NETDEV_TX_OK;
3911 }
3912
3913 /* in broadcast mode, we send everything to all usable interfaces. */
3914 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3915 {
3916         struct bonding *bond = netdev_priv(bond_dev);
3917         struct slave *slave = NULL;
3918
3919         bond_for_each_slave(bond, slave) {
3920                 if (bond_is_last_slave(bond, slave))
3921                         break;
3922                 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3923                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3924
3925                         if (!skb2) {
3926                                 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3927                                        bond_dev->name);
3928                                 continue;
3929                         }
3930                         /* bond_dev_queue_xmit always returns 0 */
3931                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3932                 }
3933         }
3934         if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3935                 bond_dev_queue_xmit(bond, skb, slave->dev);
3936         else
3937                 kfree_skb(skb);
3938
3939         return NETDEV_TX_OK;
3940 }
3941
3942 /*------------------------- Device initialization ---------------------------*/
3943
3944 static void bond_set_xmit_hash_policy(struct bonding *bond)
3945 {
3946         switch (bond->params.xmit_policy) {
3947         case BOND_XMIT_POLICY_LAYER23:
3948                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
3949                 break;
3950         case BOND_XMIT_POLICY_LAYER34:
3951                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
3952                 break;
3953         case BOND_XMIT_POLICY_LAYER2:
3954         default:
3955                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
3956                 break;
3957         }
3958 }
3959
3960 /*
3961  * Lookup the slave that corresponds to a qid
3962  */
3963 static inline int bond_slave_override(struct bonding *bond,
3964                                       struct sk_buff *skb)
3965 {
3966         struct slave *slave = NULL;
3967         struct slave *check_slave;
3968         int res = 1;
3969
3970         if (!skb->queue_mapping)
3971                 return 1;
3972
3973         /* Find out if any slaves have the same mapping as this skb. */
3974         bond_for_each_slave(bond, check_slave) {
3975                 if (check_slave->queue_id == skb->queue_mapping) {
3976                         slave = check_slave;
3977                         break;
3978                 }
3979         }
3980
3981         /* If the slave isn't UP, use default transmit policy. */
3982         if (slave && slave->queue_id && IS_UP(slave->dev) &&
3983             (slave->link == BOND_LINK_UP)) {
3984                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3985         }
3986
3987         return res;
3988 }
3989
3990
3991 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
3992 {
3993         /*
3994          * This helper function exists to help dev_pick_tx get the correct
3995          * destination queue.  Using a helper function skips a call to
3996          * skb_tx_hash and will put the skbs in the queue we expect on their
3997          * way down to the bonding driver.
3998          */
3999         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4000
4001         /*
4002          * Save the original txq to restore before passing to the driver
4003          */
4004         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4005
4006         if (unlikely(txq >= dev->real_num_tx_queues)) {
4007                 do {
4008                         txq -= dev->real_num_tx_queues;
4009                 } while (txq >= dev->real_num_tx_queues);
4010         }
4011         return txq;
4012 }
4013
4014 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4015 {
4016         struct bonding *bond = netdev_priv(dev);
4017
4018         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4019                 if (!bond_slave_override(bond, skb))
4020                         return NETDEV_TX_OK;
4021         }
4022
4023         switch (bond->params.mode) {
4024         case BOND_MODE_ROUNDROBIN:
4025                 return bond_xmit_roundrobin(skb, dev);
4026         case BOND_MODE_ACTIVEBACKUP:
4027                 return bond_xmit_activebackup(skb, dev);
4028         case BOND_MODE_XOR:
4029                 return bond_xmit_xor(skb, dev);
4030         case BOND_MODE_BROADCAST:
4031                 return bond_xmit_broadcast(skb, dev);
4032         case BOND_MODE_8023AD:
4033                 return bond_3ad_xmit_xor(skb, dev);
4034         case BOND_MODE_ALB:
4035         case BOND_MODE_TLB:
4036                 return bond_alb_xmit(skb, dev);
4037         default:
4038                 /* Should never happen, mode already checked */
4039                 pr_err("%s: Error: Unknown bonding mode %d\n",
4040                        dev->name, bond->params.mode);
4041                 WARN_ON_ONCE(1);
4042                 kfree_skb(skb);
4043                 return NETDEV_TX_OK;
4044         }
4045 }
4046
4047 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4048 {
4049         struct bonding *bond = netdev_priv(dev);
4050         netdev_tx_t ret = NETDEV_TX_OK;
4051
4052         /*
4053          * If we risk deadlock from transmitting this in the
4054          * netpoll path, tell netpoll to queue the frame for later tx
4055          */
4056         if (is_netpoll_tx_blocked(dev))
4057                 return NETDEV_TX_BUSY;
4058
4059         read_lock(&bond->lock);
4060
4061         if (!list_empty(&bond->slave_list))
4062                 ret = __bond_start_xmit(skb, dev);
4063         else
4064                 kfree_skb(skb);
4065
4066         read_unlock(&bond->lock);
4067
4068         return ret;
4069 }
4070
4071 /*
4072  * set bond mode specific net device operations
4073  */
4074 void bond_set_mode_ops(struct bonding *bond, int mode)
4075 {
4076         struct net_device *bond_dev = bond->dev;
4077
4078         switch (mode) {
4079         case BOND_MODE_ROUNDROBIN:
4080                 break;
4081         case BOND_MODE_ACTIVEBACKUP:
4082                 break;
4083         case BOND_MODE_XOR:
4084                 bond_set_xmit_hash_policy(bond);
4085                 break;
4086         case BOND_MODE_BROADCAST:
4087                 break;
4088         case BOND_MODE_8023AD:
4089                 bond_set_xmit_hash_policy(bond);
4090                 break;
4091         case BOND_MODE_ALB:
4092                 /* FALLTHRU */
4093         case BOND_MODE_TLB:
4094                 break;
4095         default:
4096                 /* Should never happen, mode already checked */
4097                 pr_err("%s: Error: Unknown bonding mode %d\n",
4098                        bond_dev->name, mode);
4099                 break;
4100         }
4101 }
4102
4103 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4104                                      struct ethtool_cmd *ecmd)
4105 {
4106         struct bonding *bond = netdev_priv(bond_dev);
4107         unsigned long speed = 0;
4108         struct slave *slave;
4109
4110         ecmd->duplex = DUPLEX_UNKNOWN;
4111         ecmd->port = PORT_OTHER;
4112
4113         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4114          * do not need to check mode.  Though link speed might not represent
4115          * the true receive or transmit bandwidth (not all modes are symmetric)
4116          * this is an accurate maximum.
4117          */
4118         read_lock(&bond->lock);
4119         bond_for_each_slave(bond, slave) {
4120                 if (SLAVE_IS_OK(slave)) {
4121                         if (slave->speed != SPEED_UNKNOWN)
4122                                 speed += slave->speed;
4123                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4124                             slave->duplex != DUPLEX_UNKNOWN)
4125                                 ecmd->duplex = slave->duplex;
4126                 }
4127         }
4128         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4129         read_unlock(&bond->lock);
4130
4131         return 0;
4132 }
4133
4134 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4135                                      struct ethtool_drvinfo *drvinfo)
4136 {
4137         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4138         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4139         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4140                  BOND_ABI_VERSION);
4141 }
4142
4143 static const struct ethtool_ops bond_ethtool_ops = {
4144         .get_drvinfo            = bond_ethtool_get_drvinfo,
4145         .get_settings           = bond_ethtool_get_settings,
4146         .get_link               = ethtool_op_get_link,
4147 };
4148
4149 static const struct net_device_ops bond_netdev_ops = {
4150         .ndo_init               = bond_init,
4151         .ndo_uninit             = bond_uninit,
4152         .ndo_open               = bond_open,
4153         .ndo_stop               = bond_close,
4154         .ndo_start_xmit         = bond_start_xmit,
4155         .ndo_select_queue       = bond_select_queue,
4156         .ndo_get_stats64        = bond_get_stats,
4157         .ndo_do_ioctl           = bond_do_ioctl,
4158         .ndo_change_rx_flags    = bond_change_rx_flags,
4159         .ndo_set_rx_mode        = bond_set_rx_mode,
4160         .ndo_change_mtu         = bond_change_mtu,
4161         .ndo_set_mac_address    = bond_set_mac_address,
4162         .ndo_neigh_setup        = bond_neigh_setup,
4163         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4164         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4165 #ifdef CONFIG_NET_POLL_CONTROLLER
4166         .ndo_netpoll_setup      = bond_netpoll_setup,
4167         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4168         .ndo_poll_controller    = bond_poll_controller,
4169 #endif
4170         .ndo_add_slave          = bond_enslave,
4171         .ndo_del_slave          = bond_release,
4172         .ndo_fix_features       = bond_fix_features,
4173 };
4174
4175 static const struct device_type bond_type = {
4176         .name = "bond",
4177 };
4178
4179 static void bond_destructor(struct net_device *bond_dev)
4180 {
4181         struct bonding *bond = netdev_priv(bond_dev);
4182         if (bond->wq)
4183                 destroy_workqueue(bond->wq);
4184         free_netdev(bond_dev);
4185 }
4186
4187 static void bond_setup(struct net_device *bond_dev)
4188 {
4189         struct bonding *bond = netdev_priv(bond_dev);
4190
4191         /* initialize rwlocks */
4192         rwlock_init(&bond->lock);
4193         rwlock_init(&bond->curr_slave_lock);
4194         INIT_LIST_HEAD(&bond->slave_list);
4195         bond->params = bonding_defaults;
4196
4197         /* Initialize pointers */
4198         bond->dev = bond_dev;
4199         INIT_LIST_HEAD(&bond->vlan_list);
4200
4201         /* Initialize the device entry points */
4202         ether_setup(bond_dev);
4203         bond_dev->netdev_ops = &bond_netdev_ops;
4204         bond_dev->ethtool_ops = &bond_ethtool_ops;
4205         bond_set_mode_ops(bond, bond->params.mode);
4206
4207         bond_dev->destructor = bond_destructor;
4208
4209         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4210
4211         /* Initialize the device options */
4212         bond_dev->tx_queue_len = 0;
4213         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4214         bond_dev->priv_flags |= IFF_BONDING;
4215         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4216
4217         /* At first, we block adding VLANs. That's the only way to
4218          * prevent problems that occur when adding VLANs over an
4219          * empty bond. The block will be removed once non-challenged
4220          * slaves are enslaved.
4221          */
4222         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4223
4224         /* don't acquire bond device's netif_tx_lock when
4225          * transmitting */
4226         bond_dev->features |= NETIF_F_LLTX;
4227
4228         /* By default, we declare the bond to be fully
4229          * VLAN hardware accelerated capable. Special
4230          * care is taken in the various xmit functions
4231          * when there are slaves that are not hw accel
4232          * capable
4233          */
4234
4235         bond_dev->hw_features = BOND_VLAN_FEATURES |
4236                                 NETIF_F_HW_VLAN_CTAG_TX |
4237                                 NETIF_F_HW_VLAN_CTAG_RX |
4238                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4239
4240         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4241         bond_dev->features |= bond_dev->hw_features;
4242 }
4243
4244 /*
4245 * Destroy a bonding device.
4246 * Must be under rtnl_lock when this function is called.
4247 */
4248 static void bond_uninit(struct net_device *bond_dev)
4249 {
4250         struct bonding *bond = netdev_priv(bond_dev);
4251         struct slave *slave, *tmp_slave;
4252         struct vlan_entry *vlan, *tmp;
4253
4254         bond_netpoll_cleanup(bond_dev);
4255
4256         /* Release the bonded slaves */
4257         list_for_each_entry_safe(slave, tmp_slave, &bond->slave_list, list)
4258                 __bond_release_one(bond_dev, slave->dev, true);
4259         pr_info("%s: released all slaves\n", bond_dev->name);
4260
4261         list_del(&bond->bond_list);
4262
4263         bond_debug_unregister(bond);
4264
4265         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4266                 list_del(&vlan->vlan_list);
4267                 kfree(vlan);
4268         }
4269 }
4270
4271 /*------------------------- Module initialization ---------------------------*/
4272
4273 /*
4274  * Convert string input module parms.  Accept either the
4275  * number of the mode or its string name.  A bit complicated because
4276  * some mode names are substrings of other names, and calls from sysfs
4277  * may have whitespace in the name (trailing newlines, for example).
4278  */
4279 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4280 {
4281         int modeint = -1, i, rv;
4282         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4283
4284         for (p = (char *)buf; *p; p++)
4285                 if (!(isdigit(*p) || isspace(*p)))
4286                         break;
4287
4288         if (*p)
4289                 rv = sscanf(buf, "%20s", modestr);
4290         else
4291                 rv = sscanf(buf, "%d", &modeint);
4292
4293         if (!rv)
4294                 return -1;
4295
4296         for (i = 0; tbl[i].modename; i++) {
4297                 if (modeint == tbl[i].mode)
4298                         return tbl[i].mode;
4299                 if (strcmp(modestr, tbl[i].modename) == 0)
4300                         return tbl[i].mode;
4301         }
4302
4303         return -1;
4304 }
4305
4306 static int bond_check_params(struct bond_params *params)
4307 {
4308         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4309         int arp_all_targets_value;
4310
4311         /*
4312          * Convert string parameters.
4313          */
4314         if (mode) {
4315                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4316                 if (bond_mode == -1) {
4317                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4318                                mode == NULL ? "NULL" : mode);
4319                         return -EINVAL;
4320                 }
4321         }
4322
4323         if (xmit_hash_policy) {
4324                 if ((bond_mode != BOND_MODE_XOR) &&
4325                     (bond_mode != BOND_MODE_8023AD)) {
4326                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4327                                bond_mode_name(bond_mode));
4328                 } else {
4329                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4330                                                         xmit_hashtype_tbl);
4331                         if (xmit_hashtype == -1) {
4332                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4333                                        xmit_hash_policy == NULL ? "NULL" :
4334                                        xmit_hash_policy);
4335                                 return -EINVAL;
4336                         }
4337                 }
4338         }
4339
4340         if (lacp_rate) {
4341                 if (bond_mode != BOND_MODE_8023AD) {
4342                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4343                                 bond_mode_name(bond_mode));
4344                 } else {
4345                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4346                         if (lacp_fast == -1) {
4347                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4348                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4349                                 return -EINVAL;
4350                         }
4351                 }
4352         }
4353
4354         if (ad_select) {
4355                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4356                 if (params->ad_select == -1) {
4357                         pr_err("Error: Invalid ad_select \"%s\"\n",
4358                                ad_select == NULL ? "NULL" : ad_select);
4359                         return -EINVAL;
4360                 }
4361
4362                 if (bond_mode != BOND_MODE_8023AD) {
4363                         pr_warning("ad_select param only affects 802.3ad mode\n");
4364                 }
4365         } else {
4366                 params->ad_select = BOND_AD_STABLE;
4367         }
4368
4369         if (max_bonds < 0) {
4370                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4371                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4372                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4373         }
4374
4375         if (miimon < 0) {
4376                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4377                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4378                 miimon = BOND_LINK_MON_INTERV;
4379         }
4380
4381         if (updelay < 0) {
4382                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4383                            updelay, INT_MAX);
4384                 updelay = 0;
4385         }
4386
4387         if (downdelay < 0) {
4388                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4389                            downdelay, INT_MAX);
4390                 downdelay = 0;
4391         }
4392
4393         if ((use_carrier != 0) && (use_carrier != 1)) {
4394                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4395                            use_carrier);
4396                 use_carrier = 1;
4397         }
4398
4399         if (num_peer_notif < 0 || num_peer_notif > 255) {
4400                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4401                            num_peer_notif);
4402                 num_peer_notif = 1;
4403         }
4404
4405         /* reset values for 802.3ad */
4406         if (bond_mode == BOND_MODE_8023AD) {
4407                 if (!miimon) {
4408                         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");
4409                         pr_warning("Forcing miimon to 100msec\n");
4410                         miimon = 100;
4411                 }
4412         }
4413
4414         if (tx_queues < 1 || tx_queues > 255) {
4415                 pr_warning("Warning: tx_queues (%d) should be between "
4416                            "1 and 255, resetting to %d\n",
4417                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4418                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4419         }
4420
4421         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4422                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4423                            "not of valid value (0/1), so it was set to "
4424                            "0\n", all_slaves_active);
4425                 all_slaves_active = 0;
4426         }
4427
4428         if (resend_igmp < 0 || resend_igmp > 255) {
4429                 pr_warning("Warning: resend_igmp (%d) should be between "
4430                            "0 and 255, resetting to %d\n",
4431                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4432                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4433         }
4434
4435         /* reset values for TLB/ALB */
4436         if ((bond_mode == BOND_MODE_TLB) ||
4437             (bond_mode == BOND_MODE_ALB)) {
4438                 if (!miimon) {
4439                         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");
4440                         pr_warning("Forcing miimon to 100msec\n");
4441                         miimon = 100;
4442                 }
4443         }
4444
4445         if (bond_mode == BOND_MODE_ALB) {
4446                 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",
4447                           updelay);
4448         }
4449
4450         if (!miimon) {
4451                 if (updelay || downdelay) {
4452                         /* just warn the user the up/down delay will have
4453                          * no effect since miimon is zero...
4454                          */
4455                         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",
4456                                    updelay, downdelay);
4457                 }
4458         } else {
4459                 /* don't allow arp monitoring */
4460                 if (arp_interval) {
4461                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4462                                    miimon, arp_interval);
4463                         arp_interval = 0;
4464                 }
4465
4466                 if ((updelay % miimon) != 0) {
4467                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4468                                    updelay, miimon,
4469                                    (updelay / miimon) * miimon);
4470                 }
4471
4472                 updelay /= miimon;
4473
4474                 if ((downdelay % miimon) != 0) {
4475                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4476                                    downdelay, miimon,
4477                                    (downdelay / miimon) * miimon);
4478                 }
4479
4480                 downdelay /= miimon;
4481         }
4482
4483         if (arp_interval < 0) {
4484                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4485                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4486                 arp_interval = BOND_LINK_ARP_INTERV;
4487         }
4488
4489         for (arp_ip_count = 0, i = 0;
4490              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4491                 /* not complete check, but should be good enough to
4492                    catch mistakes */
4493                 __be32 ip = in_aton(arp_ip_target[i]);
4494                 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4495                     ip == htonl(INADDR_BROADCAST)) {
4496                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4497                                    arp_ip_target[i]);
4498                         arp_interval = 0;
4499                 } else {
4500                         if (bond_get_targets_ip(arp_target, ip) == -1)
4501                                 arp_target[arp_ip_count++] = ip;
4502                         else
4503                                 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4504                                            &ip);
4505                 }
4506         }
4507
4508         if (arp_interval && !arp_ip_count) {
4509                 /* don't allow arping if no arp_ip_target given... */
4510                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4511                            arp_interval);
4512                 arp_interval = 0;
4513         }
4514
4515         if (arp_validate) {
4516                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4517                         pr_err("arp_validate only supported in active-backup mode\n");
4518                         return -EINVAL;
4519                 }
4520                 if (!arp_interval) {
4521                         pr_err("arp_validate requires arp_interval\n");
4522                         return -EINVAL;
4523                 }
4524
4525                 arp_validate_value = bond_parse_parm(arp_validate,
4526                                                      arp_validate_tbl);
4527                 if (arp_validate_value == -1) {
4528                         pr_err("Error: invalid arp_validate \"%s\"\n",
4529                                arp_validate == NULL ? "NULL" : arp_validate);
4530                         return -EINVAL;
4531                 }
4532         } else
4533                 arp_validate_value = 0;
4534
4535         arp_all_targets_value = 0;
4536         if (arp_all_targets) {
4537                 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4538                                                         arp_all_targets_tbl);
4539
4540                 if (arp_all_targets_value == -1) {
4541                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4542                                arp_all_targets);
4543                         arp_all_targets_value = 0;
4544                 }
4545         }
4546
4547         if (miimon) {
4548                 pr_info("MII link monitoring set to %d ms\n", miimon);
4549         } else if (arp_interval) {
4550                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4551                         arp_interval,
4552                         arp_validate_tbl[arp_validate_value].modename,
4553                         arp_ip_count);
4554
4555                 for (i = 0; i < arp_ip_count; i++)
4556                         pr_info(" %s", arp_ip_target[i]);
4557
4558                 pr_info("\n");
4559
4560         } else if (max_bonds) {
4561                 /* miimon and arp_interval not set, we need one so things
4562                  * work as expected, see bonding.txt for details
4563                  */
4564                 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");
4565         }
4566
4567         if (primary && !USES_PRIMARY(bond_mode)) {
4568                 /* currently, using a primary only makes sense
4569                  * in active backup, TLB or ALB modes
4570                  */
4571                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4572                            primary, bond_mode_name(bond_mode));
4573                 primary = NULL;
4574         }
4575
4576         if (primary && primary_reselect) {
4577                 primary_reselect_value = bond_parse_parm(primary_reselect,
4578                                                          pri_reselect_tbl);
4579                 if (primary_reselect_value == -1) {
4580                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4581                                primary_reselect ==
4582                                         NULL ? "NULL" : primary_reselect);
4583                         return -EINVAL;
4584                 }
4585         } else {
4586                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4587         }
4588
4589         if (fail_over_mac) {
4590                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4591                                                       fail_over_mac_tbl);
4592                 if (fail_over_mac_value == -1) {
4593                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4594                                arp_validate == NULL ? "NULL" : arp_validate);
4595                         return -EINVAL;
4596                 }
4597
4598                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4599                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4600         } else {
4601                 fail_over_mac_value = BOND_FOM_NONE;
4602         }
4603
4604         /* fill params struct with the proper values */
4605         params->mode = bond_mode;
4606         params->xmit_policy = xmit_hashtype;
4607         params->miimon = miimon;
4608         params->num_peer_notif = num_peer_notif;
4609         params->arp_interval = arp_interval;
4610         params->arp_validate = arp_validate_value;
4611         params->arp_all_targets = arp_all_targets_value;
4612         params->updelay = updelay;
4613         params->downdelay = downdelay;
4614         params->use_carrier = use_carrier;
4615         params->lacp_fast = lacp_fast;
4616         params->primary[0] = 0;
4617         params->primary_reselect = primary_reselect_value;
4618         params->fail_over_mac = fail_over_mac_value;
4619         params->tx_queues = tx_queues;
4620         params->all_slaves_active = all_slaves_active;
4621         params->resend_igmp = resend_igmp;
4622         params->min_links = min_links;
4623
4624         if (primary) {
4625                 strncpy(params->primary, primary, IFNAMSIZ);
4626                 params->primary[IFNAMSIZ - 1] = 0;
4627         }
4628
4629         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4630
4631         return 0;
4632 }
4633
4634 static struct lock_class_key bonding_netdev_xmit_lock_key;
4635 static struct lock_class_key bonding_netdev_addr_lock_key;
4636 static struct lock_class_key bonding_tx_busylock_key;
4637
4638 static void bond_set_lockdep_class_one(struct net_device *dev,
4639                                        struct netdev_queue *txq,
4640                                        void *_unused)
4641 {
4642         lockdep_set_class(&txq->_xmit_lock,
4643                           &bonding_netdev_xmit_lock_key);
4644 }
4645
4646 static void bond_set_lockdep_class(struct net_device *dev)
4647 {
4648         lockdep_set_class(&dev->addr_list_lock,
4649                           &bonding_netdev_addr_lock_key);
4650         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4651         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4652 }
4653
4654 /*
4655  * Called from registration process
4656  */
4657 static int bond_init(struct net_device *bond_dev)
4658 {
4659         struct bonding *bond = netdev_priv(bond_dev);
4660         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4661         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4662
4663         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4664
4665         /*
4666          * Initialize locks that may be required during
4667          * en/deslave operations.  All of the bond_open work
4668          * (of which this is part) should really be moved to
4669          * a phase prior to dev_open
4670          */
4671         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4672         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4673
4674         bond->wq = create_singlethread_workqueue(bond_dev->name);
4675         if (!bond->wq)
4676                 return -ENOMEM;
4677
4678         bond_set_lockdep_class(bond_dev);
4679
4680         list_add_tail(&bond->bond_list, &bn->dev_list);
4681
4682         bond_prepare_sysfs_group(bond);
4683
4684         bond_debug_register(bond);
4685
4686         /* Ensure valid dev_addr */
4687         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4688             bond_dev->addr_assign_type == NET_ADDR_PERM)
4689                 eth_hw_addr_random(bond_dev);
4690
4691         return 0;
4692 }
4693
4694 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4695 {
4696         if (tb[IFLA_ADDRESS]) {
4697                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4698                         return -EINVAL;
4699                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4700                         return -EADDRNOTAVAIL;
4701         }
4702         return 0;
4703 }
4704
4705 static unsigned int bond_get_num_tx_queues(void)
4706 {
4707         return tx_queues;
4708 }
4709
4710 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4711         .kind                   = "bond",
4712         .priv_size              = sizeof(struct bonding),
4713         .setup                  = bond_setup,
4714         .validate               = bond_validate,
4715         .get_num_tx_queues      = bond_get_num_tx_queues,
4716         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4717                                                              as for TX queues */
4718 };
4719
4720 /* Create a new bond based on the specified name and bonding parameters.
4721  * If name is NULL, obtain a suitable "bond%d" name for us.
4722  * Caller must NOT hold rtnl_lock; we need to release it here before we
4723  * set up our sysfs entries.
4724  */
4725 int bond_create(struct net *net, const char *name)
4726 {
4727         struct net_device *bond_dev;
4728         int res;
4729
4730         rtnl_lock();
4731
4732         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4733                                    name ? name : "bond%d",
4734                                    bond_setup, tx_queues);
4735         if (!bond_dev) {
4736                 pr_err("%s: eek! can't alloc netdev!\n", name);
4737                 rtnl_unlock();
4738                 return -ENOMEM;
4739         }
4740
4741         dev_net_set(bond_dev, net);
4742         bond_dev->rtnl_link_ops = &bond_link_ops;
4743
4744         res = register_netdevice(bond_dev);
4745
4746         netif_carrier_off(bond_dev);
4747
4748         rtnl_unlock();
4749         if (res < 0)
4750                 bond_destructor(bond_dev);
4751         return res;
4752 }
4753
4754 static int __net_init bond_net_init(struct net *net)
4755 {
4756         struct bond_net *bn = net_generic(net, bond_net_id);
4757
4758         bn->net = net;
4759         INIT_LIST_HEAD(&bn->dev_list);
4760
4761         bond_create_proc_dir(bn);
4762         bond_create_sysfs(bn);
4763
4764         return 0;
4765 }
4766
4767 static void __net_exit bond_net_exit(struct net *net)
4768 {
4769         struct bond_net *bn = net_generic(net, bond_net_id);
4770         struct bonding *bond, *tmp_bond;
4771         LIST_HEAD(list);
4772
4773         bond_destroy_sysfs(bn);
4774         bond_destroy_proc_dir(bn);
4775
4776         /* Kill off any bonds created after unregistering bond rtnl ops */
4777         rtnl_lock();
4778         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4779                 unregister_netdevice_queue(bond->dev, &list);
4780         unregister_netdevice_many(&list);
4781         rtnl_unlock();
4782 }
4783
4784 static struct pernet_operations bond_net_ops = {
4785         .init = bond_net_init,
4786         .exit = bond_net_exit,
4787         .id   = &bond_net_id,
4788         .size = sizeof(struct bond_net),
4789 };
4790
4791 static int __init bonding_init(void)
4792 {
4793         int i;
4794         int res;
4795
4796         pr_info("%s", bond_version);
4797
4798         res = bond_check_params(&bonding_defaults);
4799         if (res)
4800                 goto out;
4801
4802         res = register_pernet_subsys(&bond_net_ops);
4803         if (res)
4804                 goto out;
4805
4806         res = rtnl_link_register(&bond_link_ops);
4807         if (res)
4808                 goto err_link;
4809
4810         bond_create_debugfs();
4811
4812         for (i = 0; i < max_bonds; i++) {
4813                 res = bond_create(&init_net, NULL);
4814                 if (res)
4815                         goto err;
4816         }
4817
4818         register_netdevice_notifier(&bond_netdev_notifier);
4819 out:
4820         return res;
4821 err:
4822         rtnl_link_unregister(&bond_link_ops);
4823 err_link:
4824         unregister_pernet_subsys(&bond_net_ops);
4825         goto out;
4826
4827 }
4828
4829 static void __exit bonding_exit(void)
4830 {
4831         unregister_netdevice_notifier(&bond_netdev_notifier);
4832
4833         bond_destroy_debugfs();
4834
4835         rtnl_link_unregister(&bond_link_ops);
4836         unregister_pernet_subsys(&bond_net_ops);
4837
4838 #ifdef CONFIG_NET_POLL_CONTROLLER
4839         /*
4840          * Make sure we don't have an imbalance on our netpoll blocking
4841          */
4842         WARN_ON(atomic_read(&netpoll_block_tx));
4843 #endif
4844 }
4845
4846 module_init(bonding_init);
4847 module_exit(bonding_exit);
4848 MODULE_LICENSE("GPL");
4849 MODULE_VERSION(DRV_VERSION);
4850 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4851 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4852 MODULE_ALIAS_RTNL_LINK("bond");