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

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