drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon       = BOND_LINK_MON_INTERV;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static int all_slaves_active = 0;
 109 static struct bond_params bonding_defaults;
 110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 111
 112 module_param(max_bonds, int, 0);
 113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 114 module_param(tx_queues, int, 0);
 115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 118                                "failover event (alias of num_unsol_na)");
 119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 121                                "failover event (alias of num_grat_arp)");
 122 module_param(miimon, int, 0);
 123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 124 module_param(updelay, int, 0);
 125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 126 module_param(downdelay, int, 0);
 127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 128                             "in milliseconds");
 129 module_param(use_carrier, int, 0);
 130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 131                               "0 for off, 1 for on (default)");
 132 module_param(mode, charp, 0);
 133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 134                        "1 for active-backup, 2 for balance-xor, "
 135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 136                        "6 for balance-alb");
 137 module_param(primary, charp, 0);
 138 MODULE_PARM_DESC(primary, "Primary network device to use");
 139 module_param(primary_reselect, charp, 0);
 140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 141                                    "once it comes up; "
 142                                    "0 for always (default), "
 143                                    "1 for only if speed of primary is "
 144                                    "better, "
 145                                    "2 for only on active slave "
 146                                    "failure");
 147 module_param(lacp_rate, charp, 0);
 148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 149                             "0 for slow, 1 for fast");
 150 module_param(ad_select, charp, 0);
 151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 152                             "0 for stable (default), 1 for bandwidth, "
 153                             "2 for count");
 154 module_param(min_links, int, 0);
 155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 156
 157 module_param(xmit_hash_policy, charp, 0);
 158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 159                                    "0 for layer 2 (default), 1 for layer 3+4, "
 160                                    "2 for layer 2+3");
 161 module_param(arp_interval, int, 0);
 162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 163 module_param_array(arp_ip_target, charp, NULL, 0);
 164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 165 module_param(arp_validate, charp, 0);
 166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 167                                "0 for none (default), 1 for active, "
 168                                "2 for backup, 3 for all");
 169 module_param(fail_over_mac, charp, 0);
 170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 171                                 "the same MAC; 0 for none (default), "
 172                                 "1 for active, 2 for follow");
 173 module_param(all_slaves_active, int, 0);
 174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 175                                      "by setting active flag for all slaves; "
 176                                      "0 for never (default), 1 for always.");
 177 module_param(resend_igmp, int, 0);
 178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 179                               "link failure");
 180
 181 /*----------------------------- Global variables ----------------------------*/
 182
 183 #ifdef CONFIG_NET_POLL_CONTROLLER
 184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 185 #endif
 186
 187 int bond_net_id __read_mostly;
 188
 189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 190 static int arp_ip_count;
 191 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 193 static int lacp_fast;
 194
 195 const struct bond_parm_tbl bond_lacp_tbl[] = {
 196 {       "slow",         AD_LACP_SLOW},
 197 {       "fast",         AD_LACP_FAST},
 198 {       NULL,           -1},
 199 };
 200
 201 const struct bond_parm_tbl bond_mode_tbl[] = {
 202 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 203 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 204 {       "balance-xor",          BOND_MODE_XOR},
 205 {       "broadcast",            BOND_MODE_BROADCAST},
 206 {       "802.3ad",              BOND_MODE_8023AD},
 207 {       "balance-tlb",          BOND_MODE_TLB},
 208 {       "balance-alb",          BOND_MODE_ALB},
 209 {       NULL,                   -1},
 210 };
 211
 212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 213 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 214 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 215 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 216 {       NULL,                   -1},
 217 };
 218
 219 const struct bond_parm_tbl arp_validate_tbl[] = {
 220 {       "none",                 BOND_ARP_VALIDATE_NONE},
 221 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 222 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 223 {       "all",                  BOND_ARP_VALIDATE_ALL},
 224 {       NULL,                   -1},
 225 };
 226
 227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 228 {       "none",                 BOND_FOM_NONE},
 229 {       "active",               BOND_FOM_ACTIVE},
 230 {       "follow",               BOND_FOM_FOLLOW},
 231 {       NULL,                   -1},
 232 };
 233
 234 const struct bond_parm_tbl pri_reselect_tbl[] = {
 235 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 236 {       "better",               BOND_PRI_RESELECT_BETTER},
 237 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 238 {       NULL,                   -1},
 239 };
 240
 241 struct bond_parm_tbl ad_select_tbl[] = {
 242 {       "stable",       BOND_AD_STABLE},
 243 {       "bandwidth",    BOND_AD_BANDWIDTH},
 244 {       "count",        BOND_AD_COUNT},
 245 {       NULL,           -1},
 246 };
 247
 248 /*-------------------------- Forward declarations ---------------------------*/
 249
 250 static int bond_init(struct net_device *bond_dev);
 251 static void bond_uninit(struct net_device *bond_dev);
 252
 253 /*---------------------------- General routines -----------------------------*/
 254
 255 const char *bond_mode_name(int mode)
 256 {
 257         static const char *names[] = {
 258                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 259                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 260                 [BOND_MODE_XOR] = "load balancing (xor)",
 261                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 262                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 263                 [BOND_MODE_TLB] = "transmit load balancing",
 264                 [BOND_MODE_ALB] = "adaptive load balancing",
 265         };
 266
 267         if (mode < 0 || mode > BOND_MODE_ALB)
 268                 return "unknown";
 269
 270         return names[mode];
 271 }
 272
 273 /*---------------------------------- VLAN -----------------------------------*/
 274
 275 /**
 276  * bond_add_vlan - add a new vlan id on bond
 277  * @bond: bond that got the notification
 278  * @vlan_id: the vlan id to add
 279  *
 280  * Returns -ENOMEM if allocation failed.
 281  */
 282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 283 {
 284         struct vlan_entry *vlan;
 285
 286         pr_debug("bond: %s, vlan id %d\n",
 287                  (bond ? bond->dev->name : "None"), vlan_id);
 288
 289         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 290         if (!vlan)
 291                 return -ENOMEM;
 292
 293         INIT_LIST_HEAD(&vlan->vlan_list);
 294         vlan->vlan_id = vlan_id;
 295
 296         write_lock_bh(&bond->lock);
 297
 298         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 299
 300         write_unlock_bh(&bond->lock);
 301
 302         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 303
 304         return 0;
 305 }
 306
 307 /**
 308  * bond_del_vlan - delete a vlan id from bond
 309  * @bond: bond that got the notification
 310  * @vlan_id: the vlan id to delete
 311  *
 312  * returns -ENODEV if @vlan_id was not found in @bond.
 313  */
 314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 315 {
 316         struct vlan_entry *vlan;
 317         int res = -ENODEV;
 318
 319         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 320
 321         block_netpoll_tx();
 322         write_lock_bh(&bond->lock);
 323
 324         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 325                 if (vlan->vlan_id == vlan_id) {
 326                         list_del(&vlan->vlan_list);
 327
 328                         if (bond_is_lb(bond))
 329                                 bond_alb_clear_vlan(bond, vlan_id);
 330
 331                         pr_debug("removed VLAN ID %d from bond %s\n",
 332                                  vlan_id, bond->dev->name);
 333
 334                         kfree(vlan);
 335
 336                         res = 0;
 337                         goto out;
 338                 }
 339         }
 340
 341         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 342                  vlan_id, bond->dev->name);
 343
 344 out:
 345         write_unlock_bh(&bond->lock);
 346         unblock_netpoll_tx();
 347         return res;
 348 }
 349
 350 /**
 351  * bond_next_vlan - safely skip to the next item in the vlans list.
 352  * @bond: the bond we're working on
 353  * @curr: item we're advancing from
 354  *
 355  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 356  * or @curr->next otherwise (even if it is @curr itself again).
 357  *
 358  * Caller must hold bond->lock
 359  */
 360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 361 {
 362         struct vlan_entry *next, *last;
 363
 364         if (list_empty(&bond->vlan_list))
 365                 return NULL;
 366
 367         if (!curr) {
 368                 next = list_entry(bond->vlan_list.next,
 369                                   struct vlan_entry, vlan_list);
 370         } else {
 371                 last = list_entry(bond->vlan_list.prev,
 372                                   struct vlan_entry, vlan_list);
 373                 if (last == curr) {
 374                         next = list_entry(bond->vlan_list.next,
 375                                           struct vlan_entry, vlan_list);
 376                 } else {
 377                         next = list_entry(curr->vlan_list.next,
 378                                           struct vlan_entry, vlan_list);
 379                 }
 380         }
 381
 382         return next;
 383 }
 384
 385 /**
 386  * bond_dev_queue_xmit - Prepare skb for xmit.
 387  *
 388  * @bond: bond device that got this skb for tx.
 389  * @skb: hw accel VLAN tagged skb to transmit
 390  * @slave_dev: slave that is supposed to xmit this skbuff
 391  */
 392 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 393                         struct net_device *slave_dev)
 394 {
 395         skb->dev = slave_dev;
 396
 397         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 398                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 399         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 400
 401         if (unlikely(netpoll_tx_running(bond->dev)))
 402                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 403         else
 404                 dev_queue_xmit(skb);
 405
 406         return 0;
 407 }
 408
 409 /*
 410  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 411  * We don't protect the slave list iteration with a lock because:
 412  * a. This operation is performed in IOCTL context,
 413  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 414  * c. Holding a lock with BH disabled while directly calling a base driver
 415  *    entry point is generally a BAD idea.
 416  *
 417  * The design of synchronization/protection for this operation in the 8021q
 418  * module is good for one or more VLAN devices over a single physical device
 419  * and cannot be extended for a teaming solution like bonding, so there is a
 420  * potential race condition here where a net device from the vlan group might
 421  * be referenced (either by a base driver or the 8021q code) while it is being
 422  * removed from the system. However, it turns out we're not making matters
 423  * worse, and if it works for regular VLAN usage it will work here too.
 424 */
 425
 426 /**
 427  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 428  * @bond_dev: bonding net device that got called
 429  * @vid: vlan id being added
 430  */
 431 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 432                                 __be16 proto, u16 vid)
 433 {
 434         struct bonding *bond = netdev_priv(bond_dev);
 435         struct slave *slave, *stop_at;
 436         int i, res;
 437
 438         bond_for_each_slave(bond, slave, i) {
 439                 res = vlan_vid_add(slave->dev, proto, vid);
 440                 if (res)
 441                         goto unwind;
 442         }
 443
 444         res = bond_add_vlan(bond, vid);
 445         if (res) {
 446                 pr_err("%s: Error: Failed to add vlan id %d\n",
 447                        bond_dev->name, vid);
 448                 return res;
 449         }
 450
 451         return 0;
 452
 453 unwind:
 454         /* unwind from head to the slave that failed */
 455         stop_at = slave;
 456         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
 457                 vlan_vid_del(slave->dev, proto, vid);
 458
 459         return res;
 460 }
 461
 462 /**
 463  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 464  * @bond_dev: bonding net device that got called
 465  * @vid: vlan id being removed
 466  */
 467 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 468                                  __be16 proto, u16 vid)
 469 {
 470         struct bonding *bond = netdev_priv(bond_dev);
 471         struct slave *slave;
 472         int i, res;
 473
 474         bond_for_each_slave(bond, slave, i)
 475                 vlan_vid_del(slave->dev, proto, vid);
 476
 477         res = bond_del_vlan(bond, vid);
 478         if (res) {
 479                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 480                        bond_dev->name, vid);
 481                 return res;
 482         }
 483
 484         return 0;
 485 }
 486
 487 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 488 {
 489         struct vlan_entry *vlan;
 490         int res;
 491
 492         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 493                 res = vlan_vid_add(slave_dev, htons(ETH_P_8021Q),
 494                                    vlan->vlan_id);
 495                 if (res)
 496                         pr_warning("%s: Failed to add vlan id %d to device %s\n",
 497                                    bond->dev->name, vlan->vlan_id,
 498                                    slave_dev->name);
 499         }
 500 }
 501
 502 static void bond_del_vlans_from_slave(struct bonding *bond,
 503                                       struct net_device *slave_dev)
 504 {
 505         struct vlan_entry *vlan;
 506
 507         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 508                 if (!vlan->vlan_id)
 509                         continue;
 510                 vlan_vid_del(slave_dev, htons(ETH_P_8021Q), vlan->vlan_id);
 511         }
 512 }
 513
 514 /*------------------------------- Link status -------------------------------*/
 515
 516 /*
 517  * Set the carrier state for the master according to the state of its
 518  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 519  * do special 802.3ad magic.
 520  *
 521  * Returns zero if carrier state does not change, nonzero if it does.
 522  */
 523 static int bond_set_carrier(struct bonding *bond)
 524 {
 525         struct slave *slave;
 526         int i;
 527
 528         if (bond->slave_cnt == 0)
 529                 goto down;
 530
 531         if (bond->params.mode == BOND_MODE_8023AD)
 532                 return bond_3ad_set_carrier(bond);
 533
 534         bond_for_each_slave(bond, slave, i) {
 535                 if (slave->link == BOND_LINK_UP) {
 536                         if (!netif_carrier_ok(bond->dev)) {
 537                                 netif_carrier_on(bond->dev);
 538                                 return 1;
 539                         }
 540                         return 0;
 541                 }
 542         }
 543
 544 down:
 545         if (netif_carrier_ok(bond->dev)) {
 546                 netif_carrier_off(bond->dev);
 547                 return 1;
 548         }
 549         return 0;
 550 }
 551
 552 /*
 553  * Get link speed and duplex from the slave's base driver
 554  * using ethtool. If for some reason the call fails or the
 555  * values are invalid, set speed and duplex to -1,
 556  * and return.
 557  */
 558 static void bond_update_speed_duplex(struct slave *slave)
 559 {
 560         struct net_device *slave_dev = slave->dev;
 561         struct ethtool_cmd ecmd;
 562         u32 slave_speed;
 563         int res;
 564
 565         slave->speed = SPEED_UNKNOWN;
 566         slave->duplex = DUPLEX_UNKNOWN;
 567
 568         res = __ethtool_get_settings(slave_dev, &ecmd);
 569         if (res < 0)
 570                 return;
 571
 572         slave_speed = ethtool_cmd_speed(&ecmd);
 573         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 574                 return;
 575
 576         switch (ecmd.duplex) {
 577         case DUPLEX_FULL:
 578         case DUPLEX_HALF:
 579                 break;
 580         default:
 581                 return;
 582         }
 583
 584         slave->speed = slave_speed;
 585         slave->duplex = ecmd.duplex;
 586
 587         return;
 588 }
 589
 590 /*
 591  * if <dev> supports MII link status reporting, check its link status.
 592  *
 593  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 594  * depending upon the setting of the use_carrier parameter.
 595  *
 596  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 597  * can't tell and just pretend it is), or 0, meaning that the link is
 598  * down.
 599  *
 600  * If reporting is non-zero, instead of faking link up, return -1 if
 601  * both ETHTOOL and MII ioctls fail (meaning the device does not
 602  * support them).  If use_carrier is set, return whatever it says.
 603  * It'd be nice if there was a good way to tell if a driver supports
 604  * netif_carrier, but there really isn't.
 605  */
 606 static int bond_check_dev_link(struct bonding *bond,
 607                                struct net_device *slave_dev, int reporting)
 608 {
 609         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 610         int (*ioctl)(struct net_device *, struct ifreq *, int);
 611         struct ifreq ifr;
 612         struct mii_ioctl_data *mii;
 613
 614         if (!reporting && !netif_running(slave_dev))
 615                 return 0;
 616
 617         if (bond->params.use_carrier)
 618                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 619
 620         /* Try to get link status using Ethtool first. */
 621         if (slave_dev->ethtool_ops->get_link)
 622                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 623                         BMSR_LSTATUS : 0;
 624
 625         /* Ethtool can't be used, fallback to MII ioctls. */
 626         ioctl = slave_ops->ndo_do_ioctl;
 627         if (ioctl) {
 628                 /* TODO: set pointer to correct ioctl on a per team member */
 629                 /*       bases to make this more efficient. that is, once  */
 630                 /*       we determine the correct ioctl, we will always    */
 631                 /*       call it and not the others for that team          */
 632                 /*       member.                                           */
 633
 634                 /*
 635                  * We cannot assume that SIOCGMIIPHY will also read a
 636                  * register; not all network drivers (e.g., e100)
 637                  * support that.
 638                  */
 639
 640                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 641                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 642                 mii = if_mii(&ifr);
 643                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 644                         mii->reg_num = MII_BMSR;
 645                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 646                                 return mii->val_out & BMSR_LSTATUS;
 647                 }
 648         }
 649
 650         /*
 651          * If reporting, report that either there's no dev->do_ioctl,
 652          * or both SIOCGMIIREG and get_link failed (meaning that we
 653          * cannot report link status).  If not reporting, pretend
 654          * we're ok.
 655          */
 656         return reporting ? -1 : BMSR_LSTATUS;
 657 }
 658
 659 /*----------------------------- Multicast list ------------------------------*/
 660
 661 /*
 662  * Push the promiscuity flag down to appropriate slaves
 663  */
 664 static int bond_set_promiscuity(struct bonding *bond, int inc)
 665 {
 666         int err = 0;
 667         if (USES_PRIMARY(bond->params.mode)) {
 668                 /* write lock already acquired */
 669                 if (bond->curr_active_slave) {
 670                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 671                                                   inc);
 672                 }
 673         } else {
 674                 struct slave *slave;
 675                 int i;
 676                 bond_for_each_slave(bond, slave, i) {
 677                         err = dev_set_promiscuity(slave->dev, inc);
 678                         if (err)
 679                                 return err;
 680                 }
 681         }
 682         return err;
 683 }
 684
 685 /*
 686  * Push the allmulti flag down to all slaves
 687  */
 688 static int bond_set_allmulti(struct bonding *bond, int inc)
 689 {
 690         int err = 0;
 691         if (USES_PRIMARY(bond->params.mode)) {
 692                 /* write lock already acquired */
 693                 if (bond->curr_active_slave) {
 694                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 695                                                inc);
 696                 }
 697         } else {
 698                 struct slave *slave;
 699                 int i;
 700                 bond_for_each_slave(bond, slave, i) {
 701                         err = dev_set_allmulti(slave->dev, inc);
 702                         if (err)
 703                                 return err;
 704                 }
 705         }
 706         return err;
 707 }
 708
 709 /*
 710  * Add a Multicast address to slaves
 711  * according to mode
 712  */
 713 static void bond_mc_add(struct bonding *bond, void *addr)
 714 {
 715         if (USES_PRIMARY(bond->params.mode)) {
 716                 /* write lock already acquired */
 717                 if (bond->curr_active_slave)
 718                         dev_mc_add(bond->curr_active_slave->dev, addr);
 719         } else {
 720                 struct slave *slave;
 721                 int i;
 722
 723                 bond_for_each_slave(bond, slave, i)
 724                         dev_mc_add(slave->dev, addr);
 725         }
 726 }
 727
 728 /*
 729  * Remove a multicast address from slave
 730  * according to mode
 731  */
 732 static void bond_mc_del(struct bonding *bond, void *addr)
 733 {
 734         if (USES_PRIMARY(bond->params.mode)) {
 735                 /* write lock already acquired */
 736                 if (bond->curr_active_slave)
 737                         dev_mc_del(bond->curr_active_slave->dev, addr);
 738         } else {
 739                 struct slave *slave;
 740                 int i;
 741                 bond_for_each_slave(bond, slave, i) {
 742                         dev_mc_del(slave->dev, addr);
 743                 }
 744         }
 745 }
 746
 747
 748 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 749 {
 750         struct in_device *in_dev;
 751
 752         in_dev = __in_dev_get_rcu(dev);
 753         if (in_dev)
 754                 ip_mc_rejoin_groups(in_dev);
 755 }
 756
 757 /*
 758  * Retrieve the list of registered multicast addresses for the bonding
 759  * device and retransmit an IGMP JOIN request to the current active
 760  * slave.
 761  */
 762 static void bond_resend_igmp_join_requests(struct bonding *bond)
 763 {
 764         struct net_device *bond_dev, *vlan_dev, *upper_dev;
 765         struct vlan_entry *vlan;
 766
 767         read_lock(&bond->lock);
 768         rcu_read_lock();
 769
 770         bond_dev = bond->dev;
 771
 772         /* rejoin all groups on bond device */
 773         __bond_resend_igmp_join_requests(bond_dev);
 774
 775         /*
 776          * if bond is enslaved to a bridge,
 777          * then rejoin all groups on its master
 778          */
 779         upper_dev = netdev_master_upper_dev_get_rcu(bond_dev);
 780         if (upper_dev && upper_dev->priv_flags & IFF_EBRIDGE)
 781                 __bond_resend_igmp_join_requests(upper_dev);
 782
 783         /* rejoin all groups on vlan devices */
 784         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 785                 vlan_dev = __vlan_find_dev_deep(bond_dev, htons(ETH_P_8021Q),
 786                                                 vlan->vlan_id);
 787                 if (vlan_dev)
 788                         __bond_resend_igmp_join_requests(vlan_dev);
 789         }
 790         rcu_read_unlock();
 791
 792         /* We use curr_slave_lock to protect against concurrent access to
 793          * igmp_retrans from multiple running instances of this function and
 794          * bond_change_active_slave
 795          */
 796         write_lock_bh(&bond->curr_slave_lock);
 797         if (bond->igmp_retrans > 1) {
 798                 bond->igmp_retrans--;
 799                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 800         }
 801         write_unlock_bh(&bond->curr_slave_lock);
 802         read_unlock(&bond->lock);
 803 }
 804
 805 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 806 {
 807         struct bonding *bond = container_of(work, struct bonding,
 808                                             mcast_work.work);
 809
 810         bond_resend_igmp_join_requests(bond);
 811 }
 812
 813 /*
 814  * flush all members of flush->mc_list from device dev->mc_list
 815  */
 816 static void bond_mc_list_flush(struct net_device *bond_dev,
 817                                struct net_device *slave_dev)
 818 {
 819         struct bonding *bond = netdev_priv(bond_dev);
 820         struct netdev_hw_addr *ha;
 821
 822         netdev_for_each_mc_addr(ha, bond_dev)
 823                 dev_mc_del(slave_dev, ha->addr);
 824
 825         if (bond->params.mode == BOND_MODE_8023AD) {
 826                 /* del lacpdu mc addr from mc list */
 827                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 828
 829                 dev_mc_del(slave_dev, lacpdu_multicast);
 830         }
 831 }
 832
 833 /*--------------------------- Active slave change ---------------------------*/
 834
 835 /*
 836  * Update the mc list and multicast-related flags for the new and
 837  * old active slaves (if any) according to the multicast mode, and
 838  * promiscuous flags unconditionally.
 839  */
 840 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 841                          struct slave *old_active)
 842 {
 843         struct netdev_hw_addr *ha;
 844
 845         if (!USES_PRIMARY(bond->params.mode))
 846                 /* nothing to do -  mc list is already up-to-date on
 847                  * all slaves
 848                  */
 849                 return;
 850
 851         if (old_active) {
 852                 if (bond->dev->flags & IFF_PROMISC)
 853                         dev_set_promiscuity(old_active->dev, -1);
 854
 855                 if (bond->dev->flags & IFF_ALLMULTI)
 856                         dev_set_allmulti(old_active->dev, -1);
 857
 858                 netif_addr_lock_bh(bond->dev);
 859                 netdev_for_each_mc_addr(ha, bond->dev)
 860                         dev_mc_del(old_active->dev, ha->addr);
 861                 netif_addr_unlock_bh(bond->dev);
 862         }
 863
 864         if (new_active) {
 865                 /* FIXME: Signal errors upstream. */
 866                 if (bond->dev->flags & IFF_PROMISC)
 867                         dev_set_promiscuity(new_active->dev, 1);
 868
 869                 if (bond->dev->flags & IFF_ALLMULTI)
 870                         dev_set_allmulti(new_active->dev, 1);
 871
 872                 netif_addr_lock_bh(bond->dev);
 873                 netdev_for_each_mc_addr(ha, bond->dev)
 874                         dev_mc_add(new_active->dev, ha->addr);
 875                 netif_addr_unlock_bh(bond->dev);
 876         }
 877 }
 878
 879 /*
 880  * bond_do_fail_over_mac
 881  *
 882  * Perform special MAC address swapping for fail_over_mac settings
 883  *
 884  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 885  */
 886 static void bond_do_fail_over_mac(struct bonding *bond,
 887                                   struct slave *new_active,
 888                                   struct slave *old_active)
 889         __releases(&bond->curr_slave_lock)
 890         __releases(&bond->lock)
 891         __acquires(&bond->lock)
 892         __acquires(&bond->curr_slave_lock)
 893 {
 894         u8 tmp_mac[ETH_ALEN];
 895         struct sockaddr saddr;
 896         int rv;
 897
 898         switch (bond->params.fail_over_mac) {
 899         case BOND_FOM_ACTIVE:
 900                 if (new_active) {
 901                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 902                                new_active->dev->addr_len);
 903                         write_unlock_bh(&bond->curr_slave_lock);
 904                         read_unlock(&bond->lock);
 905                         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
 906                         read_lock(&bond->lock);
 907                         write_lock_bh(&bond->curr_slave_lock);
 908                 }
 909                 break;
 910         case BOND_FOM_FOLLOW:
 911                 /*
 912                  * if new_active && old_active, swap them
 913                  * if just old_active, do nothing (going to no active slave)
 914                  * if just new_active, set new_active to bond's MAC
 915                  */
 916                 if (!new_active)
 917                         return;
 918
 919                 write_unlock_bh(&bond->curr_slave_lock);
 920                 read_unlock(&bond->lock);
 921
 922                 if (old_active) {
 923                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 924                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 925                                ETH_ALEN);
 926                         saddr.sa_family = new_active->dev->type;
 927                 } else {
 928                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 929                         saddr.sa_family = bond->dev->type;
 930                 }
 931
 932                 rv = dev_set_mac_address(new_active->dev, &saddr);
 933                 if (rv) {
 934                         pr_err("%s: Error %d setting MAC of slave %s\n",
 935                                bond->dev->name, -rv, new_active->dev->name);
 936                         goto out;
 937                 }
 938
 939                 if (!old_active)
 940                         goto out;
 941
 942                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 943                 saddr.sa_family = old_active->dev->type;
 944
 945                 rv = dev_set_mac_address(old_active->dev, &saddr);
 946                 if (rv)
 947                         pr_err("%s: Error %d setting MAC of slave %s\n",
 948                                bond->dev->name, -rv, new_active->dev->name);
 949 out:
 950                 read_lock(&bond->lock);
 951                 write_lock_bh(&bond->curr_slave_lock);
 952                 break;
 953         default:
 954                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 955                        bond->dev->name, bond->params.fail_over_mac);
 956                 break;
 957         }
 958
 959 }
 960
 961 static bool bond_should_change_active(struct bonding *bond)
 962 {
 963         struct slave *prim = bond->primary_slave;
 964         struct slave *curr = bond->curr_active_slave;
 965
 966         if (!prim || !curr || curr->link != BOND_LINK_UP)
 967                 return true;
 968         if (bond->force_primary) {
 969                 bond->force_primary = false;
 970                 return true;
 971         }
 972         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 973             (prim->speed < curr->speed ||
 974              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 975                 return false;
 976         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 977                 return false;
 978         return true;
 979 }
 980
 981 /**
 982  * find_best_interface - select the best available slave to be the active one
 983  * @bond: our bonding struct
 984  *
 985  * Warning: Caller must hold curr_slave_lock for writing.
 986  */
 987 static struct slave *bond_find_best_slave(struct bonding *bond)
 988 {
 989         struct slave *new_active, *old_active;
 990         struct slave *bestslave = NULL;
 991         int mintime = bond->params.updelay;
 992         int i;
 993
 994         new_active = bond->curr_active_slave;
 995
 996         if (!new_active) { /* there were no active slaves left */
 997                 if (bond->slave_cnt > 0)   /* found one slave */
 998                         new_active = bond->first_slave;
 999                 else
1000                         return NULL; /* still no slave, return NULL */
1001         }
1002
1003         if ((bond->primary_slave) &&
1004             bond->primary_slave->link == BOND_LINK_UP &&
1005             bond_should_change_active(bond)) {
1006                 new_active = bond->primary_slave;
1007         }
1008
1009         /* remember where to stop iterating over the slaves */
1010         old_active = new_active;
1011
1012         bond_for_each_slave_from(bond, new_active, i, old_active) {
1013                 if (new_active->link == BOND_LINK_UP) {
1014                         return new_active;
1015                 } else if (new_active->link == BOND_LINK_BACK &&
1016                            IS_UP(new_active->dev)) {
1017                         /* link up, but waiting for stabilization */
1018                         if (new_active->delay < mintime) {
1019                                 mintime = new_active->delay;
1020                                 bestslave = new_active;
1021                         }
1022                 }
1023         }
1024
1025         return bestslave;
1026 }
1027
1028 static bool bond_should_notify_peers(struct bonding *bond)
1029 {
1030         struct slave *slave = bond->curr_active_slave;
1031
1032         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1033                  bond->dev->name, slave ? slave->dev->name : "NULL");
1034
1035         if (!slave || !bond->send_peer_notif ||
1036             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1037                 return false;
1038
1039         bond->send_peer_notif--;
1040         return true;
1041 }
1042
1043 /**
1044  * change_active_interface - change the active slave into the specified one
1045  * @bond: our bonding struct
1046  * @new: the new slave to make the active one
1047  *
1048  * Set the new slave to the bond's settings and unset them on the old
1049  * curr_active_slave.
1050  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1051  *
1052  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1053  * because it is apparently the best available slave we have, even though its
1054  * updelay hasn't timed out yet.
1055  *
1056  * If new_active is not NULL, caller must hold bond->lock for read and
1057  * curr_slave_lock for write_bh.
1058  */
1059 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1060 {
1061         struct slave *old_active = bond->curr_active_slave;
1062
1063         if (old_active == new_active)
1064                 return;
1065
1066         if (new_active) {
1067                 new_active->jiffies = jiffies;
1068
1069                 if (new_active->link == BOND_LINK_BACK) {
1070                         if (USES_PRIMARY(bond->params.mode)) {
1071                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1072                                         bond->dev->name, new_active->dev->name,
1073                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1074                         }
1075
1076                         new_active->delay = 0;
1077                         new_active->link = BOND_LINK_UP;
1078
1079                         if (bond->params.mode == BOND_MODE_8023AD)
1080                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1081
1082                         if (bond_is_lb(bond))
1083                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1084                 } else {
1085                         if (USES_PRIMARY(bond->params.mode)) {
1086                                 pr_info("%s: making interface %s the new active one.\n",
1087                                         bond->dev->name, new_active->dev->name);
1088                         }
1089                 }
1090         }
1091
1092         if (USES_PRIMARY(bond->params.mode))
1093                 bond_mc_swap(bond, new_active, old_active);
1094
1095         if (bond_is_lb(bond)) {
1096                 bond_alb_handle_active_change(bond, new_active);
1097                 if (old_active)
1098                         bond_set_slave_inactive_flags(old_active);
1099                 if (new_active)
1100                         bond_set_slave_active_flags(new_active);
1101         } else {
1102                 bond->curr_active_slave = new_active;
1103         }
1104
1105         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1106                 if (old_active)
1107                         bond_set_slave_inactive_flags(old_active);
1108
1109                 if (new_active) {
1110                         bool should_notify_peers = false;
1111
1112                         bond_set_slave_active_flags(new_active);
1113
1114                         if (bond->params.fail_over_mac)
1115                                 bond_do_fail_over_mac(bond, new_active,
1116                                                       old_active);
1117
1118                         if (netif_running(bond->dev)) {
1119                                 bond->send_peer_notif =
1120                                         bond->params.num_peer_notif;
1121                                 should_notify_peers =
1122                                         bond_should_notify_peers(bond);
1123                         }
1124
1125                         write_unlock_bh(&bond->curr_slave_lock);
1126                         read_unlock(&bond->lock);
1127
1128                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1129                         if (should_notify_peers)
1130                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1131                                                          bond->dev);
1132
1133                         read_lock(&bond->lock);
1134                         write_lock_bh(&bond->curr_slave_lock);
1135                 }
1136         }
1137
1138         /* resend IGMP joins since active slave has changed or
1139          * all were sent on curr_active_slave.
1140          * resend only if bond is brought up with the affected
1141          * bonding modes and the retransmission is enabled */
1142         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1143             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1144              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1145                 bond->igmp_retrans = bond->params.resend_igmp;
1146                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1147         }
1148 }
1149
1150 /**
1151  * bond_select_active_slave - select a new active slave, if needed
1152  * @bond: our bonding struct
1153  *
1154  * This functions should be called when one of the following occurs:
1155  * - The old curr_active_slave has been released or lost its link.
1156  * - The primary_slave has got its link back.
1157  * - A slave has got its link back and there's no old curr_active_slave.
1158  *
1159  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1160  */
1161 void bond_select_active_slave(struct bonding *bond)
1162 {
1163         struct slave *best_slave;
1164         int rv;
1165
1166         best_slave = bond_find_best_slave(bond);
1167         if (best_slave != bond->curr_active_slave) {
1168                 bond_change_active_slave(bond, best_slave);
1169                 rv = bond_set_carrier(bond);
1170                 if (!rv)
1171                         return;
1172
1173                 if (netif_carrier_ok(bond->dev)) {
1174                         pr_info("%s: first active interface up!\n",
1175                                 bond->dev->name);
1176                 } else {
1177                         pr_info("%s: now running without any active interface !\n",
1178                                 bond->dev->name);
1179                 }
1180         }
1181 }
1182
1183 /*--------------------------- slave list handling ---------------------------*/
1184
1185 /*
1186  * This function attaches the slave to the end of list.
1187  *
1188  * bond->lock held for writing by caller.
1189  */
1190 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1191 {
1192         if (bond->first_slave == NULL) { /* attaching the first slave */
1193                 new_slave->next = new_slave;
1194                 new_slave->prev = new_slave;
1195                 bond->first_slave = new_slave;
1196         } else {
1197                 new_slave->next = bond->first_slave;
1198                 new_slave->prev = bond->first_slave->prev;
1199                 new_slave->next->prev = new_slave;
1200                 new_slave->prev->next = new_slave;
1201         }
1202
1203         bond->slave_cnt++;
1204 }
1205
1206 /*
1207  * This function detaches the slave from the list.
1208  * WARNING: no check is made to verify if the slave effectively
1209  * belongs to <bond>.
1210  * Nothing is freed on return, structures are just unchained.
1211  * If any slave pointer in bond was pointing to <slave>,
1212  * it should be changed by the calling function.
1213  *
1214  * bond->lock held for writing by caller.
1215  */
1216 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1217 {
1218         if (slave->next)
1219                 slave->next->prev = slave->prev;
1220
1221         if (slave->prev)
1222                 slave->prev->next = slave->next;
1223
1224         if (bond->first_slave == slave) { /* slave is the first slave */
1225                 if (bond->slave_cnt > 1) { /* there are more slave */
1226                         bond->first_slave = slave->next;
1227                 } else {
1228                         bond->first_slave = NULL; /* slave was the last one */
1229                 }
1230         }
1231
1232         slave->next = NULL;
1233         slave->prev = NULL;
1234         bond->slave_cnt--;
1235 }
1236
1237 #ifdef CONFIG_NET_POLL_CONTROLLER
1238 static inline int slave_enable_netpoll(struct slave *slave)
1239 {
1240         struct netpoll *np;
1241         int err = 0;
1242
1243         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1244         err = -ENOMEM;
1245         if (!np)
1246                 goto out;
1247
1248         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1249         if (err) {
1250                 kfree(np);
1251                 goto out;
1252         }
1253         slave->np = np;
1254 out:
1255         return err;
1256 }
1257 static inline void slave_disable_netpoll(struct slave *slave)
1258 {
1259         struct netpoll *np = slave->np;
1260
1261         if (!np)
1262                 return;
1263
1264         slave->np = NULL;
1265         __netpoll_free_async(np);
1266 }
1267 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1268 {
1269         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1270                 return false;
1271         if (!slave_dev->netdev_ops->ndo_poll_controller)
1272                 return false;
1273         return true;
1274 }
1275
1276 static void bond_poll_controller(struct net_device *bond_dev)
1277 {
1278 }
1279
1280 static void __bond_netpoll_cleanup(struct bonding *bond)
1281 {
1282         struct slave *slave;
1283         int i;
1284
1285         bond_for_each_slave(bond, slave, i)
1286                 if (IS_UP(slave->dev))
1287                         slave_disable_netpoll(slave);
1288 }
1289 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1290 {
1291         struct bonding *bond = netdev_priv(bond_dev);
1292
1293         read_lock(&bond->lock);
1294         __bond_netpoll_cleanup(bond);
1295         read_unlock(&bond->lock);
1296 }
1297
1298 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1299 {
1300         struct bonding *bond = netdev_priv(dev);
1301         struct slave *slave;
1302         int i, err = 0;
1303
1304         read_lock(&bond->lock);
1305         bond_for_each_slave(bond, slave, i) {
1306                 err = slave_enable_netpoll(slave);
1307                 if (err) {
1308                         __bond_netpoll_cleanup(bond);
1309                         break;
1310                 }
1311         }
1312         read_unlock(&bond->lock);
1313         return err;
1314 }
1315
1316 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1317 {
1318         return bond->dev->npinfo;
1319 }
1320
1321 #else
1322 static inline int slave_enable_netpoll(struct slave *slave)
1323 {
1324         return 0;
1325 }
1326 static inline void slave_disable_netpoll(struct slave *slave)
1327 {
1328 }
1329 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1330 {
1331 }
1332 #endif
1333
1334 /*---------------------------------- IOCTL ----------------------------------*/
1335
1336 static void bond_set_dev_addr(struct net_device *bond_dev,
1337                               struct net_device *slave_dev)
1338 {
1339         pr_debug("bond_dev=%p\n", bond_dev);
1340         pr_debug("slave_dev=%p\n", slave_dev);
1341         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1342         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1343         bond_dev->addr_assign_type = NET_ADDR_SET;
1344         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
1345 }
1346
1347 static netdev_features_t bond_fix_features(struct net_device *dev,
1348         netdev_features_t features)
1349 {
1350         struct slave *slave;
1351         struct bonding *bond = netdev_priv(dev);
1352         netdev_features_t mask;
1353         int i;
1354
1355         read_lock(&bond->lock);
1356
1357         if (!bond->first_slave) {
1358                 /* Disable adding VLANs to empty bond. But why? --mq */
1359                 features |= NETIF_F_VLAN_CHALLENGED;
1360                 goto out;
1361         }
1362
1363         mask = features;
1364         features &= ~NETIF_F_ONE_FOR_ALL;
1365         features |= NETIF_F_ALL_FOR_ALL;
1366
1367         bond_for_each_slave(bond, slave, i) {
1368                 features = netdev_increment_features(features,
1369                                                      slave->dev->features,
1370                                                      mask);
1371         }
1372         features = netdev_add_tso_features(features, mask);
1373
1374 out:
1375         read_unlock(&bond->lock);
1376         return features;
1377 }
1378
1379 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1380                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1381                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1382
1383 static void bond_compute_features(struct bonding *bond)
1384 {
1385         struct slave *slave;
1386         struct net_device *bond_dev = bond->dev;
1387         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1388         unsigned short max_hard_header_len = ETH_HLEN;
1389         unsigned int gso_max_size = GSO_MAX_SIZE;
1390         u16 gso_max_segs = GSO_MAX_SEGS;
1391         int i;
1392         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1393
1394         read_lock(&bond->lock);
1395
1396         if (!bond->first_slave)
1397                 goto done;
1398
1399         bond_for_each_slave(bond, slave, i) {
1400                 vlan_features = netdev_increment_features(vlan_features,
1401                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1402
1403                 dst_release_flag &= slave->dev->priv_flags;
1404                 if (slave->dev->hard_header_len > max_hard_header_len)
1405                         max_hard_header_len = slave->dev->hard_header_len;
1406
1407                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1408                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1409         }
1410
1411 done:
1412         bond_dev->vlan_features = vlan_features;
1413         bond_dev->hard_header_len = max_hard_header_len;
1414         bond_dev->gso_max_segs = gso_max_segs;
1415         netif_set_gso_max_size(bond_dev, gso_max_size);
1416
1417         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1418         bond_dev->priv_flags = flags | dst_release_flag;
1419
1420         read_unlock(&bond->lock);
1421
1422         netdev_change_features(bond_dev);
1423 }
1424
1425 static void bond_setup_by_slave(struct net_device *bond_dev,
1426                                 struct net_device *slave_dev)
1427 {
1428         struct bonding *bond = netdev_priv(bond_dev);
1429
1430         bond_dev->header_ops        = slave_dev->header_ops;
1431
1432         bond_dev->type              = slave_dev->type;
1433         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1434         bond_dev->addr_len          = slave_dev->addr_len;
1435
1436         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1437                 slave_dev->addr_len);
1438         bond->setup_by_slave = 1;
1439 }
1440
1441 /* On bonding slaves other than the currently active slave, suppress
1442  * duplicates except for alb non-mcast/bcast.
1443  */
1444 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1445                                             struct slave *slave,
1446                                             struct bonding *bond)
1447 {
1448         if (bond_is_slave_inactive(slave)) {
1449                 if (bond->params.mode == BOND_MODE_ALB &&
1450                     skb->pkt_type != PACKET_BROADCAST &&
1451                     skb->pkt_type != PACKET_MULTICAST)
1452                         return false;
1453                 return true;
1454         }
1455         return false;
1456 }
1457
1458 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1459 {
1460         struct sk_buff *skb = *pskb;
1461         struct slave *slave;
1462         struct bonding *bond;
1463         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1464                           struct slave *);
1465         int ret = RX_HANDLER_ANOTHER;
1466
1467         skb = skb_share_check(skb, GFP_ATOMIC);
1468         if (unlikely(!skb))
1469                 return RX_HANDLER_CONSUMED;
1470
1471         *pskb = skb;
1472
1473         slave = bond_slave_get_rcu(skb->dev);
1474         bond = slave->bond;
1475
1476         if (bond->params.arp_interval)
1477                 slave->dev->last_rx = jiffies;
1478
1479         recv_probe = ACCESS_ONCE(bond->recv_probe);
1480         if (recv_probe) {
1481                 ret = recv_probe(skb, bond, slave);
1482                 if (ret == RX_HANDLER_CONSUMED) {
1483                         consume_skb(skb);
1484                         return ret;
1485                 }
1486         }
1487
1488         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1489                 return RX_HANDLER_EXACT;
1490         }
1491
1492         skb->dev = bond->dev;
1493
1494         if (bond->params.mode == BOND_MODE_ALB &&
1495             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1496             skb->pkt_type == PACKET_HOST) {
1497
1498                 if (unlikely(skb_cow_head(skb,
1499                                           skb->data - skb_mac_header(skb)))) {
1500                         kfree_skb(skb);
1501                         return RX_HANDLER_CONSUMED;
1502                 }
1503                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1504         }
1505
1506         return ret;
1507 }
1508
1509 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1510                                       struct net_device *slave_dev)
1511 {
1512         int err;
1513
1514         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1515         if (err)
1516                 return err;
1517         slave_dev->flags |= IFF_SLAVE;
1518         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1519         return 0;
1520 }
1521
1522 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1523                                   struct net_device *slave_dev)
1524 {
1525         netdev_upper_dev_unlink(slave_dev, bond_dev);
1526         slave_dev->flags &= ~IFF_SLAVE;
1527         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1528 }
1529
1530 /* enslave device <slave> to bond device <master> */
1531 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1532 {
1533         struct bonding *bond = netdev_priv(bond_dev);
1534         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1535         struct slave *new_slave = NULL;
1536         struct netdev_hw_addr *ha;
1537         struct sockaddr addr;
1538         int link_reporting;
1539         int res = 0;
1540
1541         if (!bond->params.use_carrier &&
1542             slave_dev->ethtool_ops->get_link == NULL &&
1543             slave_ops->ndo_do_ioctl == NULL) {
1544                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1545                            bond_dev->name, slave_dev->name);
1546         }
1547
1548         /* already enslaved */
1549         if (slave_dev->flags & IFF_SLAVE) {
1550                 pr_debug("Error, Device was already enslaved\n");
1551                 return -EBUSY;
1552         }
1553
1554         /* vlan challenged mutual exclusion */
1555         /* no need to lock since we're protected by rtnl_lock */
1556         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1557                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1558                 if (vlan_uses_dev(bond_dev)) {
1559                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1560                                bond_dev->name, slave_dev->name, bond_dev->name);
1561                         return -EPERM;
1562                 } else {
1563                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1564                                    bond_dev->name, slave_dev->name,
1565                                    slave_dev->name, bond_dev->name);
1566                 }
1567         } else {
1568                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1569         }
1570
1571         /*
1572          * Old ifenslave binaries are no longer supported.  These can
1573          * be identified with moderate accuracy by the state of the slave:
1574          * the current ifenslave will set the interface down prior to
1575          * enslaving it; the old ifenslave will not.
1576          */
1577         if ((slave_dev->flags & IFF_UP)) {
1578                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1579                        slave_dev->name);
1580                 res = -EPERM;
1581                 goto err_undo_flags;
1582         }
1583
1584         /* set bonding device ether type by slave - bonding netdevices are
1585          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1586          * there is a need to override some of the type dependent attribs/funcs.
1587          *
1588          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1589          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1590          */
1591         if (bond->slave_cnt == 0) {
1592                 if (bond_dev->type != slave_dev->type) {
1593                         pr_debug("%s: change device type from %d to %d\n",
1594                                  bond_dev->name,
1595                                  bond_dev->type, slave_dev->type);
1596
1597                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1598                                                        bond_dev);
1599                         res = notifier_to_errno(res);
1600                         if (res) {
1601                                 pr_err("%s: refused to change device type\n",
1602                                        bond_dev->name);
1603                                 res = -EBUSY;
1604                                 goto err_undo_flags;
1605                         }
1606
1607                         /* Flush unicast and multicast addresses */
1608                         dev_uc_flush(bond_dev);
1609                         dev_mc_flush(bond_dev);
1610
1611                         if (slave_dev->type != ARPHRD_ETHER)
1612                                 bond_setup_by_slave(bond_dev, slave_dev);
1613                         else {
1614                                 ether_setup(bond_dev);
1615                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1616                         }
1617
1618                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1619                                                  bond_dev);
1620                 }
1621         } else if (bond_dev->type != slave_dev->type) {
1622                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1623                        slave_dev->name,
1624                        slave_dev->type, bond_dev->type);
1625                 res = -EINVAL;
1626                 goto err_undo_flags;
1627         }
1628
1629         if (slave_ops->ndo_set_mac_address == NULL) {
1630                 if (bond->slave_cnt == 0) {
1631                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1632                                    bond_dev->name);
1633                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1634                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1635                         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",
1636                                bond_dev->name);
1637                         res = -EOPNOTSUPP;
1638                         goto err_undo_flags;
1639                 }
1640         }
1641
1642         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1643
1644         /* If this is the first slave, then we need to set the master's hardware
1645          * address to be the same as the slave's. */
1646         if (bond->slave_cnt == 0 && bond->dev_addr_from_first)
1647                 bond_set_dev_addr(bond->dev, slave_dev);
1648
1649         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1650         if (!new_slave) {
1651                 res = -ENOMEM;
1652                 goto err_undo_flags;
1653         }
1654
1655         /*
1656          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1657          * is set via sysfs or module option if desired.
1658          */
1659         new_slave->queue_id = 0;
1660
1661         /* Save slave's original mtu and then set it to match the bond */
1662         new_slave->original_mtu = slave_dev->mtu;
1663         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1664         if (res) {
1665                 pr_debug("Error %d calling dev_set_mtu\n", res);
1666                 goto err_free;
1667         }
1668
1669         /*
1670          * Save slave's original ("permanent") mac address for modes
1671          * that need it, and for restoring it upon release, and then
1672          * set it to the master's address
1673          */
1674         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1675
1676         if (!bond->params.fail_over_mac) {
1677                 /*
1678                  * Set slave to master's mac address.  The application already
1679                  * set the master's mac address to that of the first slave
1680                  */
1681                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1682                 addr.sa_family = slave_dev->type;
1683                 res = dev_set_mac_address(slave_dev, &addr);
1684                 if (res) {
1685                         pr_debug("Error %d calling set_mac_address\n", res);
1686                         goto err_restore_mtu;
1687                 }
1688         }
1689
1690         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1691         if (res) {
1692                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1693                 goto err_restore_mac;
1694         }
1695
1696         /* open the slave since the application closed it */
1697         res = dev_open(slave_dev);
1698         if (res) {
1699                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1700                 goto err_unset_master;
1701         }
1702
1703         new_slave->bond = bond;
1704         new_slave->dev = slave_dev;
1705         slave_dev->priv_flags |= IFF_BONDING;
1706
1707         if (bond_is_lb(bond)) {
1708                 /* bond_alb_init_slave() must be called before all other stages since
1709                  * it might fail and we do not want to have to undo everything
1710                  */
1711                 res = bond_alb_init_slave(bond, new_slave);
1712                 if (res)
1713                         goto err_close;
1714         }
1715
1716         /* If the mode USES_PRIMARY, then the new slave gets the
1717          * master's promisc (and mc) settings only if it becomes the
1718          * curr_active_slave, and that is taken care of later when calling
1719          * bond_change_active()
1720          */
1721         if (!USES_PRIMARY(bond->params.mode)) {
1722                 /* set promiscuity level to new slave */
1723                 if (bond_dev->flags & IFF_PROMISC) {
1724                         res = dev_set_promiscuity(slave_dev, 1);
1725                         if (res)
1726                                 goto err_close;
1727                 }
1728
1729                 /* set allmulti level to new slave */
1730                 if (bond_dev->flags & IFF_ALLMULTI) {
1731                         res = dev_set_allmulti(slave_dev, 1);
1732                         if (res)
1733                                 goto err_close;
1734                 }
1735
1736                 netif_addr_lock_bh(bond_dev);
1737                 /* upload master's mc_list to new slave */
1738                 netdev_for_each_mc_addr(ha, bond_dev)
1739                         dev_mc_add(slave_dev, ha->addr);
1740                 netif_addr_unlock_bh(bond_dev);
1741         }
1742
1743         if (bond->params.mode == BOND_MODE_8023AD) {
1744                 /* add lacpdu mc addr to mc list */
1745                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1746
1747                 dev_mc_add(slave_dev, lacpdu_multicast);
1748         }
1749
1750         bond_add_vlans_on_slave(bond, slave_dev);
1751
1752         write_lock_bh(&bond->lock);
1753
1754         bond_attach_slave(bond, new_slave);
1755
1756         new_slave->delay = 0;
1757         new_slave->link_failure_count = 0;
1758
1759         write_unlock_bh(&bond->lock);
1760
1761         bond_compute_features(bond);
1762
1763         bond_update_speed_duplex(new_slave);
1764
1765         read_lock(&bond->lock);
1766
1767         new_slave->last_arp_rx = jiffies -
1768                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1769
1770         if (bond->params.miimon && !bond->params.use_carrier) {
1771                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1772
1773                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1774                         /*
1775                          * miimon is set but a bonded network driver
1776                          * does not support ETHTOOL/MII and
1777                          * arp_interval is not set.  Note: if
1778                          * use_carrier is enabled, we will never go
1779                          * here (because netif_carrier is always
1780                          * supported); thus, we don't need to change
1781                          * the messages for netif_carrier.
1782                          */
1783                         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",
1784                                bond_dev->name, slave_dev->name);
1785                 } else if (link_reporting == -1) {
1786                         /* unable get link status using mii/ethtool */
1787                         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",
1788                                    bond_dev->name, slave_dev->name);
1789                 }
1790         }
1791
1792         /* check for initial state */
1793         if (bond->params.miimon) {
1794                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1795                         if (bond->params.updelay) {
1796                                 new_slave->link = BOND_LINK_BACK;
1797                                 new_slave->delay = bond->params.updelay;
1798                         } else {
1799                                 new_slave->link = BOND_LINK_UP;
1800                         }
1801                 } else {
1802                         new_slave->link = BOND_LINK_DOWN;
1803                 }
1804         } else if (bond->params.arp_interval) {
1805                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1806                         BOND_LINK_UP : BOND_LINK_DOWN);
1807         } else {
1808                 new_slave->link = BOND_LINK_UP;
1809         }
1810
1811         if (new_slave->link != BOND_LINK_DOWN)
1812                 new_slave->jiffies = jiffies;
1813         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1814                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1815                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1816
1817         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1818                 /* if there is a primary slave, remember it */
1819                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1820                         bond->primary_slave = new_slave;
1821                         bond->force_primary = true;
1822                 }
1823         }
1824
1825         write_lock_bh(&bond->curr_slave_lock);
1826
1827         switch (bond->params.mode) {
1828         case BOND_MODE_ACTIVEBACKUP:
1829                 bond_set_slave_inactive_flags(new_slave);
1830                 bond_select_active_slave(bond);
1831                 break;
1832         case BOND_MODE_8023AD:
1833                 /* in 802.3ad mode, the internal mechanism
1834                  * will activate the slaves in the selected
1835                  * aggregator
1836                  */
1837                 bond_set_slave_inactive_flags(new_slave);
1838                 /* if this is the first slave */
1839                 if (bond->slave_cnt == 1) {
1840                         SLAVE_AD_INFO(new_slave).id = 1;
1841                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1842                          * can be called only after the mac address of the bond is set
1843                          */
1844                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1845                 } else {
1846                         SLAVE_AD_INFO(new_slave).id =
1847                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1848                 }
1849
1850                 bond_3ad_bind_slave(new_slave);
1851                 break;
1852         case BOND_MODE_TLB:
1853         case BOND_MODE_ALB:
1854                 bond_set_active_slave(new_slave);
1855                 bond_set_slave_inactive_flags(new_slave);
1856                 bond_select_active_slave(bond);
1857                 break;
1858         default:
1859                 pr_debug("This slave is always active in trunk mode\n");
1860
1861                 /* always active in trunk mode */
1862                 bond_set_active_slave(new_slave);
1863
1864                 /* In trunking mode there is little meaning to curr_active_slave
1865                  * anyway (it holds no special properties of the bond device),
1866                  * so we can change it without calling change_active_interface()
1867                  */
1868                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1869                         bond->curr_active_slave = new_slave;
1870
1871                 break;
1872         } /* switch(bond_mode) */
1873
1874         write_unlock_bh(&bond->curr_slave_lock);
1875
1876         bond_set_carrier(bond);
1877
1878 #ifdef CONFIG_NET_POLL_CONTROLLER
1879         slave_dev->npinfo = bond_netpoll_info(bond);
1880         if (slave_dev->npinfo) {
1881                 if (slave_enable_netpoll(new_slave)) {
1882                         read_unlock(&bond->lock);
1883                         pr_info("Error, %s: master_dev is using netpoll, "
1884                                  "but new slave device does not support netpoll.\n",
1885                                  bond_dev->name);
1886                         res = -EBUSY;
1887                         goto err_detach;
1888                 }
1889         }
1890 #endif
1891
1892         read_unlock(&bond->lock);
1893
1894         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1895         if (res)
1896                 goto err_detach;
1897
1898         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1899                                          new_slave);
1900         if (res) {
1901                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1902                 goto err_dest_symlinks;
1903         }
1904
1905         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1906                 bond_dev->name, slave_dev->name,
1907                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1908                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1909
1910         /* enslave is successful */
1911         return 0;
1912
1913 /* Undo stages on error */
1914 err_dest_symlinks:
1915         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1916
1917 err_detach:
1918         if (!USES_PRIMARY(bond->params.mode)) {
1919                 netif_addr_lock_bh(bond_dev);
1920                 bond_mc_list_flush(bond_dev, slave_dev);
1921                 netif_addr_unlock_bh(bond_dev);
1922         }
1923         bond_del_vlans_from_slave(bond, slave_dev);
1924         write_lock_bh(&bond->lock);
1925         bond_detach_slave(bond, new_slave);
1926         if (bond->primary_slave == new_slave)
1927                 bond->primary_slave = NULL;
1928         if (bond->curr_active_slave == new_slave) {
1929                 bond_change_active_slave(bond, NULL);
1930                 write_unlock_bh(&bond->lock);
1931                 read_lock(&bond->lock);
1932                 write_lock_bh(&bond->curr_slave_lock);
1933                 bond_select_active_slave(bond);
1934                 write_unlock_bh(&bond->curr_slave_lock);
1935                 read_unlock(&bond->lock);
1936         } else {
1937                 write_unlock_bh(&bond->lock);
1938         }
1939         slave_disable_netpoll(new_slave);
1940
1941 err_close:
1942         slave_dev->priv_flags &= ~IFF_BONDING;
1943         dev_close(slave_dev);
1944
1945 err_unset_master:
1946         bond_upper_dev_unlink(bond_dev, slave_dev);
1947
1948 err_restore_mac:
1949         if (!bond->params.fail_over_mac) {
1950                 /* XXX TODO - fom follow mode needs to change master's
1951                  * MAC if this slave's MAC is in use by the bond, or at
1952                  * least print a warning.
1953                  */
1954                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1955                 addr.sa_family = slave_dev->type;
1956                 dev_set_mac_address(slave_dev, &addr);
1957         }
1958
1959 err_restore_mtu:
1960         dev_set_mtu(slave_dev, new_slave->original_mtu);
1961
1962 err_free:
1963         kfree(new_slave);
1964
1965 err_undo_flags:
1966         bond_compute_features(bond);
1967         /* Enslave of first slave has failed and we need to fix master's mac */
1968         if (bond->slave_cnt == 0 &&
1969             ether_addr_equal(bond_dev->dev_addr, slave_dev->dev_addr))
1970                 eth_hw_addr_random(bond_dev);
1971
1972         return res;
1973 }
1974
1975 /*
1976  * Try to release the slave device <slave> from the bond device <master>
1977  * It is legal to access curr_active_slave without a lock because all the function
1978  * is write-locked. If "all" is true it means that the function is being called
1979  * while destroying a bond interface and all slaves are being released.
1980  *
1981  * The rules for slave state should be:
1982  *   for Active/Backup:
1983  *     Active stays on all backups go down
1984  *   for Bonded connections:
1985  *     The first up interface should be left on and all others downed.
1986  */
1987 static int __bond_release_one(struct net_device *bond_dev,
1988                               struct net_device *slave_dev,
1989                               bool all)
1990 {
1991         struct bonding *bond = netdev_priv(bond_dev);
1992         struct slave *slave, *oldcurrent;
1993         struct sockaddr addr;
1994         netdev_features_t old_features = bond_dev->features;
1995
1996         /* slave is not a slave or master is not master of this slave */
1997         if (!(slave_dev->flags & IFF_SLAVE) ||
1998             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1999                 pr_err("%s: Error: cannot release %s.\n",
2000                        bond_dev->name, slave_dev->name);
2001                 return -EINVAL;
2002         }
2003
2004         block_netpoll_tx();
2005         write_lock_bh(&bond->lock);
2006
2007         slave = bond_get_slave_by_dev(bond, slave_dev);
2008         if (!slave) {
2009                 /* not a slave of this bond */
2010                 pr_info("%s: %s not enslaved\n",
2011                         bond_dev->name, slave_dev->name);
2012                 write_unlock_bh(&bond->lock);
2013                 unblock_netpoll_tx();
2014                 return -EINVAL;
2015         }
2016
2017         write_unlock_bh(&bond->lock);
2018         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2019          * for this slave anymore.
2020          */
2021         netdev_rx_handler_unregister(slave_dev);
2022         write_lock_bh(&bond->lock);
2023
2024         if (!all && !bond->params.fail_over_mac) {
2025                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
2026                     bond->slave_cnt > 1)
2027                         pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
2028                                    bond_dev->name, slave_dev->name,
2029                                    slave->perm_hwaddr,
2030                                    bond_dev->name, slave_dev->name);
2031         }
2032
2033         /* Inform AD package of unbinding of slave. */
2034         if (bond->params.mode == BOND_MODE_8023AD) {
2035                 /* must be called before the slave is
2036                  * detached from the list
2037                  */
2038                 bond_3ad_unbind_slave(slave);
2039         }
2040
2041         pr_info("%s: releasing %s interface %s\n",
2042                 bond_dev->name,
2043                 bond_is_active_slave(slave) ? "active" : "backup",
2044                 slave_dev->name);
2045
2046         oldcurrent = bond->curr_active_slave;
2047
2048         bond->current_arp_slave = NULL;
2049
2050         /* release the slave from its bond */
2051         bond_detach_slave(bond, slave);
2052
2053         if (bond->primary_slave == slave)
2054                 bond->primary_slave = NULL;
2055
2056         if (oldcurrent == slave)
2057                 bond_change_active_slave(bond, NULL);
2058
2059         if (bond_is_lb(bond)) {
2060                 /* Must be called only after the slave has been
2061                  * detached from the list and the curr_active_slave
2062                  * has been cleared (if our_slave == old_current),
2063                  * but before a new active slave is selected.
2064                  */
2065                 write_unlock_bh(&bond->lock);
2066                 bond_alb_deinit_slave(bond, slave);
2067                 write_lock_bh(&bond->lock);
2068         }
2069
2070         if (all) {
2071                 bond->curr_active_slave = NULL;
2072         } else if (oldcurrent == slave) {
2073                 /*
2074                  * Note that we hold RTNL over this sequence, so there
2075                  * is no concern that another slave add/remove event
2076                  * will interfere.
2077                  */
2078                 write_unlock_bh(&bond->lock);
2079                 read_lock(&bond->lock);
2080                 write_lock_bh(&bond->curr_slave_lock);
2081
2082                 bond_select_active_slave(bond);
2083
2084                 write_unlock_bh(&bond->curr_slave_lock);
2085                 read_unlock(&bond->lock);
2086                 write_lock_bh(&bond->lock);
2087         }
2088
2089         if (bond->slave_cnt == 0) {
2090                 bond_set_carrier(bond);
2091                 eth_hw_addr_random(bond_dev);
2092                 bond->dev_addr_from_first = true;
2093
2094                 if (bond_vlan_used(bond)) {
2095                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2096                                    bond_dev->name, bond_dev->name);
2097                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2098                                    bond_dev->name);
2099                 }
2100         }
2101
2102         write_unlock_bh(&bond->lock);
2103         unblock_netpoll_tx();
2104
2105         if (bond->slave_cnt == 0) {
2106                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2107                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2108         }
2109
2110         bond_compute_features(bond);
2111         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2112             (old_features & NETIF_F_VLAN_CHALLENGED))
2113                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2114                         bond_dev->name, slave_dev->name, bond_dev->name);
2115
2116         /* must do this from outside any spinlocks */
2117         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2118
2119         bond_del_vlans_from_slave(bond, slave_dev);
2120
2121         /* If the mode USES_PRIMARY, then we should only remove its
2122          * promisc and mc settings if it was the curr_active_slave, but that was
2123          * already taken care of above when we detached the slave
2124          */
2125         if (!USES_PRIMARY(bond->params.mode)) {
2126                 /* unset promiscuity level from slave */
2127                 if (bond_dev->flags & IFF_PROMISC)
2128                         dev_set_promiscuity(slave_dev, -1);
2129
2130                 /* unset allmulti level from slave */
2131                 if (bond_dev->flags & IFF_ALLMULTI)
2132                         dev_set_allmulti(slave_dev, -1);
2133
2134                 /* flush master's mc_list from slave */
2135                 netif_addr_lock_bh(bond_dev);
2136                 bond_mc_list_flush(bond_dev, slave_dev);
2137                 netif_addr_unlock_bh(bond_dev);
2138         }
2139
2140         bond_upper_dev_unlink(bond_dev, slave_dev);
2141
2142         slave_disable_netpoll(slave);
2143
2144         /* close slave before restoring its mac address */
2145         dev_close(slave_dev);
2146
2147         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2148                 /* restore original ("permanent") mac address */
2149                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2150                 addr.sa_family = slave_dev->type;
2151                 dev_set_mac_address(slave_dev, &addr);
2152         }
2153
2154         dev_set_mtu(slave_dev, slave->original_mtu);
2155
2156         slave_dev->priv_flags &= ~IFF_BONDING;
2157
2158         kfree(slave);
2159
2160         return 0;  /* deletion OK */
2161 }
2162
2163 /* A wrapper used because of ndo_del_link */
2164 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2165 {
2166         return __bond_release_one(bond_dev, slave_dev, false);
2167 }
2168
2169 /*
2170 * First release a slave and then destroy the bond if no more slaves are left.
2171 * Must be under rtnl_lock when this function is called.
2172 */
2173 static int  bond_release_and_destroy(struct net_device *bond_dev,
2174                                      struct net_device *slave_dev)
2175 {
2176         struct bonding *bond = netdev_priv(bond_dev);
2177         int ret;
2178
2179         ret = bond_release(bond_dev, slave_dev);
2180         if ((ret == 0) && (bond->slave_cnt == 0)) {
2181                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2182                 pr_info("%s: destroying bond %s.\n",
2183                         bond_dev->name, bond_dev->name);
2184                 unregister_netdevice(bond_dev);
2185         }
2186         return ret;
2187 }
2188
2189 /*
2190  * This function changes the active slave to slave <slave_dev>.
2191  * It returns -EINVAL in the following cases.
2192  *  - <slave_dev> is not found in the list.
2193  *  - There is not active slave now.
2194  *  - <slave_dev> is already active.
2195  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2196  *  - <slave_dev> is not running.
2197  * In these cases, this function does nothing.
2198  * In the other cases, current_slave pointer is changed and 0 is returned.
2199  */
2200 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2201 {
2202         struct bonding *bond = netdev_priv(bond_dev);
2203         struct slave *old_active = NULL;
2204         struct slave *new_active = NULL;
2205         int res = 0;
2206
2207         if (!USES_PRIMARY(bond->params.mode))
2208                 return -EINVAL;
2209
2210         /* Verify that bond_dev is indeed the master of slave_dev */
2211         if (!(slave_dev->flags & IFF_SLAVE) ||
2212             !netdev_has_upper_dev(slave_dev, bond_dev))
2213                 return -EINVAL;
2214
2215         read_lock(&bond->lock);
2216
2217         read_lock(&bond->curr_slave_lock);
2218         old_active = bond->curr_active_slave;
2219         read_unlock(&bond->curr_slave_lock);
2220
2221         new_active = bond_get_slave_by_dev(bond, slave_dev);
2222
2223         /*
2224          * Changing to the current active: do nothing; return success.
2225          */
2226         if (new_active && (new_active == old_active)) {
2227                 read_unlock(&bond->lock);
2228                 return 0;
2229         }
2230
2231         if ((new_active) &&
2232             (old_active) &&
2233             (new_active->link == BOND_LINK_UP) &&
2234             IS_UP(new_active->dev)) {
2235                 block_netpoll_tx();
2236                 write_lock_bh(&bond->curr_slave_lock);
2237                 bond_change_active_slave(bond, new_active);
2238                 write_unlock_bh(&bond->curr_slave_lock);
2239                 unblock_netpoll_tx();
2240         } else
2241                 res = -EINVAL;
2242
2243         read_unlock(&bond->lock);
2244
2245         return res;
2246 }
2247
2248 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2249 {
2250         struct bonding *bond = netdev_priv(bond_dev);
2251
2252         info->bond_mode = bond->params.mode;
2253         info->miimon = bond->params.miimon;
2254
2255         read_lock(&bond->lock);
2256         info->num_slaves = bond->slave_cnt;
2257         read_unlock(&bond->lock);
2258
2259         return 0;
2260 }
2261
2262 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2263 {
2264         struct bonding *bond = netdev_priv(bond_dev);
2265         struct slave *slave;
2266         int i, res = -ENODEV;
2267
2268         read_lock(&bond->lock);
2269
2270         bond_for_each_slave(bond, slave, i) {
2271                 if (i == (int)info->slave_id) {
2272                         res = 0;
2273                         strcpy(info->slave_name, slave->dev->name);
2274                         info->link = slave->link;
2275                         info->state = bond_slave_state(slave);
2276                         info->link_failure_count = slave->link_failure_count;
2277                         break;
2278                 }
2279         }
2280
2281         read_unlock(&bond->lock);
2282
2283         return res;
2284 }
2285
2286 /*-------------------------------- Monitoring -------------------------------*/
2287
2288
2289 static int bond_miimon_inspect(struct bonding *bond)
2290 {
2291         struct slave *slave;
2292         int i, link_state, commit = 0;
2293         bool ignore_updelay;
2294
2295         ignore_updelay = !bond->curr_active_slave ? true : false;
2296
2297         bond_for_each_slave(bond, slave, i) {
2298                 slave->new_link = BOND_LINK_NOCHANGE;
2299
2300                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2301
2302                 switch (slave->link) {
2303                 case BOND_LINK_UP:
2304                         if (link_state)
2305                                 continue;
2306
2307                         slave->link = BOND_LINK_FAIL;
2308                         slave->delay = bond->params.downdelay;
2309                         if (slave->delay) {
2310                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2311                                         bond->dev->name,
2312                                         (bond->params.mode ==
2313                                          BOND_MODE_ACTIVEBACKUP) ?
2314                                         (bond_is_active_slave(slave) ?
2315                                          "active " : "backup ") : "",
2316                                         slave->dev->name,
2317                                         bond->params.downdelay * bond->params.miimon);
2318                         }
2319                         /*FALLTHRU*/
2320                 case BOND_LINK_FAIL:
2321                         if (link_state) {
2322                                 /*
2323                                  * recovered before downdelay expired
2324                                  */
2325                                 slave->link = BOND_LINK_UP;
2326                                 slave->jiffies = jiffies;
2327                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2328                                         bond->dev->name,
2329                                         (bond->params.downdelay - slave->delay) *
2330                                         bond->params.miimon,
2331                                         slave->dev->name);
2332                                 continue;
2333                         }
2334
2335                         if (slave->delay <= 0) {
2336                                 slave->new_link = BOND_LINK_DOWN;
2337                                 commit++;
2338                                 continue;
2339                         }
2340
2341                         slave->delay--;
2342                         break;
2343
2344                 case BOND_LINK_DOWN:
2345                         if (!link_state)
2346                                 continue;
2347
2348                         slave->link = BOND_LINK_BACK;
2349                         slave->delay = bond->params.updelay;
2350
2351                         if (slave->delay) {
2352                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2353                                         bond->dev->name, slave->dev->name,
2354                                         ignore_updelay ? 0 :
2355                                         bond->params.updelay *
2356                                         bond->params.miimon);
2357                         }
2358                         /*FALLTHRU*/
2359                 case BOND_LINK_BACK:
2360                         if (!link_state) {
2361                                 slave->link = BOND_LINK_DOWN;
2362                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2363                                         bond->dev->name,
2364                                         (bond->params.updelay - slave->delay) *
2365                                         bond->params.miimon,
2366                                         slave->dev->name);
2367
2368                                 continue;
2369                         }
2370
2371                         if (ignore_updelay)
2372                                 slave->delay = 0;
2373
2374                         if (slave->delay <= 0) {
2375                                 slave->new_link = BOND_LINK_UP;
2376                                 commit++;
2377                                 ignore_updelay = false;
2378                                 continue;
2379                         }
2380
2381                         slave->delay--;
2382                         break;
2383                 }
2384         }
2385
2386         return commit;
2387 }
2388
2389 static void bond_miimon_commit(struct bonding *bond)
2390 {
2391         struct slave *slave;
2392         int i;
2393
2394         bond_for_each_slave(bond, slave, i) {
2395                 switch (slave->new_link) {
2396                 case BOND_LINK_NOCHANGE:
2397                         continue;
2398
2399                 case BOND_LINK_UP:
2400                         slave->link = BOND_LINK_UP;
2401                         slave->jiffies = jiffies;
2402
2403                         if (bond->params.mode == BOND_MODE_8023AD) {
2404                                 /* prevent it from being the active one */
2405                                 bond_set_backup_slave(slave);
2406                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2407                                 /* make it immediately active */
2408                                 bond_set_active_slave(slave);
2409                         } else if (slave != bond->primary_slave) {
2410                                 /* prevent it from being the active one */
2411                                 bond_set_backup_slave(slave);
2412                         }
2413
2414                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2415                                 bond->dev->name, slave->dev->name,
2416                                 slave->speed, slave->duplex ? "full" : "half");
2417
2418                         /* notify ad that the link status has changed */
2419                         if (bond->params.mode == BOND_MODE_8023AD)
2420                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2421
2422                         if (bond_is_lb(bond))
2423                                 bond_alb_handle_link_change(bond, slave,
2424                                                             BOND_LINK_UP);
2425
2426                         if (!bond->curr_active_slave ||
2427                             (slave == bond->primary_slave))
2428                                 goto do_failover;
2429
2430                         continue;
2431
2432                 case BOND_LINK_DOWN:
2433                         if (slave->link_failure_count < UINT_MAX)
2434                                 slave->link_failure_count++;
2435
2436                         slave->link = BOND_LINK_DOWN;
2437
2438                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2439                             bond->params.mode == BOND_MODE_8023AD)
2440                                 bond_set_slave_inactive_flags(slave);
2441
2442                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2443                                 bond->dev->name, slave->dev->name);
2444
2445                         if (bond->params.mode == BOND_MODE_8023AD)
2446                                 bond_3ad_handle_link_change(slave,
2447                                                             BOND_LINK_DOWN);
2448
2449                         if (bond_is_lb(bond))
2450                                 bond_alb_handle_link_change(bond, slave,
2451                                                             BOND_LINK_DOWN);
2452
2453                         if (slave == bond->curr_active_slave)
2454                                 goto do_failover;
2455
2456                         continue;
2457
2458                 default:
2459                         pr_err("%s: invalid new link %d on slave %s\n",
2460                                bond->dev->name, slave->new_link,
2461                                slave->dev->name);
2462                         slave->new_link = BOND_LINK_NOCHANGE;
2463
2464                         continue;
2465                 }
2466
2467 do_failover:
2468                 ASSERT_RTNL();
2469                 block_netpoll_tx();
2470                 write_lock_bh(&bond->curr_slave_lock);
2471                 bond_select_active_slave(bond);
2472                 write_unlock_bh(&bond->curr_slave_lock);
2473                 unblock_netpoll_tx();
2474         }
2475
2476         bond_set_carrier(bond);
2477 }
2478
2479 /*
2480  * bond_mii_monitor
2481  *
2482  * Really a wrapper that splits the mii monitor into two phases: an
2483  * inspection, then (if inspection indicates something needs to be done)
2484  * an acquisition of appropriate locks followed by a commit phase to
2485  * implement whatever link state changes are indicated.
2486  */
2487 void bond_mii_monitor(struct work_struct *work)
2488 {
2489         struct bonding *bond = container_of(work, struct bonding,
2490                                             mii_work.work);
2491         bool should_notify_peers = false;
2492         unsigned long delay;
2493
2494         read_lock(&bond->lock);
2495
2496         delay = msecs_to_jiffies(bond->params.miimon);
2497
2498         if (bond->slave_cnt == 0)
2499                 goto re_arm;
2500
2501         should_notify_peers = bond_should_notify_peers(bond);
2502
2503         if (bond_miimon_inspect(bond)) {
2504                 read_unlock(&bond->lock);
2505
2506                 /* Race avoidance with bond_close cancel of workqueue */
2507                 if (!rtnl_trylock()) {
2508                         read_lock(&bond->lock);
2509                         delay = 1;
2510                         should_notify_peers = false;
2511                         goto re_arm;
2512                 }
2513
2514                 read_lock(&bond->lock);
2515
2516                 bond_miimon_commit(bond);
2517
2518                 read_unlock(&bond->lock);
2519                 rtnl_unlock();  /* might sleep, hold no other locks */
2520                 read_lock(&bond->lock);
2521         }
2522
2523 re_arm:
2524         if (bond->params.miimon)
2525                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2526
2527         read_unlock(&bond->lock);
2528
2529         if (should_notify_peers) {
2530                 if (!rtnl_trylock()) {
2531                         read_lock(&bond->lock);
2532                         bond->send_peer_notif++;
2533                         read_unlock(&bond->lock);
2534                         return;
2535                 }
2536                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2537                 rtnl_unlock();
2538         }
2539 }
2540
2541 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2542 {
2543         struct vlan_entry *vlan;
2544         struct net_device *vlan_dev;
2545
2546         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2547                 return 1;
2548
2549         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2550                 rcu_read_lock();
2551                 vlan_dev = __vlan_find_dev_deep(bond->dev, htons(ETH_P_8021Q),
2552                                                 vlan->vlan_id);
2553                 rcu_read_unlock();
2554                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2555                         return 1;
2556         }
2557
2558         return 0;
2559 }
2560
2561 /*
2562  * We go to the (large) trouble of VLAN tagging ARP frames because
2563  * switches in VLAN mode (especially if ports are configured as
2564  * "native" to a VLAN) might not pass non-tagged frames.
2565  */
2566 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2567 {
2568         struct sk_buff *skb;
2569
2570         pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2571                  slave_dev->name, &dest_ip, &src_ip, vlan_id);
2572
2573         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2574                          NULL, slave_dev->dev_addr, NULL);
2575
2576         if (!skb) {
2577                 pr_err("ARP packet allocation failed\n");
2578                 return;
2579         }
2580         if (vlan_id) {
2581                 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2582                 if (!skb) {
2583                         pr_err("failed to insert VLAN tag\n");
2584                         return;
2585                 }
2586         }
2587         arp_xmit(skb);
2588 }
2589
2590
2591 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2592 {
2593         int i, vlan_id;
2594         __be32 *targets = bond->params.arp_targets;
2595         struct vlan_entry *vlan;
2596         struct net_device *vlan_dev = NULL;
2597         struct rtable *rt;
2598
2599         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2600                 __be32 addr;
2601                 if (!targets[i])
2602                         break;
2603                 pr_debug("basa: target %pI4\n", &targets[i]);
2604                 if (!bond_vlan_used(bond)) {
2605                         pr_debug("basa: empty vlan: arp_send\n");
2606                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2607                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2608                                       addr, 0);
2609                         continue;
2610                 }
2611
2612                 /*
2613                  * If VLANs are configured, we do a route lookup to
2614                  * determine which VLAN interface would be used, so we
2615                  * can tag the ARP with the proper VLAN tag.
2616                  */
2617                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2618                                      RTO_ONLINK, 0);
2619                 if (IS_ERR(rt)) {
2620                         if (net_ratelimit()) {
2621                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2622                                            bond->dev->name, &targets[i]);
2623                         }
2624                         continue;
2625                 }
2626
2627                 /*
2628                  * This target is not on a VLAN
2629                  */
2630                 if (rt->dst.dev == bond->dev) {
2631                         ip_rt_put(rt);
2632                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2633                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2634                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2635                                       addr, 0);
2636                         continue;
2637                 }
2638
2639                 vlan_id = 0;
2640                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2641                         rcu_read_lock();
2642                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2643                                                         htons(ETH_P_8021Q),
2644                                                         vlan->vlan_id);
2645                         rcu_read_unlock();
2646                         if (vlan_dev == rt->dst.dev) {
2647                                 vlan_id = vlan->vlan_id;
2648                                 pr_debug("basa: vlan match on %s %d\n",
2649                                        vlan_dev->name, vlan_id);
2650                                 break;
2651                         }
2652                 }
2653
2654                 if (vlan_id && vlan_dev) {
2655                         ip_rt_put(rt);
2656                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2657                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2658                                       addr, vlan_id);
2659                         continue;
2660                 }
2661
2662                 if (net_ratelimit()) {
2663                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2664                                    bond->dev->name, &targets[i],
2665                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2666                 }
2667                 ip_rt_put(rt);
2668         }
2669 }
2670
2671 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2672 {
2673         int i;
2674         __be32 *targets = bond->params.arp_targets;
2675
2676         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2677                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2678                          &sip, &tip, i, &targets[i],
2679                          bond_has_this_ip(bond, tip));
2680                 if (sip == targets[i]) {
2681                         if (bond_has_this_ip(bond, tip))
2682                                 slave->last_arp_rx = jiffies;
2683                         return;
2684                 }
2685         }
2686 }
2687
2688 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2689                         struct slave *slave)
2690 {
2691         struct arphdr *arp = (struct arphdr *)skb->data;
2692         unsigned char *arp_ptr;
2693         __be32 sip, tip;
2694         int alen;
2695
2696         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2697                 return RX_HANDLER_ANOTHER;
2698
2699         read_lock(&bond->lock);
2700         alen = arp_hdr_len(bond->dev);
2701
2702         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2703                  bond->dev->name, skb->dev->name);
2704
2705         if (alen > skb_headlen(skb)) {
2706                 arp = kmalloc(alen, GFP_ATOMIC);
2707                 if (!arp)
2708                         goto out_unlock;
2709                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2710                         goto out_unlock;
2711         }
2712
2713         if (arp->ar_hln != bond->dev->addr_len ||
2714             skb->pkt_type == PACKET_OTHERHOST ||
2715             skb->pkt_type == PACKET_LOOPBACK ||
2716             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2717             arp->ar_pro != htons(ETH_P_IP) ||
2718             arp->ar_pln != 4)
2719                 goto out_unlock;
2720
2721         arp_ptr = (unsigned char *)(arp + 1);
2722         arp_ptr += bond->dev->addr_len;
2723         memcpy(&sip, arp_ptr, 4);
2724         arp_ptr += 4 + bond->dev->addr_len;
2725         memcpy(&tip, arp_ptr, 4);
2726
2727         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2728                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2729                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2730                  &sip, &tip);
2731
2732         /*
2733          * Backup slaves won't see the ARP reply, but do come through
2734          * here for each ARP probe (so we swap the sip/tip to validate
2735          * the probe).  In a "redundant switch, common router" type of
2736          * configuration, the ARP probe will (hopefully) travel from
2737          * the active, through one switch, the router, then the other
2738          * switch before reaching the backup.
2739          */
2740         if (bond_is_active_slave(slave))
2741                 bond_validate_arp(bond, slave, sip, tip);
2742         else
2743                 bond_validate_arp(bond, slave, tip, sip);
2744
2745 out_unlock:
2746         read_unlock(&bond->lock);
2747         if (arp != (struct arphdr *)skb->data)
2748                 kfree(arp);
2749         return RX_HANDLER_ANOTHER;
2750 }
2751
2752 /*
2753  * this function is called regularly to monitor each slave's link
2754  * ensuring that traffic is being sent and received when arp monitoring
2755  * is used in load-balancing mode. if the adapter has been dormant, then an
2756  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2757  * arp monitoring in active backup mode.
2758  */
2759 void bond_loadbalance_arp_mon(struct work_struct *work)
2760 {
2761         struct bonding *bond = container_of(work, struct bonding,
2762                                             arp_work.work);
2763         struct slave *slave, *oldcurrent;
2764         int do_failover = 0;
2765         int delta_in_ticks, extra_ticks;
2766         int i;
2767
2768         read_lock(&bond->lock);
2769
2770         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2771         extra_ticks = delta_in_ticks / 2;
2772
2773         if (bond->slave_cnt == 0)
2774                 goto re_arm;
2775
2776         read_lock(&bond->curr_slave_lock);
2777         oldcurrent = bond->curr_active_slave;
2778         read_unlock(&bond->curr_slave_lock);
2779
2780         /* see if any of the previous devices are up now (i.e. they have
2781          * xmt and rcv traffic). the curr_active_slave does not come into
2782          * the picture unless it is null. also, slave->jiffies is not needed
2783          * here because we send an arp on each slave and give a slave as
2784          * long as it needs to get the tx/rx within the delta.
2785          * TODO: what about up/down delay in arp mode? it wasn't here before
2786          *       so it can wait
2787          */
2788         bond_for_each_slave(bond, slave, i) {
2789                 unsigned long trans_start = dev_trans_start(slave->dev);
2790
2791                 if (slave->link != BOND_LINK_UP) {
2792                         if (time_in_range(jiffies,
2793                                 trans_start - delta_in_ticks,
2794                                 trans_start + delta_in_ticks + extra_ticks) &&
2795                             time_in_range(jiffies,
2796                                 slave->dev->last_rx - delta_in_ticks,
2797                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2798
2799                                 slave->link  = BOND_LINK_UP;
2800                                 bond_set_active_slave(slave);
2801
2802                                 /* primary_slave has no meaning in round-robin
2803                                  * mode. the window of a slave being up and
2804                                  * curr_active_slave being null after enslaving
2805                                  * is closed.
2806                                  */
2807                                 if (!oldcurrent) {
2808                                         pr_info("%s: link status definitely up for interface %s, ",
2809                                                 bond->dev->name,
2810                                                 slave->dev->name);
2811                                         do_failover = 1;
2812                                 } else {
2813                                         pr_info("%s: interface %s is now up\n",
2814                                                 bond->dev->name,
2815                                                 slave->dev->name);
2816                                 }
2817                         }
2818                 } else {
2819                         /* slave->link == BOND_LINK_UP */
2820
2821                         /* not all switches will respond to an arp request
2822                          * when the source ip is 0, so don't take the link down
2823                          * if we don't know our ip yet
2824                          */
2825                         if (!time_in_range(jiffies,
2826                                 trans_start - delta_in_ticks,
2827                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2828                             !time_in_range(jiffies,
2829                                 slave->dev->last_rx - delta_in_ticks,
2830                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2831
2832                                 slave->link  = BOND_LINK_DOWN;
2833                                 bond_set_backup_slave(slave);
2834
2835                                 if (slave->link_failure_count < UINT_MAX)
2836                                         slave->link_failure_count++;
2837
2838                                 pr_info("%s: interface %s is now down.\n",
2839                                         bond->dev->name,
2840                                         slave->dev->name);
2841
2842                                 if (slave == oldcurrent)
2843                                         do_failover = 1;
2844                         }
2845                 }
2846
2847                 /* note: if switch is in round-robin mode, all links
2848                  * must tx arp to ensure all links rx an arp - otherwise
2849                  * links may oscillate or not come up at all; if switch is
2850                  * in something like xor mode, there is nothing we can
2851                  * do - all replies will be rx'ed on same link causing slaves
2852                  * to be unstable during low/no traffic periods
2853                  */
2854                 if (IS_UP(slave->dev))
2855                         bond_arp_send_all(bond, slave);
2856         }
2857
2858         if (do_failover) {
2859                 block_netpoll_tx();
2860                 write_lock_bh(&bond->curr_slave_lock);
2861
2862                 bond_select_active_slave(bond);
2863
2864                 write_unlock_bh(&bond->curr_slave_lock);
2865                 unblock_netpoll_tx();
2866         }
2867
2868 re_arm:
2869         if (bond->params.arp_interval)
2870                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2871
2872         read_unlock(&bond->lock);
2873 }
2874
2875 /*
2876  * Called to inspect slaves for active-backup mode ARP monitor link state
2877  * changes.  Sets new_link in slaves to specify what action should take
2878  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2879  * to link states must be committed.
2880  *
2881  * Called with bond->lock held for read.
2882  */
2883 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2884 {
2885         struct slave *slave;
2886         int i, commit = 0;
2887         unsigned long trans_start;
2888         int extra_ticks;
2889
2890         /* All the time comparisons below need some extra time. Otherwise, on
2891          * fast networks the ARP probe/reply may arrive within the same jiffy
2892          * as it was sent.  Then, the next time the ARP monitor is run, one
2893          * arp_interval will already have passed in the comparisons.
2894          */
2895         extra_ticks = delta_in_ticks / 2;
2896
2897         bond_for_each_slave(bond, slave, i) {
2898                 slave->new_link = BOND_LINK_NOCHANGE;
2899
2900                 if (slave->link != BOND_LINK_UP) {
2901                         if (time_in_range(jiffies,
2902                                 slave_last_rx(bond, slave) - delta_in_ticks,
2903                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2904
2905                                 slave->new_link = BOND_LINK_UP;
2906                                 commit++;
2907                         }
2908
2909                         continue;
2910                 }
2911
2912                 /*
2913                  * Give slaves 2*delta after being enslaved or made
2914                  * active.  This avoids bouncing, as the last receive
2915                  * times need a full ARP monitor cycle to be updated.
2916                  */
2917                 if (time_in_range(jiffies,
2918                                   slave->jiffies - delta_in_ticks,
2919                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2920                         continue;
2921
2922                 /*
2923                  * Backup slave is down if:
2924                  * - No current_arp_slave AND
2925                  * - more than 3*delta since last receive AND
2926                  * - the bond has an IP address
2927                  *
2928                  * Note: a non-null current_arp_slave indicates
2929                  * the curr_active_slave went down and we are
2930                  * searching for a new one; under this condition
2931                  * we only take the curr_active_slave down - this
2932                  * gives each slave a chance to tx/rx traffic
2933                  * before being taken out
2934                  */
2935                 if (!bond_is_active_slave(slave) &&
2936                     !bond->current_arp_slave &&
2937                     !time_in_range(jiffies,
2938                         slave_last_rx(bond, slave) - delta_in_ticks,
2939                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2940
2941                         slave->new_link = BOND_LINK_DOWN;
2942                         commit++;
2943                 }
2944
2945                 /*
2946                  * Active slave is down if:
2947                  * - more than 2*delta since transmitting OR
2948                  * - (more than 2*delta since receive AND
2949                  *    the bond has an IP address)
2950                  */
2951                 trans_start = dev_trans_start(slave->dev);
2952                 if (bond_is_active_slave(slave) &&
2953                     (!time_in_range(jiffies,
2954                         trans_start - delta_in_ticks,
2955                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
2956                      !time_in_range(jiffies,
2957                         slave_last_rx(bond, slave) - delta_in_ticks,
2958                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2959
2960                         slave->new_link = BOND_LINK_DOWN;
2961                         commit++;
2962                 }
2963         }
2964
2965         return commit;
2966 }
2967
2968 /*
2969  * Called to commit link state changes noted by inspection step of
2970  * active-backup mode ARP monitor.
2971  *
2972  * Called with RTNL and bond->lock for read.
2973  */
2974 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2975 {
2976         struct slave *slave;
2977         int i;
2978         unsigned long trans_start;
2979
2980         bond_for_each_slave(bond, slave, i) {
2981                 switch (slave->new_link) {
2982                 case BOND_LINK_NOCHANGE:
2983                         continue;
2984
2985                 case BOND_LINK_UP:
2986                         trans_start = dev_trans_start(slave->dev);
2987                         if ((!bond->curr_active_slave &&
2988                              time_in_range(jiffies,
2989                                            trans_start - delta_in_ticks,
2990                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2991                             bond->curr_active_slave != slave) {
2992                                 slave->link = BOND_LINK_UP;
2993                                 if (bond->current_arp_slave) {
2994                                         bond_set_slave_inactive_flags(
2995                                                 bond->current_arp_slave);
2996                                         bond->current_arp_slave = NULL;
2997                                 }
2998
2999                                 pr_info("%s: link status definitely up for interface %s.\n",
3000                                         bond->dev->name, slave->dev->name);
3001
3002                                 if (!bond->curr_active_slave ||
3003                                     (slave == bond->primary_slave))
3004                                         goto do_failover;
3005
3006                         }
3007
3008                         continue;
3009
3010                 case BOND_LINK_DOWN:
3011                         if (slave->link_failure_count < UINT_MAX)
3012                                 slave->link_failure_count++;
3013
3014                         slave->link = BOND_LINK_DOWN;
3015                         bond_set_slave_inactive_flags(slave);
3016
3017                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3018                                 bond->dev->name, slave->dev->name);
3019
3020                         if (slave == bond->curr_active_slave) {
3021                                 bond->current_arp_slave = NULL;
3022                                 goto do_failover;
3023                         }
3024
3025                         continue;
3026
3027                 default:
3028                         pr_err("%s: impossible: new_link %d on slave %s\n",
3029                                bond->dev->name, slave->new_link,
3030                                slave->dev->name);
3031                         continue;
3032                 }
3033
3034 do_failover:
3035                 ASSERT_RTNL();
3036                 block_netpoll_tx();
3037                 write_lock_bh(&bond->curr_slave_lock);
3038                 bond_select_active_slave(bond);
3039                 write_unlock_bh(&bond->curr_slave_lock);
3040                 unblock_netpoll_tx();
3041         }
3042
3043         bond_set_carrier(bond);
3044 }
3045
3046 /*
3047  * Send ARP probes for active-backup mode ARP monitor.
3048  *
3049  * Called with bond->lock held for read.
3050  */
3051 static void bond_ab_arp_probe(struct bonding *bond)
3052 {
3053         struct slave *slave;
3054         int i;
3055
3056         read_lock(&bond->curr_slave_lock);
3057
3058         if (bond->current_arp_slave && bond->curr_active_slave)
3059                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3060                         bond->current_arp_slave->dev->name,
3061                         bond->curr_active_slave->dev->name);
3062
3063         if (bond->curr_active_slave) {
3064                 bond_arp_send_all(bond, bond->curr_active_slave);
3065                 read_unlock(&bond->curr_slave_lock);
3066                 return;
3067         }
3068
3069         read_unlock(&bond->curr_slave_lock);
3070
3071         /* if we don't have a curr_active_slave, search for the next available
3072          * backup slave from the current_arp_slave and make it the candidate
3073          * for becoming the curr_active_slave
3074          */
3075
3076         if (!bond->current_arp_slave) {
3077                 bond->current_arp_slave = bond->first_slave;
3078                 if (!bond->current_arp_slave)
3079                         return;
3080         }
3081
3082         bond_set_slave_inactive_flags(bond->current_arp_slave);
3083
3084         /* search for next candidate */
3085         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3086                 if (IS_UP(slave->dev)) {
3087                         slave->link = BOND_LINK_BACK;
3088                         bond_set_slave_active_flags(slave);
3089                         bond_arp_send_all(bond, slave);
3090                         slave->jiffies = jiffies;
3091                         bond->current_arp_slave = slave;
3092                         break;
3093                 }
3094
3095                 /* if the link state is up at this point, we
3096                  * mark it down - this can happen if we have
3097                  * simultaneous link failures and
3098                  * reselect_active_interface doesn't make this
3099                  * one the current slave so it is still marked
3100                  * up when it is actually down
3101                  */
3102                 if (slave->link == BOND_LINK_UP) {
3103                         slave->link = BOND_LINK_DOWN;
3104                         if (slave->link_failure_count < UINT_MAX)
3105                                 slave->link_failure_count++;
3106
3107                         bond_set_slave_inactive_flags(slave);
3108
3109                         pr_info("%s: backup interface %s is now down.\n",
3110                                 bond->dev->name, slave->dev->name);
3111                 }
3112         }
3113 }
3114
3115 void bond_activebackup_arp_mon(struct work_struct *work)
3116 {
3117         struct bonding *bond = container_of(work, struct bonding,
3118                                             arp_work.work);
3119         bool should_notify_peers = false;
3120         int delta_in_ticks;
3121
3122         read_lock(&bond->lock);
3123
3124         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3125
3126         if (bond->slave_cnt == 0)
3127                 goto re_arm;
3128
3129         should_notify_peers = bond_should_notify_peers(bond);
3130
3131         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3132                 read_unlock(&bond->lock);
3133
3134                 /* Race avoidance with bond_close flush of workqueue */
3135                 if (!rtnl_trylock()) {
3136                         read_lock(&bond->lock);
3137                         delta_in_ticks = 1;
3138                         should_notify_peers = false;
3139                         goto re_arm;
3140                 }
3141
3142                 read_lock(&bond->lock);
3143
3144                 bond_ab_arp_commit(bond, delta_in_ticks);
3145
3146                 read_unlock(&bond->lock);
3147                 rtnl_unlock();
3148                 read_lock(&bond->lock);
3149         }
3150
3151         bond_ab_arp_probe(bond);
3152
3153 re_arm:
3154         if (bond->params.arp_interval)
3155                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3156
3157         read_unlock(&bond->lock);
3158
3159         if (should_notify_peers) {
3160                 if (!rtnl_trylock()) {
3161                         read_lock(&bond->lock);
3162                         bond->send_peer_notif++;
3163                         read_unlock(&bond->lock);
3164                         return;
3165                 }
3166                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3167                 rtnl_unlock();
3168         }
3169 }
3170
3171 /*-------------------------- netdev event handling --------------------------*/
3172
3173 /*
3174  * Change device name
3175  */
3176 static int bond_event_changename(struct bonding *bond)
3177 {
3178         bond_remove_proc_entry(bond);
3179         bond_create_proc_entry(bond);
3180
3181         bond_debug_reregister(bond);
3182
3183         return NOTIFY_DONE;
3184 }
3185
3186 static int bond_master_netdev_event(unsigned long event,
3187                                     struct net_device *bond_dev)
3188 {
3189         struct bonding *event_bond = netdev_priv(bond_dev);
3190
3191         switch (event) {
3192         case NETDEV_CHANGENAME:
3193                 return bond_event_changename(event_bond);
3194         case NETDEV_UNREGISTER:
3195                 bond_remove_proc_entry(event_bond);
3196                 break;
3197         case NETDEV_REGISTER:
3198                 bond_create_proc_entry(event_bond);
3199                 break;
3200         default:
3201                 break;
3202         }
3203
3204         return NOTIFY_DONE;
3205 }
3206
3207 static int bond_slave_netdev_event(unsigned long event,
3208                                    struct net_device *slave_dev)
3209 {
3210         struct slave *slave = bond_slave_get_rtnl(slave_dev);
3211         struct bonding *bond;
3212         struct net_device *bond_dev;
3213         u32 old_speed;
3214         u8 old_duplex;
3215
3216         /* A netdev event can be generated while enslaving a device
3217          * before netdev_rx_handler_register is called in which case
3218          * slave will be NULL
3219          */
3220         if (!slave)
3221                 return NOTIFY_DONE;
3222         bond_dev = slave->bond->dev;
3223         bond = slave->bond;
3224
3225         switch (event) {
3226         case NETDEV_UNREGISTER:
3227                 if (bond->setup_by_slave)
3228                         bond_release_and_destroy(bond_dev, slave_dev);
3229                 else
3230                         bond_release(bond_dev, slave_dev);
3231                 break;
3232         case NETDEV_UP:
3233         case NETDEV_CHANGE:
3234                 old_speed = slave->speed;
3235                 old_duplex = slave->duplex;
3236
3237                 bond_update_speed_duplex(slave);
3238
3239                 if (bond->params.mode == BOND_MODE_8023AD) {
3240                         if (old_speed != slave->speed)
3241                                 bond_3ad_adapter_speed_changed(slave);
3242                         if (old_duplex != slave->duplex)
3243                                 bond_3ad_adapter_duplex_changed(slave);
3244                 }
3245                 break;
3246         case NETDEV_DOWN:
3247                 /*
3248                  * ... Or is it this?
3249                  */
3250                 break;
3251         case NETDEV_CHANGEMTU:
3252                 /*
3253                  * TODO: Should slaves be allowed to
3254                  * independently alter their MTU?  For
3255                  * an active-backup bond, slaves need
3256                  * not be the same type of device, so
3257                  * MTUs may vary.  For other modes,
3258                  * slaves arguably should have the
3259                  * same MTUs. To do this, we'd need to
3260                  * take over the slave's change_mtu
3261                  * function for the duration of their
3262                  * servitude.
3263                  */
3264                 break;
3265         case NETDEV_CHANGENAME:
3266                 /*
3267                  * TODO: handle changing the primary's name
3268                  */
3269                 break;
3270         case NETDEV_FEAT_CHANGE:
3271                 bond_compute_features(bond);
3272                 break;
3273         default:
3274                 break;
3275         }
3276
3277         return NOTIFY_DONE;
3278 }
3279
3280 /*
3281  * bond_netdev_event: handle netdev notifier chain events.
3282  *
3283  * This function receives events for the netdev chain.  The caller (an
3284  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3285  * locks for us to safely manipulate the slave devices (RTNL lock,
3286  * dev_probe_lock).
3287  */
3288 static int bond_netdev_event(struct notifier_block *this,
3289                              unsigned long event, void *ptr)
3290 {
3291         struct net_device *event_dev = (struct net_device *)ptr;
3292
3293         pr_debug("event_dev: %s, event: %lx\n",
3294                  event_dev ? event_dev->name : "None",
3295                  event);
3296
3297         if (!(event_dev->priv_flags & IFF_BONDING))
3298                 return NOTIFY_DONE;
3299
3300         if (event_dev->flags & IFF_MASTER) {
3301                 pr_debug("IFF_MASTER\n");
3302                 return bond_master_netdev_event(event, event_dev);
3303         }
3304
3305         if (event_dev->flags & IFF_SLAVE) {
3306                 pr_debug("IFF_SLAVE\n");
3307                 return bond_slave_netdev_event(event, event_dev);
3308         }
3309
3310         return NOTIFY_DONE;
3311 }
3312
3313 static struct notifier_block bond_netdev_notifier = {
3314         .notifier_call = bond_netdev_event,
3315 };
3316
3317 /*---------------------------- Hashing Policies -----------------------------*/
3318
3319 /*
3320  * Hash for the output device based upon layer 2 data
3321  */
3322 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3323 {
3324         struct ethhdr *data = (struct ethhdr *)skb->data;
3325
3326         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3327                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3328
3329         return 0;
3330 }
3331
3332 /*
3333  * Hash for the output device based upon layer 2 and layer 3 data. If
3334  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3335  */
3336 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3337 {
3338         const struct ethhdr *data;
3339         const struct iphdr *iph;
3340         const struct ipv6hdr *ipv6h;
3341         u32 v6hash;
3342         const __be32 *s, *d;
3343
3344         if (skb->protocol == htons(ETH_P_IP) &&
3345             pskb_network_may_pull(skb, sizeof(*iph))) {
3346                 iph = ip_hdr(skb);
3347                 data = (struct ethhdr *)skb->data;
3348                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3349                         (data->h_dest[5] ^ data->h_source[5])) % count;
3350         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3351                    pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3352                 ipv6h = ipv6_hdr(skb);
3353                 data = (struct ethhdr *)skb->data;
3354                 s = &ipv6h->saddr.s6_addr32[0];
3355                 d = &ipv6h->daddr.s6_addr32[0];
3356                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3357                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3358                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3359         }
3360
3361         return bond_xmit_hash_policy_l2(skb, count);
3362 }
3363
3364 /*
3365  * Hash for the output device based upon layer 3 and layer 4 data. If
3366  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3367  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3368  */
3369 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3370 {
3371         u32 layer4_xor = 0;
3372         const struct iphdr *iph;
3373         const struct ipv6hdr *ipv6h;
3374         const __be32 *s, *d;
3375         const __be16 *l4 = NULL;
3376         __be16 _l4[2];
3377         int noff = skb_network_offset(skb);
3378         int poff;
3379
3380         if (skb->protocol == htons(ETH_P_IP) &&
3381             pskb_may_pull(skb, noff + sizeof(*iph))) {
3382                 iph = ip_hdr(skb);
3383                 poff = proto_ports_offset(iph->protocol);
3384
3385                 if (!ip_is_fragment(iph) && poff >= 0) {
3386                         l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3387                                                 sizeof(_l4), &_l4);
3388                         if (l4)
3389                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3390                 }
3391                 return (layer4_xor ^
3392                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3393         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3394                    pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3395                 ipv6h = ipv6_hdr(skb);
3396                 poff = proto_ports_offset(ipv6h->nexthdr);
3397                 if (poff >= 0) {
3398                         l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3399                                                 sizeof(_l4), &_l4);
3400                         if (l4)
3401                                 layer4_xor = ntohs(l4[0] ^ l4[1]);
3402                 }
3403                 s = &ipv6h->saddr.s6_addr32[0];
3404                 d = &ipv6h->daddr.s6_addr32[0];
3405                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3406                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3407                                (layer4_xor >> 8);
3408                 return layer4_xor % count;
3409         }
3410
3411         return bond_xmit_hash_policy_l2(skb, count);
3412 }
3413
3414 /*-------------------------- Device entry points ----------------------------*/
3415
3416 static void bond_work_init_all(struct bonding *bond)
3417 {
3418         INIT_DELAYED_WORK(&bond->mcast_work,
3419                           bond_resend_igmp_join_requests_delayed);
3420         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3421         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3422         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3423                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3424         else
3425                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3426         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3427 }
3428
3429 static void bond_work_cancel_all(struct bonding *bond)
3430 {
3431         cancel_delayed_work_sync(&bond->mii_work);
3432         cancel_delayed_work_sync(&bond->arp_work);
3433         cancel_delayed_work_sync(&bond->alb_work);
3434         cancel_delayed_work_sync(&bond->ad_work);
3435         cancel_delayed_work_sync(&bond->mcast_work);
3436 }
3437
3438 static int bond_open(struct net_device *bond_dev)
3439 {
3440         struct bonding *bond = netdev_priv(bond_dev);
3441         struct slave *slave;
3442         int i;
3443
3444         /* reset slave->backup and slave->inactive */
3445         read_lock(&bond->lock);
3446         if (bond->slave_cnt > 0) {
3447                 read_lock(&bond->curr_slave_lock);
3448                 bond_for_each_slave(bond, slave, i) {
3449                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3450                                 && (slave != bond->curr_active_slave)) {
3451                                 bond_set_slave_inactive_flags(slave);
3452                         } else {
3453                                 bond_set_slave_active_flags(slave);
3454                         }
3455                 }
3456                 read_unlock(&bond->curr_slave_lock);
3457         }
3458         read_unlock(&bond->lock);
3459
3460         bond_work_init_all(bond);
3461
3462         if (bond_is_lb(bond)) {
3463                 /* bond_alb_initialize must be called before the timer
3464                  * is started.
3465                  */
3466                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3467                         return -ENOMEM;
3468                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3469         }
3470
3471         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3472                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3473
3474         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3475                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3476                 if (bond->params.arp_validate)
3477                         bond->recv_probe = bond_arp_rcv;
3478         }
3479
3480         if (bond->params.mode == BOND_MODE_8023AD) {
3481                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3482                 /* register to receive LACPDUs */
3483                 bond->recv_probe = bond_3ad_lacpdu_recv;
3484                 bond_3ad_initiate_agg_selection(bond, 1);
3485         }
3486
3487         return 0;
3488 }
3489
3490 static int bond_close(struct net_device *bond_dev)
3491 {
3492         struct bonding *bond = netdev_priv(bond_dev);
3493
3494         write_lock_bh(&bond->lock);
3495         bond->send_peer_notif = 0;
3496         write_unlock_bh(&bond->lock);
3497
3498         bond_work_cancel_all(bond);
3499         if (bond_is_lb(bond)) {
3500                 /* Must be called only after all
3501                  * slaves have been released
3502                  */
3503                 bond_alb_deinitialize(bond);
3504         }
3505         bond->recv_probe = NULL;
3506
3507         return 0;
3508 }
3509
3510 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3511                                                 struct rtnl_link_stats64 *stats)
3512 {
3513         struct bonding *bond = netdev_priv(bond_dev);
3514         struct rtnl_link_stats64 temp;
3515         struct slave *slave;
3516         int i;
3517
3518         memset(stats, 0, sizeof(*stats));
3519
3520         read_lock_bh(&bond->lock);
3521
3522         bond_for_each_slave(bond, slave, i) {
3523                 const struct rtnl_link_stats64 *sstats =
3524                         dev_get_stats(slave->dev, &temp);
3525
3526                 stats->rx_packets += sstats->rx_packets;
3527                 stats->rx_bytes += sstats->rx_bytes;
3528                 stats->rx_errors += sstats->rx_errors;
3529                 stats->rx_dropped += sstats->rx_dropped;
3530
3531                 stats->tx_packets += sstats->tx_packets;
3532                 stats->tx_bytes += sstats->tx_bytes;
3533                 stats->tx_errors += sstats->tx_errors;
3534                 stats->tx_dropped += sstats->tx_dropped;
3535
3536                 stats->multicast += sstats->multicast;
3537                 stats->collisions += sstats->collisions;
3538
3539                 stats->rx_length_errors += sstats->rx_length_errors;
3540                 stats->rx_over_errors += sstats->rx_over_errors;
3541                 stats->rx_crc_errors += sstats->rx_crc_errors;
3542                 stats->rx_frame_errors += sstats->rx_frame_errors;
3543                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3544                 stats->rx_missed_errors += sstats->rx_missed_errors;
3545
3546                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3547                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3548                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3549                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3550                 stats->tx_window_errors += sstats->tx_window_errors;
3551         }
3552
3553         read_unlock_bh(&bond->lock);
3554
3555         return stats;
3556 }
3557
3558 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3559 {
3560         struct net_device *slave_dev = NULL;
3561         struct ifbond k_binfo;
3562         struct ifbond __user *u_binfo = NULL;
3563         struct ifslave k_sinfo;
3564         struct ifslave __user *u_sinfo = NULL;
3565         struct mii_ioctl_data *mii = NULL;
3566         struct net *net;
3567         int res = 0;
3568
3569         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3570
3571         switch (cmd) {
3572         case SIOCGMIIPHY:
3573                 mii = if_mii(ifr);
3574                 if (!mii)
3575                         return -EINVAL;
3576
3577                 mii->phy_id = 0;
3578                 /* Fall Through */
3579         case SIOCGMIIREG:
3580                 /*
3581                  * We do this again just in case we were called by SIOCGMIIREG
3582                  * instead of SIOCGMIIPHY.
3583                  */
3584                 mii = if_mii(ifr);
3585                 if (!mii)
3586                         return -EINVAL;
3587
3588
3589                 if (mii->reg_num == 1) {
3590                         struct bonding *bond = netdev_priv(bond_dev);
3591                         mii->val_out = 0;
3592                         read_lock(&bond->lock);
3593                         read_lock(&bond->curr_slave_lock);
3594                         if (netif_carrier_ok(bond->dev))
3595                                 mii->val_out = BMSR_LSTATUS;
3596
3597                         read_unlock(&bond->curr_slave_lock);
3598                         read_unlock(&bond->lock);
3599                 }
3600
3601                 return 0;
3602         case BOND_INFO_QUERY_OLD:
3603         case SIOCBONDINFOQUERY:
3604                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3605
3606                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3607                         return -EFAULT;
3608
3609                 res = bond_info_query(bond_dev, &k_binfo);
3610                 if (res == 0 &&
3611                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3612                         return -EFAULT;
3613
3614                 return res;
3615         case BOND_SLAVE_INFO_QUERY_OLD:
3616         case SIOCBONDSLAVEINFOQUERY:
3617                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3618
3619                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3620                         return -EFAULT;
3621
3622                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3623                 if (res == 0 &&
3624                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3625                         return -EFAULT;
3626
3627                 return res;
3628         default:
3629                 /* Go on */
3630                 break;
3631         }
3632
3633         net = dev_net(bond_dev);
3634
3635         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3636                 return -EPERM;
3637
3638         slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3639
3640         pr_debug("slave_dev=%p:\n", slave_dev);
3641
3642         if (!slave_dev)
3643                 res = -ENODEV;
3644         else {
3645                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3646                 switch (cmd) {
3647                 case BOND_ENSLAVE_OLD:
3648                 case SIOCBONDENSLAVE:
3649                         res = bond_enslave(bond_dev, slave_dev);
3650                         break;
3651                 case BOND_RELEASE_OLD:
3652                 case SIOCBONDRELEASE:
3653                         res = bond_release(bond_dev, slave_dev);
3654                         break;
3655                 case BOND_SETHWADDR_OLD:
3656                 case SIOCBONDSETHWADDR:
3657                         bond_set_dev_addr(bond_dev, slave_dev);
3658                         res = 0;
3659                         break;
3660                 case BOND_CHANGE_ACTIVE_OLD:
3661                 case SIOCBONDCHANGEACTIVE:
3662                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3663                         break;
3664                 default:
3665                         res = -EOPNOTSUPP;
3666                 }
3667
3668                 dev_put(slave_dev);
3669         }
3670
3671         return res;
3672 }
3673
3674 static bool bond_addr_in_mc_list(unsigned char *addr,
3675                                  struct netdev_hw_addr_list *list,
3676                                  int addrlen)
3677 {
3678         struct netdev_hw_addr *ha;
3679
3680         netdev_hw_addr_list_for_each(ha, list)
3681                 if (!memcmp(ha->addr, addr, addrlen))
3682                         return true;
3683
3684         return false;
3685 }
3686
3687 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3688 {
3689         struct bonding *bond = netdev_priv(bond_dev);
3690
3691         if (change & IFF_PROMISC)
3692                 bond_set_promiscuity(bond,
3693                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3694
3695         if (change & IFF_ALLMULTI)
3696                 bond_set_allmulti(bond,
3697                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3698 }
3699
3700 static void bond_set_multicast_list(struct net_device *bond_dev)
3701 {
3702         struct bonding *bond = netdev_priv(bond_dev);
3703         struct netdev_hw_addr *ha;
3704         bool found;
3705
3706         read_lock(&bond->lock);
3707
3708         /* looking for addresses to add to slaves' mc list */
3709         netdev_for_each_mc_addr(ha, bond_dev) {
3710                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3711                                              bond_dev->addr_len);
3712                 if (!found)
3713                         bond_mc_add(bond, ha->addr);
3714         }
3715
3716         /* looking for addresses to delete from slaves' list */
3717         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3718                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3719                                              bond_dev->addr_len);
3720                 if (!found)
3721                         bond_mc_del(bond, ha->addr);
3722         }
3723
3724         /* save master's multicast list */
3725         __hw_addr_flush(&bond->mc_list);
3726         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3727                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3728
3729         read_unlock(&bond->lock);
3730 }
3731
3732 static int bond_neigh_init(struct neighbour *n)
3733 {
3734         struct bonding *bond = netdev_priv(n->dev);
3735         struct slave *slave = bond->first_slave;
3736         const struct net_device_ops *slave_ops;
3737         struct neigh_parms parms;
3738         int ret;
3739
3740         if (!slave)
3741                 return 0;
3742
3743         slave_ops = slave->dev->netdev_ops;
3744
3745         if (!slave_ops->ndo_neigh_setup)
3746                 return 0;
3747
3748         parms.neigh_setup = NULL;
3749         parms.neigh_cleanup = NULL;
3750         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3751         if (ret)
3752                 return ret;
3753
3754         /*
3755          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3756          * after the last slave has been detached.  Assumes that all slaves
3757          * utilize the same neigh_cleanup (true at this writing as only user
3758          * is ipoib).
3759          */
3760         n->parms->neigh_cleanup = parms.neigh_cleanup;
3761
3762         if (!parms.neigh_setup)
3763                 return 0;
3764
3765         return parms.neigh_setup(n);
3766 }
3767
3768 /*
3769  * The bonding ndo_neigh_setup is called at init time beofre any
3770  * slave exists. So we must declare proxy setup function which will
3771  * be used at run time to resolve the actual slave neigh param setup.
3772  */
3773 static int bond_neigh_setup(struct net_device *dev,
3774                             struct neigh_parms *parms)
3775 {
3776         parms->neigh_setup   = bond_neigh_init;
3777
3778         return 0;
3779 }
3780
3781 /*
3782  * Change the MTU of all of a master's slaves to match the master
3783  */
3784 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3785 {
3786         struct bonding *bond = netdev_priv(bond_dev);
3787         struct slave *slave, *stop_at;
3788         int res = 0;
3789         int i;
3790
3791         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3792                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3793
3794         /* Can't hold bond->lock with bh disabled here since
3795          * some base drivers panic. On the other hand we can't
3796          * hold bond->lock without bh disabled because we'll
3797          * deadlock. The only solution is to rely on the fact
3798          * that we're under rtnl_lock here, and the slaves
3799          * list won't change. This doesn't solve the problem
3800          * of setting the slave's MTU while it is
3801          * transmitting, but the assumption is that the base
3802          * driver can handle that.
3803          *
3804          * TODO: figure out a way to safely iterate the slaves
3805          * list, but without holding a lock around the actual
3806          * call to the base driver.
3807          */
3808
3809         bond_for_each_slave(bond, slave, i) {
3810                 pr_debug("s %p s->p %p c_m %p\n",
3811                          slave,
3812                          slave->prev,
3813                          slave->dev->netdev_ops->ndo_change_mtu);
3814
3815                 res = dev_set_mtu(slave->dev, new_mtu);
3816
3817                 if (res) {
3818                         /* If we failed to set the slave's mtu to the new value
3819                          * we must abort the operation even in ACTIVE_BACKUP
3820                          * mode, because if we allow the backup slaves to have
3821                          * different mtu values than the active slave we'll
3822                          * need to change their mtu when doing a failover. That
3823                          * means changing their mtu from timer context, which
3824                          * is probably not a good idea.
3825                          */
3826                         pr_debug("err %d %s\n", res, slave->dev->name);
3827                         goto unwind;
3828                 }
3829         }
3830
3831         bond_dev->mtu = new_mtu;
3832
3833         return 0;
3834
3835 unwind:
3836         /* unwind from head to the slave that failed */
3837         stop_at = slave;
3838         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3839                 int tmp_res;
3840
3841                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3842                 if (tmp_res) {
3843                         pr_debug("unwind err %d dev %s\n",
3844                                  tmp_res, slave->dev->name);
3845                 }
3846         }
3847
3848         return res;
3849 }
3850
3851 /*
3852  * Change HW address
3853  *
3854  * Note that many devices must be down to change the HW address, and
3855  * downing the master releases all slaves.  We can make bonds full of
3856  * bonding devices to test this, however.
3857  */
3858 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3859 {
3860         struct bonding *bond = netdev_priv(bond_dev);
3861         struct sockaddr *sa = addr, tmp_sa;
3862         struct slave *slave, *stop_at;
3863         int res = 0;
3864         int i;
3865
3866         if (bond->params.mode == BOND_MODE_ALB)
3867                 return bond_alb_set_mac_address(bond_dev, addr);
3868
3869
3870         pr_debug("bond=%p, name=%s\n",
3871                  bond, bond_dev ? bond_dev->name : "None");
3872
3873         /*
3874          * If fail_over_mac is set to active, do nothing and return
3875          * success.  Returning an error causes ifenslave to fail.
3876          */
3877         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3878                 return 0;
3879
3880         if (!is_valid_ether_addr(sa->sa_data))
3881                 return -EADDRNOTAVAIL;
3882
3883         /* Can't hold bond->lock with bh disabled here since
3884          * some base drivers panic. On the other hand we can't
3885          * hold bond->lock without bh disabled because we'll
3886          * deadlock. The only solution is to rely on the fact
3887          * that we're under rtnl_lock here, and the slaves
3888          * list won't change. This doesn't solve the problem
3889          * of setting the slave's hw address while it is
3890          * transmitting, but the assumption is that the base
3891          * driver can handle that.
3892          *
3893          * TODO: figure out a way to safely iterate the slaves
3894          * list, but without holding a lock around the actual
3895          * call to the base driver.
3896          */
3897
3898         bond_for_each_slave(bond, slave, i) {
3899                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3900                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3901
3902                 if (slave_ops->ndo_set_mac_address == NULL) {
3903                         res = -EOPNOTSUPP;
3904                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3905                         goto unwind;
3906                 }
3907
3908                 res = dev_set_mac_address(slave->dev, addr);
3909                 if (res) {
3910                         /* TODO: consider downing the slave
3911                          * and retry ?
3912                          * User should expect communications
3913                          * breakage anyway until ARP finish
3914                          * updating, so...
3915                          */
3916                         pr_debug("err %d %s\n", res, slave->dev->name);
3917                         goto unwind;
3918                 }
3919         }
3920
3921         /* success */
3922         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3923         return 0;
3924
3925 unwind:
3926         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3927         tmp_sa.sa_family = bond_dev->type;
3928
3929         /* unwind from head to the slave that failed */
3930         stop_at = slave;
3931         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3932                 int tmp_res;
3933
3934                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3935                 if (tmp_res) {
3936                         pr_debug("unwind err %d dev %s\n",
3937                                  tmp_res, slave->dev->name);
3938                 }
3939         }
3940
3941         return res;
3942 }
3943
3944 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3945 {
3946         struct bonding *bond = netdev_priv(bond_dev);
3947         struct slave *slave, *start_at;
3948         int i, slave_no, res = 1;
3949         struct iphdr *iph = ip_hdr(skb);
3950
3951         /*
3952          * Start with the curr_active_slave that joined the bond as the
3953          * default for sending IGMP traffic.  For failover purposes one
3954          * needs to maintain some consistency for the interface that will
3955          * send the join/membership reports.  The curr_active_slave found
3956          * will send all of this type of traffic.
3957          */
3958         if ((iph->protocol == IPPROTO_IGMP) &&
3959             (skb->protocol == htons(ETH_P_IP))) {
3960
3961                 read_lock(&bond->curr_slave_lock);
3962                 slave = bond->curr_active_slave;
3963                 read_unlock(&bond->curr_slave_lock);
3964
3965                 if (!slave)
3966                         goto out;
3967         } else {
3968                 /*
3969                  * Concurrent TX may collide on rr_tx_counter; we accept
3970                  * that as being rare enough not to justify using an
3971                  * atomic op here.
3972                  */
3973                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3974
3975                 bond_for_each_slave(bond, slave, i) {
3976                         slave_no--;
3977                         if (slave_no < 0)
3978                                 break;
3979                 }
3980         }
3981
3982         start_at = slave;
3983         bond_for_each_slave_from(bond, slave, i, start_at) {
3984                 if (IS_UP(slave->dev) &&
3985                     (slave->link == BOND_LINK_UP) &&
3986                     bond_is_active_slave(slave)) {
3987                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
3988                         break;
3989                 }
3990         }
3991
3992 out:
3993         if (res) {
3994                 /* no suitable interface, frame not sent */
3995                 kfree_skb(skb);
3996         }
3997
3998         return NETDEV_TX_OK;
3999 }
4000
4001
4002 /*
4003  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4004  * the bond has a usable interface.
4005  */
4006 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4007 {
4008         struct bonding *bond = netdev_priv(bond_dev);
4009         int res = 1;
4010
4011         read_lock(&bond->curr_slave_lock);
4012
4013         if (bond->curr_active_slave)
4014                 res = bond_dev_queue_xmit(bond, skb,
4015                         bond->curr_active_slave->dev);
4016
4017         read_unlock(&bond->curr_slave_lock);
4018
4019         if (res)
4020                 /* no suitable interface, frame not sent */
4021                 kfree_skb(skb);
4022
4023         return NETDEV_TX_OK;
4024 }
4025
4026 /*
4027  * In bond_xmit_xor() , we determine the output device by using a pre-
4028  * determined xmit_hash_policy(), If the selected device is not enabled,
4029  * find the next active slave.
4030  */
4031 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4032 {
4033         struct bonding *bond = netdev_priv(bond_dev);
4034         struct slave *slave, *start_at;
4035         int slave_no;
4036         int i;
4037         int res = 1;
4038
4039         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4040
4041         bond_for_each_slave(bond, slave, i) {
4042                 slave_no--;
4043                 if (slave_no < 0)
4044                         break;
4045         }
4046
4047         start_at = slave;
4048
4049         bond_for_each_slave_from(bond, slave, i, start_at) {
4050                 if (IS_UP(slave->dev) &&
4051                     (slave->link == BOND_LINK_UP) &&
4052                     bond_is_active_slave(slave)) {
4053                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4054                         break;
4055                 }
4056         }
4057
4058         if (res) {
4059                 /* no suitable interface, frame not sent */
4060                 kfree_skb(skb);
4061         }
4062
4063         return NETDEV_TX_OK;
4064 }
4065
4066 /*
4067  * in broadcast mode, we send everything to all usable interfaces.
4068  */
4069 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4070 {
4071         struct bonding *bond = netdev_priv(bond_dev);
4072         struct slave *slave, *start_at;
4073         struct net_device *tx_dev = NULL;
4074         int i;
4075         int res = 1;
4076
4077         read_lock(&bond->curr_slave_lock);
4078         start_at = bond->curr_active_slave;
4079         read_unlock(&bond->curr_slave_lock);
4080
4081         if (!start_at)
4082                 goto out;
4083
4084         bond_for_each_slave_from(bond, slave, i, start_at) {
4085                 if (IS_UP(slave->dev) &&
4086                     (slave->link == BOND_LINK_UP) &&
4087                     bond_is_active_slave(slave)) {
4088                         if (tx_dev) {
4089                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4090                                 if (!skb2) {
4091                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4092                                                bond_dev->name);
4093                                         continue;
4094                                 }
4095
4096                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4097                                 if (res) {
4098                                         kfree_skb(skb2);
4099                                         continue;
4100                                 }
4101                         }
4102                         tx_dev = slave->dev;
4103                 }
4104         }
4105
4106         if (tx_dev)
4107                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4108
4109 out:
4110         if (res)
4111                 /* no suitable interface, frame not sent */
4112                 kfree_skb(skb);
4113
4114         /* frame sent to all suitable interfaces */
4115         return NETDEV_TX_OK;
4116 }
4117
4118 /*------------------------- Device initialization ---------------------------*/
4119
4120 static void bond_set_xmit_hash_policy(struct bonding *bond)
4121 {
4122         switch (bond->params.xmit_policy) {
4123         case BOND_XMIT_POLICY_LAYER23:
4124                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4125                 break;
4126         case BOND_XMIT_POLICY_LAYER34:
4127                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4128                 break;
4129         case BOND_XMIT_POLICY_LAYER2:
4130         default:
4131                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4132                 break;
4133         }
4134 }
4135
4136 /*
4137  * Lookup the slave that corresponds to a qid
4138  */
4139 static inline int bond_slave_override(struct bonding *bond,
4140                                       struct sk_buff *skb)
4141 {
4142         int i, res = 1;
4143         struct slave *slave = NULL;
4144         struct slave *check_slave;
4145
4146         if (!skb->queue_mapping)
4147                 return 1;
4148
4149         /* Find out if any slaves have the same mapping as this skb. */
4150         bond_for_each_slave(bond, check_slave, i) {
4151                 if (check_slave->queue_id == skb->queue_mapping) {
4152                         slave = check_slave;
4153                         break;
4154                 }
4155         }
4156
4157         /* If the slave isn't UP, use default transmit policy. */
4158         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4159             (slave->link == BOND_LINK_UP)) {
4160                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4161         }
4162
4163         return res;
4164 }
4165
4166
4167 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4168 {
4169         /*
4170          * This helper function exists to help dev_pick_tx get the correct
4171          * destination queue.  Using a helper function skips a call to
4172          * skb_tx_hash and will put the skbs in the queue we expect on their
4173          * way down to the bonding driver.
4174          */
4175         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4176
4177         /*
4178          * Save the original txq to restore before passing to the driver
4179          */
4180         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4181
4182         if (unlikely(txq >= dev->real_num_tx_queues)) {
4183                 do {
4184                         txq -= dev->real_num_tx_queues;
4185                 } while (txq >= dev->real_num_tx_queues);
4186         }
4187         return txq;
4188 }
4189
4190 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4191 {
4192         struct bonding *bond = netdev_priv(dev);
4193
4194         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4195                 if (!bond_slave_override(bond, skb))
4196                         return NETDEV_TX_OK;
4197         }
4198
4199         switch (bond->params.mode) {
4200         case BOND_MODE_ROUNDROBIN:
4201                 return bond_xmit_roundrobin(skb, dev);
4202         case BOND_MODE_ACTIVEBACKUP:
4203                 return bond_xmit_activebackup(skb, dev);
4204         case BOND_MODE_XOR:
4205                 return bond_xmit_xor(skb, dev);
4206         case BOND_MODE_BROADCAST:
4207                 return bond_xmit_broadcast(skb, dev);
4208         case BOND_MODE_8023AD:
4209                 return bond_3ad_xmit_xor(skb, dev);
4210         case BOND_MODE_ALB:
4211         case BOND_MODE_TLB:
4212                 return bond_alb_xmit(skb, dev);
4213         default:
4214                 /* Should never happen, mode already checked */
4215                 pr_err("%s: Error: Unknown bonding mode %d\n",
4216                        dev->name, bond->params.mode);
4217                 WARN_ON_ONCE(1);
4218                 kfree_skb(skb);
4219                 return NETDEV_TX_OK;
4220         }
4221 }
4222
4223 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4224 {
4225         struct bonding *bond = netdev_priv(dev);
4226         netdev_tx_t ret = NETDEV_TX_OK;
4227
4228         /*
4229          * If we risk deadlock from transmitting this in the
4230          * netpoll path, tell netpoll to queue the frame for later tx
4231          */
4232         if (is_netpoll_tx_blocked(dev))
4233                 return NETDEV_TX_BUSY;
4234
4235         read_lock(&bond->lock);
4236
4237         if (bond->slave_cnt)
4238                 ret = __bond_start_xmit(skb, dev);
4239         else
4240                 kfree_skb(skb);
4241
4242         read_unlock(&bond->lock);
4243
4244         return ret;
4245 }
4246
4247 /*
4248  * set bond mode specific net device operations
4249  */
4250 void bond_set_mode_ops(struct bonding *bond, int mode)
4251 {
4252         struct net_device *bond_dev = bond->dev;
4253
4254         switch (mode) {
4255         case BOND_MODE_ROUNDROBIN:
4256                 break;
4257         case BOND_MODE_ACTIVEBACKUP:
4258                 break;
4259         case BOND_MODE_XOR:
4260                 bond_set_xmit_hash_policy(bond);
4261                 break;
4262         case BOND_MODE_BROADCAST:
4263                 break;
4264         case BOND_MODE_8023AD:
4265                 bond_set_xmit_hash_policy(bond);
4266                 break;
4267         case BOND_MODE_ALB:
4268                 /* FALLTHRU */
4269         case BOND_MODE_TLB:
4270                 break;
4271         default:
4272                 /* Should never happen, mode already checked */
4273                 pr_err("%s: Error: Unknown bonding mode %d\n",
4274                        bond_dev->name, mode);
4275                 break;
4276         }
4277 }
4278
4279 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4280                                      struct ethtool_cmd *ecmd)
4281 {
4282         struct bonding *bond = netdev_priv(bond_dev);
4283         struct slave *slave;
4284         int i;
4285         unsigned long speed = 0;
4286
4287         ecmd->duplex = DUPLEX_UNKNOWN;
4288         ecmd->port = PORT_OTHER;
4289
4290         /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
4291          * do not need to check mode.  Though link speed might not represent
4292          * the true receive or transmit bandwidth (not all modes are symmetric)
4293          * this is an accurate maximum.
4294          */
4295         read_lock(&bond->lock);
4296         bond_for_each_slave(bond, slave, i) {
4297                 if (SLAVE_IS_OK(slave)) {
4298                         if (slave->speed != SPEED_UNKNOWN)
4299                                 speed += slave->speed;
4300                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4301                             slave->duplex != DUPLEX_UNKNOWN)
4302                                 ecmd->duplex = slave->duplex;
4303                 }
4304         }
4305         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4306         read_unlock(&bond->lock);
4307         return 0;
4308 }
4309
4310 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4311                                      struct ethtool_drvinfo *drvinfo)
4312 {
4313         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4314         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4315         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4316                  BOND_ABI_VERSION);
4317 }
4318
4319 static const struct ethtool_ops bond_ethtool_ops = {
4320         .get_drvinfo            = bond_ethtool_get_drvinfo,
4321         .get_settings           = bond_ethtool_get_settings,
4322         .get_link               = ethtool_op_get_link,
4323 };
4324
4325 static const struct net_device_ops bond_netdev_ops = {
4326         .ndo_init               = bond_init,
4327         .ndo_uninit             = bond_uninit,
4328         .ndo_open               = bond_open,
4329         .ndo_stop               = bond_close,
4330         .ndo_start_xmit         = bond_start_xmit,
4331         .ndo_select_queue       = bond_select_queue,
4332         .ndo_get_stats64        = bond_get_stats,
4333         .ndo_do_ioctl           = bond_do_ioctl,
4334         .ndo_change_rx_flags    = bond_change_rx_flags,
4335         .ndo_set_rx_mode        = bond_set_multicast_list,
4336         .ndo_change_mtu         = bond_change_mtu,
4337         .ndo_set_mac_address    = bond_set_mac_address,
4338         .ndo_neigh_setup        = bond_neigh_setup,
4339         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4340         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4341 #ifdef CONFIG_NET_POLL_CONTROLLER
4342         .ndo_netpoll_setup      = bond_netpoll_setup,
4343         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4344         .ndo_poll_controller    = bond_poll_controller,
4345 #endif
4346         .ndo_add_slave          = bond_enslave,
4347         .ndo_del_slave          = bond_release,
4348         .ndo_fix_features       = bond_fix_features,
4349 };
4350
4351 static const struct device_type bond_type = {
4352         .name = "bond",
4353 };
4354
4355 static void bond_destructor(struct net_device *bond_dev)
4356 {
4357         struct bonding *bond = netdev_priv(bond_dev);
4358         if (bond->wq)
4359                 destroy_workqueue(bond->wq);
4360         free_netdev(bond_dev);
4361 }
4362
4363 static void bond_setup(struct net_device *bond_dev)
4364 {
4365         struct bonding *bond = netdev_priv(bond_dev);
4366
4367         /* initialize rwlocks */
4368         rwlock_init(&bond->lock);
4369         rwlock_init(&bond->curr_slave_lock);
4370
4371         bond->params = bonding_defaults;
4372
4373         /* Initialize pointers */
4374         bond->dev = bond_dev;
4375         INIT_LIST_HEAD(&bond->vlan_list);
4376
4377         /* Initialize the device entry points */
4378         ether_setup(bond_dev);
4379         bond_dev->netdev_ops = &bond_netdev_ops;
4380         bond_dev->ethtool_ops = &bond_ethtool_ops;
4381         bond_set_mode_ops(bond, bond->params.mode);
4382
4383         bond_dev->destructor = bond_destructor;
4384
4385         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4386
4387         /* Initialize the device options */
4388         bond_dev->tx_queue_len = 0;
4389         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4390         bond_dev->priv_flags |= IFF_BONDING;
4391         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4392
4393         /* At first, we block adding VLANs. That's the only way to
4394          * prevent problems that occur when adding VLANs over an
4395          * empty bond. The block will be removed once non-challenged
4396          * slaves are enslaved.
4397          */
4398         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4399
4400         /* don't acquire bond device's netif_tx_lock when
4401          * transmitting */
4402         bond_dev->features |= NETIF_F_LLTX;
4403
4404         /* By default, we declare the bond to be fully
4405          * VLAN hardware accelerated capable. Special
4406          * care is taken in the various xmit functions
4407          * when there are slaves that are not hw accel
4408          * capable
4409          */
4410
4411         bond_dev->hw_features = BOND_VLAN_FEATURES |
4412                                 NETIF_F_HW_VLAN_CTAG_TX |
4413                                 NETIF_F_HW_VLAN_CTAG_RX |
4414                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4415
4416         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4417         bond_dev->features |= bond_dev->hw_features;
4418 }
4419
4420 /*
4421 * Destroy a bonding device.
4422 * Must be under rtnl_lock when this function is called.
4423 */
4424 static void bond_uninit(struct net_device *bond_dev)
4425 {
4426         struct bonding *bond = netdev_priv(bond_dev);
4427         struct vlan_entry *vlan, *tmp;
4428
4429         bond_netpoll_cleanup(bond_dev);
4430
4431         /* Release the bonded slaves */
4432         while (bond->first_slave != NULL)
4433                 __bond_release_one(bond_dev, bond->first_slave->dev, true);
4434         pr_info("%s: released all slaves\n", bond_dev->name);
4435
4436         list_del(&bond->bond_list);
4437
4438         bond_debug_unregister(bond);
4439
4440         __hw_addr_flush(&bond->mc_list);
4441
4442         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4443                 list_del(&vlan->vlan_list);
4444                 kfree(vlan);
4445         }
4446 }
4447
4448 /*------------------------- Module initialization ---------------------------*/
4449
4450 /*
4451  * Convert string input module parms.  Accept either the
4452  * number of the mode or its string name.  A bit complicated because
4453  * some mode names are substrings of other names, and calls from sysfs
4454  * may have whitespace in the name (trailing newlines, for example).
4455  */
4456 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4457 {
4458         int modeint = -1, i, rv;
4459         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4460
4461         for (p = (char *)buf; *p; p++)
4462                 if (!(isdigit(*p) || isspace(*p)))
4463                         break;
4464
4465         if (*p)
4466                 rv = sscanf(buf, "%20s", modestr);
4467         else
4468                 rv = sscanf(buf, "%d", &modeint);
4469
4470         if (!rv)
4471                 return -1;
4472
4473         for (i = 0; tbl[i].modename; i++) {
4474                 if (modeint == tbl[i].mode)
4475                         return tbl[i].mode;
4476                 if (strcmp(modestr, tbl[i].modename) == 0)
4477                         return tbl[i].mode;
4478         }
4479
4480         return -1;
4481 }
4482
4483 static int bond_check_params(struct bond_params *params)
4484 {
4485         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4486
4487         /*
4488          * Convert string parameters.
4489          */
4490         if (mode) {
4491                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4492                 if (bond_mode == -1) {
4493                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4494                                mode == NULL ? "NULL" : mode);
4495                         return -EINVAL;
4496                 }
4497         }
4498
4499         if (xmit_hash_policy) {
4500                 if ((bond_mode != BOND_MODE_XOR) &&
4501                     (bond_mode != BOND_MODE_8023AD)) {
4502                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4503                                bond_mode_name(bond_mode));
4504                 } else {
4505                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4506                                                         xmit_hashtype_tbl);
4507                         if (xmit_hashtype == -1) {
4508                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4509                                        xmit_hash_policy == NULL ? "NULL" :
4510                                        xmit_hash_policy);
4511                                 return -EINVAL;
4512                         }
4513                 }
4514         }
4515
4516         if (lacp_rate) {
4517                 if (bond_mode != BOND_MODE_8023AD) {
4518                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4519                                 bond_mode_name(bond_mode));
4520                 } else {
4521                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4522                         if (lacp_fast == -1) {
4523                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4524                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4525                                 return -EINVAL;
4526                         }
4527                 }
4528         }
4529
4530         if (ad_select) {
4531                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4532                 if (params->ad_select == -1) {
4533                         pr_err("Error: Invalid ad_select \"%s\"\n",
4534                                ad_select == NULL ? "NULL" : ad_select);
4535                         return -EINVAL;
4536                 }
4537
4538                 if (bond_mode != BOND_MODE_8023AD) {
4539                         pr_warning("ad_select param only affects 802.3ad mode\n");
4540                 }
4541         } else {
4542                 params->ad_select = BOND_AD_STABLE;
4543         }
4544
4545         if (max_bonds < 0) {
4546                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4547                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4548                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4549         }
4550
4551         if (miimon < 0) {
4552                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4553                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4554                 miimon = BOND_LINK_MON_INTERV;
4555         }
4556
4557         if (updelay < 0) {
4558                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4559                            updelay, INT_MAX);
4560                 updelay = 0;
4561         }
4562
4563         if (downdelay < 0) {
4564                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4565                            downdelay, INT_MAX);
4566                 downdelay = 0;
4567         }
4568
4569         if ((use_carrier != 0) && (use_carrier != 1)) {
4570                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4571                            use_carrier);
4572                 use_carrier = 1;
4573         }
4574
4575         if (num_peer_notif < 0 || num_peer_notif > 255) {
4576                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4577                            num_peer_notif);
4578                 num_peer_notif = 1;
4579         }
4580
4581         /* reset values for 802.3ad */
4582         if (bond_mode == BOND_MODE_8023AD) {
4583                 if (!miimon) {
4584                         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");
4585                         pr_warning("Forcing miimon to 100msec\n");
4586                         miimon = 100;
4587                 }
4588         }
4589
4590         if (tx_queues < 1 || tx_queues > 255) {
4591                 pr_warning("Warning: tx_queues (%d) should be between "
4592                            "1 and 255, resetting to %d\n",
4593                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4594                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4595         }
4596
4597         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4598                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4599                            "not of valid value (0/1), so it was set to "
4600                            "0\n", all_slaves_active);
4601                 all_slaves_active = 0;
4602         }
4603
4604         if (resend_igmp < 0 || resend_igmp > 255) {
4605                 pr_warning("Warning: resend_igmp (%d) should be between "
4606                            "0 and 255, resetting to %d\n",
4607                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4608                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4609         }
4610
4611         /* reset values for TLB/ALB */
4612         if ((bond_mode == BOND_MODE_TLB) ||
4613             (bond_mode == BOND_MODE_ALB)) {
4614                 if (!miimon) {
4615                         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");
4616                         pr_warning("Forcing miimon to 100msec\n");
4617                         miimon = 100;
4618                 }
4619         }
4620
4621         if (bond_mode == BOND_MODE_ALB) {
4622                 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",
4623                           updelay);
4624         }
4625
4626         if (!miimon) {
4627                 if (updelay || downdelay) {
4628                         /* just warn the user the up/down delay will have
4629                          * no effect since miimon is zero...
4630                          */
4631                         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",
4632                                    updelay, downdelay);
4633                 }
4634         } else {
4635                 /* don't allow arp monitoring */
4636                 if (arp_interval) {
4637                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4638                                    miimon, arp_interval);
4639                         arp_interval = 0;
4640                 }
4641
4642                 if ((updelay % miimon) != 0) {
4643                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4644                                    updelay, miimon,
4645                                    (updelay / miimon) * miimon);
4646                 }
4647
4648                 updelay /= miimon;
4649
4650                 if ((downdelay % miimon) != 0) {
4651                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4652                                    downdelay, miimon,
4653                                    (downdelay / miimon) * miimon);
4654                 }
4655
4656                 downdelay /= miimon;
4657         }
4658
4659         if (arp_interval < 0) {
4660                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4661                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4662                 arp_interval = BOND_LINK_ARP_INTERV;
4663         }
4664
4665         for (arp_ip_count = 0, i = 0;
4666              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4667                 /* not complete check, but should be good enough to
4668                    catch mistakes */
4669                 __be32 ip = in_aton(arp_ip_target[i]);
4670                 if (!isdigit(arp_ip_target[i][0]) || ip == 0 ||
4671                     ip == htonl(INADDR_BROADCAST)) {
4672                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4673                                    arp_ip_target[i]);
4674                         arp_interval = 0;
4675                 } else {
4676                         arp_target[arp_ip_count++] = ip;
4677                 }
4678         }
4679
4680         if (arp_interval && !arp_ip_count) {
4681                 /* don't allow arping if no arp_ip_target given... */
4682                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4683                            arp_interval);
4684                 arp_interval = 0;
4685         }
4686
4687         if (arp_validate) {
4688                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4689                         pr_err("arp_validate only supported in active-backup mode\n");
4690                         return -EINVAL;
4691                 }
4692                 if (!arp_interval) {
4693                         pr_err("arp_validate requires arp_interval\n");
4694                         return -EINVAL;
4695                 }
4696
4697                 arp_validate_value = bond_parse_parm(arp_validate,
4698                                                      arp_validate_tbl);
4699                 if (arp_validate_value == -1) {
4700                         pr_err("Error: invalid arp_validate \"%s\"\n",
4701                                arp_validate == NULL ? "NULL" : arp_validate);
4702                         return -EINVAL;
4703                 }
4704         } else
4705                 arp_validate_value = 0;
4706
4707         if (miimon) {
4708                 pr_info("MII link monitoring set to %d ms\n", miimon);
4709         } else if (arp_interval) {
4710                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4711                         arp_interval,
4712                         arp_validate_tbl[arp_validate_value].modename,
4713                         arp_ip_count);
4714
4715                 for (i = 0; i < arp_ip_count; i++)
4716                         pr_info(" %s", arp_ip_target[i]);
4717
4718                 pr_info("\n");
4719
4720         } else if (max_bonds) {
4721                 /* miimon and arp_interval not set, we need one so things
4722                  * work as expected, see bonding.txt for details
4723                  */
4724                 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");
4725         }
4726
4727         if (primary && !USES_PRIMARY(bond_mode)) {
4728                 /* currently, using a primary only makes sense
4729                  * in active backup, TLB or ALB modes
4730                  */
4731                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4732                            primary, bond_mode_name(bond_mode));
4733                 primary = NULL;
4734         }
4735
4736         if (primary && primary_reselect) {
4737                 primary_reselect_value = bond_parse_parm(primary_reselect,
4738                                                          pri_reselect_tbl);
4739                 if (primary_reselect_value == -1) {
4740                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4741                                primary_reselect ==
4742                                         NULL ? "NULL" : primary_reselect);
4743                         return -EINVAL;
4744                 }
4745         } else {
4746                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4747         }
4748
4749         if (fail_over_mac) {
4750                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4751                                                       fail_over_mac_tbl);
4752                 if (fail_over_mac_value == -1) {
4753                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4754                                arp_validate == NULL ? "NULL" : arp_validate);
4755                         return -EINVAL;
4756                 }
4757
4758                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4759                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4760         } else {
4761                 fail_over_mac_value = BOND_FOM_NONE;
4762         }
4763
4764         /* fill params struct with the proper values */
4765         params->mode = bond_mode;
4766         params->xmit_policy = xmit_hashtype;
4767         params->miimon = miimon;
4768         params->num_peer_notif = num_peer_notif;
4769         params->arp_interval = arp_interval;
4770         params->arp_validate = arp_validate_value;
4771         params->updelay = updelay;
4772         params->downdelay = downdelay;
4773         params->use_carrier = use_carrier;
4774         params->lacp_fast = lacp_fast;
4775         params->primary[0] = 0;
4776         params->primary_reselect = primary_reselect_value;
4777         params->fail_over_mac = fail_over_mac_value;
4778         params->tx_queues = tx_queues;
4779         params->all_slaves_active = all_slaves_active;
4780         params->resend_igmp = resend_igmp;
4781         params->min_links = min_links;
4782
4783         if (primary) {
4784                 strncpy(params->primary, primary, IFNAMSIZ);
4785                 params->primary[IFNAMSIZ - 1] = 0;
4786         }
4787
4788         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4789
4790         return 0;
4791 }
4792
4793 static struct lock_class_key bonding_netdev_xmit_lock_key;
4794 static struct lock_class_key bonding_netdev_addr_lock_key;
4795 static struct lock_class_key bonding_tx_busylock_key;
4796
4797 static void bond_set_lockdep_class_one(struct net_device *dev,
4798                                        struct netdev_queue *txq,
4799                                        void *_unused)
4800 {
4801         lockdep_set_class(&txq->_xmit_lock,
4802                           &bonding_netdev_xmit_lock_key);
4803 }
4804
4805 static void bond_set_lockdep_class(struct net_device *dev)
4806 {
4807         lockdep_set_class(&dev->addr_list_lock,
4808                           &bonding_netdev_addr_lock_key);
4809         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4810         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4811 }
4812
4813 /*
4814  * Called from registration process
4815  */
4816 static int bond_init(struct net_device *bond_dev)
4817 {
4818         struct bonding *bond = netdev_priv(bond_dev);
4819         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4820         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4821
4822         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4823
4824         /*
4825          * Initialize locks that may be required during
4826          * en/deslave operations.  All of the bond_open work
4827          * (of which this is part) should really be moved to
4828          * a phase prior to dev_open
4829          */
4830         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4831         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4832
4833         bond->wq = create_singlethread_workqueue(bond_dev->name);
4834         if (!bond->wq)
4835                 return -ENOMEM;
4836
4837         bond_set_lockdep_class(bond_dev);
4838
4839         list_add_tail(&bond->bond_list, &bn->dev_list);
4840
4841         bond_prepare_sysfs_group(bond);
4842
4843         bond_debug_register(bond);
4844
4845         /* Ensure valid dev_addr */
4846         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4847             bond_dev->addr_assign_type == NET_ADDR_PERM) {
4848                 eth_hw_addr_random(bond_dev);
4849                 bond->dev_addr_from_first = true;
4850         }
4851
4852         __hw_addr_init(&bond->mc_list);
4853         return 0;
4854 }
4855
4856 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4857 {
4858         if (tb[IFLA_ADDRESS]) {
4859                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4860                         return -EINVAL;
4861                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4862                         return -EADDRNOTAVAIL;
4863         }
4864         return 0;
4865 }
4866
4867 static unsigned int bond_get_num_tx_queues(void)
4868 {
4869         return tx_queues;
4870 }
4871
4872 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4873         .kind                   = "bond",
4874         .priv_size              = sizeof(struct bonding),
4875         .setup                  = bond_setup,
4876         .validate               = bond_validate,
4877         .get_num_tx_queues      = bond_get_num_tx_queues,
4878         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4879                                                              as for TX queues */
4880 };
4881
4882 /* Create a new bond based on the specified name and bonding parameters.
4883  * If name is NULL, obtain a suitable "bond%d" name for us.
4884  * Caller must NOT hold rtnl_lock; we need to release it here before we
4885  * set up our sysfs entries.
4886  */
4887 int bond_create(struct net *net, const char *name)
4888 {
4889         struct net_device *bond_dev;
4890         int res;
4891
4892         rtnl_lock();
4893
4894         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4895                                    name ? name : "bond%d",
4896                                    bond_setup, tx_queues);
4897         if (!bond_dev) {
4898                 pr_err("%s: eek! can't alloc netdev!\n", name);
4899                 rtnl_unlock();
4900                 return -ENOMEM;
4901         }
4902
4903         dev_net_set(bond_dev, net);
4904         bond_dev->rtnl_link_ops = &bond_link_ops;
4905
4906         res = register_netdevice(bond_dev);
4907
4908         netif_carrier_off(bond_dev);
4909
4910         rtnl_unlock();
4911         if (res < 0)
4912                 bond_destructor(bond_dev);
4913         return res;
4914 }
4915
4916 static int __net_init bond_net_init(struct net *net)
4917 {
4918         struct bond_net *bn = net_generic(net, bond_net_id);
4919
4920         bn->net = net;
4921         INIT_LIST_HEAD(&bn->dev_list);
4922
4923         bond_create_proc_dir(bn);
4924         bond_create_sysfs(bn);
4925
4926         return 0;
4927 }
4928
4929 static void __net_exit bond_net_exit(struct net *net)
4930 {
4931         struct bond_net *bn = net_generic(net, bond_net_id);
4932         struct bonding *bond, *tmp_bond;
4933         LIST_HEAD(list);
4934
4935         bond_destroy_sysfs(bn);
4936         bond_destroy_proc_dir(bn);
4937
4938         /* Kill off any bonds created after unregistering bond rtnl ops */
4939         rtnl_lock();
4940         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4941                 unregister_netdevice_queue(bond->dev, &list);
4942         unregister_netdevice_many(&list);
4943         rtnl_unlock();
4944 }
4945
4946 static struct pernet_operations bond_net_ops = {
4947         .init = bond_net_init,
4948         .exit = bond_net_exit,
4949         .id   = &bond_net_id,
4950         .size = sizeof(struct bond_net),
4951 };
4952
4953 static int __init bonding_init(void)
4954 {
4955         int i;
4956         int res;
4957
4958         pr_info("%s", bond_version);
4959
4960         res = bond_check_params(&bonding_defaults);
4961         if (res)
4962                 goto out;
4963
4964         res = register_pernet_subsys(&bond_net_ops);
4965         if (res)
4966                 goto out;
4967
4968         res = rtnl_link_register(&bond_link_ops);
4969         if (res)
4970                 goto err_link;
4971
4972         bond_create_debugfs();
4973
4974         for (i = 0; i < max_bonds; i++) {
4975                 res = bond_create(&init_net, NULL);
4976                 if (res)
4977                         goto err;
4978         }
4979
4980         register_netdevice_notifier(&bond_netdev_notifier);
4981 out:
4982         return res;
4983 err:
4984         rtnl_link_unregister(&bond_link_ops);
4985 err_link:
4986         unregister_pernet_subsys(&bond_net_ops);
4987         goto out;
4988
4989 }
4990
4991 static void __exit bonding_exit(void)
4992 {
4993         unregister_netdevice_notifier(&bond_netdev_notifier);
4994
4995         bond_destroy_debugfs();
4996
4997         rtnl_link_unregister(&bond_link_ops);
4998         unregister_pernet_subsys(&bond_net_ops);
4999
5000 #ifdef CONFIG_NET_POLL_CONTROLLER
5001         /*
5002          * Make sure we don't have an imbalance on our netpoll blocking
5003          */
5004         WARN_ON(atomic_read(&netpoll_block_tx));
5005 #endif
5006 }
5007
5008 module_init(bonding_init);
5009 module_exit(bonding_exit);
5010 MODULE_LICENSE("GPL");
5011 MODULE_VERSION(DRV_VERSION);
5012 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5013 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5014 MODULE_ALIAS_RTNL_LINK("bond");