drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  35
  36 #include <linux/kernel.h>
  37 #include <linux/module.h>
  38 #include <linux/types.h>
  39 #include <linux/fcntl.h>
  40 #include <linux/interrupt.h>
  41 #include <linux/ptrace.h>
  42 #include <linux/ioport.h>
  43 #include <linux/in.h>
  44 #include <net/ip.h>
  45 #include <linux/ip.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include "bonding.h"
  81 #include "bond_3ad.h"
  82 #include "bond_alb.h"
  83
  84 /*---------------------------- Module parameters ----------------------------*/
  85
  86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  87 #define BOND_LINK_MON_INTERV    0
  88 #define BOND_LINK_ARP_INTERV    0
  89
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon       = BOND_LINK_MON_INTERV;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval = BOND_LINK_ARP_INTERV;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *fail_over_mac;
 108 static int all_slaves_active = 0;
 109 static struct bond_params bonding_defaults;
 110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 111
 112 module_param(max_bonds, int, 0);
 113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 114 module_param(tx_queues, int, 0);
 115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 118                                "failover event (alias of num_unsol_na)");
 119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 121                                "failover event (alias of num_grat_arp)");
 122 module_param(miimon, int, 0);
 123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 124 module_param(updelay, int, 0);
 125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 126 module_param(downdelay, int, 0);
 127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 128                             "in milliseconds");
 129 module_param(use_carrier, int, 0);
 130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 131                               "0 for off, 1 for on (default)");
 132 module_param(mode, charp, 0);
 133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 134                        "1 for active-backup, 2 for balance-xor, "
 135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 136                        "6 for balance-alb");
 137 module_param(primary, charp, 0);
 138 MODULE_PARM_DESC(primary, "Primary network device to use");
 139 module_param(primary_reselect, charp, 0);
 140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 141                                    "once it comes up; "
 142                                    "0 for always (default), "
 143                                    "1 for only if speed of primary is "
 144                                    "better, "
 145                                    "2 for only on active slave "
 146                                    "failure");
 147 module_param(lacp_rate, charp, 0);
 148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 149                             "0 for slow, 1 for fast");
 150 module_param(ad_select, charp, 0);
 151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 152                             "0 for stable (default), 1 for bandwidth, "
 153                             "2 for count");
 154 module_param(min_links, int, 0);
 155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 156
 157 module_param(xmit_hash_policy, charp, 0);
 158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 159                                    "0 for layer 2 (default), 1 for layer 3+4, "
 160                                    "2 for layer 2+3");
 161 module_param(arp_interval, int, 0);
 162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 163 module_param_array(arp_ip_target, charp, NULL, 0);
 164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 165 module_param(arp_validate, charp, 0);
 166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 167                                "0 for none (default), 1 for active, "
 168                                "2 for backup, 3 for all");
 169 module_param(fail_over_mac, charp, 0);
 170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 171                                 "the same MAC; 0 for none (default), "
 172                                 "1 for active, 2 for follow");
 173 module_param(all_slaves_active, int, 0);
 174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
 175                                      "by setting active flag for all slaves; "
 176                                      "0 for never (default), 1 for always.");
 177 module_param(resend_igmp, int, 0);
 178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 179                               "link failure");
 180
 181 /*----------------------------- Global variables ----------------------------*/
 182
 183 #ifdef CONFIG_NET_POLL_CONTROLLER
 184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 185 #endif
 186
 187 int bond_net_id __read_mostly;
 188
 189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 190 static int arp_ip_count;
 191 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 193 static int lacp_fast;
 194
 195 const struct bond_parm_tbl bond_lacp_tbl[] = {
 196 {       "slow",         AD_LACP_SLOW},
 197 {       "fast",         AD_LACP_FAST},
 198 {       NULL,           -1},
 199 };
 200
 201 const struct bond_parm_tbl bond_mode_tbl[] = {
 202 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
 203 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
 204 {       "balance-xor",          BOND_MODE_XOR},
 205 {       "broadcast",            BOND_MODE_BROADCAST},
 206 {       "802.3ad",              BOND_MODE_8023AD},
 207 {       "balance-tlb",          BOND_MODE_TLB},
 208 {       "balance-alb",          BOND_MODE_ALB},
 209 {       NULL,                   -1},
 210 };
 211
 212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
 213 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
 214 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
 215 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
 216 {       NULL,                   -1},
 217 };
 218
 219 const struct bond_parm_tbl arp_validate_tbl[] = {
 220 {       "none",                 BOND_ARP_VALIDATE_NONE},
 221 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
 222 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
 223 {       "all",                  BOND_ARP_VALIDATE_ALL},
 224 {       NULL,                   -1},
 225 };
 226
 227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
 228 {       "none",                 BOND_FOM_NONE},
 229 {       "active",               BOND_FOM_ACTIVE},
 230 {       "follow",               BOND_FOM_FOLLOW},
 231 {       NULL,                   -1},
 232 };
 233
 234 const struct bond_parm_tbl pri_reselect_tbl[] = {
 235 {       "always",               BOND_PRI_RESELECT_ALWAYS},
 236 {       "better",               BOND_PRI_RESELECT_BETTER},
 237 {       "failure",              BOND_PRI_RESELECT_FAILURE},
 238 {       NULL,                   -1},
 239 };
 240
 241 struct bond_parm_tbl ad_select_tbl[] = {
 242 {       "stable",       BOND_AD_STABLE},
 243 {       "bandwidth",    BOND_AD_BANDWIDTH},
 244 {       "count",        BOND_AD_COUNT},
 245 {       NULL,           -1},
 246 };
 247
 248 /*-------------------------- Forward declarations ---------------------------*/
 249
 250 static int bond_init(struct net_device *bond_dev);
 251 static void bond_uninit(struct net_device *bond_dev);
 252
 253 /*---------------------------- General routines -----------------------------*/
 254
 255 const char *bond_mode_name(int mode)
 256 {
 257         static const char *names[] = {
 258                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 259                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 260                 [BOND_MODE_XOR] = "load balancing (xor)",
 261                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 262                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 263                 [BOND_MODE_TLB] = "transmit load balancing",
 264                 [BOND_MODE_ALB] = "adaptive load balancing",
 265         };
 266
 267         if (mode < 0 || mode > BOND_MODE_ALB)
 268                 return "unknown";
 269
 270         return names[mode];
 271 }
 272
 273 /*---------------------------------- VLAN -----------------------------------*/
 274
 275 /**
 276  * bond_add_vlan - add a new vlan id on bond
 277  * @bond: bond that got the notification
 278  * @vlan_id: the vlan id to add
 279  *
 280  * Returns -ENOMEM if allocation failed.
 281  */
 282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
 283 {
 284         struct vlan_entry *vlan;
 285
 286         pr_debug("bond: %s, vlan id %d\n",
 287                  (bond ? bond->dev->name : "None"), vlan_id);
 288
 289         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
 290         if (!vlan)
 291                 return -ENOMEM;
 292
 293         INIT_LIST_HEAD(&vlan->vlan_list);
 294         vlan->vlan_id = vlan_id;
 295
 296         write_lock_bh(&bond->lock);
 297
 298         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
 299
 300         write_unlock_bh(&bond->lock);
 301
 302         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
 303
 304         return 0;
 305 }
 306
 307 /**
 308  * bond_del_vlan - delete a vlan id from bond
 309  * @bond: bond that got the notification
 310  * @vlan_id: the vlan id to delete
 311  *
 312  * returns -ENODEV if @vlan_id was not found in @bond.
 313  */
 314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
 315 {
 316         struct vlan_entry *vlan;
 317         int res = -ENODEV;
 318
 319         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
 320
 321         block_netpoll_tx();
 322         write_lock_bh(&bond->lock);
 323
 324         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 325                 if (vlan->vlan_id == vlan_id) {
 326                         list_del(&vlan->vlan_list);
 327
 328                         if (bond_is_lb(bond))
 329                                 bond_alb_clear_vlan(bond, vlan_id);
 330
 331                         pr_debug("removed VLAN ID %d from bond %s\n",
 332                                  vlan_id, bond->dev->name);
 333
 334                         kfree(vlan);
 335
 336                         res = 0;
 337                         goto out;
 338                 }
 339         }
 340
 341         pr_debug("couldn't find VLAN ID %d in bond %s\n",
 342                  vlan_id, bond->dev->name);
 343
 344 out:
 345         write_unlock_bh(&bond->lock);
 346         unblock_netpoll_tx();
 347         return res;
 348 }
 349
 350 /**
 351  * bond_next_vlan - safely skip to the next item in the vlans list.
 352  * @bond: the bond we're working on
 353  * @curr: item we're advancing from
 354  *
 355  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
 356  * or @curr->next otherwise (even if it is @curr itself again).
 357  *
 358  * Caller must hold bond->lock
 359  */
 360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
 361 {
 362         struct vlan_entry *next, *last;
 363
 364         if (list_empty(&bond->vlan_list))
 365                 return NULL;
 366
 367         if (!curr) {
 368                 next = list_entry(bond->vlan_list.next,
 369                                   struct vlan_entry, vlan_list);
 370         } else {
 371                 last = list_entry(bond->vlan_list.prev,
 372                                   struct vlan_entry, vlan_list);
 373                 if (last == curr) {
 374                         next = list_entry(bond->vlan_list.next,
 375                                           struct vlan_entry, vlan_list);
 376                 } else {
 377                         next = list_entry(curr->vlan_list.next,
 378                                           struct vlan_entry, vlan_list);
 379                 }
 380         }
 381
 382         return next;
 383 }
 384
 385 /**
 386  * bond_dev_queue_xmit - Prepare skb for xmit.
 387  *
 388  * @bond: bond device that got this skb for tx.
 389  * @skb: hw accel VLAN tagged skb to transmit
 390  * @slave_dev: slave that is supposed to xmit this skbuff
 391  */
 392 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 393                         struct net_device *slave_dev)
 394 {
 395         skb->dev = slave_dev;
 396
 397         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 398                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 399         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 400
 401         if (unlikely(netpoll_tx_running(bond->dev)))
 402                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 403         else
 404                 dev_queue_xmit(skb);
 405
 406         return 0;
 407 }
 408
 409 /*
 410  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 411  * We don't protect the slave list iteration with a lock because:
 412  * a. This operation is performed in IOCTL context,
 413  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 414  * c. Holding a lock with BH disabled while directly calling a base driver
 415  *    entry point is generally a BAD idea.
 416  *
 417  * The design of synchronization/protection for this operation in the 8021q
 418  * module is good for one or more VLAN devices over a single physical device
 419  * and cannot be extended for a teaming solution like bonding, so there is a
 420  * potential race condition here where a net device from the vlan group might
 421  * be referenced (either by a base driver or the 8021q code) while it is being
 422  * removed from the system. However, it turns out we're not making matters
 423  * worse, and if it works for regular VLAN usage it will work here too.
 424 */
 425
 426 /**
 427  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 428  * @bond_dev: bonding net device that got called
 429  * @vid: vlan id being added
 430  */
 431 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
 432 {
 433         struct bonding *bond = netdev_priv(bond_dev);
 434         struct slave *slave, *stop_at;
 435         int i, res;
 436
 437         bond_for_each_slave(bond, slave, i) {
 438                 res = vlan_vid_add(slave->dev, vid);
 439                 if (res)
 440                         goto unwind;
 441         }
 442
 443         res = bond_add_vlan(bond, vid);
 444         if (res) {
 445                 pr_err("%s: Error: Failed to add vlan id %d\n",
 446                        bond_dev->name, vid);
 447                 return res;
 448         }
 449
 450         return 0;
 451
 452 unwind:
 453         /* unwind from head to the slave that failed */
 454         stop_at = slave;
 455         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
 456                 vlan_vid_del(slave->dev, vid);
 457
 458         return res;
 459 }
 460
 461 /**
 462  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 463  * @bond_dev: bonding net device that got called
 464  * @vid: vlan id being removed
 465  */
 466 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
 467 {
 468         struct bonding *bond = netdev_priv(bond_dev);
 469         struct slave *slave;
 470         int i, res;
 471
 472         bond_for_each_slave(bond, slave, i)
 473                 vlan_vid_del(slave->dev, vid);
 474
 475         res = bond_del_vlan(bond, vid);
 476         if (res) {
 477                 pr_err("%s: Error: Failed to remove vlan id %d\n",
 478                        bond_dev->name, vid);
 479                 return res;
 480         }
 481
 482         return 0;
 483 }
 484
 485 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
 486 {
 487         struct vlan_entry *vlan;
 488         int res;
 489
 490         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 491                 res = vlan_vid_add(slave_dev, vlan->vlan_id);
 492                 if (res)
 493                         pr_warning("%s: Failed to add vlan id %d to device %s\n",
 494                                    bond->dev->name, vlan->vlan_id,
 495                                    slave_dev->name);
 496         }
 497 }
 498
 499 static void bond_del_vlans_from_slave(struct bonding *bond,
 500                                       struct net_device *slave_dev)
 501 {
 502         struct vlan_entry *vlan;
 503
 504         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 505                 if (!vlan->vlan_id)
 506                         continue;
 507                 vlan_vid_del(slave_dev, vlan->vlan_id);
 508         }
 509 }
 510
 511 /*------------------------------- Link status -------------------------------*/
 512
 513 /*
 514  * Set the carrier state for the master according to the state of its
 515  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 516  * do special 802.3ad magic.
 517  *
 518  * Returns zero if carrier state does not change, nonzero if it does.
 519  */
 520 static int bond_set_carrier(struct bonding *bond)
 521 {
 522         struct slave *slave;
 523         int i;
 524
 525         if (bond->slave_cnt == 0)
 526                 goto down;
 527
 528         if (bond->params.mode == BOND_MODE_8023AD)
 529                 return bond_3ad_set_carrier(bond);
 530
 531         bond_for_each_slave(bond, slave, i) {
 532                 if (slave->link == BOND_LINK_UP) {
 533                         if (!netif_carrier_ok(bond->dev)) {
 534                                 netif_carrier_on(bond->dev);
 535                                 return 1;
 536                         }
 537                         return 0;
 538                 }
 539         }
 540
 541 down:
 542         if (netif_carrier_ok(bond->dev)) {
 543                 netif_carrier_off(bond->dev);
 544                 return 1;
 545         }
 546         return 0;
 547 }
 548
 549 /*
 550  * Get link speed and duplex from the slave's base driver
 551  * using ethtool. If for some reason the call fails or the
 552  * values are invalid, set speed and duplex to -1,
 553  * and return.
 554  */
 555 static void bond_update_speed_duplex(struct slave *slave)
 556 {
 557         struct net_device *slave_dev = slave->dev;
 558         struct ethtool_cmd ecmd;
 559         u32 slave_speed;
 560         int res;
 561
 562         slave->speed = SPEED_UNKNOWN;
 563         slave->duplex = DUPLEX_UNKNOWN;
 564
 565         res = __ethtool_get_settings(slave_dev, &ecmd);
 566         if (res < 0)
 567                 return;
 568
 569         slave_speed = ethtool_cmd_speed(&ecmd);
 570         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 571                 return;
 572
 573         switch (ecmd.duplex) {
 574         case DUPLEX_FULL:
 575         case DUPLEX_HALF:
 576                 break;
 577         default:
 578                 return;
 579         }
 580
 581         slave->speed = slave_speed;
 582         slave->duplex = ecmd.duplex;
 583
 584         return;
 585 }
 586
 587 /*
 588  * if <dev> supports MII link status reporting, check its link status.
 589  *
 590  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 591  * depending upon the setting of the use_carrier parameter.
 592  *
 593  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 594  * can't tell and just pretend it is), or 0, meaning that the link is
 595  * down.
 596  *
 597  * If reporting is non-zero, instead of faking link up, return -1 if
 598  * both ETHTOOL and MII ioctls fail (meaning the device does not
 599  * support them).  If use_carrier is set, return whatever it says.
 600  * It'd be nice if there was a good way to tell if a driver supports
 601  * netif_carrier, but there really isn't.
 602  */
 603 static int bond_check_dev_link(struct bonding *bond,
 604                                struct net_device *slave_dev, int reporting)
 605 {
 606         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 607         int (*ioctl)(struct net_device *, struct ifreq *, int);
 608         struct ifreq ifr;
 609         struct mii_ioctl_data *mii;
 610
 611         if (!reporting && !netif_running(slave_dev))
 612                 return 0;
 613
 614         if (bond->params.use_carrier)
 615                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 616
 617         /* Try to get link status using Ethtool first. */
 618         if (slave_dev->ethtool_ops) {
 619                 if (slave_dev->ethtool_ops->get_link) {
 620                         u32 link;
 621
 622                         link = slave_dev->ethtool_ops->get_link(slave_dev);
 623
 624                         return link ? BMSR_LSTATUS : 0;
 625                 }
 626         }
 627
 628         /* Ethtool can't be used, fallback to MII ioctls. */
 629         ioctl = slave_ops->ndo_do_ioctl;
 630         if (ioctl) {
 631                 /* TODO: set pointer to correct ioctl on a per team member */
 632                 /*       bases to make this more efficient. that is, once  */
 633                 /*       we determine the correct ioctl, we will always    */
 634                 /*       call it and not the others for that team          */
 635                 /*       member.                                           */
 636
 637                 /*
 638                  * We cannot assume that SIOCGMIIPHY will also read a
 639                  * register; not all network drivers (e.g., e100)
 640                  * support that.
 641                  */
 642
 643                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 644                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 645                 mii = if_mii(&ifr);
 646                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 647                         mii->reg_num = MII_BMSR;
 648                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 649                                 return mii->val_out & BMSR_LSTATUS;
 650                 }
 651         }
 652
 653         /*
 654          * If reporting, report that either there's no dev->do_ioctl,
 655          * or both SIOCGMIIREG and get_link failed (meaning that we
 656          * cannot report link status).  If not reporting, pretend
 657          * we're ok.
 658          */
 659         return reporting ? -1 : BMSR_LSTATUS;
 660 }
 661
 662 /*----------------------------- Multicast list ------------------------------*/
 663
 664 /*
 665  * Push the promiscuity flag down to appropriate slaves
 666  */
 667 static int bond_set_promiscuity(struct bonding *bond, int inc)
 668 {
 669         int err = 0;
 670         if (USES_PRIMARY(bond->params.mode)) {
 671                 /* write lock already acquired */
 672                 if (bond->curr_active_slave) {
 673                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
 674                                                   inc);
 675                 }
 676         } else {
 677                 struct slave *slave;
 678                 int i;
 679                 bond_for_each_slave(bond, slave, i) {
 680                         err = dev_set_promiscuity(slave->dev, inc);
 681                         if (err)
 682                                 return err;
 683                 }
 684         }
 685         return err;
 686 }
 687
 688 /*
 689  * Push the allmulti flag down to all slaves
 690  */
 691 static int bond_set_allmulti(struct bonding *bond, int inc)
 692 {
 693         int err = 0;
 694         if (USES_PRIMARY(bond->params.mode)) {
 695                 /* write lock already acquired */
 696                 if (bond->curr_active_slave) {
 697                         err = dev_set_allmulti(bond->curr_active_slave->dev,
 698                                                inc);
 699                 }
 700         } else {
 701                 struct slave *slave;
 702                 int i;
 703                 bond_for_each_slave(bond, slave, i) {
 704                         err = dev_set_allmulti(slave->dev, inc);
 705                         if (err)
 706                                 return err;
 707                 }
 708         }
 709         return err;
 710 }
 711
 712 /*
 713  * Add a Multicast address to slaves
 714  * according to mode
 715  */
 716 static void bond_mc_add(struct bonding *bond, void *addr)
 717 {
 718         if (USES_PRIMARY(bond->params.mode)) {
 719                 /* write lock already acquired */
 720                 if (bond->curr_active_slave)
 721                         dev_mc_add(bond->curr_active_slave->dev, addr);
 722         } else {
 723                 struct slave *slave;
 724                 int i;
 725
 726                 bond_for_each_slave(bond, slave, i)
 727                         dev_mc_add(slave->dev, addr);
 728         }
 729 }
 730
 731 /*
 732  * Remove a multicast address from slave
 733  * according to mode
 734  */
 735 static void bond_mc_del(struct bonding *bond, void *addr)
 736 {
 737         if (USES_PRIMARY(bond->params.mode)) {
 738                 /* write lock already acquired */
 739                 if (bond->curr_active_slave)
 740                         dev_mc_del(bond->curr_active_slave->dev, addr);
 741         } else {
 742                 struct slave *slave;
 743                 int i;
 744                 bond_for_each_slave(bond, slave, i) {
 745                         dev_mc_del(slave->dev, addr);
 746                 }
 747         }
 748 }
 749
 750
 751 static void __bond_resend_igmp_join_requests(struct net_device *dev)
 752 {
 753         struct in_device *in_dev;
 754
 755         rcu_read_lock();
 756         in_dev = __in_dev_get_rcu(dev);
 757         if (in_dev)
 758                 ip_mc_rejoin_groups(in_dev);
 759         rcu_read_unlock();
 760 }
 761
 762 /*
 763  * Retrieve the list of registered multicast addresses for the bonding
 764  * device and retransmit an IGMP JOIN request to the current active
 765  * slave.
 766  */
 767 static void bond_resend_igmp_join_requests(struct bonding *bond)
 768 {
 769         struct net_device *bond_dev, *vlan_dev, *master_dev;
 770         struct vlan_entry *vlan;
 771
 772         read_lock(&bond->lock);
 773
 774         bond_dev = bond->dev;
 775
 776         /* rejoin all groups on bond device */
 777         __bond_resend_igmp_join_requests(bond_dev);
 778
 779         /*
 780          * if bond is enslaved to a bridge,
 781          * then rejoin all groups on its master
 782          */
 783         master_dev = bond_dev->master;
 784         if (master_dev)
 785                 if ((master_dev->priv_flags & IFF_EBRIDGE)
 786                         && (bond_dev->priv_flags & IFF_BRIDGE_PORT))
 787                         __bond_resend_igmp_join_requests(master_dev);
 788
 789         /* rejoin all groups on vlan devices */
 790         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
 791                 rcu_read_lock();
 792                 vlan_dev = __vlan_find_dev_deep(bond_dev,
 793                                                 vlan->vlan_id);
 794                 rcu_read_unlock();
 795                 if (vlan_dev)
 796                         __bond_resend_igmp_join_requests(vlan_dev);
 797         }
 798
 799         if (--bond->igmp_retrans > 0)
 800                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 801
 802         read_unlock(&bond->lock);
 803 }
 804
 805 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 806 {
 807         struct bonding *bond = container_of(work, struct bonding,
 808                                             mcast_work.work);
 809         bond_resend_igmp_join_requests(bond);
 810 }
 811
 812 /*
 813  * flush all members of flush->mc_list from device dev->mc_list
 814  */
 815 static void bond_mc_list_flush(struct net_device *bond_dev,
 816                                struct net_device *slave_dev)
 817 {
 818         struct bonding *bond = netdev_priv(bond_dev);
 819         struct netdev_hw_addr *ha;
 820
 821         netdev_for_each_mc_addr(ha, bond_dev)
 822                 dev_mc_del(slave_dev, ha->addr);
 823
 824         if (bond->params.mode == BOND_MODE_8023AD) {
 825                 /* del lacpdu mc addr from mc list */
 826                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 827
 828                 dev_mc_del(slave_dev, lacpdu_multicast);
 829         }
 830 }
 831
 832 /*--------------------------- Active slave change ---------------------------*/
 833
 834 /*
 835  * Update the mc list and multicast-related flags for the new and
 836  * old active slaves (if any) according to the multicast mode, and
 837  * promiscuous flags unconditionally.
 838  */
 839 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
 840                          struct slave *old_active)
 841 {
 842         struct netdev_hw_addr *ha;
 843
 844         if (!USES_PRIMARY(bond->params.mode))
 845                 /* nothing to do -  mc list is already up-to-date on
 846                  * all slaves
 847                  */
 848                 return;
 849
 850         if (old_active) {
 851                 if (bond->dev->flags & IFF_PROMISC)
 852                         dev_set_promiscuity(old_active->dev, -1);
 853
 854                 if (bond->dev->flags & IFF_ALLMULTI)
 855                         dev_set_allmulti(old_active->dev, -1);
 856
 857                 netdev_for_each_mc_addr(ha, bond->dev)
 858                         dev_mc_del(old_active->dev, ha->addr);
 859         }
 860
 861         if (new_active) {
 862                 /* FIXME: Signal errors upstream. */
 863                 if (bond->dev->flags & IFF_PROMISC)
 864                         dev_set_promiscuity(new_active->dev, 1);
 865
 866                 if (bond->dev->flags & IFF_ALLMULTI)
 867                         dev_set_allmulti(new_active->dev, 1);
 868
 869                 netdev_for_each_mc_addr(ha, bond->dev)
 870                         dev_mc_add(new_active->dev, ha->addr);
 871         }
 872 }
 873
 874 /*
 875  * bond_do_fail_over_mac
 876  *
 877  * Perform special MAC address swapping for fail_over_mac settings
 878  *
 879  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
 880  */
 881 static void bond_do_fail_over_mac(struct bonding *bond,
 882                                   struct slave *new_active,
 883                                   struct slave *old_active)
 884         __releases(&bond->curr_slave_lock)
 885         __releases(&bond->lock)
 886         __acquires(&bond->lock)
 887         __acquires(&bond->curr_slave_lock)
 888 {
 889         u8 tmp_mac[ETH_ALEN];
 890         struct sockaddr saddr;
 891         int rv;
 892
 893         switch (bond->params.fail_over_mac) {
 894         case BOND_FOM_ACTIVE:
 895                 if (new_active) {
 896                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
 897                                new_active->dev->addr_len);
 898                         write_unlock_bh(&bond->curr_slave_lock);
 899                         read_unlock(&bond->lock);
 900                         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
 901                         read_lock(&bond->lock);
 902                         write_lock_bh(&bond->curr_slave_lock);
 903                 }
 904                 break;
 905         case BOND_FOM_FOLLOW:
 906                 /*
 907                  * if new_active && old_active, swap them
 908                  * if just old_active, do nothing (going to no active slave)
 909                  * if just new_active, set new_active to bond's MAC
 910                  */
 911                 if (!new_active)
 912                         return;
 913
 914                 write_unlock_bh(&bond->curr_slave_lock);
 915                 read_unlock(&bond->lock);
 916
 917                 if (old_active) {
 918                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
 919                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
 920                                ETH_ALEN);
 921                         saddr.sa_family = new_active->dev->type;
 922                 } else {
 923                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
 924                         saddr.sa_family = bond->dev->type;
 925                 }
 926
 927                 rv = dev_set_mac_address(new_active->dev, &saddr);
 928                 if (rv) {
 929                         pr_err("%s: Error %d setting MAC of slave %s\n",
 930                                bond->dev->name, -rv, new_active->dev->name);
 931                         goto out;
 932                 }
 933
 934                 if (!old_active)
 935                         goto out;
 936
 937                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
 938                 saddr.sa_family = old_active->dev->type;
 939
 940                 rv = dev_set_mac_address(old_active->dev, &saddr);
 941                 if (rv)
 942                         pr_err("%s: Error %d setting MAC of slave %s\n",
 943                                bond->dev->name, -rv, new_active->dev->name);
 944 out:
 945                 read_lock(&bond->lock);
 946                 write_lock_bh(&bond->curr_slave_lock);
 947                 break;
 948         default:
 949                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
 950                        bond->dev->name, bond->params.fail_over_mac);
 951                 break;
 952         }
 953
 954 }
 955
 956 static bool bond_should_change_active(struct bonding *bond)
 957 {
 958         struct slave *prim = bond->primary_slave;
 959         struct slave *curr = bond->curr_active_slave;
 960
 961         if (!prim || !curr || curr->link != BOND_LINK_UP)
 962                 return true;
 963         if (bond->force_primary) {
 964                 bond->force_primary = false;
 965                 return true;
 966         }
 967         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
 968             (prim->speed < curr->speed ||
 969              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
 970                 return false;
 971         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
 972                 return false;
 973         return true;
 974 }
 975
 976 /**
 977  * find_best_interface - select the best available slave to be the active one
 978  * @bond: our bonding struct
 979  *
 980  * Warning: Caller must hold curr_slave_lock for writing.
 981  */
 982 static struct slave *bond_find_best_slave(struct bonding *bond)
 983 {
 984         struct slave *new_active, *old_active;
 985         struct slave *bestslave = NULL;
 986         int mintime = bond->params.updelay;
 987         int i;
 988
 989         new_active = bond->curr_active_slave;
 990
 991         if (!new_active) { /* there were no active slaves left */
 992                 if (bond->slave_cnt > 0)   /* found one slave */
 993                         new_active = bond->first_slave;
 994                 else
 995                         return NULL; /* still no slave, return NULL */
 996         }
 997
 998         if ((bond->primary_slave) &&
 999             bond->primary_slave->link == BOND_LINK_UP &&
1000             bond_should_change_active(bond)) {
1001                 new_active = bond->primary_slave;
1002         }
1003
1004         /* remember where to stop iterating over the slaves */
1005         old_active = new_active;
1006
1007         bond_for_each_slave_from(bond, new_active, i, old_active) {
1008                 if (new_active->link == BOND_LINK_UP) {
1009                         return new_active;
1010                 } else if (new_active->link == BOND_LINK_BACK &&
1011                            IS_UP(new_active->dev)) {
1012                         /* link up, but waiting for stabilization */
1013                         if (new_active->delay < mintime) {
1014                                 mintime = new_active->delay;
1015                                 bestslave = new_active;
1016                         }
1017                 }
1018         }
1019
1020         return bestslave;
1021 }
1022
1023 static bool bond_should_notify_peers(struct bonding *bond)
1024 {
1025         struct slave *slave = bond->curr_active_slave;
1026
1027         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1028                  bond->dev->name, slave ? slave->dev->name : "NULL");
1029
1030         if (!slave || !bond->send_peer_notif ||
1031             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1032                 return false;
1033
1034         bond->send_peer_notif--;
1035         return true;
1036 }
1037
1038 /**
1039  * change_active_interface - change the active slave into the specified one
1040  * @bond: our bonding struct
1041  * @new: the new slave to make the active one
1042  *
1043  * Set the new slave to the bond's settings and unset them on the old
1044  * curr_active_slave.
1045  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1046  *
1047  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1048  * because it is apparently the best available slave we have, even though its
1049  * updelay hasn't timed out yet.
1050  *
1051  * If new_active is not NULL, caller must hold bond->lock for read and
1052  * curr_slave_lock for write_bh.
1053  */
1054 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1055 {
1056         struct slave *old_active = bond->curr_active_slave;
1057
1058         if (old_active == new_active)
1059                 return;
1060
1061         if (new_active) {
1062                 new_active->jiffies = jiffies;
1063
1064                 if (new_active->link == BOND_LINK_BACK) {
1065                         if (USES_PRIMARY(bond->params.mode)) {
1066                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1067                                         bond->dev->name, new_active->dev->name,
1068                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1069                         }
1070
1071                         new_active->delay = 0;
1072                         new_active->link = BOND_LINK_UP;
1073
1074                         if (bond->params.mode == BOND_MODE_8023AD)
1075                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1076
1077                         if (bond_is_lb(bond))
1078                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1079                 } else {
1080                         if (USES_PRIMARY(bond->params.mode)) {
1081                                 pr_info("%s: making interface %s the new active one.\n",
1082                                         bond->dev->name, new_active->dev->name);
1083                         }
1084                 }
1085         }
1086
1087         if (USES_PRIMARY(bond->params.mode))
1088                 bond_mc_swap(bond, new_active, old_active);
1089
1090         if (bond_is_lb(bond)) {
1091                 bond_alb_handle_active_change(bond, new_active);
1092                 if (old_active)
1093                         bond_set_slave_inactive_flags(old_active);
1094                 if (new_active)
1095                         bond_set_slave_active_flags(new_active);
1096         } else {
1097                 bond->curr_active_slave = new_active;
1098         }
1099
1100         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1101                 if (old_active)
1102                         bond_set_slave_inactive_flags(old_active);
1103
1104                 if (new_active) {
1105                         bool should_notify_peers = false;
1106
1107                         bond_set_slave_active_flags(new_active);
1108
1109                         if (bond->params.fail_over_mac)
1110                                 bond_do_fail_over_mac(bond, new_active,
1111                                                       old_active);
1112
1113                         if (netif_running(bond->dev)) {
1114                                 bond->send_peer_notif =
1115                                         bond->params.num_peer_notif;
1116                                 should_notify_peers =
1117                                         bond_should_notify_peers(bond);
1118                         }
1119
1120                         write_unlock_bh(&bond->curr_slave_lock);
1121                         read_unlock(&bond->lock);
1122
1123                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1124                         if (should_notify_peers)
1125                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1126                                                          bond->dev);
1127
1128                         read_lock(&bond->lock);
1129                         write_lock_bh(&bond->curr_slave_lock);
1130                 }
1131         }
1132
1133         /* resend IGMP joins since active slave has changed or
1134          * all were sent on curr_active_slave.
1135          * resend only if bond is brought up with the affected
1136          * bonding modes and the retransmission is enabled */
1137         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1138             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1139              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1140                 bond->igmp_retrans = bond->params.resend_igmp;
1141                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1142         }
1143 }
1144
1145 /**
1146  * bond_select_active_slave - select a new active slave, if needed
1147  * @bond: our bonding struct
1148  *
1149  * This functions should be called when one of the following occurs:
1150  * - The old curr_active_slave has been released or lost its link.
1151  * - The primary_slave has got its link back.
1152  * - A slave has got its link back and there's no old curr_active_slave.
1153  *
1154  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1155  */
1156 void bond_select_active_slave(struct bonding *bond)
1157 {
1158         struct slave *best_slave;
1159         int rv;
1160
1161         best_slave = bond_find_best_slave(bond);
1162         if (best_slave != bond->curr_active_slave) {
1163                 bond_change_active_slave(bond, best_slave);
1164                 rv = bond_set_carrier(bond);
1165                 if (!rv)
1166                         return;
1167
1168                 if (netif_carrier_ok(bond->dev)) {
1169                         pr_info("%s: first active interface up!\n",
1170                                 bond->dev->name);
1171                 } else {
1172                         pr_info("%s: now running without any active interface !\n",
1173                                 bond->dev->name);
1174                 }
1175         }
1176 }
1177
1178 /*--------------------------- slave list handling ---------------------------*/
1179
1180 /*
1181  * This function attaches the slave to the end of list.
1182  *
1183  * bond->lock held for writing by caller.
1184  */
1185 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1186 {
1187         if (bond->first_slave == NULL) { /* attaching the first slave */
1188                 new_slave->next = new_slave;
1189                 new_slave->prev = new_slave;
1190                 bond->first_slave = new_slave;
1191         } else {
1192                 new_slave->next = bond->first_slave;
1193                 new_slave->prev = bond->first_slave->prev;
1194                 new_slave->next->prev = new_slave;
1195                 new_slave->prev->next = new_slave;
1196         }
1197
1198         bond->slave_cnt++;
1199 }
1200
1201 /*
1202  * This function detaches the slave from the list.
1203  * WARNING: no check is made to verify if the slave effectively
1204  * belongs to <bond>.
1205  * Nothing is freed on return, structures are just unchained.
1206  * If any slave pointer in bond was pointing to <slave>,
1207  * it should be changed by the calling function.
1208  *
1209  * bond->lock held for writing by caller.
1210  */
1211 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1212 {
1213         if (slave->next)
1214                 slave->next->prev = slave->prev;
1215
1216         if (slave->prev)
1217                 slave->prev->next = slave->next;
1218
1219         if (bond->first_slave == slave) { /* slave is the first slave */
1220                 if (bond->slave_cnt > 1) { /* there are more slave */
1221                         bond->first_slave = slave->next;
1222                 } else {
1223                         bond->first_slave = NULL; /* slave was the last one */
1224                 }
1225         }
1226
1227         slave->next = NULL;
1228         slave->prev = NULL;
1229         bond->slave_cnt--;
1230 }
1231
1232 #ifdef CONFIG_NET_POLL_CONTROLLER
1233 static inline int slave_enable_netpoll(struct slave *slave)
1234 {
1235         struct netpoll *np;
1236         int err = 0;
1237
1238         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1239         err = -ENOMEM;
1240         if (!np)
1241                 goto out;
1242
1243         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1244         if (err) {
1245                 kfree(np);
1246                 goto out;
1247         }
1248         slave->np = np;
1249 out:
1250         return err;
1251 }
1252 static inline void slave_disable_netpoll(struct slave *slave)
1253 {
1254         struct netpoll *np = slave->np;
1255
1256         if (!np)
1257                 return;
1258
1259         slave->np = NULL;
1260         __netpoll_free_rcu(np);
1261 }
1262 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1263 {
1264         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1265                 return false;
1266         if (!slave_dev->netdev_ops->ndo_poll_controller)
1267                 return false;
1268         return true;
1269 }
1270
1271 static void bond_poll_controller(struct net_device *bond_dev)
1272 {
1273 }
1274
1275 static void __bond_netpoll_cleanup(struct bonding *bond)
1276 {
1277         struct slave *slave;
1278         int i;
1279
1280         bond_for_each_slave(bond, slave, i)
1281                 if (IS_UP(slave->dev))
1282                         slave_disable_netpoll(slave);
1283 }
1284 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1285 {
1286         struct bonding *bond = netdev_priv(bond_dev);
1287
1288         read_lock(&bond->lock);
1289         __bond_netpoll_cleanup(bond);
1290         read_unlock(&bond->lock);
1291 }
1292
1293 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1294 {
1295         struct bonding *bond = netdev_priv(dev);
1296         struct slave *slave;
1297         int i, err = 0;
1298
1299         read_lock(&bond->lock);
1300         bond_for_each_slave(bond, slave, i) {
1301                 err = slave_enable_netpoll(slave);
1302                 if (err) {
1303                         __bond_netpoll_cleanup(bond);
1304                         break;
1305                 }
1306         }
1307         read_unlock(&bond->lock);
1308         return err;
1309 }
1310
1311 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1312 {
1313         return bond->dev->npinfo;
1314 }
1315
1316 #else
1317 static inline int slave_enable_netpoll(struct slave *slave)
1318 {
1319         return 0;
1320 }
1321 static inline void slave_disable_netpoll(struct slave *slave)
1322 {
1323 }
1324 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1325 {
1326 }
1327 #endif
1328
1329 /*---------------------------------- IOCTL ----------------------------------*/
1330
1331 static int bond_sethwaddr(struct net_device *bond_dev,
1332                           struct net_device *slave_dev)
1333 {
1334         pr_debug("bond_dev=%p\n", bond_dev);
1335         pr_debug("slave_dev=%p\n", slave_dev);
1336         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1337         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1338         return 0;
1339 }
1340
1341 static netdev_features_t bond_fix_features(struct net_device *dev,
1342         netdev_features_t features)
1343 {
1344         struct slave *slave;
1345         struct bonding *bond = netdev_priv(dev);
1346         netdev_features_t mask;
1347         int i;
1348
1349         read_lock(&bond->lock);
1350
1351         if (!bond->first_slave) {
1352                 /* Disable adding VLANs to empty bond. But why? --mq */
1353                 features |= NETIF_F_VLAN_CHALLENGED;
1354                 goto out;
1355         }
1356
1357         mask = features;
1358         features &= ~NETIF_F_ONE_FOR_ALL;
1359         features |= NETIF_F_ALL_FOR_ALL;
1360
1361         bond_for_each_slave(bond, slave, i) {
1362                 features = netdev_increment_features(features,
1363                                                      slave->dev->features,
1364                                                      mask);
1365         }
1366
1367 out:
1368         read_unlock(&bond->lock);
1369         return features;
1370 }
1371
1372 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1373                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1374                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1375
1376 static void bond_compute_features(struct bonding *bond)
1377 {
1378         struct slave *slave;
1379         struct net_device *bond_dev = bond->dev;
1380         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1381         unsigned short max_hard_header_len = ETH_HLEN;
1382         unsigned int gso_max_size = GSO_MAX_SIZE;
1383         u16 gso_max_segs = GSO_MAX_SEGS;
1384         int i;
1385         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1386
1387         read_lock(&bond->lock);
1388
1389         if (!bond->first_slave)
1390                 goto done;
1391
1392         bond_for_each_slave(bond, slave, i) {
1393                 vlan_features = netdev_increment_features(vlan_features,
1394                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1395
1396                 dst_release_flag &= slave->dev->priv_flags;
1397                 if (slave->dev->hard_header_len > max_hard_header_len)
1398                         max_hard_header_len = slave->dev->hard_header_len;
1399
1400                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1401                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1402         }
1403
1404 done:
1405         bond_dev->vlan_features = vlan_features;
1406         bond_dev->hard_header_len = max_hard_header_len;
1407         bond_dev->gso_max_segs = gso_max_segs;
1408         netif_set_gso_max_size(bond_dev, gso_max_size);
1409
1410         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1411         bond_dev->priv_flags = flags | dst_release_flag;
1412
1413         read_unlock(&bond->lock);
1414
1415         netdev_change_features(bond_dev);
1416 }
1417
1418 static void bond_setup_by_slave(struct net_device *bond_dev,
1419                                 struct net_device *slave_dev)
1420 {
1421         struct bonding *bond = netdev_priv(bond_dev);
1422
1423         bond_dev->header_ops        = slave_dev->header_ops;
1424
1425         bond_dev->type              = slave_dev->type;
1426         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1427         bond_dev->addr_len          = slave_dev->addr_len;
1428
1429         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1430                 slave_dev->addr_len);
1431         bond->setup_by_slave = 1;
1432 }
1433
1434 /* On bonding slaves other than the currently active slave, suppress
1435  * duplicates except for alb non-mcast/bcast.
1436  */
1437 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1438                                             struct slave *slave,
1439                                             struct bonding *bond)
1440 {
1441         if (bond_is_slave_inactive(slave)) {
1442                 if (bond->params.mode == BOND_MODE_ALB &&
1443                     skb->pkt_type != PACKET_BROADCAST &&
1444                     skb->pkt_type != PACKET_MULTICAST)
1445                         return false;
1446                 return true;
1447         }
1448         return false;
1449 }
1450
1451 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1452 {
1453         struct sk_buff *skb = *pskb;
1454         struct slave *slave;
1455         struct bonding *bond;
1456         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1457                           struct slave *);
1458         int ret = RX_HANDLER_ANOTHER;
1459
1460         skb = skb_share_check(skb, GFP_ATOMIC);
1461         if (unlikely(!skb))
1462                 return RX_HANDLER_CONSUMED;
1463
1464         *pskb = skb;
1465
1466         slave = bond_slave_get_rcu(skb->dev);
1467         bond = slave->bond;
1468
1469         if (bond->params.arp_interval)
1470                 slave->dev->last_rx = jiffies;
1471
1472         recv_probe = ACCESS_ONCE(bond->recv_probe);
1473         if (recv_probe) {
1474                 ret = recv_probe(skb, bond, slave);
1475                 if (ret == RX_HANDLER_CONSUMED) {
1476                         consume_skb(skb);
1477                         return ret;
1478                 }
1479         }
1480
1481         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1482                 return RX_HANDLER_EXACT;
1483         }
1484
1485         skb->dev = bond->dev;
1486
1487         if (bond->params.mode == BOND_MODE_ALB &&
1488             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1489             skb->pkt_type == PACKET_HOST) {
1490
1491                 if (unlikely(skb_cow_head(skb,
1492                                           skb->data - skb_mac_header(skb)))) {
1493                         kfree_skb(skb);
1494                         return RX_HANDLER_CONSUMED;
1495                 }
1496                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1497         }
1498
1499         return ret;
1500 }
1501
1502 /* enslave device <slave> to bond device <master> */
1503 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1504 {
1505         struct bonding *bond = netdev_priv(bond_dev);
1506         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1507         struct slave *new_slave = NULL;
1508         struct netdev_hw_addr *ha;
1509         struct sockaddr addr;
1510         int link_reporting;
1511         int res = 0;
1512
1513         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1514                 slave_ops->ndo_do_ioctl == NULL) {
1515                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1516                            bond_dev->name, slave_dev->name);
1517         }
1518
1519         /* already enslaved */
1520         if (slave_dev->flags & IFF_SLAVE) {
1521                 pr_debug("Error, Device was already enslaved\n");
1522                 return -EBUSY;
1523         }
1524
1525         /* vlan challenged mutual exclusion */
1526         /* no need to lock since we're protected by rtnl_lock */
1527         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1528                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1529                 if (vlan_uses_dev(bond_dev)) {
1530                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1531                                bond_dev->name, slave_dev->name, bond_dev->name);
1532                         return -EPERM;
1533                 } else {
1534                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1535                                    bond_dev->name, slave_dev->name,
1536                                    slave_dev->name, bond_dev->name);
1537                 }
1538         } else {
1539                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1540         }
1541
1542         /*
1543          * Old ifenslave binaries are no longer supported.  These can
1544          * be identified with moderate accuracy by the state of the slave:
1545          * the current ifenslave will set the interface down prior to
1546          * enslaving it; the old ifenslave will not.
1547          */
1548         if ((slave_dev->flags & IFF_UP)) {
1549                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1550                        slave_dev->name);
1551                 res = -EPERM;
1552                 goto err_undo_flags;
1553         }
1554
1555         /* set bonding device ether type by slave - bonding netdevices are
1556          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1557          * there is a need to override some of the type dependent attribs/funcs.
1558          *
1559          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1560          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1561          */
1562         if (bond->slave_cnt == 0) {
1563                 if (bond_dev->type != slave_dev->type) {
1564                         pr_debug("%s: change device type from %d to %d\n",
1565                                  bond_dev->name,
1566                                  bond_dev->type, slave_dev->type);
1567
1568                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1569                                                        bond_dev);
1570                         res = notifier_to_errno(res);
1571                         if (res) {
1572                                 pr_err("%s: refused to change device type\n",
1573                                        bond_dev->name);
1574                                 res = -EBUSY;
1575                                 goto err_undo_flags;
1576                         }
1577
1578                         /* Flush unicast and multicast addresses */
1579                         dev_uc_flush(bond_dev);
1580                         dev_mc_flush(bond_dev);
1581
1582                         if (slave_dev->type != ARPHRD_ETHER)
1583                                 bond_setup_by_slave(bond_dev, slave_dev);
1584                         else {
1585                                 ether_setup(bond_dev);
1586                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1587                         }
1588
1589                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1590                                                  bond_dev);
1591                 }
1592         } else if (bond_dev->type != slave_dev->type) {
1593                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1594                        slave_dev->name,
1595                        slave_dev->type, bond_dev->type);
1596                 res = -EINVAL;
1597                 goto err_undo_flags;
1598         }
1599
1600         if (slave_ops->ndo_set_mac_address == NULL) {
1601                 if (bond->slave_cnt == 0) {
1602                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1603                                    bond_dev->name);
1604                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1605                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1606                         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",
1607                                bond_dev->name);
1608                         res = -EOPNOTSUPP;
1609                         goto err_undo_flags;
1610                 }
1611         }
1612
1613         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1614
1615         /* If this is the first slave, then we need to set the master's hardware
1616          * address to be the same as the slave's. */
1617         if (is_zero_ether_addr(bond->dev->dev_addr))
1618                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1619                        slave_dev->addr_len);
1620
1621
1622         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1623         if (!new_slave) {
1624                 res = -ENOMEM;
1625                 goto err_undo_flags;
1626         }
1627
1628         /*
1629          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1630          * is set via sysfs or module option if desired.
1631          */
1632         new_slave->queue_id = 0;
1633
1634         /* Save slave's original mtu and then set it to match the bond */
1635         new_slave->original_mtu = slave_dev->mtu;
1636         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1637         if (res) {
1638                 pr_debug("Error %d calling dev_set_mtu\n", res);
1639                 goto err_free;
1640         }
1641
1642         /*
1643          * Save slave's original ("permanent") mac address for modes
1644          * that need it, and for restoring it upon release, and then
1645          * set it to the master's address
1646          */
1647         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1648
1649         if (!bond->params.fail_over_mac) {
1650                 /*
1651                  * Set slave to master's mac address.  The application already
1652                  * set the master's mac address to that of the first slave
1653                  */
1654                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1655                 addr.sa_family = slave_dev->type;
1656                 res = dev_set_mac_address(slave_dev, &addr);
1657                 if (res) {
1658                         pr_debug("Error %d calling set_mac_address\n", res);
1659                         goto err_restore_mtu;
1660                 }
1661         }
1662
1663         res = netdev_set_bond_master(slave_dev, bond_dev);
1664         if (res) {
1665                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1666                 goto err_restore_mac;
1667         }
1668
1669         /* open the slave since the application closed it */
1670         res = dev_open(slave_dev);
1671         if (res) {
1672                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1673                 goto err_unset_master;
1674         }
1675
1676         new_slave->bond = bond;
1677         new_slave->dev = slave_dev;
1678         slave_dev->priv_flags |= IFF_BONDING;
1679
1680         if (bond_is_lb(bond)) {
1681                 /* bond_alb_init_slave() must be called before all other stages since
1682                  * it might fail and we do not want to have to undo everything
1683                  */
1684                 res = bond_alb_init_slave(bond, new_slave);
1685                 if (res)
1686                         goto err_close;
1687         }
1688
1689         /* If the mode USES_PRIMARY, then the new slave gets the
1690          * master's promisc (and mc) settings only if it becomes the
1691          * curr_active_slave, and that is taken care of later when calling
1692          * bond_change_active()
1693          */
1694         if (!USES_PRIMARY(bond->params.mode)) {
1695                 /* set promiscuity level to new slave */
1696                 if (bond_dev->flags & IFF_PROMISC) {
1697                         res = dev_set_promiscuity(slave_dev, 1);
1698                         if (res)
1699                                 goto err_close;
1700                 }
1701
1702                 /* set allmulti level to new slave */
1703                 if (bond_dev->flags & IFF_ALLMULTI) {
1704                         res = dev_set_allmulti(slave_dev, 1);
1705                         if (res)
1706                                 goto err_close;
1707                 }
1708
1709                 netif_addr_lock_bh(bond_dev);
1710                 /* upload master's mc_list to new slave */
1711                 netdev_for_each_mc_addr(ha, bond_dev)
1712                         dev_mc_add(slave_dev, ha->addr);
1713                 netif_addr_unlock_bh(bond_dev);
1714         }
1715
1716         if (bond->params.mode == BOND_MODE_8023AD) {
1717                 /* add lacpdu mc addr to mc list */
1718                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1719
1720                 dev_mc_add(slave_dev, lacpdu_multicast);
1721         }
1722
1723         bond_add_vlans_on_slave(bond, slave_dev);
1724
1725         write_lock_bh(&bond->lock);
1726
1727         bond_attach_slave(bond, new_slave);
1728
1729         new_slave->delay = 0;
1730         new_slave->link_failure_count = 0;
1731
1732         write_unlock_bh(&bond->lock);
1733
1734         bond_compute_features(bond);
1735
1736         read_lock(&bond->lock);
1737
1738         new_slave->last_arp_rx = jiffies -
1739                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1740
1741         if (bond->params.miimon && !bond->params.use_carrier) {
1742                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1743
1744                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1745                         /*
1746                          * miimon is set but a bonded network driver
1747                          * does not support ETHTOOL/MII and
1748                          * arp_interval is not set.  Note: if
1749                          * use_carrier is enabled, we will never go
1750                          * here (because netif_carrier is always
1751                          * supported); thus, we don't need to change
1752                          * the messages for netif_carrier.
1753                          */
1754                         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",
1755                                bond_dev->name, slave_dev->name);
1756                 } else if (link_reporting == -1) {
1757                         /* unable get link status using mii/ethtool */
1758                         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",
1759                                    bond_dev->name, slave_dev->name);
1760                 }
1761         }
1762
1763         /* check for initial state */
1764         if (bond->params.miimon) {
1765                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1766                         if (bond->params.updelay) {
1767                                 new_slave->link = BOND_LINK_BACK;
1768                                 new_slave->delay = bond->params.updelay;
1769                         } else {
1770                                 new_slave->link = BOND_LINK_UP;
1771                         }
1772                 } else {
1773                         new_slave->link = BOND_LINK_DOWN;
1774                 }
1775         } else if (bond->params.arp_interval) {
1776                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1777                         BOND_LINK_UP : BOND_LINK_DOWN);
1778         } else {
1779                 new_slave->link = BOND_LINK_UP;
1780         }
1781
1782         if (new_slave->link != BOND_LINK_DOWN)
1783                 new_slave->jiffies = jiffies;
1784         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1785                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1786                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1787
1788         bond_update_speed_duplex(new_slave);
1789
1790         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1791                 /* if there is a primary slave, remember it */
1792                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1793                         bond->primary_slave = new_slave;
1794                         bond->force_primary = true;
1795                 }
1796         }
1797
1798         write_lock_bh(&bond->curr_slave_lock);
1799
1800         switch (bond->params.mode) {
1801         case BOND_MODE_ACTIVEBACKUP:
1802                 bond_set_slave_inactive_flags(new_slave);
1803                 bond_select_active_slave(bond);
1804                 break;
1805         case BOND_MODE_8023AD:
1806                 /* in 802.3ad mode, the internal mechanism
1807                  * will activate the slaves in the selected
1808                  * aggregator
1809                  */
1810                 bond_set_slave_inactive_flags(new_slave);
1811                 /* if this is the first slave */
1812                 if (bond->slave_cnt == 1) {
1813                         SLAVE_AD_INFO(new_slave).id = 1;
1814                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1815                          * can be called only after the mac address of the bond is set
1816                          */
1817                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1818                 } else {
1819                         SLAVE_AD_INFO(new_slave).id =
1820                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1821                 }
1822
1823                 bond_3ad_bind_slave(new_slave);
1824                 break;
1825         case BOND_MODE_TLB:
1826         case BOND_MODE_ALB:
1827                 bond_set_active_slave(new_slave);
1828                 bond_set_slave_inactive_flags(new_slave);
1829                 bond_select_active_slave(bond);
1830                 break;
1831         default:
1832                 pr_debug("This slave is always active in trunk mode\n");
1833
1834                 /* always active in trunk mode */
1835                 bond_set_active_slave(new_slave);
1836
1837                 /* In trunking mode there is little meaning to curr_active_slave
1838                  * anyway (it holds no special properties of the bond device),
1839                  * so we can change it without calling change_active_interface()
1840                  */
1841                 if (!bond->curr_active_slave)
1842                         bond->curr_active_slave = new_slave;
1843
1844                 break;
1845         } /* switch(bond_mode) */
1846
1847         write_unlock_bh(&bond->curr_slave_lock);
1848
1849         bond_set_carrier(bond);
1850
1851 #ifdef CONFIG_NET_POLL_CONTROLLER
1852         slave_dev->npinfo = bond_netpoll_info(bond);
1853         if (slave_dev->npinfo) {
1854                 if (slave_enable_netpoll(new_slave)) {
1855                         read_unlock(&bond->lock);
1856                         pr_info("Error, %s: master_dev is using netpoll, "
1857                                  "but new slave device does not support netpoll.\n",
1858                                  bond_dev->name);
1859                         res = -EBUSY;
1860                         goto err_detach;
1861                 }
1862         }
1863 #endif
1864
1865         read_unlock(&bond->lock);
1866
1867         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1868         if (res)
1869                 goto err_detach;
1870
1871         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1872                                          new_slave);
1873         if (res) {
1874                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1875                 goto err_dest_symlinks;
1876         }
1877
1878         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1879                 bond_dev->name, slave_dev->name,
1880                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1881                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1882
1883         /* enslave is successful */
1884         return 0;
1885
1886 /* Undo stages on error */
1887 err_dest_symlinks:
1888         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1889
1890 err_detach:
1891         write_lock_bh(&bond->lock);
1892         bond_detach_slave(bond, new_slave);
1893         write_unlock_bh(&bond->lock);
1894
1895 err_close:
1896         dev_close(slave_dev);
1897
1898 err_unset_master:
1899         netdev_set_bond_master(slave_dev, NULL);
1900
1901 err_restore_mac:
1902         if (!bond->params.fail_over_mac) {
1903                 /* XXX TODO - fom follow mode needs to change master's
1904                  * MAC if this slave's MAC is in use by the bond, or at
1905                  * least print a warning.
1906                  */
1907                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1908                 addr.sa_family = slave_dev->type;
1909                 dev_set_mac_address(slave_dev, &addr);
1910         }
1911
1912 err_restore_mtu:
1913         dev_set_mtu(slave_dev, new_slave->original_mtu);
1914
1915 err_free:
1916         kfree(new_slave);
1917
1918 err_undo_flags:
1919         bond_compute_features(bond);
1920
1921         return res;
1922 }
1923
1924 /*
1925  * Try to release the slave device <slave> from the bond device <master>
1926  * It is legal to access curr_active_slave without a lock because all the function
1927  * is write-locked.
1928  *
1929  * The rules for slave state should be:
1930  *   for Active/Backup:
1931  *     Active stays on all backups go down
1932  *   for Bonded connections:
1933  *     The first up interface should be left on and all others downed.
1934  */
1935 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1936 {
1937         struct bonding *bond = netdev_priv(bond_dev);
1938         struct slave *slave, *oldcurrent;
1939         struct sockaddr addr;
1940         netdev_features_t old_features = bond_dev->features;
1941
1942         /* slave is not a slave or master is not master of this slave */
1943         if (!(slave_dev->flags & IFF_SLAVE) ||
1944             (slave_dev->master != bond_dev)) {
1945                 pr_err("%s: Error: cannot release %s.\n",
1946                        bond_dev->name, slave_dev->name);
1947                 return -EINVAL;
1948         }
1949
1950         block_netpoll_tx();
1951         call_netdevice_notifiers(NETDEV_RELEASE, bond_dev);
1952         write_lock_bh(&bond->lock);
1953
1954         slave = bond_get_slave_by_dev(bond, slave_dev);
1955         if (!slave) {
1956                 /* not a slave of this bond */
1957                 pr_info("%s: %s not enslaved\n",
1958                         bond_dev->name, slave_dev->name);
1959                 write_unlock_bh(&bond->lock);
1960                 unblock_netpoll_tx();
1961                 return -EINVAL;
1962         }
1963
1964         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1965          * for this slave anymore.
1966          */
1967         netdev_rx_handler_unregister(slave_dev);
1968         write_unlock_bh(&bond->lock);
1969         synchronize_net();
1970         write_lock_bh(&bond->lock);
1971
1972         if (!bond->params.fail_over_mac) {
1973                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1974                     bond->slave_cnt > 1)
1975                         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",
1976                                    bond_dev->name, slave_dev->name,
1977                                    slave->perm_hwaddr,
1978                                    bond_dev->name, slave_dev->name);
1979         }
1980
1981         /* Inform AD package of unbinding of slave. */
1982         if (bond->params.mode == BOND_MODE_8023AD) {
1983                 /* must be called before the slave is
1984                  * detached from the list
1985                  */
1986                 bond_3ad_unbind_slave(slave);
1987         }
1988
1989         pr_info("%s: releasing %s interface %s\n",
1990                 bond_dev->name,
1991                 bond_is_active_slave(slave) ? "active" : "backup",
1992                 slave_dev->name);
1993
1994         oldcurrent = bond->curr_active_slave;
1995
1996         bond->current_arp_slave = NULL;
1997
1998         /* release the slave from its bond */
1999         bond_detach_slave(bond, slave);
2000
2001         if (bond->primary_slave == slave)
2002                 bond->primary_slave = NULL;
2003
2004         if (oldcurrent == slave)
2005                 bond_change_active_slave(bond, NULL);
2006
2007         if (bond_is_lb(bond)) {
2008                 /* Must be called only after the slave has been
2009                  * detached from the list and the curr_active_slave
2010                  * has been cleared (if our_slave == old_current),
2011                  * but before a new active slave is selected.
2012                  */
2013                 write_unlock_bh(&bond->lock);
2014                 bond_alb_deinit_slave(bond, slave);
2015                 write_lock_bh(&bond->lock);
2016         }
2017
2018         if (oldcurrent == slave) {
2019                 /*
2020                  * Note that we hold RTNL over this sequence, so there
2021                  * is no concern that another slave add/remove event
2022                  * will interfere.
2023                  */
2024                 write_unlock_bh(&bond->lock);
2025                 read_lock(&bond->lock);
2026                 write_lock_bh(&bond->curr_slave_lock);
2027
2028                 bond_select_active_slave(bond);
2029
2030                 write_unlock_bh(&bond->curr_slave_lock);
2031                 read_unlock(&bond->lock);
2032                 write_lock_bh(&bond->lock);
2033         }
2034
2035         if (bond->slave_cnt == 0) {
2036                 bond_set_carrier(bond);
2037
2038                 /* if the last slave was removed, zero the mac address
2039                  * of the master so it will be set by the application
2040                  * to the mac address of the first slave
2041                  */
2042                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2043
2044                 if (bond_vlan_used(bond)) {
2045                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2046                                    bond_dev->name, bond_dev->name);
2047                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2048                                    bond_dev->name);
2049                 }
2050         }
2051
2052         write_unlock_bh(&bond->lock);
2053         unblock_netpoll_tx();
2054
2055         if (bond->slave_cnt == 0)
2056                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2057
2058         bond_compute_features(bond);
2059         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2060             (old_features & NETIF_F_VLAN_CHALLENGED))
2061                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2062                         bond_dev->name, slave_dev->name, bond_dev->name);
2063
2064         /* must do this from outside any spinlocks */
2065         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2066
2067         bond_del_vlans_from_slave(bond, slave_dev);
2068
2069         /* If the mode USES_PRIMARY, then we should only remove its
2070          * promisc and mc settings if it was the curr_active_slave, but that was
2071          * already taken care of above when we detached the slave
2072          */
2073         if (!USES_PRIMARY(bond->params.mode)) {
2074                 /* unset promiscuity level from slave */
2075                 if (bond_dev->flags & IFF_PROMISC)
2076                         dev_set_promiscuity(slave_dev, -1);
2077
2078                 /* unset allmulti level from slave */
2079                 if (bond_dev->flags & IFF_ALLMULTI)
2080                         dev_set_allmulti(slave_dev, -1);
2081
2082                 /* flush master's mc_list from slave */
2083                 netif_addr_lock_bh(bond_dev);
2084                 bond_mc_list_flush(bond_dev, slave_dev);
2085                 netif_addr_unlock_bh(bond_dev);
2086         }
2087
2088         netdev_set_bond_master(slave_dev, NULL);
2089
2090         slave_disable_netpoll(slave);
2091
2092         /* close slave before restoring its mac address */
2093         dev_close(slave_dev);
2094
2095         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2096                 /* restore original ("permanent") mac address */
2097                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2098                 addr.sa_family = slave_dev->type;
2099                 dev_set_mac_address(slave_dev, &addr);
2100         }
2101
2102         dev_set_mtu(slave_dev, slave->original_mtu);
2103
2104         slave_dev->priv_flags &= ~IFF_BONDING;
2105
2106         kfree(slave);
2107
2108         return 0;  /* deletion OK */
2109 }
2110
2111 /*
2112 * First release a slave and then destroy the bond if no more slaves are left.
2113 * Must be under rtnl_lock when this function is called.
2114 */
2115 static int  bond_release_and_destroy(struct net_device *bond_dev,
2116                                      struct net_device *slave_dev)
2117 {
2118         struct bonding *bond = netdev_priv(bond_dev);
2119         int ret;
2120
2121         ret = bond_release(bond_dev, slave_dev);
2122         if ((ret == 0) && (bond->slave_cnt == 0)) {
2123                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2124                 pr_info("%s: destroying bond %s.\n",
2125                         bond_dev->name, bond_dev->name);
2126                 unregister_netdevice(bond_dev);
2127         }
2128         return ret;
2129 }
2130
2131 /*
2132  * This function releases all slaves.
2133  */
2134 static int bond_release_all(struct net_device *bond_dev)
2135 {
2136         struct bonding *bond = netdev_priv(bond_dev);
2137         struct slave *slave;
2138         struct net_device *slave_dev;
2139         struct sockaddr addr;
2140
2141         write_lock_bh(&bond->lock);
2142
2143         netif_carrier_off(bond_dev);
2144
2145         if (bond->slave_cnt == 0)
2146                 goto out;
2147
2148         bond->current_arp_slave = NULL;
2149         bond->primary_slave = NULL;
2150         bond_change_active_slave(bond, NULL);
2151
2152         while ((slave = bond->first_slave) != NULL) {
2153                 /* Inform AD package of unbinding of slave
2154                  * before slave is detached from the list.
2155                  */
2156                 if (bond->params.mode == BOND_MODE_8023AD)
2157                         bond_3ad_unbind_slave(slave);
2158
2159                 slave_dev = slave->dev;
2160                 bond_detach_slave(bond, slave);
2161
2162                 /* now that the slave is detached, unlock and perform
2163                  * all the undo steps that should not be called from
2164                  * within a lock.
2165                  */
2166                 write_unlock_bh(&bond->lock);
2167
2168                 /* unregister rx_handler early so bond_handle_frame wouldn't
2169                  * be called for this slave anymore.
2170                  */
2171                 netdev_rx_handler_unregister(slave_dev);
2172                 synchronize_net();
2173
2174                 if (bond_is_lb(bond)) {
2175                         /* must be called only after the slave
2176                          * has been detached from the list
2177                          */
2178                         bond_alb_deinit_slave(bond, slave);
2179                 }
2180
2181                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2182                 bond_del_vlans_from_slave(bond, slave_dev);
2183
2184                 /* If the mode USES_PRIMARY, then we should only remove its
2185                  * promisc and mc settings if it was the curr_active_slave, but that was
2186                  * already taken care of above when we detached the slave
2187                  */
2188                 if (!USES_PRIMARY(bond->params.mode)) {
2189                         /* unset promiscuity level from slave */
2190                         if (bond_dev->flags & IFF_PROMISC)
2191                                 dev_set_promiscuity(slave_dev, -1);
2192
2193                         /* unset allmulti level from slave */
2194                         if (bond_dev->flags & IFF_ALLMULTI)
2195                                 dev_set_allmulti(slave_dev, -1);
2196
2197                         /* flush master's mc_list from slave */
2198                         netif_addr_lock_bh(bond_dev);
2199                         bond_mc_list_flush(bond_dev, slave_dev);
2200                         netif_addr_unlock_bh(bond_dev);
2201                 }
2202
2203                 netdev_set_bond_master(slave_dev, NULL);
2204
2205                 slave_disable_netpoll(slave);
2206
2207                 /* close slave before restoring its mac address */
2208                 dev_close(slave_dev);
2209
2210                 if (!bond->params.fail_over_mac) {
2211                         /* restore original ("permanent") mac address*/
2212                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2213                         addr.sa_family = slave_dev->type;
2214                         dev_set_mac_address(slave_dev, &addr);
2215                 }
2216
2217                 kfree(slave);
2218
2219                 /* re-acquire the lock before getting the next slave */
2220                 write_lock_bh(&bond->lock);
2221         }
2222
2223         /* zero the mac address of the master so it will be
2224          * set by the application to the mac address of the
2225          * first slave
2226          */
2227         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2228
2229         if (bond_vlan_used(bond)) {
2230                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2231                            bond_dev->name, bond_dev->name);
2232                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2233                            bond_dev->name);
2234         }
2235
2236         pr_info("%s: released all slaves\n", bond_dev->name);
2237
2238 out:
2239         write_unlock_bh(&bond->lock);
2240
2241         bond_compute_features(bond);
2242
2243         return 0;
2244 }
2245
2246 /*
2247  * This function changes the active slave to slave <slave_dev>.
2248  * It returns -EINVAL in the following cases.
2249  *  - <slave_dev> is not found in the list.
2250  *  - There is not active slave now.
2251  *  - <slave_dev> is already active.
2252  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2253  *  - <slave_dev> is not running.
2254  * In these cases, this function does nothing.
2255  * In the other cases, current_slave pointer is changed and 0 is returned.
2256  */
2257 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2258 {
2259         struct bonding *bond = netdev_priv(bond_dev);
2260         struct slave *old_active = NULL;
2261         struct slave *new_active = NULL;
2262         int res = 0;
2263
2264         if (!USES_PRIMARY(bond->params.mode))
2265                 return -EINVAL;
2266
2267         /* Verify that master_dev is indeed the master of slave_dev */
2268         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2269                 return -EINVAL;
2270
2271         read_lock(&bond->lock);
2272
2273         read_lock(&bond->curr_slave_lock);
2274         old_active = bond->curr_active_slave;
2275         read_unlock(&bond->curr_slave_lock);
2276
2277         new_active = bond_get_slave_by_dev(bond, slave_dev);
2278
2279         /*
2280          * Changing to the current active: do nothing; return success.
2281          */
2282         if (new_active && (new_active == old_active)) {
2283                 read_unlock(&bond->lock);
2284                 return 0;
2285         }
2286
2287         if ((new_active) &&
2288             (old_active) &&
2289             (new_active->link == BOND_LINK_UP) &&
2290             IS_UP(new_active->dev)) {
2291                 block_netpoll_tx();
2292                 write_lock_bh(&bond->curr_slave_lock);
2293                 bond_change_active_slave(bond, new_active);
2294                 write_unlock_bh(&bond->curr_slave_lock);
2295                 unblock_netpoll_tx();
2296         } else
2297                 res = -EINVAL;
2298
2299         read_unlock(&bond->lock);
2300
2301         return res;
2302 }
2303
2304 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2305 {
2306         struct bonding *bond = netdev_priv(bond_dev);
2307
2308         info->bond_mode = bond->params.mode;
2309         info->miimon = bond->params.miimon;
2310
2311         read_lock(&bond->lock);
2312         info->num_slaves = bond->slave_cnt;
2313         read_unlock(&bond->lock);
2314
2315         return 0;
2316 }
2317
2318 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2319 {
2320         struct bonding *bond = netdev_priv(bond_dev);
2321         struct slave *slave;
2322         int i, res = -ENODEV;
2323
2324         read_lock(&bond->lock);
2325
2326         bond_for_each_slave(bond, slave, i) {
2327                 if (i == (int)info->slave_id) {
2328                         res = 0;
2329                         strcpy(info->slave_name, slave->dev->name);
2330                         info->link = slave->link;
2331                         info->state = bond_slave_state(slave);
2332                         info->link_failure_count = slave->link_failure_count;
2333                         break;
2334                 }
2335         }
2336
2337         read_unlock(&bond->lock);
2338
2339         return res;
2340 }
2341
2342 /*-------------------------------- Monitoring -------------------------------*/
2343
2344
2345 static int bond_miimon_inspect(struct bonding *bond)
2346 {
2347         struct slave *slave;
2348         int i, link_state, commit = 0;
2349         bool ignore_updelay;
2350
2351         ignore_updelay = !bond->curr_active_slave ? true : false;
2352
2353         bond_for_each_slave(bond, slave, i) {
2354                 slave->new_link = BOND_LINK_NOCHANGE;
2355
2356                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2357
2358                 switch (slave->link) {
2359                 case BOND_LINK_UP:
2360                         if (link_state)
2361                                 continue;
2362
2363                         slave->link = BOND_LINK_FAIL;
2364                         slave->delay = bond->params.downdelay;
2365                         if (slave->delay) {
2366                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2367                                         bond->dev->name,
2368                                         (bond->params.mode ==
2369                                          BOND_MODE_ACTIVEBACKUP) ?
2370                                         (bond_is_active_slave(slave) ?
2371                                          "active " : "backup ") : "",
2372                                         slave->dev->name,
2373                                         bond->params.downdelay * bond->params.miimon);
2374                         }
2375                         /*FALLTHRU*/
2376                 case BOND_LINK_FAIL:
2377                         if (link_state) {
2378                                 /*
2379                                  * recovered before downdelay expired
2380                                  */
2381                                 slave->link = BOND_LINK_UP;
2382                                 slave->jiffies = jiffies;
2383                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2384                                         bond->dev->name,
2385                                         (bond->params.downdelay - slave->delay) *
2386                                         bond->params.miimon,
2387                                         slave->dev->name);
2388                                 continue;
2389                         }
2390
2391                         if (slave->delay <= 0) {
2392                                 slave->new_link = BOND_LINK_DOWN;
2393                                 commit++;
2394                                 continue;
2395                         }
2396
2397                         slave->delay--;
2398                         break;
2399
2400                 case BOND_LINK_DOWN:
2401                         if (!link_state)
2402                                 continue;
2403
2404                         slave->link = BOND_LINK_BACK;
2405                         slave->delay = bond->params.updelay;
2406
2407                         if (slave->delay) {
2408                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2409                                         bond->dev->name, slave->dev->name,
2410                                         ignore_updelay ? 0 :
2411                                         bond->params.updelay *
2412                                         bond->params.miimon);
2413                         }
2414                         /*FALLTHRU*/
2415                 case BOND_LINK_BACK:
2416                         if (!link_state) {
2417                                 slave->link = BOND_LINK_DOWN;
2418                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2419                                         bond->dev->name,
2420                                         (bond->params.updelay - slave->delay) *
2421                                         bond->params.miimon,
2422                                         slave->dev->name);
2423
2424                                 continue;
2425                         }
2426
2427                         if (ignore_updelay)
2428                                 slave->delay = 0;
2429
2430                         if (slave->delay <= 0) {
2431                                 slave->new_link = BOND_LINK_UP;
2432                                 commit++;
2433                                 ignore_updelay = false;
2434                                 continue;
2435                         }
2436
2437                         slave->delay--;
2438                         break;
2439                 }
2440         }
2441
2442         return commit;
2443 }
2444
2445 static void bond_miimon_commit(struct bonding *bond)
2446 {
2447         struct slave *slave;
2448         int i;
2449
2450         bond_for_each_slave(bond, slave, i) {
2451                 switch (slave->new_link) {
2452                 case BOND_LINK_NOCHANGE:
2453                         continue;
2454
2455                 case BOND_LINK_UP:
2456                         slave->link = BOND_LINK_UP;
2457                         slave->jiffies = jiffies;
2458
2459                         if (bond->params.mode == BOND_MODE_8023AD) {
2460                                 /* prevent it from being the active one */
2461                                 bond_set_backup_slave(slave);
2462                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2463                                 /* make it immediately active */
2464                                 bond_set_active_slave(slave);
2465                         } else if (slave != bond->primary_slave) {
2466                                 /* prevent it from being the active one */
2467                                 bond_set_backup_slave(slave);
2468                         }
2469
2470                         bond_update_speed_duplex(slave);
2471
2472                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2473                                 bond->dev->name, slave->dev->name,
2474                                 slave->speed, slave->duplex ? "full" : "half");
2475
2476                         /* notify ad that the link status has changed */
2477                         if (bond->params.mode == BOND_MODE_8023AD)
2478                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2479
2480                         if (bond_is_lb(bond))
2481                                 bond_alb_handle_link_change(bond, slave,
2482                                                             BOND_LINK_UP);
2483
2484                         if (!bond->curr_active_slave ||
2485                             (slave == bond->primary_slave))
2486                                 goto do_failover;
2487
2488                         continue;
2489
2490                 case BOND_LINK_DOWN:
2491                         if (slave->link_failure_count < UINT_MAX)
2492                                 slave->link_failure_count++;
2493
2494                         slave->link = BOND_LINK_DOWN;
2495
2496                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2497                             bond->params.mode == BOND_MODE_8023AD)
2498                                 bond_set_slave_inactive_flags(slave);
2499
2500                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2501                                 bond->dev->name, slave->dev->name);
2502
2503                         if (bond->params.mode == BOND_MODE_8023AD)
2504                                 bond_3ad_handle_link_change(slave,
2505                                                             BOND_LINK_DOWN);
2506
2507                         if (bond_is_lb(bond))
2508                                 bond_alb_handle_link_change(bond, slave,
2509                                                             BOND_LINK_DOWN);
2510
2511                         if (slave == bond->curr_active_slave)
2512                                 goto do_failover;
2513
2514                         continue;
2515
2516                 default:
2517                         pr_err("%s: invalid new link %d on slave %s\n",
2518                                bond->dev->name, slave->new_link,
2519                                slave->dev->name);
2520                         slave->new_link = BOND_LINK_NOCHANGE;
2521
2522                         continue;
2523                 }
2524
2525 do_failover:
2526                 ASSERT_RTNL();
2527                 block_netpoll_tx();
2528                 write_lock_bh(&bond->curr_slave_lock);
2529                 bond_select_active_slave(bond);
2530                 write_unlock_bh(&bond->curr_slave_lock);
2531                 unblock_netpoll_tx();
2532         }
2533
2534         bond_set_carrier(bond);
2535 }
2536
2537 /*
2538  * bond_mii_monitor
2539  *
2540  * Really a wrapper that splits the mii monitor into two phases: an
2541  * inspection, then (if inspection indicates something needs to be done)
2542  * an acquisition of appropriate locks followed by a commit phase to
2543  * implement whatever link state changes are indicated.
2544  */
2545 void bond_mii_monitor(struct work_struct *work)
2546 {
2547         struct bonding *bond = container_of(work, struct bonding,
2548                                             mii_work.work);
2549         bool should_notify_peers = false;
2550         unsigned long delay;
2551
2552         read_lock(&bond->lock);
2553
2554         delay = msecs_to_jiffies(bond->params.miimon);
2555
2556         if (bond->slave_cnt == 0)
2557                 goto re_arm;
2558
2559         should_notify_peers = bond_should_notify_peers(bond);
2560
2561         if (bond_miimon_inspect(bond)) {
2562                 read_unlock(&bond->lock);
2563
2564                 /* Race avoidance with bond_close cancel of workqueue */
2565                 if (!rtnl_trylock()) {
2566                         read_lock(&bond->lock);
2567                         delay = 1;
2568                         should_notify_peers = false;
2569                         goto re_arm;
2570                 }
2571
2572                 read_lock(&bond->lock);
2573
2574                 bond_miimon_commit(bond);
2575
2576                 read_unlock(&bond->lock);
2577                 rtnl_unlock();  /* might sleep, hold no other locks */
2578                 read_lock(&bond->lock);
2579         }
2580
2581 re_arm:
2582         if (bond->params.miimon)
2583                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2584
2585         read_unlock(&bond->lock);
2586
2587         if (should_notify_peers) {
2588                 if (!rtnl_trylock()) {
2589                         read_lock(&bond->lock);
2590                         bond->send_peer_notif++;
2591                         read_unlock(&bond->lock);
2592                         return;
2593                 }
2594                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2595                 rtnl_unlock();
2596         }
2597 }
2598
2599 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2600 {
2601         struct vlan_entry *vlan;
2602         struct net_device *vlan_dev;
2603
2604         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2605                 return 1;
2606
2607         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2608                 rcu_read_lock();
2609                 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2610                 rcu_read_unlock();
2611                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2612                         return 1;
2613         }
2614
2615         return 0;
2616 }
2617
2618 /*
2619  * We go to the (large) trouble of VLAN tagging ARP frames because
2620  * switches in VLAN mode (especially if ports are configured as
2621  * "native" to a VLAN) might not pass non-tagged frames.
2622  */
2623 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2624 {
2625         struct sk_buff *skb;
2626
2627         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2628                  slave_dev->name, dest_ip, src_ip, vlan_id);
2629
2630         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2631                          NULL, slave_dev->dev_addr, NULL);
2632
2633         if (!skb) {
2634                 pr_err("ARP packet allocation failed\n");
2635                 return;
2636         }
2637         if (vlan_id) {
2638                 skb = vlan_put_tag(skb, vlan_id);
2639                 if (!skb) {
2640                         pr_err("failed to insert VLAN tag\n");
2641                         return;
2642                 }
2643         }
2644         arp_xmit(skb);
2645 }
2646
2647
2648 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2649 {
2650         int i, vlan_id;
2651         __be32 *targets = bond->params.arp_targets;
2652         struct vlan_entry *vlan;
2653         struct net_device *vlan_dev = NULL;
2654         struct rtable *rt;
2655
2656         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2657                 __be32 addr;
2658                 if (!targets[i])
2659                         break;
2660                 pr_debug("basa: target %x\n", targets[i]);
2661                 if (!bond_vlan_used(bond)) {
2662                         pr_debug("basa: empty vlan: arp_send\n");
2663                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2664                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2665                                       addr, 0);
2666                         continue;
2667                 }
2668
2669                 /*
2670                  * If VLANs are configured, we do a route lookup to
2671                  * determine which VLAN interface would be used, so we
2672                  * can tag the ARP with the proper VLAN tag.
2673                  */
2674                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2675                                      RTO_ONLINK, 0);
2676                 if (IS_ERR(rt)) {
2677                         if (net_ratelimit()) {
2678                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2679                                            bond->dev->name, &targets[i]);
2680                         }
2681                         continue;
2682                 }
2683
2684                 /*
2685                  * This target is not on a VLAN
2686                  */
2687                 if (rt->dst.dev == bond->dev) {
2688                         ip_rt_put(rt);
2689                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2690                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2691                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2692                                       addr, 0);
2693                         continue;
2694                 }
2695
2696                 vlan_id = 0;
2697                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2698                         rcu_read_lock();
2699                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2700                                                         vlan->vlan_id);
2701                         rcu_read_unlock();
2702                         if (vlan_dev == rt->dst.dev) {
2703                                 vlan_id = vlan->vlan_id;
2704                                 pr_debug("basa: vlan match on %s %d\n",
2705                                        vlan_dev->name, vlan_id);
2706                                 break;
2707                         }
2708                 }
2709
2710                 if (vlan_id && vlan_dev) {
2711                         ip_rt_put(rt);
2712                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2713                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2714                                       addr, vlan_id);
2715                         continue;
2716                 }
2717
2718                 if (net_ratelimit()) {
2719                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2720                                    bond->dev->name, &targets[i],
2721                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2722                 }
2723                 ip_rt_put(rt);
2724         }
2725 }
2726
2727 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2728 {
2729         int i;
2730         __be32 *targets = bond->params.arp_targets;
2731
2732         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2733                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2734                          &sip, &tip, i, &targets[i],
2735                          bond_has_this_ip(bond, tip));
2736                 if (sip == targets[i]) {
2737                         if (bond_has_this_ip(bond, tip))
2738                                 slave->last_arp_rx = jiffies;
2739                         return;
2740                 }
2741         }
2742 }
2743
2744 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2745                         struct slave *slave)
2746 {
2747         struct arphdr *arp = (struct arphdr *)skb->data;
2748         unsigned char *arp_ptr;
2749         __be32 sip, tip;
2750         int alen;
2751
2752         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2753                 return RX_HANDLER_ANOTHER;
2754
2755         read_lock(&bond->lock);
2756         alen = arp_hdr_len(bond->dev);
2757
2758         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2759                  bond->dev->name, skb->dev->name);
2760
2761         if (alen > skb_headlen(skb)) {
2762                 arp = kmalloc(alen, GFP_ATOMIC);
2763                 if (!arp)
2764                         goto out_unlock;
2765                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2766                         goto out_unlock;
2767         }
2768
2769         if (arp->ar_hln != bond->dev->addr_len ||
2770             skb->pkt_type == PACKET_OTHERHOST ||
2771             skb->pkt_type == PACKET_LOOPBACK ||
2772             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2773             arp->ar_pro != htons(ETH_P_IP) ||
2774             arp->ar_pln != 4)
2775                 goto out_unlock;
2776
2777         arp_ptr = (unsigned char *)(arp + 1);
2778         arp_ptr += bond->dev->addr_len;
2779         memcpy(&sip, arp_ptr, 4);
2780         arp_ptr += 4 + bond->dev->addr_len;
2781         memcpy(&tip, arp_ptr, 4);
2782
2783         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2784                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2785                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2786                  &sip, &tip);
2787
2788         /*
2789          * Backup slaves won't see the ARP reply, but do come through
2790          * here for each ARP probe (so we swap the sip/tip to validate
2791          * the probe).  In a "redundant switch, common router" type of
2792          * configuration, the ARP probe will (hopefully) travel from
2793          * the active, through one switch, the router, then the other
2794          * switch before reaching the backup.
2795          */
2796         if (bond_is_active_slave(slave))
2797                 bond_validate_arp(bond, slave, sip, tip);
2798         else
2799                 bond_validate_arp(bond, slave, tip, sip);
2800
2801 out_unlock:
2802         read_unlock(&bond->lock);
2803         if (arp != (struct arphdr *)skb->data)
2804                 kfree(arp);
2805         return RX_HANDLER_ANOTHER;
2806 }
2807
2808 /*
2809  * this function is called regularly to monitor each slave's link
2810  * ensuring that traffic is being sent and received when arp monitoring
2811  * is used in load-balancing mode. if the adapter has been dormant, then an
2812  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2813  * arp monitoring in active backup mode.
2814  */
2815 void bond_loadbalance_arp_mon(struct work_struct *work)
2816 {
2817         struct bonding *bond = container_of(work, struct bonding,
2818                                             arp_work.work);
2819         struct slave *slave, *oldcurrent;
2820         int do_failover = 0;
2821         int delta_in_ticks, extra_ticks;
2822         int i;
2823
2824         read_lock(&bond->lock);
2825
2826         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2827         extra_ticks = delta_in_ticks / 2;
2828
2829         if (bond->slave_cnt == 0)
2830                 goto re_arm;
2831
2832         read_lock(&bond->curr_slave_lock);
2833         oldcurrent = bond->curr_active_slave;
2834         read_unlock(&bond->curr_slave_lock);
2835
2836         /* see if any of the previous devices are up now (i.e. they have
2837          * xmt and rcv traffic). the curr_active_slave does not come into
2838          * the picture unless it is null. also, slave->jiffies is not needed
2839          * here because we send an arp on each slave and give a slave as
2840          * long as it needs to get the tx/rx within the delta.
2841          * TODO: what about up/down delay in arp mode? it wasn't here before
2842          *       so it can wait
2843          */
2844         bond_for_each_slave(bond, slave, i) {
2845                 unsigned long trans_start = dev_trans_start(slave->dev);
2846
2847                 if (slave->link != BOND_LINK_UP) {
2848                         if (time_in_range(jiffies,
2849                                 trans_start - delta_in_ticks,
2850                                 trans_start + delta_in_ticks + extra_ticks) &&
2851                             time_in_range(jiffies,
2852                                 slave->dev->last_rx - delta_in_ticks,
2853                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2854
2855                                 slave->link  = BOND_LINK_UP;
2856                                 bond_set_active_slave(slave);
2857
2858                                 /* primary_slave has no meaning in round-robin
2859                                  * mode. the window of a slave being up and
2860                                  * curr_active_slave being null after enslaving
2861                                  * is closed.
2862                                  */
2863                                 if (!oldcurrent) {
2864                                         pr_info("%s: link status definitely up for interface %s, ",
2865                                                 bond->dev->name,
2866                                                 slave->dev->name);
2867                                         do_failover = 1;
2868                                 } else {
2869                                         pr_info("%s: interface %s is now up\n",
2870                                                 bond->dev->name,
2871                                                 slave->dev->name);
2872                                 }
2873                         }
2874                 } else {
2875                         /* slave->link == BOND_LINK_UP */
2876
2877                         /* not all switches will respond to an arp request
2878                          * when the source ip is 0, so don't take the link down
2879                          * if we don't know our ip yet
2880                          */
2881                         if (!time_in_range(jiffies,
2882                                 trans_start - delta_in_ticks,
2883                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2884                             !time_in_range(jiffies,
2885                                 slave->dev->last_rx - delta_in_ticks,
2886                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2887
2888                                 slave->link  = BOND_LINK_DOWN;
2889                                 bond_set_backup_slave(slave);
2890
2891                                 if (slave->link_failure_count < UINT_MAX)
2892                                         slave->link_failure_count++;
2893
2894                                 pr_info("%s: interface %s is now down.\n",
2895                                         bond->dev->name,
2896                                         slave->dev->name);
2897
2898                                 if (slave == oldcurrent)
2899                                         do_failover = 1;
2900                         }
2901                 }
2902
2903                 /* note: if switch is in round-robin mode, all links
2904                  * must tx arp to ensure all links rx an arp - otherwise
2905                  * links may oscillate or not come up at all; if switch is
2906                  * in something like xor mode, there is nothing we can
2907                  * do - all replies will be rx'ed on same link causing slaves
2908                  * to be unstable during low/no traffic periods
2909                  */
2910                 if (IS_UP(slave->dev))
2911                         bond_arp_send_all(bond, slave);
2912         }
2913
2914         if (do_failover) {
2915                 block_netpoll_tx();
2916                 write_lock_bh(&bond->curr_slave_lock);
2917
2918                 bond_select_active_slave(bond);
2919
2920                 write_unlock_bh(&bond->curr_slave_lock);
2921                 unblock_netpoll_tx();
2922         }
2923
2924 re_arm:
2925         if (bond->params.arp_interval)
2926                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2927
2928         read_unlock(&bond->lock);
2929 }
2930
2931 /*
2932  * Called to inspect slaves for active-backup mode ARP monitor link state
2933  * changes.  Sets new_link in slaves to specify what action should take
2934  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2935  * to link states must be committed.
2936  *
2937  * Called with bond->lock held for read.
2938  */
2939 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2940 {
2941         struct slave *slave;
2942         int i, commit = 0;
2943         unsigned long trans_start;
2944         int extra_ticks;
2945
2946         /* All the time comparisons below need some extra time. Otherwise, on
2947          * fast networks the ARP probe/reply may arrive within the same jiffy
2948          * as it was sent.  Then, the next time the ARP monitor is run, one
2949          * arp_interval will already have passed in the comparisons.
2950          */
2951         extra_ticks = delta_in_ticks / 2;
2952
2953         bond_for_each_slave(bond, slave, i) {
2954                 slave->new_link = BOND_LINK_NOCHANGE;
2955
2956                 if (slave->link != BOND_LINK_UP) {
2957                         if (time_in_range(jiffies,
2958                                 slave_last_rx(bond, slave) - delta_in_ticks,
2959                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2960
2961                                 slave->new_link = BOND_LINK_UP;
2962                                 commit++;
2963                         }
2964
2965                         continue;
2966                 }
2967
2968                 /*
2969                  * Give slaves 2*delta after being enslaved or made
2970                  * active.  This avoids bouncing, as the last receive
2971                  * times need a full ARP monitor cycle to be updated.
2972                  */
2973                 if (time_in_range(jiffies,
2974                                   slave->jiffies - delta_in_ticks,
2975                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2976                         continue;
2977
2978                 /*
2979                  * Backup slave is down if:
2980                  * - No current_arp_slave AND
2981                  * - more than 3*delta since last receive AND
2982                  * - the bond has an IP address
2983                  *
2984                  * Note: a non-null current_arp_slave indicates
2985                  * the curr_active_slave went down and we are
2986                  * searching for a new one; under this condition
2987                  * we only take the curr_active_slave down - this
2988                  * gives each slave a chance to tx/rx traffic
2989                  * before being taken out
2990                  */
2991                 if (!bond_is_active_slave(slave) &&
2992                     !bond->current_arp_slave &&
2993                     !time_in_range(jiffies,
2994                         slave_last_rx(bond, slave) - delta_in_ticks,
2995                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2996
2997                         slave->new_link = BOND_LINK_DOWN;
2998                         commit++;
2999                 }
3000
3001                 /*
3002                  * Active slave is down if:
3003                  * - more than 2*delta since transmitting OR
3004                  * - (more than 2*delta since receive AND
3005                  *    the bond has an IP address)
3006                  */
3007                 trans_start = dev_trans_start(slave->dev);
3008                 if (bond_is_active_slave(slave) &&
3009                     (!time_in_range(jiffies,
3010                         trans_start - delta_in_ticks,
3011                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
3012                      !time_in_range(jiffies,
3013                         slave_last_rx(bond, slave) - delta_in_ticks,
3014                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
3015
3016                         slave->new_link = BOND_LINK_DOWN;
3017                         commit++;
3018                 }
3019         }
3020
3021         return commit;
3022 }
3023
3024 /*
3025  * Called to commit link state changes noted by inspection step of
3026  * active-backup mode ARP monitor.
3027  *
3028  * Called with RTNL and bond->lock for read.
3029  */
3030 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3031 {
3032         struct slave *slave;
3033         int i;
3034         unsigned long trans_start;
3035
3036         bond_for_each_slave(bond, slave, i) {
3037                 switch (slave->new_link) {
3038                 case BOND_LINK_NOCHANGE:
3039                         continue;
3040
3041                 case BOND_LINK_UP:
3042                         trans_start = dev_trans_start(slave->dev);
3043                         if ((!bond->curr_active_slave &&
3044                              time_in_range(jiffies,
3045                                            trans_start - delta_in_ticks,
3046                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
3047                             bond->curr_active_slave != slave) {
3048                                 slave->link = BOND_LINK_UP;
3049                                 if (bond->current_arp_slave) {
3050                                         bond_set_slave_inactive_flags(
3051                                                 bond->current_arp_slave);
3052                                         bond->current_arp_slave = NULL;
3053                                 }
3054
3055                                 pr_info("%s: link status definitely up for interface %s.\n",
3056                                         bond->dev->name, slave->dev->name);
3057
3058                                 if (!bond->curr_active_slave ||
3059                                     (slave == bond->primary_slave))
3060                                         goto do_failover;
3061
3062                         }
3063
3064                         continue;
3065
3066                 case BOND_LINK_DOWN:
3067                         if (slave->link_failure_count < UINT_MAX)
3068                                 slave->link_failure_count++;
3069
3070                         slave->link = BOND_LINK_DOWN;
3071                         bond_set_slave_inactive_flags(slave);
3072
3073                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3074                                 bond->dev->name, slave->dev->name);
3075
3076                         if (slave == bond->curr_active_slave) {
3077                                 bond->current_arp_slave = NULL;
3078                                 goto do_failover;
3079                         }
3080
3081                         continue;
3082
3083                 default:
3084                         pr_err("%s: impossible: new_link %d on slave %s\n",
3085                                bond->dev->name, slave->new_link,
3086                                slave->dev->name);
3087                         continue;
3088                 }
3089
3090 do_failover:
3091                 ASSERT_RTNL();
3092                 block_netpoll_tx();
3093                 write_lock_bh(&bond->curr_slave_lock);
3094                 bond_select_active_slave(bond);
3095                 write_unlock_bh(&bond->curr_slave_lock);
3096                 unblock_netpoll_tx();
3097         }
3098
3099         bond_set_carrier(bond);
3100 }
3101
3102 /*
3103  * Send ARP probes for active-backup mode ARP monitor.
3104  *
3105  * Called with bond->lock held for read.
3106  */
3107 static void bond_ab_arp_probe(struct bonding *bond)
3108 {
3109         struct slave *slave;
3110         int i;
3111
3112         read_lock(&bond->curr_slave_lock);
3113
3114         if (bond->current_arp_slave && bond->curr_active_slave)
3115                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3116                         bond->current_arp_slave->dev->name,
3117                         bond->curr_active_slave->dev->name);
3118
3119         if (bond->curr_active_slave) {
3120                 bond_arp_send_all(bond, bond->curr_active_slave);
3121                 read_unlock(&bond->curr_slave_lock);
3122                 return;
3123         }
3124
3125         read_unlock(&bond->curr_slave_lock);
3126
3127         /* if we don't have a curr_active_slave, search for the next available
3128          * backup slave from the current_arp_slave and make it the candidate
3129          * for becoming the curr_active_slave
3130          */
3131
3132         if (!bond->current_arp_slave) {
3133                 bond->current_arp_slave = bond->first_slave;
3134                 if (!bond->current_arp_slave)
3135                         return;
3136         }
3137
3138         bond_set_slave_inactive_flags(bond->current_arp_slave);
3139
3140         /* search for next candidate */
3141         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3142                 if (IS_UP(slave->dev)) {
3143                         slave->link = BOND_LINK_BACK;
3144                         bond_set_slave_active_flags(slave);
3145                         bond_arp_send_all(bond, slave);
3146                         slave->jiffies = jiffies;
3147                         bond->current_arp_slave = slave;
3148                         break;
3149                 }
3150
3151                 /* if the link state is up at this point, we
3152                  * mark it down - this can happen if we have
3153                  * simultaneous link failures and
3154                  * reselect_active_interface doesn't make this
3155                  * one the current slave so it is still marked
3156                  * up when it is actually down
3157                  */
3158                 if (slave->link == BOND_LINK_UP) {
3159                         slave->link = BOND_LINK_DOWN;
3160                         if (slave->link_failure_count < UINT_MAX)
3161                                 slave->link_failure_count++;
3162
3163                         bond_set_slave_inactive_flags(slave);
3164
3165                         pr_info("%s: backup interface %s is now down.\n",
3166                                 bond->dev->name, slave->dev->name);
3167                 }
3168         }
3169 }
3170
3171 void bond_activebackup_arp_mon(struct work_struct *work)
3172 {
3173         struct bonding *bond = container_of(work, struct bonding,
3174                                             arp_work.work);
3175         bool should_notify_peers = false;
3176         int delta_in_ticks;
3177
3178         read_lock(&bond->lock);
3179
3180         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3181
3182         if (bond->slave_cnt == 0)
3183                 goto re_arm;
3184
3185         should_notify_peers = bond_should_notify_peers(bond);
3186
3187         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3188                 read_unlock(&bond->lock);
3189
3190                 /* Race avoidance with bond_close flush of workqueue */
3191                 if (!rtnl_trylock()) {
3192                         read_lock(&bond->lock);
3193                         delta_in_ticks = 1;
3194                         should_notify_peers = false;
3195                         goto re_arm;
3196                 }
3197
3198                 read_lock(&bond->lock);
3199
3200                 bond_ab_arp_commit(bond, delta_in_ticks);
3201
3202                 read_unlock(&bond->lock);
3203                 rtnl_unlock();
3204                 read_lock(&bond->lock);
3205         }
3206
3207         bond_ab_arp_probe(bond);
3208
3209 re_arm:
3210         if (bond->params.arp_interval)
3211                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3212
3213         read_unlock(&bond->lock);
3214
3215         if (should_notify_peers) {
3216                 if (!rtnl_trylock()) {
3217                         read_lock(&bond->lock);
3218                         bond->send_peer_notif++;
3219                         read_unlock(&bond->lock);
3220                         return;
3221                 }
3222                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3223                 rtnl_unlock();
3224         }
3225 }
3226
3227 /*-------------------------- netdev event handling --------------------------*/
3228
3229 /*
3230  * Change device name
3231  */
3232 static int bond_event_changename(struct bonding *bond)
3233 {
3234         bond_remove_proc_entry(bond);
3235         bond_create_proc_entry(bond);
3236
3237         bond_debug_reregister(bond);
3238
3239         return NOTIFY_DONE;
3240 }
3241
3242 static int bond_master_netdev_event(unsigned long event,
3243                                     struct net_device *bond_dev)
3244 {
3245         struct bonding *event_bond = netdev_priv(bond_dev);
3246
3247         switch (event) {
3248         case NETDEV_CHANGENAME:
3249                 return bond_event_changename(event_bond);
3250         case NETDEV_UNREGISTER:
3251                 bond_remove_proc_entry(event_bond);
3252                 break;
3253         case NETDEV_REGISTER:
3254                 bond_create_proc_entry(event_bond);
3255                 break;
3256         default:
3257                 break;
3258         }
3259
3260         return NOTIFY_DONE;
3261 }
3262
3263 static int bond_slave_netdev_event(unsigned long event,
3264                                    struct net_device *slave_dev)
3265 {
3266         struct net_device *bond_dev = slave_dev->master;
3267         struct bonding *bond = netdev_priv(bond_dev);
3268         struct slave *slave = NULL;
3269
3270         switch (event) {
3271         case NETDEV_UNREGISTER:
3272                 if (bond_dev) {
3273                         if (bond->setup_by_slave)
3274                                 bond_release_and_destroy(bond_dev, slave_dev);
3275                         else
3276                                 bond_release(bond_dev, slave_dev);
3277                 }
3278                 break;
3279         case NETDEV_UP:
3280         case NETDEV_CHANGE:
3281                 slave = bond_get_slave_by_dev(bond, slave_dev);
3282                 if (slave) {
3283                         u32 old_speed = slave->speed;
3284                         u8  old_duplex = slave->duplex;
3285
3286                         bond_update_speed_duplex(slave);
3287
3288                         if (bond->params.mode == BOND_MODE_8023AD) {
3289                                 if (old_speed != slave->speed)
3290                                         bond_3ad_adapter_speed_changed(slave);
3291                                 if (old_duplex != slave->duplex)
3292                                         bond_3ad_adapter_duplex_changed(slave);
3293                         }
3294                 }
3295
3296                 break;
3297         case NETDEV_DOWN:
3298                 /*
3299                  * ... Or is it this?
3300                  */
3301                 break;
3302         case NETDEV_CHANGEMTU:
3303                 /*
3304                  * TODO: Should slaves be allowed to
3305                  * independently alter their MTU?  For
3306                  * an active-backup bond, slaves need
3307                  * not be the same type of device, so
3308                  * MTUs may vary.  For other modes,
3309                  * slaves arguably should have the
3310                  * same MTUs. To do this, we'd need to
3311                  * take over the slave's change_mtu
3312                  * function for the duration of their
3313                  * servitude.
3314                  */
3315                 break;
3316         case NETDEV_CHANGENAME:
3317                 /*
3318                  * TODO: handle changing the primary's name
3319                  */
3320                 break;
3321         case NETDEV_FEAT_CHANGE:
3322                 bond_compute_features(bond);
3323                 break;
3324         default:
3325                 break;
3326         }
3327
3328         return NOTIFY_DONE;
3329 }
3330
3331 /*
3332  * bond_netdev_event: handle netdev notifier chain events.
3333  *
3334  * This function receives events for the netdev chain.  The caller (an
3335  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3336  * locks for us to safely manipulate the slave devices (RTNL lock,
3337  * dev_probe_lock).
3338  */
3339 static int bond_netdev_event(struct notifier_block *this,
3340                              unsigned long event, void *ptr)
3341 {
3342         struct net_device *event_dev = (struct net_device *)ptr;
3343
3344         pr_debug("event_dev: %s, event: %lx\n",
3345                  event_dev ? event_dev->name : "None",
3346                  event);
3347
3348         if (!(event_dev->priv_flags & IFF_BONDING))
3349                 return NOTIFY_DONE;
3350
3351         if (event_dev->flags & IFF_MASTER) {
3352                 pr_debug("IFF_MASTER\n");
3353                 return bond_master_netdev_event(event, event_dev);
3354         }
3355
3356         if (event_dev->flags & IFF_SLAVE) {
3357                 pr_debug("IFF_SLAVE\n");
3358                 return bond_slave_netdev_event(event, event_dev);
3359         }
3360
3361         return NOTIFY_DONE;
3362 }
3363
3364 static struct notifier_block bond_netdev_notifier = {
3365         .notifier_call = bond_netdev_event,
3366 };
3367
3368 /*---------------------------- Hashing Policies -----------------------------*/
3369
3370 /*
3371  * Hash for the output device based upon layer 2 data
3372  */
3373 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3374 {
3375         struct ethhdr *data = (struct ethhdr *)skb->data;
3376
3377         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3378                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3379
3380         return 0;
3381 }
3382
3383 /*
3384  * Hash for the output device based upon layer 2 and layer 3 data. If
3385  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3386  */
3387 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3388 {
3389         struct ethhdr *data = (struct ethhdr *)skb->data;
3390         struct iphdr *iph;
3391         struct ipv6hdr *ipv6h;
3392         u32 v6hash;
3393         __be32 *s, *d;
3394
3395         if (skb->protocol == htons(ETH_P_IP) &&
3396             skb_network_header_len(skb) >= sizeof(*iph)) {
3397                 iph = ip_hdr(skb);
3398                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3399                         (data->h_dest[5] ^ data->h_source[5])) % count;
3400         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3401                    skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3402                 ipv6h = ipv6_hdr(skb);
3403                 s = &ipv6h->saddr.s6_addr32[0];
3404                 d = &ipv6h->daddr.s6_addr32[0];
3405                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3406                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3407                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3408         }
3409
3410         return bond_xmit_hash_policy_l2(skb, count);
3411 }
3412
3413 /*
3414  * Hash for the output device based upon layer 3 and layer 4 data. If
3415  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3416  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3417  */
3418 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3419 {
3420         u32 layer4_xor = 0;
3421         struct iphdr *iph;
3422         struct ipv6hdr *ipv6h;
3423         __be32 *s, *d;
3424         __be16 *layer4hdr;
3425
3426         if (skb->protocol == htons(ETH_P_IP) &&
3427             skb_network_header_len(skb) >= sizeof(*iph)) {
3428                 iph = ip_hdr(skb);
3429                 if (!ip_is_fragment(iph) &&
3430                     (iph->protocol == IPPROTO_TCP ||
3431                      iph->protocol == IPPROTO_UDP) &&
3432                     (skb_headlen(skb) - skb_network_offset(skb) >=
3433                      iph->ihl * sizeof(u32) + sizeof(*layer4hdr) * 2)) {
3434                         layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3435                         layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3436                 }
3437                 return (layer4_xor ^
3438                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3439         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3440                    skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3441                 ipv6h = ipv6_hdr(skb);
3442                 if ((ipv6h->nexthdr == IPPROTO_TCP ||
3443                      ipv6h->nexthdr == IPPROTO_UDP) &&
3444                     (skb_headlen(skb) - skb_network_offset(skb) >=
3445                      sizeof(*ipv6h) + sizeof(*layer4hdr) * 2)) {
3446                         layer4hdr = (__be16 *)(ipv6h + 1);
3447                         layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3448                 }
3449                 s = &ipv6h->saddr.s6_addr32[0];
3450                 d = &ipv6h->daddr.s6_addr32[0];
3451                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3452                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3453                                (layer4_xor >> 8);
3454                 return layer4_xor % count;
3455         }
3456
3457         return bond_xmit_hash_policy_l2(skb, count);
3458 }
3459
3460 /*-------------------------- Device entry points ----------------------------*/
3461
3462 static void bond_work_init_all(struct bonding *bond)
3463 {
3464         INIT_DELAYED_WORK(&bond->mcast_work,
3465                           bond_resend_igmp_join_requests_delayed);
3466         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3467         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3468         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3469                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3470         else
3471                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3472         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3473 }
3474
3475 static void bond_work_cancel_all(struct bonding *bond)
3476 {
3477         cancel_delayed_work_sync(&bond->mii_work);
3478         cancel_delayed_work_sync(&bond->arp_work);
3479         cancel_delayed_work_sync(&bond->alb_work);
3480         cancel_delayed_work_sync(&bond->ad_work);
3481         cancel_delayed_work_sync(&bond->mcast_work);
3482 }
3483
3484 static int bond_open(struct net_device *bond_dev)
3485 {
3486         struct bonding *bond = netdev_priv(bond_dev);
3487         struct slave *slave;
3488         int i;
3489
3490         /* reset slave->backup and slave->inactive */
3491         read_lock(&bond->lock);
3492         if (bond->slave_cnt > 0) {
3493                 read_lock(&bond->curr_slave_lock);
3494                 bond_for_each_slave(bond, slave, i) {
3495                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3496                                 && (slave != bond->curr_active_slave)) {
3497                                 bond_set_slave_inactive_flags(slave);
3498                         } else {
3499                                 bond_set_slave_active_flags(slave);
3500                         }
3501                 }
3502                 read_unlock(&bond->curr_slave_lock);
3503         }
3504         read_unlock(&bond->lock);
3505
3506         bond_work_init_all(bond);
3507
3508         if (bond_is_lb(bond)) {
3509                 /* bond_alb_initialize must be called before the timer
3510                  * is started.
3511                  */
3512                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3513                         return -ENOMEM;
3514                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3515         }
3516
3517         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3518                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3519
3520         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3521                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3522                 if (bond->params.arp_validate)
3523                         bond->recv_probe = bond_arp_rcv;
3524         }
3525
3526         if (bond->params.mode == BOND_MODE_8023AD) {
3527                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3528                 /* register to receive LACPDUs */
3529                 bond->recv_probe = bond_3ad_lacpdu_recv;
3530                 bond_3ad_initiate_agg_selection(bond, 1);
3531         }
3532
3533         return 0;
3534 }
3535
3536 static int bond_close(struct net_device *bond_dev)
3537 {
3538         struct bonding *bond = netdev_priv(bond_dev);
3539
3540         write_lock_bh(&bond->lock);
3541         bond->send_peer_notif = 0;
3542         write_unlock_bh(&bond->lock);
3543
3544         bond_work_cancel_all(bond);
3545         if (bond_is_lb(bond)) {
3546                 /* Must be called only after all
3547                  * slaves have been released
3548                  */
3549                 bond_alb_deinitialize(bond);
3550         }
3551         bond->recv_probe = NULL;
3552
3553         return 0;
3554 }
3555
3556 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3557                                                 struct rtnl_link_stats64 *stats)
3558 {
3559         struct bonding *bond = netdev_priv(bond_dev);
3560         struct rtnl_link_stats64 temp;
3561         struct slave *slave;
3562         int i;
3563
3564         memset(stats, 0, sizeof(*stats));
3565
3566         read_lock_bh(&bond->lock);
3567
3568         bond_for_each_slave(bond, slave, i) {
3569                 const struct rtnl_link_stats64 *sstats =
3570                         dev_get_stats(slave->dev, &temp);
3571
3572                 stats->rx_packets += sstats->rx_packets;
3573                 stats->rx_bytes += sstats->rx_bytes;
3574                 stats->rx_errors += sstats->rx_errors;
3575                 stats->rx_dropped += sstats->rx_dropped;
3576
3577                 stats->tx_packets += sstats->tx_packets;
3578                 stats->tx_bytes += sstats->tx_bytes;
3579                 stats->tx_errors += sstats->tx_errors;
3580                 stats->tx_dropped += sstats->tx_dropped;
3581
3582                 stats->multicast += sstats->multicast;
3583                 stats->collisions += sstats->collisions;
3584
3585                 stats->rx_length_errors += sstats->rx_length_errors;
3586                 stats->rx_over_errors += sstats->rx_over_errors;
3587                 stats->rx_crc_errors += sstats->rx_crc_errors;
3588                 stats->rx_frame_errors += sstats->rx_frame_errors;
3589                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3590                 stats->rx_missed_errors += sstats->rx_missed_errors;
3591
3592                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3593                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3594                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3595                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3596                 stats->tx_window_errors += sstats->tx_window_errors;
3597         }
3598
3599         read_unlock_bh(&bond->lock);
3600
3601         return stats;
3602 }
3603
3604 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3605 {
3606         struct net_device *slave_dev = NULL;
3607         struct ifbond k_binfo;
3608         struct ifbond __user *u_binfo = NULL;
3609         struct ifslave k_sinfo;
3610         struct ifslave __user *u_sinfo = NULL;
3611         struct mii_ioctl_data *mii = NULL;
3612         int res = 0;
3613
3614         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3615
3616         switch (cmd) {
3617         case SIOCGMIIPHY:
3618                 mii = if_mii(ifr);
3619                 if (!mii)
3620                         return -EINVAL;
3621
3622                 mii->phy_id = 0;
3623                 /* Fall Through */
3624         case SIOCGMIIREG:
3625                 /*
3626                  * We do this again just in case we were called by SIOCGMIIREG
3627                  * instead of SIOCGMIIPHY.
3628                  */
3629                 mii = if_mii(ifr);
3630                 if (!mii)
3631                         return -EINVAL;
3632
3633
3634                 if (mii->reg_num == 1) {
3635                         struct bonding *bond = netdev_priv(bond_dev);
3636                         mii->val_out = 0;
3637                         read_lock(&bond->lock);
3638                         read_lock(&bond->curr_slave_lock);
3639                         if (netif_carrier_ok(bond->dev))
3640                                 mii->val_out = BMSR_LSTATUS;
3641
3642                         read_unlock(&bond->curr_slave_lock);
3643                         read_unlock(&bond->lock);
3644                 }
3645
3646                 return 0;
3647         case BOND_INFO_QUERY_OLD:
3648         case SIOCBONDINFOQUERY:
3649                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3650
3651                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3652                         return -EFAULT;
3653
3654                 res = bond_info_query(bond_dev, &k_binfo);
3655                 if (res == 0 &&
3656                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3657                         return -EFAULT;
3658
3659                 return res;
3660         case BOND_SLAVE_INFO_QUERY_OLD:
3661         case SIOCBONDSLAVEINFOQUERY:
3662                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3663
3664                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3665                         return -EFAULT;
3666
3667                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3668                 if (res == 0 &&
3669                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3670                         return -EFAULT;
3671
3672                 return res;
3673         default:
3674                 /* Go on */
3675                 break;
3676         }
3677
3678         if (!capable(CAP_NET_ADMIN))
3679                 return -EPERM;
3680
3681         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3682
3683         pr_debug("slave_dev=%p:\n", slave_dev);
3684
3685         if (!slave_dev)
3686                 res = -ENODEV;
3687         else {
3688                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3689                 switch (cmd) {
3690                 case BOND_ENSLAVE_OLD:
3691                 case SIOCBONDENSLAVE:
3692                         res = bond_enslave(bond_dev, slave_dev);
3693                         break;
3694                 case BOND_RELEASE_OLD:
3695                 case SIOCBONDRELEASE:
3696                         res = bond_release(bond_dev, slave_dev);
3697                         break;
3698                 case BOND_SETHWADDR_OLD:
3699                 case SIOCBONDSETHWADDR:
3700                         res = bond_sethwaddr(bond_dev, slave_dev);
3701                         break;
3702                 case BOND_CHANGE_ACTIVE_OLD:
3703                 case SIOCBONDCHANGEACTIVE:
3704                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3705                         break;
3706                 default:
3707                         res = -EOPNOTSUPP;
3708                 }
3709
3710                 dev_put(slave_dev);
3711         }
3712
3713         return res;
3714 }
3715
3716 static bool bond_addr_in_mc_list(unsigned char *addr,
3717                                  struct netdev_hw_addr_list *list,
3718                                  int addrlen)
3719 {
3720         struct netdev_hw_addr *ha;
3721
3722         netdev_hw_addr_list_for_each(ha, list)
3723                 if (!memcmp(ha->addr, addr, addrlen))
3724                         return true;
3725
3726         return false;
3727 }
3728
3729 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3730 {
3731         struct bonding *bond = netdev_priv(bond_dev);
3732
3733         if (change & IFF_PROMISC)
3734                 bond_set_promiscuity(bond,
3735                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3736
3737         if (change & IFF_ALLMULTI)
3738                 bond_set_allmulti(bond,
3739                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3740 }
3741
3742 static void bond_set_multicast_list(struct net_device *bond_dev)
3743 {
3744         struct bonding *bond = netdev_priv(bond_dev);
3745         struct netdev_hw_addr *ha;
3746         bool found;
3747
3748         read_lock(&bond->lock);
3749
3750         /* looking for addresses to add to slaves' mc list */
3751         netdev_for_each_mc_addr(ha, bond_dev) {
3752                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3753                                              bond_dev->addr_len);
3754                 if (!found)
3755                         bond_mc_add(bond, ha->addr);
3756         }
3757
3758         /* looking for addresses to delete from slaves' list */
3759         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3760                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3761                                              bond_dev->addr_len);
3762                 if (!found)
3763                         bond_mc_del(bond, ha->addr);
3764         }
3765
3766         /* save master's multicast list */
3767         __hw_addr_flush(&bond->mc_list);
3768         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3769                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3770
3771         read_unlock(&bond->lock);
3772 }
3773
3774 static int bond_neigh_init(struct neighbour *n)
3775 {
3776         struct bonding *bond = netdev_priv(n->dev);
3777         struct slave *slave = bond->first_slave;
3778         const struct net_device_ops *slave_ops;
3779         struct neigh_parms parms;
3780         int ret;
3781
3782         if (!slave)
3783                 return 0;
3784
3785         slave_ops = slave->dev->netdev_ops;
3786
3787         if (!slave_ops->ndo_neigh_setup)
3788                 return 0;
3789
3790         parms.neigh_setup = NULL;
3791         parms.neigh_cleanup = NULL;
3792         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3793         if (ret)
3794                 return ret;
3795
3796         /*
3797          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3798          * after the last slave has been detached.  Assumes that all slaves
3799          * utilize the same neigh_cleanup (true at this writing as only user
3800          * is ipoib).
3801          */
3802         n->parms->neigh_cleanup = parms.neigh_cleanup;
3803
3804         if (!parms.neigh_setup)
3805                 return 0;
3806
3807         return parms.neigh_setup(n);
3808 }
3809
3810 /*
3811  * The bonding ndo_neigh_setup is called at init time beofre any
3812  * slave exists. So we must declare proxy setup function which will
3813  * be used at run time to resolve the actual slave neigh param setup.
3814  */
3815 static int bond_neigh_setup(struct net_device *dev,
3816                             struct neigh_parms *parms)
3817 {
3818         parms->neigh_setup   = bond_neigh_init;
3819
3820         return 0;
3821 }
3822
3823 /*
3824  * Change the MTU of all of a master's slaves to match the master
3825  */
3826 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3827 {
3828         struct bonding *bond = netdev_priv(bond_dev);
3829         struct slave *slave, *stop_at;
3830         int res = 0;
3831         int i;
3832
3833         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3834                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3835
3836         /* Can't hold bond->lock with bh disabled here since
3837          * some base drivers panic. On the other hand we can't
3838          * hold bond->lock without bh disabled because we'll
3839          * deadlock. The only solution is to rely on the fact
3840          * that we're under rtnl_lock here, and the slaves
3841          * list won't change. This doesn't solve the problem
3842          * of setting the slave's MTU while it is
3843          * transmitting, but the assumption is that the base
3844          * driver can handle that.
3845          *
3846          * TODO: figure out a way to safely iterate the slaves
3847          * list, but without holding a lock around the actual
3848          * call to the base driver.
3849          */
3850
3851         bond_for_each_slave(bond, slave, i) {
3852                 pr_debug("s %p s->p %p c_m %p\n",
3853                          slave,
3854                          slave->prev,
3855                          slave->dev->netdev_ops->ndo_change_mtu);
3856
3857                 res = dev_set_mtu(slave->dev, new_mtu);
3858
3859                 if (res) {
3860                         /* If we failed to set the slave's mtu to the new value
3861                          * we must abort the operation even in ACTIVE_BACKUP
3862                          * mode, because if we allow the backup slaves to have
3863                          * different mtu values than the active slave we'll
3864                          * need to change their mtu when doing a failover. That
3865                          * means changing their mtu from timer context, which
3866                          * is probably not a good idea.
3867                          */
3868                         pr_debug("err %d %s\n", res, slave->dev->name);
3869                         goto unwind;
3870                 }
3871         }
3872
3873         bond_dev->mtu = new_mtu;
3874
3875         return 0;
3876
3877 unwind:
3878         /* unwind from head to the slave that failed */
3879         stop_at = slave;
3880         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3881                 int tmp_res;
3882
3883                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3884                 if (tmp_res) {
3885                         pr_debug("unwind err %d dev %s\n",
3886                                  tmp_res, slave->dev->name);
3887                 }
3888         }
3889
3890         return res;
3891 }
3892
3893 /*
3894  * Change HW address
3895  *
3896  * Note that many devices must be down to change the HW address, and
3897  * downing the master releases all slaves.  We can make bonds full of
3898  * bonding devices to test this, however.
3899  */
3900 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3901 {
3902         struct bonding *bond = netdev_priv(bond_dev);
3903         struct sockaddr *sa = addr, tmp_sa;
3904         struct slave *slave, *stop_at;
3905         int res = 0;
3906         int i;
3907
3908         if (bond->params.mode == BOND_MODE_ALB)
3909                 return bond_alb_set_mac_address(bond_dev, addr);
3910
3911
3912         pr_debug("bond=%p, name=%s\n",
3913                  bond, bond_dev ? bond_dev->name : "None");
3914
3915         /*
3916          * If fail_over_mac is set to active, do nothing and return
3917          * success.  Returning an error causes ifenslave to fail.
3918          */
3919         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3920                 return 0;
3921
3922         if (!is_valid_ether_addr(sa->sa_data))
3923                 return -EADDRNOTAVAIL;
3924
3925         /* Can't hold bond->lock with bh disabled here since
3926          * some base drivers panic. On the other hand we can't
3927          * hold bond->lock without bh disabled because we'll
3928          * deadlock. The only solution is to rely on the fact
3929          * that we're under rtnl_lock here, and the slaves
3930          * list won't change. This doesn't solve the problem
3931          * of setting the slave's hw address while it is
3932          * transmitting, but the assumption is that the base
3933          * driver can handle that.
3934          *
3935          * TODO: figure out a way to safely iterate the slaves
3936          * list, but without holding a lock around the actual
3937          * call to the base driver.
3938          */
3939
3940         bond_for_each_slave(bond, slave, i) {
3941                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3942                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3943
3944                 if (slave_ops->ndo_set_mac_address == NULL) {
3945                         res = -EOPNOTSUPP;
3946                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3947                         goto unwind;
3948                 }
3949
3950                 res = dev_set_mac_address(slave->dev, addr);
3951                 if (res) {
3952                         /* TODO: consider downing the slave
3953                          * and retry ?
3954                          * User should expect communications
3955                          * breakage anyway until ARP finish
3956                          * updating, so...
3957                          */
3958                         pr_debug("err %d %s\n", res, slave->dev->name);
3959                         goto unwind;
3960                 }
3961         }
3962
3963         /* success */
3964         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3965         return 0;
3966
3967 unwind:
3968         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3969         tmp_sa.sa_family = bond_dev->type;
3970
3971         /* unwind from head to the slave that failed */
3972         stop_at = slave;
3973         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3974                 int tmp_res;
3975
3976                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3977                 if (tmp_res) {
3978                         pr_debug("unwind err %d dev %s\n",
3979                                  tmp_res, slave->dev->name);
3980                 }
3981         }
3982
3983         return res;
3984 }
3985
3986 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3987 {
3988         struct bonding *bond = netdev_priv(bond_dev);
3989         struct slave *slave, *start_at;
3990         int i, slave_no, res = 1;
3991         struct iphdr *iph = ip_hdr(skb);
3992
3993         /*
3994          * Start with the curr_active_slave that joined the bond as the
3995          * default for sending IGMP traffic.  For failover purposes one
3996          * needs to maintain some consistency for the interface that will
3997          * send the join/membership reports.  The curr_active_slave found
3998          * will send all of this type of traffic.
3999          */
4000         if ((iph->protocol == IPPROTO_IGMP) &&
4001             (skb->protocol == htons(ETH_P_IP))) {
4002
4003                 read_lock(&bond->curr_slave_lock);
4004                 slave = bond->curr_active_slave;
4005                 read_unlock(&bond->curr_slave_lock);
4006
4007                 if (!slave)
4008                         goto out;
4009         } else {
4010                 /*
4011                  * Concurrent TX may collide on rr_tx_counter; we accept
4012                  * that as being rare enough not to justify using an
4013                  * atomic op here.
4014                  */
4015                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4016
4017                 bond_for_each_slave(bond, slave, i) {
4018                         slave_no--;
4019                         if (slave_no < 0)
4020                                 break;
4021                 }
4022         }
4023
4024         start_at = slave;
4025         bond_for_each_slave_from(bond, slave, i, start_at) {
4026                 if (IS_UP(slave->dev) &&
4027                     (slave->link == BOND_LINK_UP) &&
4028                     bond_is_active_slave(slave)) {
4029                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4030                         break;
4031                 }
4032         }
4033
4034 out:
4035         if (res) {
4036                 /* no suitable interface, frame not sent */
4037                 kfree_skb(skb);
4038         }
4039
4040         return NETDEV_TX_OK;
4041 }
4042
4043
4044 /*
4045  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4046  * the bond has a usable interface.
4047  */
4048 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4049 {
4050         struct bonding *bond = netdev_priv(bond_dev);
4051         int res = 1;
4052
4053         read_lock(&bond->curr_slave_lock);
4054
4055         if (bond->curr_active_slave)
4056                 res = bond_dev_queue_xmit(bond, skb,
4057                         bond->curr_active_slave->dev);
4058
4059         read_unlock(&bond->curr_slave_lock);
4060
4061         if (res)
4062                 /* no suitable interface, frame not sent */
4063                 kfree_skb(skb);
4064
4065         return NETDEV_TX_OK;
4066 }
4067
4068 /*
4069  * In bond_xmit_xor() , we determine the output device by using a pre-
4070  * determined xmit_hash_policy(), If the selected device is not enabled,
4071  * find the next active slave.
4072  */
4073 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4074 {
4075         struct bonding *bond = netdev_priv(bond_dev);
4076         struct slave *slave, *start_at;
4077         int slave_no;
4078         int i;
4079         int res = 1;
4080
4081         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4082
4083         bond_for_each_slave(bond, slave, i) {
4084                 slave_no--;
4085                 if (slave_no < 0)
4086                         break;
4087         }
4088
4089         start_at = slave;
4090
4091         bond_for_each_slave_from(bond, slave, i, start_at) {
4092                 if (IS_UP(slave->dev) &&
4093                     (slave->link == BOND_LINK_UP) &&
4094                     bond_is_active_slave(slave)) {
4095                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4096                         break;
4097                 }
4098         }
4099
4100         if (res) {
4101                 /* no suitable interface, frame not sent */
4102                 kfree_skb(skb);
4103         }
4104
4105         return NETDEV_TX_OK;
4106 }
4107
4108 /*
4109  * in broadcast mode, we send everything to all usable interfaces.
4110  */
4111 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4112 {
4113         struct bonding *bond = netdev_priv(bond_dev);
4114         struct slave *slave, *start_at;
4115         struct net_device *tx_dev = NULL;
4116         int i;
4117         int res = 1;
4118
4119         read_lock(&bond->curr_slave_lock);
4120         start_at = bond->curr_active_slave;
4121         read_unlock(&bond->curr_slave_lock);
4122
4123         if (!start_at)
4124                 goto out;
4125
4126         bond_for_each_slave_from(bond, slave, i, start_at) {
4127                 if (IS_UP(slave->dev) &&
4128                     (slave->link == BOND_LINK_UP) &&
4129                     bond_is_active_slave(slave)) {
4130                         if (tx_dev) {
4131                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4132                                 if (!skb2) {
4133                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4134                                                bond_dev->name);
4135                                         continue;
4136                                 }
4137
4138                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4139                                 if (res) {
4140                                         kfree_skb(skb2);
4141                                         continue;
4142                                 }
4143                         }
4144                         tx_dev = slave->dev;
4145                 }
4146         }
4147
4148         if (tx_dev)
4149                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4150
4151 out:
4152         if (res)
4153                 /* no suitable interface, frame not sent */
4154                 kfree_skb(skb);
4155
4156         /* frame sent to all suitable interfaces */
4157         return NETDEV_TX_OK;
4158 }
4159
4160 /*------------------------- Device initialization ---------------------------*/
4161
4162 static void bond_set_xmit_hash_policy(struct bonding *bond)
4163 {
4164         switch (bond->params.xmit_policy) {
4165         case BOND_XMIT_POLICY_LAYER23:
4166                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4167                 break;
4168         case BOND_XMIT_POLICY_LAYER34:
4169                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4170                 break;
4171         case BOND_XMIT_POLICY_LAYER2:
4172         default:
4173                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4174                 break;
4175         }
4176 }
4177
4178 /*
4179  * Lookup the slave that corresponds to a qid
4180  */
4181 static inline int bond_slave_override(struct bonding *bond,
4182                                       struct sk_buff *skb)
4183 {
4184         int i, res = 1;
4185         struct slave *slave = NULL;
4186         struct slave *check_slave;
4187
4188         if (!skb->queue_mapping)
4189                 return 1;
4190
4191         /* Find out if any slaves have the same mapping as this skb. */
4192         bond_for_each_slave(bond, check_slave, i) {
4193                 if (check_slave->queue_id == skb->queue_mapping) {
4194                         slave = check_slave;
4195                         break;
4196                 }
4197         }
4198
4199         /* If the slave isn't UP, use default transmit policy. */
4200         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4201             (slave->link == BOND_LINK_UP)) {
4202                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4203         }
4204
4205         return res;
4206 }
4207
4208
4209 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4210 {
4211         /*
4212          * This helper function exists to help dev_pick_tx get the correct
4213          * destination queue.  Using a helper function skips a call to
4214          * skb_tx_hash and will put the skbs in the queue we expect on their
4215          * way down to the bonding driver.
4216          */
4217         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4218
4219         /*
4220          * Save the original txq to restore before passing to the driver
4221          */
4222         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4223
4224         if (unlikely(txq >= dev->real_num_tx_queues)) {
4225                 do {
4226                         txq -= dev->real_num_tx_queues;
4227                 } while (txq >= dev->real_num_tx_queues);
4228         }
4229         return txq;
4230 }
4231
4232 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4233 {
4234         struct bonding *bond = netdev_priv(dev);
4235
4236         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4237                 if (!bond_slave_override(bond, skb))
4238                         return NETDEV_TX_OK;
4239         }
4240
4241         switch (bond->params.mode) {
4242         case BOND_MODE_ROUNDROBIN:
4243                 return bond_xmit_roundrobin(skb, dev);
4244         case BOND_MODE_ACTIVEBACKUP:
4245                 return bond_xmit_activebackup(skb, dev);
4246         case BOND_MODE_XOR:
4247                 return bond_xmit_xor(skb, dev);
4248         case BOND_MODE_BROADCAST:
4249                 return bond_xmit_broadcast(skb, dev);
4250         case BOND_MODE_8023AD:
4251                 return bond_3ad_xmit_xor(skb, dev);
4252         case BOND_MODE_ALB:
4253         case BOND_MODE_TLB:
4254                 return bond_alb_xmit(skb, dev);
4255         default:
4256                 /* Should never happen, mode already checked */
4257                 pr_err("%s: Error: Unknown bonding mode %d\n",
4258                        dev->name, bond->params.mode);
4259                 WARN_ON_ONCE(1);
4260                 kfree_skb(skb);
4261                 return NETDEV_TX_OK;
4262         }
4263 }
4264
4265 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4266 {
4267         struct bonding *bond = netdev_priv(dev);
4268         netdev_tx_t ret = NETDEV_TX_OK;
4269
4270         /*
4271          * If we risk deadlock from transmitting this in the
4272          * netpoll path, tell netpoll to queue the frame for later tx
4273          */
4274         if (is_netpoll_tx_blocked(dev))
4275                 return NETDEV_TX_BUSY;
4276
4277         read_lock(&bond->lock);
4278
4279         if (bond->slave_cnt)
4280                 ret = __bond_start_xmit(skb, dev);
4281         else
4282                 kfree_skb(skb);
4283
4284         read_unlock(&bond->lock);
4285
4286         return ret;
4287 }
4288
4289 /*
4290  * set bond mode specific net device operations
4291  */
4292 void bond_set_mode_ops(struct bonding *bond, int mode)
4293 {
4294         struct net_device *bond_dev = bond->dev;
4295
4296         switch (mode) {
4297         case BOND_MODE_ROUNDROBIN:
4298                 break;
4299         case BOND_MODE_ACTIVEBACKUP:
4300                 break;
4301         case BOND_MODE_XOR:
4302                 bond_set_xmit_hash_policy(bond);
4303                 break;
4304         case BOND_MODE_BROADCAST:
4305                 break;
4306         case BOND_MODE_8023AD:
4307                 bond_set_xmit_hash_policy(bond);
4308                 break;
4309         case BOND_MODE_ALB:
4310                 /* FALLTHRU */
4311         case BOND_MODE_TLB:
4312                 break;
4313         default:
4314                 /* Should never happen, mode already checked */
4315                 pr_err("%s: Error: Unknown bonding mode %d\n",
4316                        bond_dev->name, mode);
4317                 break;
4318         }
4319 }
4320
4321 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4322                                     struct ethtool_drvinfo *drvinfo)
4323 {
4324         strncpy(drvinfo->driver, DRV_NAME, 32);
4325         strncpy(drvinfo->version, DRV_VERSION, 32);
4326         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4327 }
4328
4329 static const struct ethtool_ops bond_ethtool_ops = {
4330         .get_drvinfo            = bond_ethtool_get_drvinfo,
4331         .get_link               = ethtool_op_get_link,
4332 };
4333
4334 static const struct net_device_ops bond_netdev_ops = {
4335         .ndo_init               = bond_init,
4336         .ndo_uninit             = bond_uninit,
4337         .ndo_open               = bond_open,
4338         .ndo_stop               = bond_close,
4339         .ndo_start_xmit         = bond_start_xmit,
4340         .ndo_select_queue       = bond_select_queue,
4341         .ndo_get_stats64        = bond_get_stats,
4342         .ndo_do_ioctl           = bond_do_ioctl,
4343         .ndo_change_rx_flags    = bond_change_rx_flags,
4344         .ndo_set_rx_mode        = bond_set_multicast_list,
4345         .ndo_change_mtu         = bond_change_mtu,
4346         .ndo_set_mac_address    = bond_set_mac_address,
4347         .ndo_neigh_setup        = bond_neigh_setup,
4348         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4349         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4350 #ifdef CONFIG_NET_POLL_CONTROLLER
4351         .ndo_netpoll_setup      = bond_netpoll_setup,
4352         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4353         .ndo_poll_controller    = bond_poll_controller,
4354 #endif
4355         .ndo_add_slave          = bond_enslave,
4356         .ndo_del_slave          = bond_release,
4357         .ndo_fix_features       = bond_fix_features,
4358 };
4359
4360 static void bond_destructor(struct net_device *bond_dev)
4361 {
4362         struct bonding *bond = netdev_priv(bond_dev);
4363         if (bond->wq)
4364                 destroy_workqueue(bond->wq);
4365         free_netdev(bond_dev);
4366 }
4367
4368 static void bond_setup(struct net_device *bond_dev)
4369 {
4370         struct bonding *bond = netdev_priv(bond_dev);
4371
4372         /* initialize rwlocks */
4373         rwlock_init(&bond->lock);
4374         rwlock_init(&bond->curr_slave_lock);
4375
4376         bond->params = bonding_defaults;
4377
4378         /* Initialize pointers */
4379         bond->dev = bond_dev;
4380         INIT_LIST_HEAD(&bond->vlan_list);
4381
4382         /* Initialize the device entry points */
4383         ether_setup(bond_dev);
4384         bond_dev->netdev_ops = &bond_netdev_ops;
4385         bond_dev->ethtool_ops = &bond_ethtool_ops;
4386         bond_set_mode_ops(bond, bond->params.mode);
4387
4388         bond_dev->destructor = bond_destructor;
4389
4390         /* Initialize the device options */
4391         bond_dev->tx_queue_len = 0;
4392         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4393         bond_dev->priv_flags |= IFF_BONDING;
4394         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4395
4396         /* At first, we block adding VLANs. That's the only way to
4397          * prevent problems that occur when adding VLANs over an
4398          * empty bond. The block will be removed once non-challenged
4399          * slaves are enslaved.
4400          */
4401         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4402
4403         /* don't acquire bond device's netif_tx_lock when
4404          * transmitting */
4405         bond_dev->features |= NETIF_F_LLTX;
4406
4407         /* By default, we declare the bond to be fully
4408          * VLAN hardware accelerated capable. Special
4409          * care is taken in the various xmit functions
4410          * when there are slaves that are not hw accel
4411          * capable
4412          */
4413
4414         bond_dev->hw_features = BOND_VLAN_FEATURES |
4415                                 NETIF_F_HW_VLAN_TX |
4416                                 NETIF_F_HW_VLAN_RX |
4417                                 NETIF_F_HW_VLAN_FILTER;
4418
4419         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4420         bond_dev->features |= bond_dev->hw_features;
4421 }
4422
4423 /*
4424 * Destroy a bonding device.
4425 * Must be under rtnl_lock when this function is called.
4426 */
4427 static void bond_uninit(struct net_device *bond_dev)
4428 {
4429         struct bonding *bond = netdev_priv(bond_dev);
4430         struct vlan_entry *vlan, *tmp;
4431
4432         bond_netpoll_cleanup(bond_dev);
4433
4434         /* Release the bonded slaves */
4435         bond_release_all(bond_dev);
4436
4437         list_del(&bond->bond_list);
4438
4439         bond_work_cancel_all(bond);
4440
4441         bond_debug_unregister(bond);
4442
4443         __hw_addr_flush(&bond->mc_list);
4444
4445         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4446                 list_del(&vlan->vlan_list);
4447                 kfree(vlan);
4448         }
4449 }
4450
4451 /*------------------------- Module initialization ---------------------------*/
4452
4453 /*
4454  * Convert string input module parms.  Accept either the
4455  * number of the mode or its string name.  A bit complicated because
4456  * some mode names are substrings of other names, and calls from sysfs
4457  * may have whitespace in the name (trailing newlines, for example).
4458  */
4459 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4460 {
4461         int modeint = -1, i, rv;
4462         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4463
4464         for (p = (char *)buf; *p; p++)
4465                 if (!(isdigit(*p) || isspace(*p)))
4466                         break;
4467
4468         if (*p)
4469                 rv = sscanf(buf, "%20s", modestr);
4470         else
4471                 rv = sscanf(buf, "%d", &modeint);
4472
4473         if (!rv)
4474                 return -1;
4475
4476         for (i = 0; tbl[i].modename; i++) {
4477                 if (modeint == tbl[i].mode)
4478                         return tbl[i].mode;
4479                 if (strcmp(modestr, tbl[i].modename) == 0)
4480                         return tbl[i].mode;
4481         }
4482
4483         return -1;
4484 }
4485
4486 static int bond_check_params(struct bond_params *params)
4487 {
4488         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4489
4490         /*
4491          * Convert string parameters.
4492          */
4493         if (mode) {
4494                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4495                 if (bond_mode == -1) {
4496                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4497                                mode == NULL ? "NULL" : mode);
4498                         return -EINVAL;
4499                 }
4500         }
4501
4502         if (xmit_hash_policy) {
4503                 if ((bond_mode != BOND_MODE_XOR) &&
4504                     (bond_mode != BOND_MODE_8023AD)) {
4505                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4506                                bond_mode_name(bond_mode));
4507                 } else {
4508                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4509                                                         xmit_hashtype_tbl);
4510                         if (xmit_hashtype == -1) {
4511                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4512                                        xmit_hash_policy == NULL ? "NULL" :
4513                                        xmit_hash_policy);
4514                                 return -EINVAL;
4515                         }
4516                 }
4517         }
4518
4519         if (lacp_rate) {
4520                 if (bond_mode != BOND_MODE_8023AD) {
4521                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4522                                 bond_mode_name(bond_mode));
4523                 } else {
4524                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4525                         if (lacp_fast == -1) {
4526                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4527                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4528                                 return -EINVAL;
4529                         }
4530                 }
4531         }
4532
4533         if (ad_select) {
4534                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4535                 if (params->ad_select == -1) {
4536                         pr_err("Error: Invalid ad_select \"%s\"\n",
4537                                ad_select == NULL ? "NULL" : ad_select);
4538                         return -EINVAL;
4539                 }
4540
4541                 if (bond_mode != BOND_MODE_8023AD) {
4542                         pr_warning("ad_select param only affects 802.3ad mode\n");
4543                 }
4544         } else {
4545                 params->ad_select = BOND_AD_STABLE;
4546         }
4547
4548         if (max_bonds < 0) {
4549                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4550                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4551                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4552         }
4553
4554         if (miimon < 0) {
4555                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4556                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4557                 miimon = BOND_LINK_MON_INTERV;
4558         }
4559
4560         if (updelay < 0) {
4561                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4562                            updelay, INT_MAX);
4563                 updelay = 0;
4564         }
4565
4566         if (downdelay < 0) {
4567                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4568                            downdelay, INT_MAX);
4569                 downdelay = 0;
4570         }
4571
4572         if ((use_carrier != 0) && (use_carrier != 1)) {
4573                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4574                            use_carrier);
4575                 use_carrier = 1;
4576         }
4577
4578         if (num_peer_notif < 0 || num_peer_notif > 255) {
4579                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4580                            num_peer_notif);
4581                 num_peer_notif = 1;
4582         }
4583
4584         /* reset values for 802.3ad */
4585         if (bond_mode == BOND_MODE_8023AD) {
4586                 if (!miimon) {
4587                         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");
4588                         pr_warning("Forcing miimon to 100msec\n");
4589                         miimon = 100;
4590                 }
4591         }
4592
4593         if (tx_queues < 1 || tx_queues > 255) {
4594                 pr_warning("Warning: tx_queues (%d) should be between "
4595                            "1 and 255, resetting to %d\n",
4596                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4597                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4598         }
4599
4600         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4601                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4602                            "not of valid value (0/1), so it was set to "
4603                            "0\n", all_slaves_active);
4604                 all_slaves_active = 0;
4605         }
4606
4607         if (resend_igmp < 0 || resend_igmp > 255) {
4608                 pr_warning("Warning: resend_igmp (%d) should be between "
4609                            "0 and 255, resetting to %d\n",
4610                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4611                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4612         }
4613
4614         /* reset values for TLB/ALB */
4615         if ((bond_mode == BOND_MODE_TLB) ||
4616             (bond_mode == BOND_MODE_ALB)) {
4617                 if (!miimon) {
4618                         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");
4619                         pr_warning("Forcing miimon to 100msec\n");
4620                         miimon = 100;
4621                 }
4622         }
4623
4624         if (bond_mode == BOND_MODE_ALB) {
4625                 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",
4626                           updelay);
4627         }
4628
4629         if (!miimon) {
4630                 if (updelay || downdelay) {
4631                         /* just warn the user the up/down delay will have
4632                          * no effect since miimon is zero...
4633                          */
4634                         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",
4635                                    updelay, downdelay);
4636                 }
4637         } else {
4638                 /* don't allow arp monitoring */
4639                 if (arp_interval) {
4640                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4641                                    miimon, arp_interval);
4642                         arp_interval = 0;
4643                 }
4644
4645                 if ((updelay % miimon) != 0) {
4646                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4647                                    updelay, miimon,
4648                                    (updelay / miimon) * miimon);
4649                 }
4650
4651                 updelay /= miimon;
4652
4653                 if ((downdelay % miimon) != 0) {
4654                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4655                                    downdelay, miimon,
4656                                    (downdelay / miimon) * miimon);
4657                 }
4658
4659                 downdelay /= miimon;
4660         }
4661
4662         if (arp_interval < 0) {
4663                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4664                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4665                 arp_interval = BOND_LINK_ARP_INTERV;
4666         }
4667
4668         for (arp_ip_count = 0;
4669              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4670              arp_ip_count++) {
4671                 /* not complete check, but should be good enough to
4672                    catch mistakes */
4673                 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4674                 if (!isdigit(arp_ip_target[arp_ip_count][0]) ||
4675                     ip == 0 || ip == htonl(INADDR_BROADCAST)) {
4676                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4677                                    arp_ip_target[arp_ip_count]);
4678                         arp_interval = 0;
4679                 } else {
4680                         arp_target[arp_ip_count] = ip;
4681                 }
4682         }
4683
4684         if (arp_interval && !arp_ip_count) {
4685                 /* don't allow arping if no arp_ip_target given... */
4686                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4687                            arp_interval);
4688                 arp_interval = 0;
4689         }
4690
4691         if (arp_validate) {
4692                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4693                         pr_err("arp_validate only supported in active-backup mode\n");
4694                         return -EINVAL;
4695                 }
4696                 if (!arp_interval) {
4697                         pr_err("arp_validate requires arp_interval\n");
4698                         return -EINVAL;
4699                 }
4700
4701                 arp_validate_value = bond_parse_parm(arp_validate,
4702                                                      arp_validate_tbl);
4703                 if (arp_validate_value == -1) {
4704                         pr_err("Error: invalid arp_validate \"%s\"\n",
4705                                arp_validate == NULL ? "NULL" : arp_validate);
4706                         return -EINVAL;
4707                 }
4708         } else
4709                 arp_validate_value = 0;
4710
4711         if (miimon) {
4712                 pr_info("MII link monitoring set to %d ms\n", miimon);
4713         } else if (arp_interval) {
4714                 int i;
4715
4716                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4717                         arp_interval,
4718                         arp_validate_tbl[arp_validate_value].modename,
4719                         arp_ip_count);
4720
4721                 for (i = 0; i < arp_ip_count; i++)
4722                         pr_info(" %s", arp_ip_target[i]);
4723
4724                 pr_info("\n");
4725
4726         } else if (max_bonds) {
4727                 /* miimon and arp_interval not set, we need one so things
4728                  * work as expected, see bonding.txt for details
4729                  */
4730                 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");
4731         }
4732
4733         if (primary && !USES_PRIMARY(bond_mode)) {
4734                 /* currently, using a primary only makes sense
4735                  * in active backup, TLB or ALB modes
4736                  */
4737                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4738                            primary, bond_mode_name(bond_mode));
4739                 primary = NULL;
4740         }
4741
4742         if (primary && primary_reselect) {
4743                 primary_reselect_value = bond_parse_parm(primary_reselect,
4744                                                          pri_reselect_tbl);
4745                 if (primary_reselect_value == -1) {
4746                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4747                                primary_reselect ==
4748                                         NULL ? "NULL" : primary_reselect);
4749                         return -EINVAL;
4750                 }
4751         } else {
4752                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4753         }
4754
4755         if (fail_over_mac) {
4756                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4757                                                       fail_over_mac_tbl);
4758                 if (fail_over_mac_value == -1) {
4759                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4760                                arp_validate == NULL ? "NULL" : arp_validate);
4761                         return -EINVAL;
4762                 }
4763
4764                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4765                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4766         } else {
4767                 fail_over_mac_value = BOND_FOM_NONE;
4768         }
4769
4770         /* fill params struct with the proper values */
4771         params->mode = bond_mode;
4772         params->xmit_policy = xmit_hashtype;
4773         params->miimon = miimon;
4774         params->num_peer_notif = num_peer_notif;
4775         params->arp_interval = arp_interval;
4776         params->arp_validate = arp_validate_value;
4777         params->updelay = updelay;
4778         params->downdelay = downdelay;
4779         params->use_carrier = use_carrier;
4780         params->lacp_fast = lacp_fast;
4781         params->primary[0] = 0;
4782         params->primary_reselect = primary_reselect_value;
4783         params->fail_over_mac = fail_over_mac_value;
4784         params->tx_queues = tx_queues;
4785         params->all_slaves_active = all_slaves_active;
4786         params->resend_igmp = resend_igmp;
4787         params->min_links = min_links;
4788
4789         if (primary) {
4790                 strncpy(params->primary, primary, IFNAMSIZ);
4791                 params->primary[IFNAMSIZ - 1] = 0;
4792         }
4793
4794         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4795
4796         return 0;
4797 }
4798
4799 static struct lock_class_key bonding_netdev_xmit_lock_key;
4800 static struct lock_class_key bonding_netdev_addr_lock_key;
4801 static struct lock_class_key bonding_tx_busylock_key;
4802
4803 static void bond_set_lockdep_class_one(struct net_device *dev,
4804                                        struct netdev_queue *txq,
4805                                        void *_unused)
4806 {
4807         lockdep_set_class(&txq->_xmit_lock,
4808                           &bonding_netdev_xmit_lock_key);
4809 }
4810
4811 static void bond_set_lockdep_class(struct net_device *dev)
4812 {
4813         lockdep_set_class(&dev->addr_list_lock,
4814                           &bonding_netdev_addr_lock_key);
4815         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4816         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4817 }
4818
4819 /*
4820  * Called from registration process
4821  */
4822 static int bond_init(struct net_device *bond_dev)
4823 {
4824         struct bonding *bond = netdev_priv(bond_dev);
4825         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4826         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4827
4828         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4829
4830         /*
4831          * Initialize locks that may be required during
4832          * en/deslave operations.  All of the bond_open work
4833          * (of which this is part) should really be moved to
4834          * a phase prior to dev_open
4835          */
4836         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4837         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4838
4839         bond->wq = create_singlethread_workqueue(bond_dev->name);
4840         if (!bond->wq)
4841                 return -ENOMEM;
4842
4843         bond_set_lockdep_class(bond_dev);
4844
4845         list_add_tail(&bond->bond_list, &bn->dev_list);
4846
4847         bond_prepare_sysfs_group(bond);
4848
4849         bond_debug_register(bond);
4850
4851         __hw_addr_init(&bond->mc_list);
4852         return 0;
4853 }
4854
4855 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4856 {
4857         if (tb[IFLA_ADDRESS]) {
4858                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4859                         return -EINVAL;
4860                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4861                         return -EADDRNOTAVAIL;
4862         }
4863         return 0;
4864 }
4865
4866 static unsigned int bond_get_num_tx_queues(void)
4867 {
4868         return tx_queues;
4869 }
4870
4871 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4872         .kind                   = "bond",
4873         .priv_size              = sizeof(struct bonding),
4874         .setup                  = bond_setup,
4875         .validate               = bond_validate,
4876         .get_num_tx_queues      = bond_get_num_tx_queues,
4877         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4878                                                              as for TX queues */
4879 };
4880
4881 /* Create a new bond based on the specified name and bonding parameters.
4882  * If name is NULL, obtain a suitable "bond%d" name for us.
4883  * Caller must NOT hold rtnl_lock; we need to release it here before we
4884  * set up our sysfs entries.
4885  */
4886 int bond_create(struct net *net, const char *name)
4887 {
4888         struct net_device *bond_dev;
4889         int res;
4890
4891         rtnl_lock();
4892
4893         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4894                                    name ? name : "bond%d",
4895                                    bond_setup, tx_queues);
4896         if (!bond_dev) {
4897                 pr_err("%s: eek! can't alloc netdev!\n", name);
4898                 rtnl_unlock();
4899                 return -ENOMEM;
4900         }
4901
4902         dev_net_set(bond_dev, net);
4903         bond_dev->rtnl_link_ops = &bond_link_ops;
4904
4905         res = register_netdevice(bond_dev);
4906
4907         netif_carrier_off(bond_dev);
4908
4909         rtnl_unlock();
4910         if (res < 0)
4911                 bond_destructor(bond_dev);
4912         return res;
4913 }
4914
4915 static int __net_init bond_net_init(struct net *net)
4916 {
4917         struct bond_net *bn = net_generic(net, bond_net_id);
4918
4919         bn->net = net;
4920         INIT_LIST_HEAD(&bn->dev_list);
4921
4922         bond_create_proc_dir(bn);
4923         bond_create_sysfs(bn);
4924
4925         return 0;
4926 }
4927
4928 static void __net_exit bond_net_exit(struct net *net)
4929 {
4930         struct bond_net *bn = net_generic(net, bond_net_id);
4931
4932         bond_destroy_sysfs(bn);
4933         bond_destroy_proc_dir(bn);
4934 }
4935
4936 static struct pernet_operations bond_net_ops = {
4937         .init = bond_net_init,
4938         .exit = bond_net_exit,
4939         .id   = &bond_net_id,
4940         .size = sizeof(struct bond_net),
4941 };
4942
4943 static int __init bonding_init(void)
4944 {
4945         int i;
4946         int res;
4947
4948         pr_info("%s", bond_version);
4949
4950         res = bond_check_params(&bonding_defaults);
4951         if (res)
4952                 goto out;
4953
4954         res = register_pernet_subsys(&bond_net_ops);
4955         if (res)
4956                 goto out;
4957
4958         res = rtnl_link_register(&bond_link_ops);
4959         if (res)
4960                 goto err_link;
4961
4962         bond_create_debugfs();
4963
4964         for (i = 0; i < max_bonds; i++) {
4965                 res = bond_create(&init_net, NULL);
4966                 if (res)
4967                         goto err;
4968         }
4969
4970         register_netdevice_notifier(&bond_netdev_notifier);
4971 out:
4972         return res;
4973 err:
4974         rtnl_link_unregister(&bond_link_ops);
4975 err_link:
4976         unregister_pernet_subsys(&bond_net_ops);
4977         goto out;
4978
4979 }
4980
4981 static void __exit bonding_exit(void)
4982 {
4983         unregister_netdevice_notifier(&bond_netdev_notifier);
4984
4985         bond_destroy_debugfs();
4986
4987         rtnl_link_unregister(&bond_link_ops);
4988         unregister_pernet_subsys(&bond_net_ops);
4989
4990 #ifdef CONFIG_NET_POLL_CONTROLLER
4991         /*
4992          * Make sure we don't have an imbalance on our netpoll blocking
4993          */
4994         WARN_ON(atomic_read(&netpoll_block_tx));
4995 #endif
4996 }
4997
4998 module_init(bonding_init);
4999 module_exit(bonding_exit);
5000 MODULE_LICENSE("GPL");
5001 MODULE_VERSION(DRV_VERSION);
5002 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5003 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5004 MODULE_ALIAS_RTNL_LINK("bond");