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         int i;
1383         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1384
1385         read_lock(&bond->lock);
1386
1387         if (!bond->first_slave)
1388                 goto done;
1389
1390         bond_for_each_slave(bond, slave, i) {
1391                 vlan_features = netdev_increment_features(vlan_features,
1392                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1393
1394                 dst_release_flag &= slave->dev->priv_flags;
1395                 if (slave->dev->hard_header_len > max_hard_header_len)
1396                         max_hard_header_len = slave->dev->hard_header_len;
1397         }
1398
1399 done:
1400         bond_dev->vlan_features = vlan_features;
1401         bond_dev->hard_header_len = max_hard_header_len;
1402
1403         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1404         bond_dev->priv_flags = flags | dst_release_flag;
1405
1406         read_unlock(&bond->lock);
1407
1408         netdev_change_features(bond_dev);
1409 }
1410
1411 static void bond_setup_by_slave(struct net_device *bond_dev,
1412                                 struct net_device *slave_dev)
1413 {
1414         struct bonding *bond = netdev_priv(bond_dev);
1415
1416         bond_dev->header_ops        = slave_dev->header_ops;
1417
1418         bond_dev->type              = slave_dev->type;
1419         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1420         bond_dev->addr_len          = slave_dev->addr_len;
1421
1422         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1423                 slave_dev->addr_len);
1424         bond->setup_by_slave = 1;
1425 }
1426
1427 /* On bonding slaves other than the currently active slave, suppress
1428  * duplicates except for alb non-mcast/bcast.
1429  */
1430 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1431                                             struct slave *slave,
1432                                             struct bonding *bond)
1433 {
1434         if (bond_is_slave_inactive(slave)) {
1435                 if (bond->params.mode == BOND_MODE_ALB &&
1436                     skb->pkt_type != PACKET_BROADCAST &&
1437                     skb->pkt_type != PACKET_MULTICAST)
1438                         return false;
1439                 return true;
1440         }
1441         return false;
1442 }
1443
1444 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1445 {
1446         struct sk_buff *skb = *pskb;
1447         struct slave *slave;
1448         struct bonding *bond;
1449         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1450                           struct slave *);
1451         int ret = RX_HANDLER_ANOTHER;
1452
1453         skb = skb_share_check(skb, GFP_ATOMIC);
1454         if (unlikely(!skb))
1455                 return RX_HANDLER_CONSUMED;
1456
1457         *pskb = skb;
1458
1459         slave = bond_slave_get_rcu(skb->dev);
1460         bond = slave->bond;
1461
1462         if (bond->params.arp_interval)
1463                 slave->dev->last_rx = jiffies;
1464
1465         recv_probe = ACCESS_ONCE(bond->recv_probe);
1466         if (recv_probe) {
1467                 ret = recv_probe(skb, bond, slave);
1468                 if (ret == RX_HANDLER_CONSUMED) {
1469                         consume_skb(skb);
1470                         return ret;
1471                 }
1472         }
1473
1474         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1475                 return RX_HANDLER_EXACT;
1476         }
1477
1478         skb->dev = bond->dev;
1479
1480         if (bond->params.mode == BOND_MODE_ALB &&
1481             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1482             skb->pkt_type == PACKET_HOST) {
1483
1484                 if (unlikely(skb_cow_head(skb,
1485                                           skb->data - skb_mac_header(skb)))) {
1486                         kfree_skb(skb);
1487                         return RX_HANDLER_CONSUMED;
1488                 }
1489                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1490         }
1491
1492         return ret;
1493 }
1494
1495 /* enslave device <slave> to bond device <master> */
1496 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1497 {
1498         struct bonding *bond = netdev_priv(bond_dev);
1499         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1500         struct slave *new_slave = NULL;
1501         struct netdev_hw_addr *ha;
1502         struct sockaddr addr;
1503         int link_reporting;
1504         int res = 0;
1505
1506         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1507                 slave_ops->ndo_do_ioctl == NULL) {
1508                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1509                            bond_dev->name, slave_dev->name);
1510         }
1511
1512         /* already enslaved */
1513         if (slave_dev->flags & IFF_SLAVE) {
1514                 pr_debug("Error, Device was already enslaved\n");
1515                 return -EBUSY;
1516         }
1517
1518         /* vlan challenged mutual exclusion */
1519         /* no need to lock since we're protected by rtnl_lock */
1520         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1521                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1522                 if (vlan_uses_dev(bond_dev)) {
1523                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1524                                bond_dev->name, slave_dev->name, bond_dev->name);
1525                         return -EPERM;
1526                 } else {
1527                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1528                                    bond_dev->name, slave_dev->name,
1529                                    slave_dev->name, bond_dev->name);
1530                 }
1531         } else {
1532                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1533         }
1534
1535         /*
1536          * Old ifenslave binaries are no longer supported.  These can
1537          * be identified with moderate accuracy by the state of the slave:
1538          * the current ifenslave will set the interface down prior to
1539          * enslaving it; the old ifenslave will not.
1540          */
1541         if ((slave_dev->flags & IFF_UP)) {
1542                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1543                        slave_dev->name);
1544                 res = -EPERM;
1545                 goto err_undo_flags;
1546         }
1547
1548         /* set bonding device ether type by slave - bonding netdevices are
1549          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1550          * there is a need to override some of the type dependent attribs/funcs.
1551          *
1552          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1553          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1554          */
1555         if (bond->slave_cnt == 0) {
1556                 if (bond_dev->type != slave_dev->type) {
1557                         pr_debug("%s: change device type from %d to %d\n",
1558                                  bond_dev->name,
1559                                  bond_dev->type, slave_dev->type);
1560
1561                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1562                                                        bond_dev);
1563                         res = notifier_to_errno(res);
1564                         if (res) {
1565                                 pr_err("%s: refused to change device type\n",
1566                                        bond_dev->name);
1567                                 res = -EBUSY;
1568                                 goto err_undo_flags;
1569                         }
1570
1571                         /* Flush unicast and multicast addresses */
1572                         dev_uc_flush(bond_dev);
1573                         dev_mc_flush(bond_dev);
1574
1575                         if (slave_dev->type != ARPHRD_ETHER)
1576                                 bond_setup_by_slave(bond_dev, slave_dev);
1577                         else {
1578                                 ether_setup(bond_dev);
1579                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1580                         }
1581
1582                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1583                                                  bond_dev);
1584                 }
1585         } else if (bond_dev->type != slave_dev->type) {
1586                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1587                        slave_dev->name,
1588                        slave_dev->type, bond_dev->type);
1589                 res = -EINVAL;
1590                 goto err_undo_flags;
1591         }
1592
1593         if (slave_ops->ndo_set_mac_address == NULL) {
1594                 if (bond->slave_cnt == 0) {
1595                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1596                                    bond_dev->name);
1597                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1598                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1599                         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",
1600                                bond_dev->name);
1601                         res = -EOPNOTSUPP;
1602                         goto err_undo_flags;
1603                 }
1604         }
1605
1606         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1607
1608         /* If this is the first slave, then we need to set the master's hardware
1609          * address to be the same as the slave's. */
1610         if (is_zero_ether_addr(bond->dev->dev_addr))
1611                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1612                        slave_dev->addr_len);
1613
1614
1615         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1616         if (!new_slave) {
1617                 res = -ENOMEM;
1618                 goto err_undo_flags;
1619         }
1620
1621         /*
1622          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1623          * is set via sysfs or module option if desired.
1624          */
1625         new_slave->queue_id = 0;
1626
1627         /* Save slave's original mtu and then set it to match the bond */
1628         new_slave->original_mtu = slave_dev->mtu;
1629         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1630         if (res) {
1631                 pr_debug("Error %d calling dev_set_mtu\n", res);
1632                 goto err_free;
1633         }
1634
1635         /*
1636          * Save slave's original ("permanent") mac address for modes
1637          * that need it, and for restoring it upon release, and then
1638          * set it to the master's address
1639          */
1640         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1641
1642         if (!bond->params.fail_over_mac) {
1643                 /*
1644                  * Set slave to master's mac address.  The application already
1645                  * set the master's mac address to that of the first slave
1646                  */
1647                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1648                 addr.sa_family = slave_dev->type;
1649                 res = dev_set_mac_address(slave_dev, &addr);
1650                 if (res) {
1651                         pr_debug("Error %d calling set_mac_address\n", res);
1652                         goto err_restore_mtu;
1653                 }
1654         }
1655
1656         res = netdev_set_bond_master(slave_dev, bond_dev);
1657         if (res) {
1658                 pr_debug("Error %d calling netdev_set_bond_master\n", res);
1659                 goto err_restore_mac;
1660         }
1661
1662         /* open the slave since the application closed it */
1663         res = dev_open(slave_dev);
1664         if (res) {
1665                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1666                 goto err_unset_master;
1667         }
1668
1669         new_slave->bond = bond;
1670         new_slave->dev = slave_dev;
1671         slave_dev->priv_flags |= IFF_BONDING;
1672
1673         if (bond_is_lb(bond)) {
1674                 /* bond_alb_init_slave() must be called before all other stages since
1675                  * it might fail and we do not want to have to undo everything
1676                  */
1677                 res = bond_alb_init_slave(bond, new_slave);
1678                 if (res)
1679                         goto err_close;
1680         }
1681
1682         /* If the mode USES_PRIMARY, then the new slave gets the
1683          * master's promisc (and mc) settings only if it becomes the
1684          * curr_active_slave, and that is taken care of later when calling
1685          * bond_change_active()
1686          */
1687         if (!USES_PRIMARY(bond->params.mode)) {
1688                 /* set promiscuity level to new slave */
1689                 if (bond_dev->flags & IFF_PROMISC) {
1690                         res = dev_set_promiscuity(slave_dev, 1);
1691                         if (res)
1692                                 goto err_close;
1693                 }
1694
1695                 /* set allmulti level to new slave */
1696                 if (bond_dev->flags & IFF_ALLMULTI) {
1697                         res = dev_set_allmulti(slave_dev, 1);
1698                         if (res)
1699                                 goto err_close;
1700                 }
1701
1702                 netif_addr_lock_bh(bond_dev);
1703                 /* upload master's mc_list to new slave */
1704                 netdev_for_each_mc_addr(ha, bond_dev)
1705                         dev_mc_add(slave_dev, ha->addr);
1706                 netif_addr_unlock_bh(bond_dev);
1707         }
1708
1709         if (bond->params.mode == BOND_MODE_8023AD) {
1710                 /* add lacpdu mc addr to mc list */
1711                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1712
1713                 dev_mc_add(slave_dev, lacpdu_multicast);
1714         }
1715
1716         bond_add_vlans_on_slave(bond, slave_dev);
1717
1718         write_lock_bh(&bond->lock);
1719
1720         bond_attach_slave(bond, new_slave);
1721
1722         new_slave->delay = 0;
1723         new_slave->link_failure_count = 0;
1724
1725         write_unlock_bh(&bond->lock);
1726
1727         bond_compute_features(bond);
1728
1729         read_lock(&bond->lock);
1730
1731         new_slave->last_arp_rx = jiffies -
1732                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1733
1734         if (bond->params.miimon && !bond->params.use_carrier) {
1735                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1736
1737                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1738                         /*
1739                          * miimon is set but a bonded network driver
1740                          * does not support ETHTOOL/MII and
1741                          * arp_interval is not set.  Note: if
1742                          * use_carrier is enabled, we will never go
1743                          * here (because netif_carrier is always
1744                          * supported); thus, we don't need to change
1745                          * the messages for netif_carrier.
1746                          */
1747                         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",
1748                                bond_dev->name, slave_dev->name);
1749                 } else if (link_reporting == -1) {
1750                         /* unable get link status using mii/ethtool */
1751                         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",
1752                                    bond_dev->name, slave_dev->name);
1753                 }
1754         }
1755
1756         /* check for initial state */
1757         if (bond->params.miimon) {
1758                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1759                         if (bond->params.updelay) {
1760                                 new_slave->link = BOND_LINK_BACK;
1761                                 new_slave->delay = bond->params.updelay;
1762                         } else {
1763                                 new_slave->link = BOND_LINK_UP;
1764                         }
1765                 } else {
1766                         new_slave->link = BOND_LINK_DOWN;
1767                 }
1768         } else if (bond->params.arp_interval) {
1769                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1770                         BOND_LINK_UP : BOND_LINK_DOWN);
1771         } else {
1772                 new_slave->link = BOND_LINK_UP;
1773         }
1774
1775         if (new_slave->link != BOND_LINK_DOWN)
1776                 new_slave->jiffies = jiffies;
1777         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1778                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1779                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1780
1781         bond_update_speed_duplex(new_slave);
1782
1783         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1784                 /* if there is a primary slave, remember it */
1785                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1786                         bond->primary_slave = new_slave;
1787                         bond->force_primary = true;
1788                 }
1789         }
1790
1791         write_lock_bh(&bond->curr_slave_lock);
1792
1793         switch (bond->params.mode) {
1794         case BOND_MODE_ACTIVEBACKUP:
1795                 bond_set_slave_inactive_flags(new_slave);
1796                 bond_select_active_slave(bond);
1797                 break;
1798         case BOND_MODE_8023AD:
1799                 /* in 802.3ad mode, the internal mechanism
1800                  * will activate the slaves in the selected
1801                  * aggregator
1802                  */
1803                 bond_set_slave_inactive_flags(new_slave);
1804                 /* if this is the first slave */
1805                 if (bond->slave_cnt == 1) {
1806                         SLAVE_AD_INFO(new_slave).id = 1;
1807                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1808                          * can be called only after the mac address of the bond is set
1809                          */
1810                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1811                 } else {
1812                         SLAVE_AD_INFO(new_slave).id =
1813                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1814                 }
1815
1816                 bond_3ad_bind_slave(new_slave);
1817                 break;
1818         case BOND_MODE_TLB:
1819         case BOND_MODE_ALB:
1820                 bond_set_active_slave(new_slave);
1821                 bond_set_slave_inactive_flags(new_slave);
1822                 bond_select_active_slave(bond);
1823                 break;
1824         default:
1825                 pr_debug("This slave is always active in trunk mode\n");
1826
1827                 /* always active in trunk mode */
1828                 bond_set_active_slave(new_slave);
1829
1830                 /* In trunking mode there is little meaning to curr_active_slave
1831                  * anyway (it holds no special properties of the bond device),
1832                  * so we can change it without calling change_active_interface()
1833                  */
1834                 if (!bond->curr_active_slave)
1835                         bond->curr_active_slave = new_slave;
1836
1837                 break;
1838         } /* switch(bond_mode) */
1839
1840         write_unlock_bh(&bond->curr_slave_lock);
1841
1842         bond_set_carrier(bond);
1843
1844 #ifdef CONFIG_NET_POLL_CONTROLLER
1845         slave_dev->npinfo = bond_netpoll_info(bond);
1846         if (slave_dev->npinfo) {
1847                 if (slave_enable_netpoll(new_slave)) {
1848                         read_unlock(&bond->lock);
1849                         pr_info("Error, %s: master_dev is using netpoll, "
1850                                  "but new slave device does not support netpoll.\n",
1851                                  bond_dev->name);
1852                         res = -EBUSY;
1853                         goto err_detach;
1854                 }
1855         }
1856 #endif
1857
1858         read_unlock(&bond->lock);
1859
1860         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1861         if (res)
1862                 goto err_detach;
1863
1864         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1865                                          new_slave);
1866         if (res) {
1867                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1868                 goto err_dest_symlinks;
1869         }
1870
1871         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1872                 bond_dev->name, slave_dev->name,
1873                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1874                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1875
1876         /* enslave is successful */
1877         return 0;
1878
1879 /* Undo stages on error */
1880 err_dest_symlinks:
1881         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1882
1883 err_detach:
1884         write_lock_bh(&bond->lock);
1885         bond_detach_slave(bond, new_slave);
1886         write_unlock_bh(&bond->lock);
1887
1888 err_close:
1889         dev_close(slave_dev);
1890
1891 err_unset_master:
1892         netdev_set_bond_master(slave_dev, NULL);
1893
1894 err_restore_mac:
1895         if (!bond->params.fail_over_mac) {
1896                 /* XXX TODO - fom follow mode needs to change master's
1897                  * MAC if this slave's MAC is in use by the bond, or at
1898                  * least print a warning.
1899                  */
1900                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1901                 addr.sa_family = slave_dev->type;
1902                 dev_set_mac_address(slave_dev, &addr);
1903         }
1904
1905 err_restore_mtu:
1906         dev_set_mtu(slave_dev, new_slave->original_mtu);
1907
1908 err_free:
1909         kfree(new_slave);
1910
1911 err_undo_flags:
1912         bond_compute_features(bond);
1913
1914         return res;
1915 }
1916
1917 /*
1918  * Try to release the slave device <slave> from the bond device <master>
1919  * It is legal to access curr_active_slave without a lock because all the function
1920  * is write-locked.
1921  *
1922  * The rules for slave state should be:
1923  *   for Active/Backup:
1924  *     Active stays on all backups go down
1925  *   for Bonded connections:
1926  *     The first up interface should be left on and all others downed.
1927  */
1928 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1929 {
1930         struct bonding *bond = netdev_priv(bond_dev);
1931         struct slave *slave, *oldcurrent;
1932         struct sockaddr addr;
1933         netdev_features_t old_features = bond_dev->features;
1934
1935         /* slave is not a slave or master is not master of this slave */
1936         if (!(slave_dev->flags & IFF_SLAVE) ||
1937             (slave_dev->master != bond_dev)) {
1938                 pr_err("%s: Error: cannot release %s.\n",
1939                        bond_dev->name, slave_dev->name);
1940                 return -EINVAL;
1941         }
1942
1943         block_netpoll_tx();
1944         call_netdevice_notifiers(NETDEV_RELEASE, bond_dev);
1945         write_lock_bh(&bond->lock);
1946
1947         slave = bond_get_slave_by_dev(bond, slave_dev);
1948         if (!slave) {
1949                 /* not a slave of this bond */
1950                 pr_info("%s: %s not enslaved\n",
1951                         bond_dev->name, slave_dev->name);
1952                 write_unlock_bh(&bond->lock);
1953                 unblock_netpoll_tx();
1954                 return -EINVAL;
1955         }
1956
1957         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1958          * for this slave anymore.
1959          */
1960         netdev_rx_handler_unregister(slave_dev);
1961         write_unlock_bh(&bond->lock);
1962         synchronize_net();
1963         write_lock_bh(&bond->lock);
1964
1965         if (!bond->params.fail_over_mac) {
1966                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1967                     bond->slave_cnt > 1)
1968                         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",
1969                                    bond_dev->name, slave_dev->name,
1970                                    slave->perm_hwaddr,
1971                                    bond_dev->name, slave_dev->name);
1972         }
1973
1974         /* Inform AD package of unbinding of slave. */
1975         if (bond->params.mode == BOND_MODE_8023AD) {
1976                 /* must be called before the slave is
1977                  * detached from the list
1978                  */
1979                 bond_3ad_unbind_slave(slave);
1980         }
1981
1982         pr_info("%s: releasing %s interface %s\n",
1983                 bond_dev->name,
1984                 bond_is_active_slave(slave) ? "active" : "backup",
1985                 slave_dev->name);
1986
1987         oldcurrent = bond->curr_active_slave;
1988
1989         bond->current_arp_slave = NULL;
1990
1991         /* release the slave from its bond */
1992         bond_detach_slave(bond, slave);
1993
1994         if (bond->primary_slave == slave)
1995                 bond->primary_slave = NULL;
1996
1997         if (oldcurrent == slave)
1998                 bond_change_active_slave(bond, NULL);
1999
2000         if (bond_is_lb(bond)) {
2001                 /* Must be called only after the slave has been
2002                  * detached from the list and the curr_active_slave
2003                  * has been cleared (if our_slave == old_current),
2004                  * but before a new active slave is selected.
2005                  */
2006                 write_unlock_bh(&bond->lock);
2007                 bond_alb_deinit_slave(bond, slave);
2008                 write_lock_bh(&bond->lock);
2009         }
2010
2011         if (oldcurrent == slave) {
2012                 /*
2013                  * Note that we hold RTNL over this sequence, so there
2014                  * is no concern that another slave add/remove event
2015                  * will interfere.
2016                  */
2017                 write_unlock_bh(&bond->lock);
2018                 read_lock(&bond->lock);
2019                 write_lock_bh(&bond->curr_slave_lock);
2020
2021                 bond_select_active_slave(bond);
2022
2023                 write_unlock_bh(&bond->curr_slave_lock);
2024                 read_unlock(&bond->lock);
2025                 write_lock_bh(&bond->lock);
2026         }
2027
2028         if (bond->slave_cnt == 0) {
2029                 bond_set_carrier(bond);
2030
2031                 /* if the last slave was removed, zero the mac address
2032                  * of the master so it will be set by the application
2033                  * to the mac address of the first slave
2034                  */
2035                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2036
2037                 if (bond_vlan_used(bond)) {
2038                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2039                                    bond_dev->name, bond_dev->name);
2040                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2041                                    bond_dev->name);
2042                 }
2043         }
2044
2045         write_unlock_bh(&bond->lock);
2046         unblock_netpoll_tx();
2047
2048         if (bond->slave_cnt == 0)
2049                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2050
2051         bond_compute_features(bond);
2052         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2053             (old_features & NETIF_F_VLAN_CHALLENGED))
2054                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2055                         bond_dev->name, slave_dev->name, bond_dev->name);
2056
2057         /* must do this from outside any spinlocks */
2058         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2059
2060         bond_del_vlans_from_slave(bond, slave_dev);
2061
2062         /* If the mode USES_PRIMARY, then we should only remove its
2063          * promisc and mc settings if it was the curr_active_slave, but that was
2064          * already taken care of above when we detached the slave
2065          */
2066         if (!USES_PRIMARY(bond->params.mode)) {
2067                 /* unset promiscuity level from slave */
2068                 if (bond_dev->flags & IFF_PROMISC)
2069                         dev_set_promiscuity(slave_dev, -1);
2070
2071                 /* unset allmulti level from slave */
2072                 if (bond_dev->flags & IFF_ALLMULTI)
2073                         dev_set_allmulti(slave_dev, -1);
2074
2075                 /* flush master's mc_list from slave */
2076                 netif_addr_lock_bh(bond_dev);
2077                 bond_mc_list_flush(bond_dev, slave_dev);
2078                 netif_addr_unlock_bh(bond_dev);
2079         }
2080
2081         netdev_set_bond_master(slave_dev, NULL);
2082
2083         slave_disable_netpoll(slave);
2084
2085         /* close slave before restoring its mac address */
2086         dev_close(slave_dev);
2087
2088         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2089                 /* restore original ("permanent") mac address */
2090                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2091                 addr.sa_family = slave_dev->type;
2092                 dev_set_mac_address(slave_dev, &addr);
2093         }
2094
2095         dev_set_mtu(slave_dev, slave->original_mtu);
2096
2097         slave_dev->priv_flags &= ~IFF_BONDING;
2098
2099         kfree(slave);
2100
2101         return 0;  /* deletion OK */
2102 }
2103
2104 /*
2105 * First release a slave and then destroy the bond if no more slaves are left.
2106 * Must be under rtnl_lock when this function is called.
2107 */
2108 static int  bond_release_and_destroy(struct net_device *bond_dev,
2109                                      struct net_device *slave_dev)
2110 {
2111         struct bonding *bond = netdev_priv(bond_dev);
2112         int ret;
2113
2114         ret = bond_release(bond_dev, slave_dev);
2115         if ((ret == 0) && (bond->slave_cnt == 0)) {
2116                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2117                 pr_info("%s: destroying bond %s.\n",
2118                         bond_dev->name, bond_dev->name);
2119                 unregister_netdevice(bond_dev);
2120         }
2121         return ret;
2122 }
2123
2124 /*
2125  * This function releases all slaves.
2126  */
2127 static int bond_release_all(struct net_device *bond_dev)
2128 {
2129         struct bonding *bond = netdev_priv(bond_dev);
2130         struct slave *slave;
2131         struct net_device *slave_dev;
2132         struct sockaddr addr;
2133
2134         write_lock_bh(&bond->lock);
2135
2136         netif_carrier_off(bond_dev);
2137
2138         if (bond->slave_cnt == 0)
2139                 goto out;
2140
2141         bond->current_arp_slave = NULL;
2142         bond->primary_slave = NULL;
2143         bond_change_active_slave(bond, NULL);
2144
2145         while ((slave = bond->first_slave) != NULL) {
2146                 /* Inform AD package of unbinding of slave
2147                  * before slave is detached from the list.
2148                  */
2149                 if (bond->params.mode == BOND_MODE_8023AD)
2150                         bond_3ad_unbind_slave(slave);
2151
2152                 slave_dev = slave->dev;
2153                 bond_detach_slave(bond, slave);
2154
2155                 /* now that the slave is detached, unlock and perform
2156                  * all the undo steps that should not be called from
2157                  * within a lock.
2158                  */
2159                 write_unlock_bh(&bond->lock);
2160
2161                 /* unregister rx_handler early so bond_handle_frame wouldn't
2162                  * be called for this slave anymore.
2163                  */
2164                 netdev_rx_handler_unregister(slave_dev);
2165                 synchronize_net();
2166
2167                 if (bond_is_lb(bond)) {
2168                         /* must be called only after the slave
2169                          * has been detached from the list
2170                          */
2171                         bond_alb_deinit_slave(bond, slave);
2172                 }
2173
2174                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2175                 bond_del_vlans_from_slave(bond, slave_dev);
2176
2177                 /* If the mode USES_PRIMARY, then we should only remove its
2178                  * promisc and mc settings if it was the curr_active_slave, but that was
2179                  * already taken care of above when we detached the slave
2180                  */
2181                 if (!USES_PRIMARY(bond->params.mode)) {
2182                         /* unset promiscuity level from slave */
2183                         if (bond_dev->flags & IFF_PROMISC)
2184                                 dev_set_promiscuity(slave_dev, -1);
2185
2186                         /* unset allmulti level from slave */
2187                         if (bond_dev->flags & IFF_ALLMULTI)
2188                                 dev_set_allmulti(slave_dev, -1);
2189
2190                         /* flush master's mc_list from slave */
2191                         netif_addr_lock_bh(bond_dev);
2192                         bond_mc_list_flush(bond_dev, slave_dev);
2193                         netif_addr_unlock_bh(bond_dev);
2194                 }
2195
2196                 netdev_set_bond_master(slave_dev, NULL);
2197
2198                 slave_disable_netpoll(slave);
2199
2200                 /* close slave before restoring its mac address */
2201                 dev_close(slave_dev);
2202
2203                 if (!bond->params.fail_over_mac) {
2204                         /* restore original ("permanent") mac address*/
2205                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2206                         addr.sa_family = slave_dev->type;
2207                         dev_set_mac_address(slave_dev, &addr);
2208                 }
2209
2210                 kfree(slave);
2211
2212                 /* re-acquire the lock before getting the next slave */
2213                 write_lock_bh(&bond->lock);
2214         }
2215
2216         /* zero the mac address of the master so it will be
2217          * set by the application to the mac address of the
2218          * first slave
2219          */
2220         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2221
2222         if (bond_vlan_used(bond)) {
2223                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2224                            bond_dev->name, bond_dev->name);
2225                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2226                            bond_dev->name);
2227         }
2228
2229         pr_info("%s: released all slaves\n", bond_dev->name);
2230
2231 out:
2232         write_unlock_bh(&bond->lock);
2233
2234         bond_compute_features(bond);
2235
2236         return 0;
2237 }
2238
2239 /*
2240  * This function changes the active slave to slave <slave_dev>.
2241  * It returns -EINVAL in the following cases.
2242  *  - <slave_dev> is not found in the list.
2243  *  - There is not active slave now.
2244  *  - <slave_dev> is already active.
2245  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2246  *  - <slave_dev> is not running.
2247  * In these cases, this function does nothing.
2248  * In the other cases, current_slave pointer is changed and 0 is returned.
2249  */
2250 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2251 {
2252         struct bonding *bond = netdev_priv(bond_dev);
2253         struct slave *old_active = NULL;
2254         struct slave *new_active = NULL;
2255         int res = 0;
2256
2257         if (!USES_PRIMARY(bond->params.mode))
2258                 return -EINVAL;
2259
2260         /* Verify that master_dev is indeed the master of slave_dev */
2261         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2262                 return -EINVAL;
2263
2264         read_lock(&bond->lock);
2265
2266         read_lock(&bond->curr_slave_lock);
2267         old_active = bond->curr_active_slave;
2268         read_unlock(&bond->curr_slave_lock);
2269
2270         new_active = bond_get_slave_by_dev(bond, slave_dev);
2271
2272         /*
2273          * Changing to the current active: do nothing; return success.
2274          */
2275         if (new_active && (new_active == old_active)) {
2276                 read_unlock(&bond->lock);
2277                 return 0;
2278         }
2279
2280         if ((new_active) &&
2281             (old_active) &&
2282             (new_active->link == BOND_LINK_UP) &&
2283             IS_UP(new_active->dev)) {
2284                 block_netpoll_tx();
2285                 write_lock_bh(&bond->curr_slave_lock);
2286                 bond_change_active_slave(bond, new_active);
2287                 write_unlock_bh(&bond->curr_slave_lock);
2288                 unblock_netpoll_tx();
2289         } else
2290                 res = -EINVAL;
2291
2292         read_unlock(&bond->lock);
2293
2294         return res;
2295 }
2296
2297 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2298 {
2299         struct bonding *bond = netdev_priv(bond_dev);
2300
2301         info->bond_mode = bond->params.mode;
2302         info->miimon = bond->params.miimon;
2303
2304         read_lock(&bond->lock);
2305         info->num_slaves = bond->slave_cnt;
2306         read_unlock(&bond->lock);
2307
2308         return 0;
2309 }
2310
2311 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2312 {
2313         struct bonding *bond = netdev_priv(bond_dev);
2314         struct slave *slave;
2315         int i, res = -ENODEV;
2316
2317         read_lock(&bond->lock);
2318
2319         bond_for_each_slave(bond, slave, i) {
2320                 if (i == (int)info->slave_id) {
2321                         res = 0;
2322                         strcpy(info->slave_name, slave->dev->name);
2323                         info->link = slave->link;
2324                         info->state = bond_slave_state(slave);
2325                         info->link_failure_count = slave->link_failure_count;
2326                         break;
2327                 }
2328         }
2329
2330         read_unlock(&bond->lock);
2331
2332         return res;
2333 }
2334
2335 /*-------------------------------- Monitoring -------------------------------*/
2336
2337
2338 static int bond_miimon_inspect(struct bonding *bond)
2339 {
2340         struct slave *slave;
2341         int i, link_state, commit = 0;
2342         bool ignore_updelay;
2343
2344         ignore_updelay = !bond->curr_active_slave ? true : false;
2345
2346         bond_for_each_slave(bond, slave, i) {
2347                 slave->new_link = BOND_LINK_NOCHANGE;
2348
2349                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2350
2351                 switch (slave->link) {
2352                 case BOND_LINK_UP:
2353                         if (link_state)
2354                                 continue;
2355
2356                         slave->link = BOND_LINK_FAIL;
2357                         slave->delay = bond->params.downdelay;
2358                         if (slave->delay) {
2359                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2360                                         bond->dev->name,
2361                                         (bond->params.mode ==
2362                                          BOND_MODE_ACTIVEBACKUP) ?
2363                                         (bond_is_active_slave(slave) ?
2364                                          "active " : "backup ") : "",
2365                                         slave->dev->name,
2366                                         bond->params.downdelay * bond->params.miimon);
2367                         }
2368                         /*FALLTHRU*/
2369                 case BOND_LINK_FAIL:
2370                         if (link_state) {
2371                                 /*
2372                                  * recovered before downdelay expired
2373                                  */
2374                                 slave->link = BOND_LINK_UP;
2375                                 slave->jiffies = jiffies;
2376                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2377                                         bond->dev->name,
2378                                         (bond->params.downdelay - slave->delay) *
2379                                         bond->params.miimon,
2380                                         slave->dev->name);
2381                                 continue;
2382                         }
2383
2384                         if (slave->delay <= 0) {
2385                                 slave->new_link = BOND_LINK_DOWN;
2386                                 commit++;
2387                                 continue;
2388                         }
2389
2390                         slave->delay--;
2391                         break;
2392
2393                 case BOND_LINK_DOWN:
2394                         if (!link_state)
2395                                 continue;
2396
2397                         slave->link = BOND_LINK_BACK;
2398                         slave->delay = bond->params.updelay;
2399
2400                         if (slave->delay) {
2401                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2402                                         bond->dev->name, slave->dev->name,
2403                                         ignore_updelay ? 0 :
2404                                         bond->params.updelay *
2405                                         bond->params.miimon);
2406                         }
2407                         /*FALLTHRU*/
2408                 case BOND_LINK_BACK:
2409                         if (!link_state) {
2410                                 slave->link = BOND_LINK_DOWN;
2411                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2412                                         bond->dev->name,
2413                                         (bond->params.updelay - slave->delay) *
2414                                         bond->params.miimon,
2415                                         slave->dev->name);
2416
2417                                 continue;
2418                         }
2419
2420                         if (ignore_updelay)
2421                                 slave->delay = 0;
2422
2423                         if (slave->delay <= 0) {
2424                                 slave->new_link = BOND_LINK_UP;
2425                                 commit++;
2426                                 ignore_updelay = false;
2427                                 continue;
2428                         }
2429
2430                         slave->delay--;
2431                         break;
2432                 }
2433         }
2434
2435         return commit;
2436 }
2437
2438 static void bond_miimon_commit(struct bonding *bond)
2439 {
2440         struct slave *slave;
2441         int i;
2442
2443         bond_for_each_slave(bond, slave, i) {
2444                 switch (slave->new_link) {
2445                 case BOND_LINK_NOCHANGE:
2446                         continue;
2447
2448                 case BOND_LINK_UP:
2449                         slave->link = BOND_LINK_UP;
2450                         slave->jiffies = jiffies;
2451
2452                         if (bond->params.mode == BOND_MODE_8023AD) {
2453                                 /* prevent it from being the active one */
2454                                 bond_set_backup_slave(slave);
2455                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2456                                 /* make it immediately active */
2457                                 bond_set_active_slave(slave);
2458                         } else if (slave != bond->primary_slave) {
2459                                 /* prevent it from being the active one */
2460                                 bond_set_backup_slave(slave);
2461                         }
2462
2463                         bond_update_speed_duplex(slave);
2464
2465                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2466                                 bond->dev->name, slave->dev->name,
2467                                 slave->speed, slave->duplex ? "full" : "half");
2468
2469                         /* notify ad that the link status has changed */
2470                         if (bond->params.mode == BOND_MODE_8023AD)
2471                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2472
2473                         if (bond_is_lb(bond))
2474                                 bond_alb_handle_link_change(bond, slave,
2475                                                             BOND_LINK_UP);
2476
2477                         if (!bond->curr_active_slave ||
2478                             (slave == bond->primary_slave))
2479                                 goto do_failover;
2480
2481                         continue;
2482
2483                 case BOND_LINK_DOWN:
2484                         if (slave->link_failure_count < UINT_MAX)
2485                                 slave->link_failure_count++;
2486
2487                         slave->link = BOND_LINK_DOWN;
2488
2489                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2490                             bond->params.mode == BOND_MODE_8023AD)
2491                                 bond_set_slave_inactive_flags(slave);
2492
2493                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2494                                 bond->dev->name, slave->dev->name);
2495
2496                         if (bond->params.mode == BOND_MODE_8023AD)
2497                                 bond_3ad_handle_link_change(slave,
2498                                                             BOND_LINK_DOWN);
2499
2500                         if (bond_is_lb(bond))
2501                                 bond_alb_handle_link_change(bond, slave,
2502                                                             BOND_LINK_DOWN);
2503
2504                         if (slave == bond->curr_active_slave)
2505                                 goto do_failover;
2506
2507                         continue;
2508
2509                 default:
2510                         pr_err("%s: invalid new link %d on slave %s\n",
2511                                bond->dev->name, slave->new_link,
2512                                slave->dev->name);
2513                         slave->new_link = BOND_LINK_NOCHANGE;
2514
2515                         continue;
2516                 }
2517
2518 do_failover:
2519                 ASSERT_RTNL();
2520                 block_netpoll_tx();
2521                 write_lock_bh(&bond->curr_slave_lock);
2522                 bond_select_active_slave(bond);
2523                 write_unlock_bh(&bond->curr_slave_lock);
2524                 unblock_netpoll_tx();
2525         }
2526
2527         bond_set_carrier(bond);
2528 }
2529
2530 /*
2531  * bond_mii_monitor
2532  *
2533  * Really a wrapper that splits the mii monitor into two phases: an
2534  * inspection, then (if inspection indicates something needs to be done)
2535  * an acquisition of appropriate locks followed by a commit phase to
2536  * implement whatever link state changes are indicated.
2537  */
2538 void bond_mii_monitor(struct work_struct *work)
2539 {
2540         struct bonding *bond = container_of(work, struct bonding,
2541                                             mii_work.work);
2542         bool should_notify_peers = false;
2543         unsigned long delay;
2544
2545         read_lock(&bond->lock);
2546
2547         delay = msecs_to_jiffies(bond->params.miimon);
2548
2549         if (bond->slave_cnt == 0)
2550                 goto re_arm;
2551
2552         should_notify_peers = bond_should_notify_peers(bond);
2553
2554         if (bond_miimon_inspect(bond)) {
2555                 read_unlock(&bond->lock);
2556
2557                 /* Race avoidance with bond_close cancel of workqueue */
2558                 if (!rtnl_trylock()) {
2559                         read_lock(&bond->lock);
2560                         delay = 1;
2561                         should_notify_peers = false;
2562                         goto re_arm;
2563                 }
2564
2565                 read_lock(&bond->lock);
2566
2567                 bond_miimon_commit(bond);
2568
2569                 read_unlock(&bond->lock);
2570                 rtnl_unlock();  /* might sleep, hold no other locks */
2571                 read_lock(&bond->lock);
2572         }
2573
2574 re_arm:
2575         if (bond->params.miimon)
2576                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2577
2578         read_unlock(&bond->lock);
2579
2580         if (should_notify_peers) {
2581                 if (!rtnl_trylock()) {
2582                         read_lock(&bond->lock);
2583                         bond->send_peer_notif++;
2584                         read_unlock(&bond->lock);
2585                         return;
2586                 }
2587                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2588                 rtnl_unlock();
2589         }
2590 }
2591
2592 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2593 {
2594         struct vlan_entry *vlan;
2595         struct net_device *vlan_dev;
2596
2597         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2598                 return 1;
2599
2600         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2601                 rcu_read_lock();
2602                 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2603                 rcu_read_unlock();
2604                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2605                         return 1;
2606         }
2607
2608         return 0;
2609 }
2610
2611 /*
2612  * We go to the (large) trouble of VLAN tagging ARP frames because
2613  * switches in VLAN mode (especially if ports are configured as
2614  * "native" to a VLAN) might not pass non-tagged frames.
2615  */
2616 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2617 {
2618         struct sk_buff *skb;
2619
2620         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2621                  slave_dev->name, dest_ip, src_ip, vlan_id);
2622
2623         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2624                          NULL, slave_dev->dev_addr, NULL);
2625
2626         if (!skb) {
2627                 pr_err("ARP packet allocation failed\n");
2628                 return;
2629         }
2630         if (vlan_id) {
2631                 skb = vlan_put_tag(skb, vlan_id);
2632                 if (!skb) {
2633                         pr_err("failed to insert VLAN tag\n");
2634                         return;
2635                 }
2636         }
2637         arp_xmit(skb);
2638 }
2639
2640
2641 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2642 {
2643         int i, vlan_id;
2644         __be32 *targets = bond->params.arp_targets;
2645         struct vlan_entry *vlan;
2646         struct net_device *vlan_dev = NULL;
2647         struct rtable *rt;
2648
2649         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2650                 __be32 addr;
2651                 if (!targets[i])
2652                         break;
2653                 pr_debug("basa: target %x\n", targets[i]);
2654                 if (!bond_vlan_used(bond)) {
2655                         pr_debug("basa: empty vlan: arp_send\n");
2656                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2657                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2658                                       addr, 0);
2659                         continue;
2660                 }
2661
2662                 /*
2663                  * If VLANs are configured, we do a route lookup to
2664                  * determine which VLAN interface would be used, so we
2665                  * can tag the ARP with the proper VLAN tag.
2666                  */
2667                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2668                                      RTO_ONLINK, 0);
2669                 if (IS_ERR(rt)) {
2670                         if (net_ratelimit()) {
2671                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2672                                            bond->dev->name, &targets[i]);
2673                         }
2674                         continue;
2675                 }
2676
2677                 /*
2678                  * This target is not on a VLAN
2679                  */
2680                 if (rt->dst.dev == bond->dev) {
2681                         ip_rt_put(rt);
2682                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2683                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2684                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2685                                       addr, 0);
2686                         continue;
2687                 }
2688
2689                 vlan_id = 0;
2690                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2691                         rcu_read_lock();
2692                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2693                                                         vlan->vlan_id);
2694                         rcu_read_unlock();
2695                         if (vlan_dev == rt->dst.dev) {
2696                                 vlan_id = vlan->vlan_id;
2697                                 pr_debug("basa: vlan match on %s %d\n",
2698                                        vlan_dev->name, vlan_id);
2699                                 break;
2700                         }
2701                 }
2702
2703                 if (vlan_id && vlan_dev) {
2704                         ip_rt_put(rt);
2705                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2706                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2707                                       addr, vlan_id);
2708                         continue;
2709                 }
2710
2711                 if (net_ratelimit()) {
2712                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2713                                    bond->dev->name, &targets[i],
2714                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2715                 }
2716                 ip_rt_put(rt);
2717         }
2718 }
2719
2720 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2721 {
2722         int i;
2723         __be32 *targets = bond->params.arp_targets;
2724
2725         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2726                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2727                          &sip, &tip, i, &targets[i],
2728                          bond_has_this_ip(bond, tip));
2729                 if (sip == targets[i]) {
2730                         if (bond_has_this_ip(bond, tip))
2731                                 slave->last_arp_rx = jiffies;
2732                         return;
2733                 }
2734         }
2735 }
2736
2737 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2738                         struct slave *slave)
2739 {
2740         struct arphdr *arp = (struct arphdr *)skb->data;
2741         unsigned char *arp_ptr;
2742         __be32 sip, tip;
2743         int alen;
2744
2745         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2746                 return RX_HANDLER_ANOTHER;
2747
2748         read_lock(&bond->lock);
2749         alen = arp_hdr_len(bond->dev);
2750
2751         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2752                  bond->dev->name, skb->dev->name);
2753
2754         if (alen > skb_headlen(skb)) {
2755                 arp = kmalloc(alen, GFP_ATOMIC);
2756                 if (!arp)
2757                         goto out_unlock;
2758                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2759                         goto out_unlock;
2760         }
2761
2762         if (arp->ar_hln != bond->dev->addr_len ||
2763             skb->pkt_type == PACKET_OTHERHOST ||
2764             skb->pkt_type == PACKET_LOOPBACK ||
2765             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2766             arp->ar_pro != htons(ETH_P_IP) ||
2767             arp->ar_pln != 4)
2768                 goto out_unlock;
2769
2770         arp_ptr = (unsigned char *)(arp + 1);
2771         arp_ptr += bond->dev->addr_len;
2772         memcpy(&sip, arp_ptr, 4);
2773         arp_ptr += 4 + bond->dev->addr_len;
2774         memcpy(&tip, arp_ptr, 4);
2775
2776         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2777                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2778                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2779                  &sip, &tip);
2780
2781         /*
2782          * Backup slaves won't see the ARP reply, but do come through
2783          * here for each ARP probe (so we swap the sip/tip to validate
2784          * the probe).  In a "redundant switch, common router" type of
2785          * configuration, the ARP probe will (hopefully) travel from
2786          * the active, through one switch, the router, then the other
2787          * switch before reaching the backup.
2788          */
2789         if (bond_is_active_slave(slave))
2790                 bond_validate_arp(bond, slave, sip, tip);
2791         else
2792                 bond_validate_arp(bond, slave, tip, sip);
2793
2794 out_unlock:
2795         read_unlock(&bond->lock);
2796         if (arp != (struct arphdr *)skb->data)
2797                 kfree(arp);
2798         return RX_HANDLER_ANOTHER;
2799 }
2800
2801 /*
2802  * this function is called regularly to monitor each slave's link
2803  * ensuring that traffic is being sent and received when arp monitoring
2804  * is used in load-balancing mode. if the adapter has been dormant, then an
2805  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2806  * arp monitoring in active backup mode.
2807  */
2808 void bond_loadbalance_arp_mon(struct work_struct *work)
2809 {
2810         struct bonding *bond = container_of(work, struct bonding,
2811                                             arp_work.work);
2812         struct slave *slave, *oldcurrent;
2813         int do_failover = 0;
2814         int delta_in_ticks, extra_ticks;
2815         int i;
2816
2817         read_lock(&bond->lock);
2818
2819         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2820         extra_ticks = delta_in_ticks / 2;
2821
2822         if (bond->slave_cnt == 0)
2823                 goto re_arm;
2824
2825         read_lock(&bond->curr_slave_lock);
2826         oldcurrent = bond->curr_active_slave;
2827         read_unlock(&bond->curr_slave_lock);
2828
2829         /* see if any of the previous devices are up now (i.e. they have
2830          * xmt and rcv traffic). the curr_active_slave does not come into
2831          * the picture unless it is null. also, slave->jiffies is not needed
2832          * here because we send an arp on each slave and give a slave as
2833          * long as it needs to get the tx/rx within the delta.
2834          * TODO: what about up/down delay in arp mode? it wasn't here before
2835          *       so it can wait
2836          */
2837         bond_for_each_slave(bond, slave, i) {
2838                 unsigned long trans_start = dev_trans_start(slave->dev);
2839
2840                 if (slave->link != BOND_LINK_UP) {
2841                         if (time_in_range(jiffies,
2842                                 trans_start - delta_in_ticks,
2843                                 trans_start + delta_in_ticks + extra_ticks) &&
2844                             time_in_range(jiffies,
2845                                 slave->dev->last_rx - delta_in_ticks,
2846                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2847
2848                                 slave->link  = BOND_LINK_UP;
2849                                 bond_set_active_slave(slave);
2850
2851                                 /* primary_slave has no meaning in round-robin
2852                                  * mode. the window of a slave being up and
2853                                  * curr_active_slave being null after enslaving
2854                                  * is closed.
2855                                  */
2856                                 if (!oldcurrent) {
2857                                         pr_info("%s: link status definitely up for interface %s, ",
2858                                                 bond->dev->name,
2859                                                 slave->dev->name);
2860                                         do_failover = 1;
2861                                 } else {
2862                                         pr_info("%s: interface %s is now up\n",
2863                                                 bond->dev->name,
2864                                                 slave->dev->name);
2865                                 }
2866                         }
2867                 } else {
2868                         /* slave->link == BOND_LINK_UP */
2869
2870                         /* not all switches will respond to an arp request
2871                          * when the source ip is 0, so don't take the link down
2872                          * if we don't know our ip yet
2873                          */
2874                         if (!time_in_range(jiffies,
2875                                 trans_start - delta_in_ticks,
2876                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2877                             !time_in_range(jiffies,
2878                                 slave->dev->last_rx - delta_in_ticks,
2879                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2880
2881                                 slave->link  = BOND_LINK_DOWN;
2882                                 bond_set_backup_slave(slave);
2883
2884                                 if (slave->link_failure_count < UINT_MAX)
2885                                         slave->link_failure_count++;
2886
2887                                 pr_info("%s: interface %s is now down.\n",
2888                                         bond->dev->name,
2889                                         slave->dev->name);
2890
2891                                 if (slave == oldcurrent)
2892                                         do_failover = 1;
2893                         }
2894                 }
2895
2896                 /* note: if switch is in round-robin mode, all links
2897                  * must tx arp to ensure all links rx an arp - otherwise
2898                  * links may oscillate or not come up at all; if switch is
2899                  * in something like xor mode, there is nothing we can
2900                  * do - all replies will be rx'ed on same link causing slaves
2901                  * to be unstable during low/no traffic periods
2902                  */
2903                 if (IS_UP(slave->dev))
2904                         bond_arp_send_all(bond, slave);
2905         }
2906
2907         if (do_failover) {
2908                 block_netpoll_tx();
2909                 write_lock_bh(&bond->curr_slave_lock);
2910
2911                 bond_select_active_slave(bond);
2912
2913                 write_unlock_bh(&bond->curr_slave_lock);
2914                 unblock_netpoll_tx();
2915         }
2916
2917 re_arm:
2918         if (bond->params.arp_interval)
2919                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2920
2921         read_unlock(&bond->lock);
2922 }
2923
2924 /*
2925  * Called to inspect slaves for active-backup mode ARP monitor link state
2926  * changes.  Sets new_link in slaves to specify what action should take
2927  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2928  * to link states must be committed.
2929  *
2930  * Called with bond->lock held for read.
2931  */
2932 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2933 {
2934         struct slave *slave;
2935         int i, commit = 0;
2936         unsigned long trans_start;
2937         int extra_ticks;
2938
2939         /* All the time comparisons below need some extra time. Otherwise, on
2940          * fast networks the ARP probe/reply may arrive within the same jiffy
2941          * as it was sent.  Then, the next time the ARP monitor is run, one
2942          * arp_interval will already have passed in the comparisons.
2943          */
2944         extra_ticks = delta_in_ticks / 2;
2945
2946         bond_for_each_slave(bond, slave, i) {
2947                 slave->new_link = BOND_LINK_NOCHANGE;
2948
2949                 if (slave->link != BOND_LINK_UP) {
2950                         if (time_in_range(jiffies,
2951                                 slave_last_rx(bond, slave) - delta_in_ticks,
2952                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2953
2954                                 slave->new_link = BOND_LINK_UP;
2955                                 commit++;
2956                         }
2957
2958                         continue;
2959                 }
2960
2961                 /*
2962                  * Give slaves 2*delta after being enslaved or made
2963                  * active.  This avoids bouncing, as the last receive
2964                  * times need a full ARP monitor cycle to be updated.
2965                  */
2966                 if (time_in_range(jiffies,
2967                                   slave->jiffies - delta_in_ticks,
2968                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2969                         continue;
2970
2971                 /*
2972                  * Backup slave is down if:
2973                  * - No current_arp_slave AND
2974                  * - more than 3*delta since last receive AND
2975                  * - the bond has an IP address
2976                  *
2977                  * Note: a non-null current_arp_slave indicates
2978                  * the curr_active_slave went down and we are
2979                  * searching for a new one; under this condition
2980                  * we only take the curr_active_slave down - this
2981                  * gives each slave a chance to tx/rx traffic
2982                  * before being taken out
2983                  */
2984                 if (!bond_is_active_slave(slave) &&
2985                     !bond->current_arp_slave &&
2986                     !time_in_range(jiffies,
2987                         slave_last_rx(bond, slave) - delta_in_ticks,
2988                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2989
2990                         slave->new_link = BOND_LINK_DOWN;
2991                         commit++;
2992                 }
2993
2994                 /*
2995                  * Active slave is down if:
2996                  * - more than 2*delta since transmitting OR
2997                  * - (more than 2*delta since receive AND
2998                  *    the bond has an IP address)
2999                  */
3000                 trans_start = dev_trans_start(slave->dev);
3001                 if (bond_is_active_slave(slave) &&
3002                     (!time_in_range(jiffies,
3003                         trans_start - delta_in_ticks,
3004                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
3005                      !time_in_range(jiffies,
3006                         slave_last_rx(bond, slave) - delta_in_ticks,
3007                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
3008
3009                         slave->new_link = BOND_LINK_DOWN;
3010                         commit++;
3011                 }
3012         }
3013
3014         return commit;
3015 }
3016
3017 /*
3018  * Called to commit link state changes noted by inspection step of
3019  * active-backup mode ARP monitor.
3020  *
3021  * Called with RTNL and bond->lock for read.
3022  */
3023 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3024 {
3025         struct slave *slave;
3026         int i;
3027         unsigned long trans_start;
3028
3029         bond_for_each_slave(bond, slave, i) {
3030                 switch (slave->new_link) {
3031                 case BOND_LINK_NOCHANGE:
3032                         continue;
3033
3034                 case BOND_LINK_UP:
3035                         trans_start = dev_trans_start(slave->dev);
3036                         if ((!bond->curr_active_slave &&
3037                              time_in_range(jiffies,
3038                                            trans_start - delta_in_ticks,
3039                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
3040                             bond->curr_active_slave != slave) {
3041                                 slave->link = BOND_LINK_UP;
3042                                 if (bond->current_arp_slave) {
3043                                         bond_set_slave_inactive_flags(
3044                                                 bond->current_arp_slave);
3045                                         bond->current_arp_slave = NULL;
3046                                 }
3047
3048                                 pr_info("%s: link status definitely up for interface %s.\n",
3049                                         bond->dev->name, slave->dev->name);
3050
3051                                 if (!bond->curr_active_slave ||
3052                                     (slave == bond->primary_slave))
3053                                         goto do_failover;
3054
3055                         }
3056
3057                         continue;
3058
3059                 case BOND_LINK_DOWN:
3060                         if (slave->link_failure_count < UINT_MAX)
3061                                 slave->link_failure_count++;
3062
3063                         slave->link = BOND_LINK_DOWN;
3064                         bond_set_slave_inactive_flags(slave);
3065
3066                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3067                                 bond->dev->name, slave->dev->name);
3068
3069                         if (slave == bond->curr_active_slave) {
3070                                 bond->current_arp_slave = NULL;
3071                                 goto do_failover;
3072                         }
3073
3074                         continue;
3075
3076                 default:
3077                         pr_err("%s: impossible: new_link %d on slave %s\n",
3078                                bond->dev->name, slave->new_link,
3079                                slave->dev->name);
3080                         continue;
3081                 }
3082
3083 do_failover:
3084                 ASSERT_RTNL();
3085                 block_netpoll_tx();
3086                 write_lock_bh(&bond->curr_slave_lock);
3087                 bond_select_active_slave(bond);
3088                 write_unlock_bh(&bond->curr_slave_lock);
3089                 unblock_netpoll_tx();
3090         }
3091
3092         bond_set_carrier(bond);
3093 }
3094
3095 /*
3096  * Send ARP probes for active-backup mode ARP monitor.
3097  *
3098  * Called with bond->lock held for read.
3099  */
3100 static void bond_ab_arp_probe(struct bonding *bond)
3101 {
3102         struct slave *slave;
3103         int i;
3104
3105         read_lock(&bond->curr_slave_lock);
3106
3107         if (bond->current_arp_slave && bond->curr_active_slave)
3108                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3109                         bond->current_arp_slave->dev->name,
3110                         bond->curr_active_slave->dev->name);
3111
3112         if (bond->curr_active_slave) {
3113                 bond_arp_send_all(bond, bond->curr_active_slave);
3114                 read_unlock(&bond->curr_slave_lock);
3115                 return;
3116         }
3117
3118         read_unlock(&bond->curr_slave_lock);
3119
3120         /* if we don't have a curr_active_slave, search for the next available
3121          * backup slave from the current_arp_slave and make it the candidate
3122          * for becoming the curr_active_slave
3123          */
3124
3125         if (!bond->current_arp_slave) {
3126                 bond->current_arp_slave = bond->first_slave;
3127                 if (!bond->current_arp_slave)
3128                         return;
3129         }
3130
3131         bond_set_slave_inactive_flags(bond->current_arp_slave);
3132
3133         /* search for next candidate */
3134         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3135                 if (IS_UP(slave->dev)) {
3136                         slave->link = BOND_LINK_BACK;
3137                         bond_set_slave_active_flags(slave);
3138                         bond_arp_send_all(bond, slave);
3139                         slave->jiffies = jiffies;
3140                         bond->current_arp_slave = slave;
3141                         break;
3142                 }
3143
3144                 /* if the link state is up at this point, we
3145                  * mark it down - this can happen if we have
3146                  * simultaneous link failures and
3147                  * reselect_active_interface doesn't make this
3148                  * one the current slave so it is still marked
3149                  * up when it is actually down
3150                  */
3151                 if (slave->link == BOND_LINK_UP) {
3152                         slave->link = BOND_LINK_DOWN;
3153                         if (slave->link_failure_count < UINT_MAX)
3154                                 slave->link_failure_count++;
3155
3156                         bond_set_slave_inactive_flags(slave);
3157
3158                         pr_info("%s: backup interface %s is now down.\n",
3159                                 bond->dev->name, slave->dev->name);
3160                 }
3161         }
3162 }
3163
3164 void bond_activebackup_arp_mon(struct work_struct *work)
3165 {
3166         struct bonding *bond = container_of(work, struct bonding,
3167                                             arp_work.work);
3168         bool should_notify_peers = false;
3169         int delta_in_ticks;
3170
3171         read_lock(&bond->lock);
3172
3173         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3174
3175         if (bond->slave_cnt == 0)
3176                 goto re_arm;
3177
3178         should_notify_peers = bond_should_notify_peers(bond);
3179
3180         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3181                 read_unlock(&bond->lock);
3182
3183                 /* Race avoidance with bond_close flush of workqueue */
3184                 if (!rtnl_trylock()) {
3185                         read_lock(&bond->lock);
3186                         delta_in_ticks = 1;
3187                         should_notify_peers = false;
3188                         goto re_arm;
3189                 }
3190
3191                 read_lock(&bond->lock);
3192
3193                 bond_ab_arp_commit(bond, delta_in_ticks);
3194
3195                 read_unlock(&bond->lock);
3196                 rtnl_unlock();
3197                 read_lock(&bond->lock);
3198         }
3199
3200         bond_ab_arp_probe(bond);
3201
3202 re_arm:
3203         if (bond->params.arp_interval)
3204                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3205
3206         read_unlock(&bond->lock);
3207
3208         if (should_notify_peers) {
3209                 if (!rtnl_trylock()) {
3210                         read_lock(&bond->lock);
3211                         bond->send_peer_notif++;
3212                         read_unlock(&bond->lock);
3213                         return;
3214                 }
3215                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3216                 rtnl_unlock();
3217         }
3218 }
3219
3220 /*-------------------------- netdev event handling --------------------------*/
3221
3222 /*
3223  * Change device name
3224  */
3225 static int bond_event_changename(struct bonding *bond)
3226 {
3227         bond_remove_proc_entry(bond);
3228         bond_create_proc_entry(bond);
3229
3230         bond_debug_reregister(bond);
3231
3232         return NOTIFY_DONE;
3233 }
3234
3235 static int bond_master_netdev_event(unsigned long event,
3236                                     struct net_device *bond_dev)
3237 {
3238         struct bonding *event_bond = netdev_priv(bond_dev);
3239
3240         switch (event) {
3241         case NETDEV_CHANGENAME:
3242                 return bond_event_changename(event_bond);
3243         case NETDEV_UNREGISTER:
3244                 bond_remove_proc_entry(event_bond);
3245                 break;
3246         case NETDEV_REGISTER:
3247                 bond_create_proc_entry(event_bond);
3248                 break;
3249         default:
3250                 break;
3251         }
3252
3253         return NOTIFY_DONE;
3254 }
3255
3256 static int bond_slave_netdev_event(unsigned long event,
3257                                    struct net_device *slave_dev)
3258 {
3259         struct net_device *bond_dev = slave_dev->master;
3260         struct bonding *bond = netdev_priv(bond_dev);
3261         struct slave *slave = NULL;
3262
3263         switch (event) {
3264         case NETDEV_UNREGISTER:
3265                 if (bond_dev) {
3266                         if (bond->setup_by_slave)
3267                                 bond_release_and_destroy(bond_dev, slave_dev);
3268                         else
3269                                 bond_release(bond_dev, slave_dev);
3270                 }
3271                 break;
3272         case NETDEV_UP:
3273         case NETDEV_CHANGE:
3274                 slave = bond_get_slave_by_dev(bond, slave_dev);
3275                 if (slave) {
3276                         u32 old_speed = slave->speed;
3277                         u8  old_duplex = slave->duplex;
3278
3279                         bond_update_speed_duplex(slave);
3280
3281                         if (bond->params.mode == BOND_MODE_8023AD) {
3282                                 if (old_speed != slave->speed)
3283                                         bond_3ad_adapter_speed_changed(slave);
3284                                 if (old_duplex != slave->duplex)
3285                                         bond_3ad_adapter_duplex_changed(slave);
3286                         }
3287                 }
3288
3289                 break;
3290         case NETDEV_DOWN:
3291                 /*
3292                  * ... Or is it this?
3293                  */
3294                 break;
3295         case NETDEV_CHANGEMTU:
3296                 /*
3297                  * TODO: Should slaves be allowed to
3298                  * independently alter their MTU?  For
3299                  * an active-backup bond, slaves need
3300                  * not be the same type of device, so
3301                  * MTUs may vary.  For other modes,
3302                  * slaves arguably should have the
3303                  * same MTUs. To do this, we'd need to
3304                  * take over the slave's change_mtu
3305                  * function for the duration of their
3306                  * servitude.
3307                  */
3308                 break;
3309         case NETDEV_CHANGENAME:
3310                 /*
3311                  * TODO: handle changing the primary's name
3312                  */
3313                 break;
3314         case NETDEV_FEAT_CHANGE:
3315                 bond_compute_features(bond);
3316                 break;
3317         default:
3318                 break;
3319         }
3320
3321         return NOTIFY_DONE;
3322 }
3323
3324 /*
3325  * bond_netdev_event: handle netdev notifier chain events.
3326  *
3327  * This function receives events for the netdev chain.  The caller (an
3328  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3329  * locks for us to safely manipulate the slave devices (RTNL lock,
3330  * dev_probe_lock).
3331  */
3332 static int bond_netdev_event(struct notifier_block *this,
3333                              unsigned long event, void *ptr)
3334 {
3335         struct net_device *event_dev = (struct net_device *)ptr;
3336
3337         pr_debug("event_dev: %s, event: %lx\n",
3338                  event_dev ? event_dev->name : "None",
3339                  event);
3340
3341         if (!(event_dev->priv_flags & IFF_BONDING))
3342                 return NOTIFY_DONE;
3343
3344         if (event_dev->flags & IFF_MASTER) {
3345                 pr_debug("IFF_MASTER\n");
3346                 return bond_master_netdev_event(event, event_dev);
3347         }
3348
3349         if (event_dev->flags & IFF_SLAVE) {
3350                 pr_debug("IFF_SLAVE\n");
3351                 return bond_slave_netdev_event(event, event_dev);
3352         }
3353
3354         return NOTIFY_DONE;
3355 }
3356
3357 static struct notifier_block bond_netdev_notifier = {
3358         .notifier_call = bond_netdev_event,
3359 };
3360
3361 /*---------------------------- Hashing Policies -----------------------------*/
3362
3363 /*
3364  * Hash for the output device based upon layer 2 data
3365  */
3366 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3367 {
3368         struct ethhdr *data = (struct ethhdr *)skb->data;
3369
3370         if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3371                 return (data->h_dest[5] ^ data->h_source[5]) % count;
3372
3373         return 0;
3374 }
3375
3376 /*
3377  * Hash for the output device based upon layer 2 and layer 3 data. If
3378  * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3379  */
3380 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3381 {
3382         struct ethhdr *data = (struct ethhdr *)skb->data;
3383         struct iphdr *iph;
3384         struct ipv6hdr *ipv6h;
3385         u32 v6hash;
3386         __be32 *s, *d;
3387
3388         if (skb->protocol == htons(ETH_P_IP) &&
3389             skb_network_header_len(skb) >= sizeof(*iph)) {
3390                 iph = ip_hdr(skb);
3391                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3392                         (data->h_dest[5] ^ data->h_source[5])) % count;
3393         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3394                    skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3395                 ipv6h = ipv6_hdr(skb);
3396                 s = &ipv6h->saddr.s6_addr32[0];
3397                 d = &ipv6h->daddr.s6_addr32[0];
3398                 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3399                 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3400                 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3401         }
3402
3403         return bond_xmit_hash_policy_l2(skb, count);
3404 }
3405
3406 /*
3407  * Hash for the output device based upon layer 3 and layer 4 data. If
3408  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3409  * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3410  */
3411 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3412 {
3413         u32 layer4_xor = 0;
3414         struct iphdr *iph;
3415         struct ipv6hdr *ipv6h;
3416         __be32 *s, *d;
3417         __be16 *layer4hdr;
3418
3419         if (skb->protocol == htons(ETH_P_IP) &&
3420             skb_network_header_len(skb) >= sizeof(*iph)) {
3421                 iph = ip_hdr(skb);
3422                 if (!ip_is_fragment(iph) &&
3423                     (iph->protocol == IPPROTO_TCP ||
3424                      iph->protocol == IPPROTO_UDP) &&
3425                     (skb_headlen(skb) - skb_network_offset(skb) >=
3426                      iph->ihl * sizeof(u32) + sizeof(*layer4hdr) * 2)) {
3427                         layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3428                         layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3429                 }
3430                 return (layer4_xor ^
3431                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3432         } else if (skb->protocol == htons(ETH_P_IPV6) &&
3433                    skb_network_header_len(skb) >= sizeof(*ipv6h)) {
3434                 ipv6h = ipv6_hdr(skb);
3435                 if ((ipv6h->nexthdr == IPPROTO_TCP ||
3436                      ipv6h->nexthdr == IPPROTO_UDP) &&
3437                     (skb_headlen(skb) - skb_network_offset(skb) >=
3438                      sizeof(*ipv6h) + sizeof(*layer4hdr) * 2)) {
3439                         layer4hdr = (__be16 *)(ipv6h + 1);
3440                         layer4_xor = ntohs(*layer4hdr ^ *(layer4hdr + 1));
3441                 }
3442                 s = &ipv6h->saddr.s6_addr32[0];
3443                 d = &ipv6h->daddr.s6_addr32[0];
3444                 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3445                 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3446                                (layer4_xor >> 8);
3447                 return layer4_xor % count;
3448         }
3449
3450         return bond_xmit_hash_policy_l2(skb, count);
3451 }
3452
3453 /*-------------------------- Device entry points ----------------------------*/
3454
3455 static int bond_open(struct net_device *bond_dev)
3456 {
3457         struct bonding *bond = netdev_priv(bond_dev);
3458         struct slave *slave;
3459         int i;
3460
3461         /* reset slave->backup and slave->inactive */
3462         read_lock(&bond->lock);
3463         if (bond->slave_cnt > 0) {
3464                 read_lock(&bond->curr_slave_lock);
3465                 bond_for_each_slave(bond, slave, i) {
3466                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3467                                 && (slave != bond->curr_active_slave)) {
3468                                 bond_set_slave_inactive_flags(slave);
3469                         } else {
3470                                 bond_set_slave_active_flags(slave);
3471                         }
3472                 }
3473                 read_unlock(&bond->curr_slave_lock);
3474         }
3475         read_unlock(&bond->lock);
3476
3477         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3478
3479         if (bond_is_lb(bond)) {
3480                 /* bond_alb_initialize must be called before the timer
3481                  * is started.
3482                  */
3483                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3484                         /* something went wrong - fail the open operation */
3485                         return -ENOMEM;
3486                 }
3487
3488                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3489                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3490         }
3491
3492         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3493                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3494                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3495         }
3496
3497         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3498                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3499                         INIT_DELAYED_WORK(&bond->arp_work,
3500                                           bond_activebackup_arp_mon);
3501                 else
3502                         INIT_DELAYED_WORK(&bond->arp_work,
3503                                           bond_loadbalance_arp_mon);
3504
3505                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3506                 if (bond->params.arp_validate)
3507                         bond->recv_probe = bond_arp_rcv;
3508         }
3509
3510         if (bond->params.mode == BOND_MODE_8023AD) {
3511                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3512                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3513                 /* register to receive LACPDUs */
3514                 bond->recv_probe = bond_3ad_lacpdu_recv;
3515                 bond_3ad_initiate_agg_selection(bond, 1);
3516         }
3517
3518         return 0;
3519 }
3520
3521 static int bond_close(struct net_device *bond_dev)
3522 {
3523         struct bonding *bond = netdev_priv(bond_dev);
3524
3525         write_lock_bh(&bond->lock);
3526
3527         bond->send_peer_notif = 0;
3528
3529         write_unlock_bh(&bond->lock);
3530
3531         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3532                 cancel_delayed_work_sync(&bond->mii_work);
3533         }
3534
3535         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3536                 cancel_delayed_work_sync(&bond->arp_work);
3537         }
3538
3539         switch (bond->params.mode) {
3540         case BOND_MODE_8023AD:
3541                 cancel_delayed_work_sync(&bond->ad_work);
3542                 break;
3543         case BOND_MODE_TLB:
3544         case BOND_MODE_ALB:
3545                 cancel_delayed_work_sync(&bond->alb_work);
3546                 break;
3547         default:
3548                 break;
3549         }
3550
3551         if (delayed_work_pending(&bond->mcast_work))
3552                 cancel_delayed_work_sync(&bond->mcast_work);
3553
3554         if (bond_is_lb(bond)) {
3555                 /* Must be called only after all
3556                  * slaves have been released
3557                  */
3558                 bond_alb_deinitialize(bond);
3559         }
3560         bond->recv_probe = NULL;
3561
3562         return 0;
3563 }
3564
3565 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3566                                                 struct rtnl_link_stats64 *stats)
3567 {
3568         struct bonding *bond = netdev_priv(bond_dev);
3569         struct rtnl_link_stats64 temp;
3570         struct slave *slave;
3571         int i;
3572
3573         memset(stats, 0, sizeof(*stats));
3574
3575         read_lock_bh(&bond->lock);
3576
3577         bond_for_each_slave(bond, slave, i) {
3578                 const struct rtnl_link_stats64 *sstats =
3579                         dev_get_stats(slave->dev, &temp);
3580
3581                 stats->rx_packets += sstats->rx_packets;
3582                 stats->rx_bytes += sstats->rx_bytes;
3583                 stats->rx_errors += sstats->rx_errors;
3584                 stats->rx_dropped += sstats->rx_dropped;
3585
3586                 stats->tx_packets += sstats->tx_packets;
3587                 stats->tx_bytes += sstats->tx_bytes;
3588                 stats->tx_errors += sstats->tx_errors;
3589                 stats->tx_dropped += sstats->tx_dropped;
3590
3591                 stats->multicast += sstats->multicast;
3592                 stats->collisions += sstats->collisions;
3593
3594                 stats->rx_length_errors += sstats->rx_length_errors;
3595                 stats->rx_over_errors += sstats->rx_over_errors;
3596                 stats->rx_crc_errors += sstats->rx_crc_errors;
3597                 stats->rx_frame_errors += sstats->rx_frame_errors;
3598                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3599                 stats->rx_missed_errors += sstats->rx_missed_errors;
3600
3601                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3602                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3603                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3604                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3605                 stats->tx_window_errors += sstats->tx_window_errors;
3606         }
3607
3608         read_unlock_bh(&bond->lock);
3609
3610         return stats;
3611 }
3612
3613 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3614 {
3615         struct net_device *slave_dev = NULL;
3616         struct ifbond k_binfo;
3617         struct ifbond __user *u_binfo = NULL;
3618         struct ifslave k_sinfo;
3619         struct ifslave __user *u_sinfo = NULL;
3620         struct mii_ioctl_data *mii = NULL;
3621         int res = 0;
3622
3623         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3624
3625         switch (cmd) {
3626         case SIOCGMIIPHY:
3627                 mii = if_mii(ifr);
3628                 if (!mii)
3629                         return -EINVAL;
3630
3631                 mii->phy_id = 0;
3632                 /* Fall Through */
3633         case SIOCGMIIREG:
3634                 /*
3635                  * We do this again just in case we were called by SIOCGMIIREG
3636                  * instead of SIOCGMIIPHY.
3637                  */
3638                 mii = if_mii(ifr);
3639                 if (!mii)
3640                         return -EINVAL;
3641
3642
3643                 if (mii->reg_num == 1) {
3644                         struct bonding *bond = netdev_priv(bond_dev);
3645                         mii->val_out = 0;
3646                         read_lock(&bond->lock);
3647                         read_lock(&bond->curr_slave_lock);
3648                         if (netif_carrier_ok(bond->dev))
3649                                 mii->val_out = BMSR_LSTATUS;
3650
3651                         read_unlock(&bond->curr_slave_lock);
3652                         read_unlock(&bond->lock);
3653                 }
3654
3655                 return 0;
3656         case BOND_INFO_QUERY_OLD:
3657         case SIOCBONDINFOQUERY:
3658                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3659
3660                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3661                         return -EFAULT;
3662
3663                 res = bond_info_query(bond_dev, &k_binfo);
3664                 if (res == 0 &&
3665                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3666                         return -EFAULT;
3667
3668                 return res;
3669         case BOND_SLAVE_INFO_QUERY_OLD:
3670         case SIOCBONDSLAVEINFOQUERY:
3671                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3672
3673                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3674                         return -EFAULT;
3675
3676                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3677                 if (res == 0 &&
3678                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3679                         return -EFAULT;
3680
3681                 return res;
3682         default:
3683                 /* Go on */
3684                 break;
3685         }
3686
3687         if (!capable(CAP_NET_ADMIN))
3688                 return -EPERM;
3689
3690         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3691
3692         pr_debug("slave_dev=%p:\n", slave_dev);
3693
3694         if (!slave_dev)
3695                 res = -ENODEV;
3696         else {
3697                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3698                 switch (cmd) {
3699                 case BOND_ENSLAVE_OLD:
3700                 case SIOCBONDENSLAVE:
3701                         res = bond_enslave(bond_dev, slave_dev);
3702                         break;
3703                 case BOND_RELEASE_OLD:
3704                 case SIOCBONDRELEASE:
3705                         res = bond_release(bond_dev, slave_dev);
3706                         break;
3707                 case BOND_SETHWADDR_OLD:
3708                 case SIOCBONDSETHWADDR:
3709                         res = bond_sethwaddr(bond_dev, slave_dev);
3710                         break;
3711                 case BOND_CHANGE_ACTIVE_OLD:
3712                 case SIOCBONDCHANGEACTIVE:
3713                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3714                         break;
3715                 default:
3716                         res = -EOPNOTSUPP;
3717                 }
3718
3719                 dev_put(slave_dev);
3720         }
3721
3722         return res;
3723 }
3724
3725 static bool bond_addr_in_mc_list(unsigned char *addr,
3726                                  struct netdev_hw_addr_list *list,
3727                                  int addrlen)
3728 {
3729         struct netdev_hw_addr *ha;
3730
3731         netdev_hw_addr_list_for_each(ha, list)
3732                 if (!memcmp(ha->addr, addr, addrlen))
3733                         return true;
3734
3735         return false;
3736 }
3737
3738 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3739 {
3740         struct bonding *bond = netdev_priv(bond_dev);
3741
3742         if (change & IFF_PROMISC)
3743                 bond_set_promiscuity(bond,
3744                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3745
3746         if (change & IFF_ALLMULTI)
3747                 bond_set_allmulti(bond,
3748                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3749 }
3750
3751 static void bond_set_multicast_list(struct net_device *bond_dev)
3752 {
3753         struct bonding *bond = netdev_priv(bond_dev);
3754         struct netdev_hw_addr *ha;
3755         bool found;
3756
3757         read_lock(&bond->lock);
3758
3759         /* looking for addresses to add to slaves' mc list */
3760         netdev_for_each_mc_addr(ha, bond_dev) {
3761                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3762                                              bond_dev->addr_len);
3763                 if (!found)
3764                         bond_mc_add(bond, ha->addr);
3765         }
3766
3767         /* looking for addresses to delete from slaves' list */
3768         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3769                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3770                                              bond_dev->addr_len);
3771                 if (!found)
3772                         bond_mc_del(bond, ha->addr);
3773         }
3774
3775         /* save master's multicast list */
3776         __hw_addr_flush(&bond->mc_list);
3777         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3778                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3779
3780         read_unlock(&bond->lock);
3781 }
3782
3783 static int bond_neigh_init(struct neighbour *n)
3784 {
3785         struct bonding *bond = netdev_priv(n->dev);
3786         struct slave *slave = bond->first_slave;
3787         const struct net_device_ops *slave_ops;
3788         struct neigh_parms parms;
3789         int ret;
3790
3791         if (!slave)
3792                 return 0;
3793
3794         slave_ops = slave->dev->netdev_ops;
3795
3796         if (!slave_ops->ndo_neigh_setup)
3797                 return 0;
3798
3799         parms.neigh_setup = NULL;
3800         parms.neigh_cleanup = NULL;
3801         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3802         if (ret)
3803                 return ret;
3804
3805         /*
3806          * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3807          * after the last slave has been detached.  Assumes that all slaves
3808          * utilize the same neigh_cleanup (true at this writing as only user
3809          * is ipoib).
3810          */
3811         n->parms->neigh_cleanup = parms.neigh_cleanup;
3812
3813         if (!parms.neigh_setup)
3814                 return 0;
3815
3816         return parms.neigh_setup(n);
3817 }
3818
3819 /*
3820  * The bonding ndo_neigh_setup is called at init time beofre any
3821  * slave exists. So we must declare proxy setup function which will
3822  * be used at run time to resolve the actual slave neigh param setup.
3823  */
3824 static int bond_neigh_setup(struct net_device *dev,
3825                             struct neigh_parms *parms)
3826 {
3827         parms->neigh_setup   = bond_neigh_init;
3828
3829         return 0;
3830 }
3831
3832 /*
3833  * Change the MTU of all of a master's slaves to match the master
3834  */
3835 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3836 {
3837         struct bonding *bond = netdev_priv(bond_dev);
3838         struct slave *slave, *stop_at;
3839         int res = 0;
3840         int i;
3841
3842         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3843                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3844
3845         /* Can't hold bond->lock with bh disabled here since
3846          * some base drivers panic. On the other hand we can't
3847          * hold bond->lock without bh disabled because we'll
3848          * deadlock. The only solution is to rely on the fact
3849          * that we're under rtnl_lock here, and the slaves
3850          * list won't change. This doesn't solve the problem
3851          * of setting the slave's MTU while it is
3852          * transmitting, but the assumption is that the base
3853          * driver can handle that.
3854          *
3855          * TODO: figure out a way to safely iterate the slaves
3856          * list, but without holding a lock around the actual
3857          * call to the base driver.
3858          */
3859
3860         bond_for_each_slave(bond, slave, i) {
3861                 pr_debug("s %p s->p %p c_m %p\n",
3862                          slave,
3863                          slave->prev,
3864                          slave->dev->netdev_ops->ndo_change_mtu);
3865
3866                 res = dev_set_mtu(slave->dev, new_mtu);
3867
3868                 if (res) {
3869                         /* If we failed to set the slave's mtu to the new value
3870                          * we must abort the operation even in ACTIVE_BACKUP
3871                          * mode, because if we allow the backup slaves to have
3872                          * different mtu values than the active slave we'll
3873                          * need to change their mtu when doing a failover. That
3874                          * means changing their mtu from timer context, which
3875                          * is probably not a good idea.
3876                          */
3877                         pr_debug("err %d %s\n", res, slave->dev->name);
3878                         goto unwind;
3879                 }
3880         }
3881
3882         bond_dev->mtu = new_mtu;
3883
3884         return 0;
3885
3886 unwind:
3887         /* unwind from head to the slave that failed */
3888         stop_at = slave;
3889         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3890                 int tmp_res;
3891
3892                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3893                 if (tmp_res) {
3894                         pr_debug("unwind err %d dev %s\n",
3895                                  tmp_res, slave->dev->name);
3896                 }
3897         }
3898
3899         return res;
3900 }
3901
3902 /*
3903  * Change HW address
3904  *
3905  * Note that many devices must be down to change the HW address, and
3906  * downing the master releases all slaves.  We can make bonds full of
3907  * bonding devices to test this, however.
3908  */
3909 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3910 {
3911         struct bonding *bond = netdev_priv(bond_dev);
3912         struct sockaddr *sa = addr, tmp_sa;
3913         struct slave *slave, *stop_at;
3914         int res = 0;
3915         int i;
3916
3917         if (bond->params.mode == BOND_MODE_ALB)
3918                 return bond_alb_set_mac_address(bond_dev, addr);
3919
3920
3921         pr_debug("bond=%p, name=%s\n",
3922                  bond, bond_dev ? bond_dev->name : "None");
3923
3924         /*
3925          * If fail_over_mac is set to active, do nothing and return
3926          * success.  Returning an error causes ifenslave to fail.
3927          */
3928         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3929                 return 0;
3930
3931         if (!is_valid_ether_addr(sa->sa_data))
3932                 return -EADDRNOTAVAIL;
3933
3934         /* Can't hold bond->lock with bh disabled here since
3935          * some base drivers panic. On the other hand we can't
3936          * hold bond->lock without bh disabled because we'll
3937          * deadlock. The only solution is to rely on the fact
3938          * that we're under rtnl_lock here, and the slaves
3939          * list won't change. This doesn't solve the problem
3940          * of setting the slave's hw address while it is
3941          * transmitting, but the assumption is that the base
3942          * driver can handle that.
3943          *
3944          * TODO: figure out a way to safely iterate the slaves
3945          * list, but without holding a lock around the actual
3946          * call to the base driver.
3947          */
3948
3949         bond_for_each_slave(bond, slave, i) {
3950                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3951                 pr_debug("slave %p %s\n", slave, slave->dev->name);
3952
3953                 if (slave_ops->ndo_set_mac_address == NULL) {
3954                         res = -EOPNOTSUPP;
3955                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3956                         goto unwind;
3957                 }
3958
3959                 res = dev_set_mac_address(slave->dev, addr);
3960                 if (res) {
3961                         /* TODO: consider downing the slave
3962                          * and retry ?
3963                          * User should expect communications
3964                          * breakage anyway until ARP finish
3965                          * updating, so...
3966                          */
3967                         pr_debug("err %d %s\n", res, slave->dev->name);
3968                         goto unwind;
3969                 }
3970         }
3971
3972         /* success */
3973         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3974         return 0;
3975
3976 unwind:
3977         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3978         tmp_sa.sa_family = bond_dev->type;
3979
3980         /* unwind from head to the slave that failed */
3981         stop_at = slave;
3982         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3983                 int tmp_res;
3984
3985                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3986                 if (tmp_res) {
3987                         pr_debug("unwind err %d dev %s\n",
3988                                  tmp_res, slave->dev->name);
3989                 }
3990         }
3991
3992         return res;
3993 }
3994
3995 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3996 {
3997         struct bonding *bond = netdev_priv(bond_dev);
3998         struct slave *slave, *start_at;
3999         int i, slave_no, res = 1;
4000         struct iphdr *iph = ip_hdr(skb);
4001
4002         /*
4003          * Start with the curr_active_slave that joined the bond as the
4004          * default for sending IGMP traffic.  For failover purposes one
4005          * needs to maintain some consistency for the interface that will
4006          * send the join/membership reports.  The curr_active_slave found
4007          * will send all of this type of traffic.
4008          */
4009         if ((iph->protocol == IPPROTO_IGMP) &&
4010             (skb->protocol == htons(ETH_P_IP))) {
4011
4012                 read_lock(&bond->curr_slave_lock);
4013                 slave = bond->curr_active_slave;
4014                 read_unlock(&bond->curr_slave_lock);
4015
4016                 if (!slave)
4017                         goto out;
4018         } else {
4019                 /*
4020                  * Concurrent TX may collide on rr_tx_counter; we accept
4021                  * that as being rare enough not to justify using an
4022                  * atomic op here.
4023                  */
4024                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4025
4026                 bond_for_each_slave(bond, slave, i) {
4027                         slave_no--;
4028                         if (slave_no < 0)
4029                                 break;
4030                 }
4031         }
4032
4033         start_at = slave;
4034         bond_for_each_slave_from(bond, slave, i, start_at) {
4035                 if (IS_UP(slave->dev) &&
4036                     (slave->link == BOND_LINK_UP) &&
4037                     bond_is_active_slave(slave)) {
4038                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4039                         break;
4040                 }
4041         }
4042
4043 out:
4044         if (res) {
4045                 /* no suitable interface, frame not sent */
4046                 kfree_skb(skb);
4047         }
4048
4049         return NETDEV_TX_OK;
4050 }
4051
4052
4053 /*
4054  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4055  * the bond has a usable interface.
4056  */
4057 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4058 {
4059         struct bonding *bond = netdev_priv(bond_dev);
4060         int res = 1;
4061
4062         read_lock(&bond->curr_slave_lock);
4063
4064         if (bond->curr_active_slave)
4065                 res = bond_dev_queue_xmit(bond, skb,
4066                         bond->curr_active_slave->dev);
4067
4068         read_unlock(&bond->curr_slave_lock);
4069
4070         if (res)
4071                 /* no suitable interface, frame not sent */
4072                 kfree_skb(skb);
4073
4074         return NETDEV_TX_OK;
4075 }
4076
4077 /*
4078  * In bond_xmit_xor() , we determine the output device by using a pre-
4079  * determined xmit_hash_policy(), If the selected device is not enabled,
4080  * find the next active slave.
4081  */
4082 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4083 {
4084         struct bonding *bond = netdev_priv(bond_dev);
4085         struct slave *slave, *start_at;
4086         int slave_no;
4087         int i;
4088         int res = 1;
4089
4090         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4091
4092         bond_for_each_slave(bond, slave, i) {
4093                 slave_no--;
4094                 if (slave_no < 0)
4095                         break;
4096         }
4097
4098         start_at = slave;
4099
4100         bond_for_each_slave_from(bond, slave, i, start_at) {
4101                 if (IS_UP(slave->dev) &&
4102                     (slave->link == BOND_LINK_UP) &&
4103                     bond_is_active_slave(slave)) {
4104                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4105                         break;
4106                 }
4107         }
4108
4109         if (res) {
4110                 /* no suitable interface, frame not sent */
4111                 kfree_skb(skb);
4112         }
4113
4114         return NETDEV_TX_OK;
4115 }
4116
4117 /*
4118  * in broadcast mode, we send everything to all usable interfaces.
4119  */
4120 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4121 {
4122         struct bonding *bond = netdev_priv(bond_dev);
4123         struct slave *slave, *start_at;
4124         struct net_device *tx_dev = NULL;
4125         int i;
4126         int res = 1;
4127
4128         read_lock(&bond->curr_slave_lock);
4129         start_at = bond->curr_active_slave;
4130         read_unlock(&bond->curr_slave_lock);
4131
4132         if (!start_at)
4133                 goto out;
4134
4135         bond_for_each_slave_from(bond, slave, i, start_at) {
4136                 if (IS_UP(slave->dev) &&
4137                     (slave->link == BOND_LINK_UP) &&
4138                     bond_is_active_slave(slave)) {
4139                         if (tx_dev) {
4140                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4141                                 if (!skb2) {
4142                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4143                                                bond_dev->name);
4144                                         continue;
4145                                 }
4146
4147                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4148                                 if (res) {
4149                                         kfree_skb(skb2);
4150                                         continue;
4151                                 }
4152                         }
4153                         tx_dev = slave->dev;
4154                 }
4155         }
4156
4157         if (tx_dev)
4158                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4159
4160 out:
4161         if (res)
4162                 /* no suitable interface, frame not sent */
4163                 kfree_skb(skb);
4164
4165         /* frame sent to all suitable interfaces */
4166         return NETDEV_TX_OK;
4167 }
4168
4169 /*------------------------- Device initialization ---------------------------*/
4170
4171 static void bond_set_xmit_hash_policy(struct bonding *bond)
4172 {
4173         switch (bond->params.xmit_policy) {
4174         case BOND_XMIT_POLICY_LAYER23:
4175                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4176                 break;
4177         case BOND_XMIT_POLICY_LAYER34:
4178                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4179                 break;
4180         case BOND_XMIT_POLICY_LAYER2:
4181         default:
4182                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4183                 break;
4184         }
4185 }
4186
4187 /*
4188  * Lookup the slave that corresponds to a qid
4189  */
4190 static inline int bond_slave_override(struct bonding *bond,
4191                                       struct sk_buff *skb)
4192 {
4193         int i, res = 1;
4194         struct slave *slave = NULL;
4195         struct slave *check_slave;
4196
4197         if (!skb->queue_mapping)
4198                 return 1;
4199
4200         /* Find out if any slaves have the same mapping as this skb. */
4201         bond_for_each_slave(bond, check_slave, i) {
4202                 if (check_slave->queue_id == skb->queue_mapping) {
4203                         slave = check_slave;
4204                         break;
4205                 }
4206         }
4207
4208         /* If the slave isn't UP, use default transmit policy. */
4209         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4210             (slave->link == BOND_LINK_UP)) {
4211                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4212         }
4213
4214         return res;
4215 }
4216
4217
4218 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4219 {
4220         /*
4221          * This helper function exists to help dev_pick_tx get the correct
4222          * destination queue.  Using a helper function skips a call to
4223          * skb_tx_hash and will put the skbs in the queue we expect on their
4224          * way down to the bonding driver.
4225          */
4226         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4227
4228         /*
4229          * Save the original txq to restore before passing to the driver
4230          */
4231         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4232
4233         if (unlikely(txq >= dev->real_num_tx_queues)) {
4234                 do {
4235                         txq -= dev->real_num_tx_queues;
4236                 } while (txq >= dev->real_num_tx_queues);
4237         }
4238         return txq;
4239 }
4240
4241 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4242 {
4243         struct bonding *bond = netdev_priv(dev);
4244
4245         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4246                 if (!bond_slave_override(bond, skb))
4247                         return NETDEV_TX_OK;
4248         }
4249
4250         switch (bond->params.mode) {
4251         case BOND_MODE_ROUNDROBIN:
4252                 return bond_xmit_roundrobin(skb, dev);
4253         case BOND_MODE_ACTIVEBACKUP:
4254                 return bond_xmit_activebackup(skb, dev);
4255         case BOND_MODE_XOR:
4256                 return bond_xmit_xor(skb, dev);
4257         case BOND_MODE_BROADCAST:
4258                 return bond_xmit_broadcast(skb, dev);
4259         case BOND_MODE_8023AD:
4260                 return bond_3ad_xmit_xor(skb, dev);
4261         case BOND_MODE_ALB:
4262         case BOND_MODE_TLB:
4263                 return bond_alb_xmit(skb, dev);
4264         default:
4265                 /* Should never happen, mode already checked */
4266                 pr_err("%s: Error: Unknown bonding mode %d\n",
4267                        dev->name, bond->params.mode);
4268                 WARN_ON_ONCE(1);
4269                 kfree_skb(skb);
4270                 return NETDEV_TX_OK;
4271         }
4272 }
4273
4274 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4275 {
4276         struct bonding *bond = netdev_priv(dev);
4277         netdev_tx_t ret = NETDEV_TX_OK;
4278
4279         /*
4280          * If we risk deadlock from transmitting this in the
4281          * netpoll path, tell netpoll to queue the frame for later tx
4282          */
4283         if (is_netpoll_tx_blocked(dev))
4284                 return NETDEV_TX_BUSY;
4285
4286         read_lock(&bond->lock);
4287
4288         if (bond->slave_cnt)
4289                 ret = __bond_start_xmit(skb, dev);
4290         else
4291                 kfree_skb(skb);
4292
4293         read_unlock(&bond->lock);
4294
4295         return ret;
4296 }
4297
4298 /*
4299  * set bond mode specific net device operations
4300  */
4301 void bond_set_mode_ops(struct bonding *bond, int mode)
4302 {
4303         struct net_device *bond_dev = bond->dev;
4304
4305         switch (mode) {
4306         case BOND_MODE_ROUNDROBIN:
4307                 break;
4308         case BOND_MODE_ACTIVEBACKUP:
4309                 break;
4310         case BOND_MODE_XOR:
4311                 bond_set_xmit_hash_policy(bond);
4312                 break;
4313         case BOND_MODE_BROADCAST:
4314                 break;
4315         case BOND_MODE_8023AD:
4316                 bond_set_xmit_hash_policy(bond);
4317                 break;
4318         case BOND_MODE_ALB:
4319                 /* FALLTHRU */
4320         case BOND_MODE_TLB:
4321                 break;
4322         default:
4323                 /* Should never happen, mode already checked */
4324                 pr_err("%s: Error: Unknown bonding mode %d\n",
4325                        bond_dev->name, mode);
4326                 break;
4327         }
4328 }
4329
4330 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4331                                     struct ethtool_drvinfo *drvinfo)
4332 {
4333         strncpy(drvinfo->driver, DRV_NAME, 32);
4334         strncpy(drvinfo->version, DRV_VERSION, 32);
4335         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4336 }
4337
4338 static const struct ethtool_ops bond_ethtool_ops = {
4339         .get_drvinfo            = bond_ethtool_get_drvinfo,
4340         .get_link               = ethtool_op_get_link,
4341 };
4342
4343 static const struct net_device_ops bond_netdev_ops = {
4344         .ndo_init               = bond_init,
4345         .ndo_uninit             = bond_uninit,
4346         .ndo_open               = bond_open,
4347         .ndo_stop               = bond_close,
4348         .ndo_start_xmit         = bond_start_xmit,
4349         .ndo_select_queue       = bond_select_queue,
4350         .ndo_get_stats64        = bond_get_stats,
4351         .ndo_do_ioctl           = bond_do_ioctl,
4352         .ndo_change_rx_flags    = bond_change_rx_flags,
4353         .ndo_set_rx_mode        = bond_set_multicast_list,
4354         .ndo_change_mtu         = bond_change_mtu,
4355         .ndo_set_mac_address    = bond_set_mac_address,
4356         .ndo_neigh_setup        = bond_neigh_setup,
4357         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4358         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4359 #ifdef CONFIG_NET_POLL_CONTROLLER
4360         .ndo_netpoll_setup      = bond_netpoll_setup,
4361         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4362         .ndo_poll_controller    = bond_poll_controller,
4363 #endif
4364         .ndo_add_slave          = bond_enslave,
4365         .ndo_del_slave          = bond_release,
4366         .ndo_fix_features       = bond_fix_features,
4367 };
4368
4369 static void bond_destructor(struct net_device *bond_dev)
4370 {
4371         struct bonding *bond = netdev_priv(bond_dev);
4372         if (bond->wq)
4373                 destroy_workqueue(bond->wq);
4374         free_netdev(bond_dev);
4375 }
4376
4377 static void bond_setup(struct net_device *bond_dev)
4378 {
4379         struct bonding *bond = netdev_priv(bond_dev);
4380
4381         /* initialize rwlocks */
4382         rwlock_init(&bond->lock);
4383         rwlock_init(&bond->curr_slave_lock);
4384
4385         bond->params = bonding_defaults;
4386
4387         /* Initialize pointers */
4388         bond->dev = bond_dev;
4389         INIT_LIST_HEAD(&bond->vlan_list);
4390
4391         /* Initialize the device entry points */
4392         ether_setup(bond_dev);
4393         bond_dev->netdev_ops = &bond_netdev_ops;
4394         bond_dev->ethtool_ops = &bond_ethtool_ops;
4395         bond_set_mode_ops(bond, bond->params.mode);
4396
4397         bond_dev->destructor = bond_destructor;
4398
4399         /* Initialize the device options */
4400         bond_dev->tx_queue_len = 0;
4401         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4402         bond_dev->priv_flags |= IFF_BONDING;
4403         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4404
4405         /* At first, we block adding VLANs. That's the only way to
4406          * prevent problems that occur when adding VLANs over an
4407          * empty bond. The block will be removed once non-challenged
4408          * slaves are enslaved.
4409          */
4410         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4411
4412         /* don't acquire bond device's netif_tx_lock when
4413          * transmitting */
4414         bond_dev->features |= NETIF_F_LLTX;
4415
4416         /* By default, we declare the bond to be fully
4417          * VLAN hardware accelerated capable. Special
4418          * care is taken in the various xmit functions
4419          * when there are slaves that are not hw accel
4420          * capable
4421          */
4422
4423         bond_dev->hw_features = BOND_VLAN_FEATURES |
4424                                 NETIF_F_HW_VLAN_TX |
4425                                 NETIF_F_HW_VLAN_RX |
4426                                 NETIF_F_HW_VLAN_FILTER;
4427
4428         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4429         bond_dev->features |= bond_dev->hw_features;
4430 }
4431
4432 static void bond_work_cancel_all(struct bonding *bond)
4433 {
4434         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4435                 cancel_delayed_work_sync(&bond->mii_work);
4436
4437         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4438                 cancel_delayed_work_sync(&bond->arp_work);
4439
4440         if (bond->params.mode == BOND_MODE_ALB &&
4441             delayed_work_pending(&bond->alb_work))
4442                 cancel_delayed_work_sync(&bond->alb_work);
4443
4444         if (bond->params.mode == BOND_MODE_8023AD &&
4445             delayed_work_pending(&bond->ad_work))
4446                 cancel_delayed_work_sync(&bond->ad_work);
4447
4448         if (delayed_work_pending(&bond->mcast_work))
4449                 cancel_delayed_work_sync(&bond->mcast_work);
4450 }
4451
4452 /*
4453 * Destroy a bonding device.
4454 * Must be under rtnl_lock when this function is called.
4455 */
4456 static void bond_uninit(struct net_device *bond_dev)
4457 {
4458         struct bonding *bond = netdev_priv(bond_dev);
4459         struct vlan_entry *vlan, *tmp;
4460
4461         bond_netpoll_cleanup(bond_dev);
4462
4463         /* Release the bonded slaves */
4464         bond_release_all(bond_dev);
4465
4466         list_del(&bond->bond_list);
4467
4468         bond_work_cancel_all(bond);
4469
4470         bond_debug_unregister(bond);
4471
4472         __hw_addr_flush(&bond->mc_list);
4473
4474         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4475                 list_del(&vlan->vlan_list);
4476                 kfree(vlan);
4477         }
4478 }
4479
4480 /*------------------------- Module initialization ---------------------------*/
4481
4482 /*
4483  * Convert string input module parms.  Accept either the
4484  * number of the mode or its string name.  A bit complicated because
4485  * some mode names are substrings of other names, and calls from sysfs
4486  * may have whitespace in the name (trailing newlines, for example).
4487  */
4488 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4489 {
4490         int modeint = -1, i, rv;
4491         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4492
4493         for (p = (char *)buf; *p; p++)
4494                 if (!(isdigit(*p) || isspace(*p)))
4495                         break;
4496
4497         if (*p)
4498                 rv = sscanf(buf, "%20s", modestr);
4499         else
4500                 rv = sscanf(buf, "%d", &modeint);
4501
4502         if (!rv)
4503                 return -1;
4504
4505         for (i = 0; tbl[i].modename; i++) {
4506                 if (modeint == tbl[i].mode)
4507                         return tbl[i].mode;
4508                 if (strcmp(modestr, tbl[i].modename) == 0)
4509                         return tbl[i].mode;
4510         }
4511
4512         return -1;
4513 }
4514
4515 static int bond_check_params(struct bond_params *params)
4516 {
4517         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4518
4519         /*
4520          * Convert string parameters.
4521          */
4522         if (mode) {
4523                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4524                 if (bond_mode == -1) {
4525                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4526                                mode == NULL ? "NULL" : mode);
4527                         return -EINVAL;
4528                 }
4529         }
4530
4531         if (xmit_hash_policy) {
4532                 if ((bond_mode != BOND_MODE_XOR) &&
4533                     (bond_mode != BOND_MODE_8023AD)) {
4534                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4535                                bond_mode_name(bond_mode));
4536                 } else {
4537                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4538                                                         xmit_hashtype_tbl);
4539                         if (xmit_hashtype == -1) {
4540                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4541                                        xmit_hash_policy == NULL ? "NULL" :
4542                                        xmit_hash_policy);
4543                                 return -EINVAL;
4544                         }
4545                 }
4546         }
4547
4548         if (lacp_rate) {
4549                 if (bond_mode != BOND_MODE_8023AD) {
4550                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4551                                 bond_mode_name(bond_mode));
4552                 } else {
4553                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4554                         if (lacp_fast == -1) {
4555                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4556                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4557                                 return -EINVAL;
4558                         }
4559                 }
4560         }
4561
4562         if (ad_select) {
4563                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4564                 if (params->ad_select == -1) {
4565                         pr_err("Error: Invalid ad_select \"%s\"\n",
4566                                ad_select == NULL ? "NULL" : ad_select);
4567                         return -EINVAL;
4568                 }
4569
4570                 if (bond_mode != BOND_MODE_8023AD) {
4571                         pr_warning("ad_select param only affects 802.3ad mode\n");
4572                 }
4573         } else {
4574                 params->ad_select = BOND_AD_STABLE;
4575         }
4576
4577         if (max_bonds < 0) {
4578                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4579                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4580                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4581         }
4582
4583         if (miimon < 0) {
4584                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4585                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4586                 miimon = BOND_LINK_MON_INTERV;
4587         }
4588
4589         if (updelay < 0) {
4590                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4591                            updelay, INT_MAX);
4592                 updelay = 0;
4593         }
4594
4595         if (downdelay < 0) {
4596                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4597                            downdelay, INT_MAX);
4598                 downdelay = 0;
4599         }
4600
4601         if ((use_carrier != 0) && (use_carrier != 1)) {
4602                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4603                            use_carrier);
4604                 use_carrier = 1;
4605         }
4606
4607         if (num_peer_notif < 0 || num_peer_notif > 255) {
4608                 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4609                            num_peer_notif);
4610                 num_peer_notif = 1;
4611         }
4612
4613         /* reset values for 802.3ad */
4614         if (bond_mode == BOND_MODE_8023AD) {
4615                 if (!miimon) {
4616                         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");
4617                         pr_warning("Forcing miimon to 100msec\n");
4618                         miimon = 100;
4619                 }
4620         }
4621
4622         if (tx_queues < 1 || tx_queues > 255) {
4623                 pr_warning("Warning: tx_queues (%d) should be between "
4624                            "1 and 255, resetting to %d\n",
4625                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4626                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4627         }
4628
4629         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4630                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4631                            "not of valid value (0/1), so it was set to "
4632                            "0\n", all_slaves_active);
4633                 all_slaves_active = 0;
4634         }
4635
4636         if (resend_igmp < 0 || resend_igmp > 255) {
4637                 pr_warning("Warning: resend_igmp (%d) should be between "
4638                            "0 and 255, resetting to %d\n",
4639                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4640                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4641         }
4642
4643         /* reset values for TLB/ALB */
4644         if ((bond_mode == BOND_MODE_TLB) ||
4645             (bond_mode == BOND_MODE_ALB)) {
4646                 if (!miimon) {
4647                         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");
4648                         pr_warning("Forcing miimon to 100msec\n");
4649                         miimon = 100;
4650                 }
4651         }
4652
4653         if (bond_mode == BOND_MODE_ALB) {
4654                 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",
4655                           updelay);
4656         }
4657
4658         if (!miimon) {
4659                 if (updelay || downdelay) {
4660                         /* just warn the user the up/down delay will have
4661                          * no effect since miimon is zero...
4662                          */
4663                         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",
4664                                    updelay, downdelay);
4665                 }
4666         } else {
4667                 /* don't allow arp monitoring */
4668                 if (arp_interval) {
4669                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4670                                    miimon, arp_interval);
4671                         arp_interval = 0;
4672                 }
4673
4674                 if ((updelay % miimon) != 0) {
4675                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4676                                    updelay, miimon,
4677                                    (updelay / miimon) * miimon);
4678                 }
4679
4680                 updelay /= miimon;
4681
4682                 if ((downdelay % miimon) != 0) {
4683                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4684                                    downdelay, miimon,
4685                                    (downdelay / miimon) * miimon);
4686                 }
4687
4688                 downdelay /= miimon;
4689         }
4690
4691         if (arp_interval < 0) {
4692                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4693                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4694                 arp_interval = BOND_LINK_ARP_INTERV;
4695         }
4696
4697         for (arp_ip_count = 0;
4698              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4699              arp_ip_count++) {
4700                 /* not complete check, but should be good enough to
4701                    catch mistakes */
4702                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4703                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4704                                    arp_ip_target[arp_ip_count]);
4705                         arp_interval = 0;
4706                 } else {
4707                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4708                         arp_target[arp_ip_count] = ip;
4709                 }
4710         }
4711
4712         if (arp_interval && !arp_ip_count) {
4713                 /* don't allow arping if no arp_ip_target given... */
4714                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4715                            arp_interval);
4716                 arp_interval = 0;
4717         }
4718
4719         if (arp_validate) {
4720                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4721                         pr_err("arp_validate only supported in active-backup mode\n");
4722                         return -EINVAL;
4723                 }
4724                 if (!arp_interval) {
4725                         pr_err("arp_validate requires arp_interval\n");
4726                         return -EINVAL;
4727                 }
4728
4729                 arp_validate_value = bond_parse_parm(arp_validate,
4730                                                      arp_validate_tbl);
4731                 if (arp_validate_value == -1) {
4732                         pr_err("Error: invalid arp_validate \"%s\"\n",
4733                                arp_validate == NULL ? "NULL" : arp_validate);
4734                         return -EINVAL;
4735                 }
4736         } else
4737                 arp_validate_value = 0;
4738
4739         if (miimon) {
4740                 pr_info("MII link monitoring set to %d ms\n", miimon);
4741         } else if (arp_interval) {
4742                 int i;
4743
4744                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4745                         arp_interval,
4746                         arp_validate_tbl[arp_validate_value].modename,
4747                         arp_ip_count);
4748
4749                 for (i = 0; i < arp_ip_count; i++)
4750                         pr_info(" %s", arp_ip_target[i]);
4751
4752                 pr_info("\n");
4753
4754         } else if (max_bonds) {
4755                 /* miimon and arp_interval not set, we need one so things
4756                  * work as expected, see bonding.txt for details
4757                  */
4758                 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");
4759         }
4760
4761         if (primary && !USES_PRIMARY(bond_mode)) {
4762                 /* currently, using a primary only makes sense
4763                  * in active backup, TLB or ALB modes
4764                  */
4765                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4766                            primary, bond_mode_name(bond_mode));
4767                 primary = NULL;
4768         }
4769
4770         if (primary && primary_reselect) {
4771                 primary_reselect_value = bond_parse_parm(primary_reselect,
4772                                                          pri_reselect_tbl);
4773                 if (primary_reselect_value == -1) {
4774                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4775                                primary_reselect ==
4776                                         NULL ? "NULL" : primary_reselect);
4777                         return -EINVAL;
4778                 }
4779         } else {
4780                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4781         }
4782
4783         if (fail_over_mac) {
4784                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4785                                                       fail_over_mac_tbl);
4786                 if (fail_over_mac_value == -1) {
4787                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4788                                arp_validate == NULL ? "NULL" : arp_validate);
4789                         return -EINVAL;
4790                 }
4791
4792                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4793                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4794         } else {
4795                 fail_over_mac_value = BOND_FOM_NONE;
4796         }
4797
4798         /* fill params struct with the proper values */
4799         params->mode = bond_mode;
4800         params->xmit_policy = xmit_hashtype;
4801         params->miimon = miimon;
4802         params->num_peer_notif = num_peer_notif;
4803         params->arp_interval = arp_interval;
4804         params->arp_validate = arp_validate_value;
4805         params->updelay = updelay;
4806         params->downdelay = downdelay;
4807         params->use_carrier = use_carrier;
4808         params->lacp_fast = lacp_fast;
4809         params->primary[0] = 0;
4810         params->primary_reselect = primary_reselect_value;
4811         params->fail_over_mac = fail_over_mac_value;
4812         params->tx_queues = tx_queues;
4813         params->all_slaves_active = all_slaves_active;
4814         params->resend_igmp = resend_igmp;
4815         params->min_links = min_links;
4816
4817         if (primary) {
4818                 strncpy(params->primary, primary, IFNAMSIZ);
4819                 params->primary[IFNAMSIZ - 1] = 0;
4820         }
4821
4822         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4823
4824         return 0;
4825 }
4826
4827 static struct lock_class_key bonding_netdev_xmit_lock_key;
4828 static struct lock_class_key bonding_netdev_addr_lock_key;
4829 static struct lock_class_key bonding_tx_busylock_key;
4830
4831 static void bond_set_lockdep_class_one(struct net_device *dev,
4832                                        struct netdev_queue *txq,
4833                                        void *_unused)
4834 {
4835         lockdep_set_class(&txq->_xmit_lock,
4836                           &bonding_netdev_xmit_lock_key);
4837 }
4838
4839 static void bond_set_lockdep_class(struct net_device *dev)
4840 {
4841         lockdep_set_class(&dev->addr_list_lock,
4842                           &bonding_netdev_addr_lock_key);
4843         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4844         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4845 }
4846
4847 /*
4848  * Called from registration process
4849  */
4850 static int bond_init(struct net_device *bond_dev)
4851 {
4852         struct bonding *bond = netdev_priv(bond_dev);
4853         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4854         struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4855
4856         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4857
4858         /*
4859          * Initialize locks that may be required during
4860          * en/deslave operations.  All of the bond_open work
4861          * (of which this is part) should really be moved to
4862          * a phase prior to dev_open
4863          */
4864         spin_lock_init(&(bond_info->tx_hashtbl_lock));
4865         spin_lock_init(&(bond_info->rx_hashtbl_lock));
4866
4867         bond->wq = create_singlethread_workqueue(bond_dev->name);
4868         if (!bond->wq)
4869                 return -ENOMEM;
4870
4871         bond_set_lockdep_class(bond_dev);
4872
4873         list_add_tail(&bond->bond_list, &bn->dev_list);
4874
4875         bond_prepare_sysfs_group(bond);
4876
4877         bond_debug_register(bond);
4878
4879         __hw_addr_init(&bond->mc_list);
4880         return 0;
4881 }
4882
4883 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4884 {
4885         if (tb[IFLA_ADDRESS]) {
4886                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4887                         return -EINVAL;
4888                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4889                         return -EADDRNOTAVAIL;
4890         }
4891         return 0;
4892 }
4893
4894 static unsigned int bond_get_num_tx_queues(void)
4895 {
4896         return tx_queues;
4897 }
4898
4899 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4900         .kind                   = "bond",
4901         .priv_size              = sizeof(struct bonding),
4902         .setup                  = bond_setup,
4903         .validate               = bond_validate,
4904         .get_num_tx_queues      = bond_get_num_tx_queues,
4905         .get_num_rx_queues      = bond_get_num_tx_queues, /* Use the same number
4906                                                              as for TX queues */
4907 };
4908
4909 /* Create a new bond based on the specified name and bonding parameters.
4910  * If name is NULL, obtain a suitable "bond%d" name for us.
4911  * Caller must NOT hold rtnl_lock; we need to release it here before we
4912  * set up our sysfs entries.
4913  */
4914 int bond_create(struct net *net, const char *name)
4915 {
4916         struct net_device *bond_dev;
4917         int res;
4918
4919         rtnl_lock();
4920
4921         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4922                                    name ? name : "bond%d",
4923                                    bond_setup, tx_queues);
4924         if (!bond_dev) {
4925                 pr_err("%s: eek! can't alloc netdev!\n", name);
4926                 rtnl_unlock();
4927                 return -ENOMEM;
4928         }
4929
4930         dev_net_set(bond_dev, net);
4931         bond_dev->rtnl_link_ops = &bond_link_ops;
4932
4933         res = register_netdevice(bond_dev);
4934
4935         netif_carrier_off(bond_dev);
4936
4937         rtnl_unlock();
4938         if (res < 0)
4939                 bond_destructor(bond_dev);
4940         return res;
4941 }
4942
4943 static int __net_init bond_net_init(struct net *net)
4944 {
4945         struct bond_net *bn = net_generic(net, bond_net_id);
4946
4947         bn->net = net;
4948         INIT_LIST_HEAD(&bn->dev_list);
4949
4950         bond_create_proc_dir(bn);
4951         bond_create_sysfs(bn);
4952
4953         return 0;
4954 }
4955
4956 static void __net_exit bond_net_exit(struct net *net)
4957 {
4958         struct bond_net *bn = net_generic(net, bond_net_id);
4959
4960         bond_destroy_sysfs(bn);
4961         bond_destroy_proc_dir(bn);
4962 }
4963
4964 static struct pernet_operations bond_net_ops = {
4965         .init = bond_net_init,
4966         .exit = bond_net_exit,
4967         .id   = &bond_net_id,
4968         .size = sizeof(struct bond_net),
4969 };
4970
4971 static int __init bonding_init(void)
4972 {
4973         int i;
4974         int res;
4975
4976         pr_info("%s", bond_version);
4977
4978         res = bond_check_params(&bonding_defaults);
4979         if (res)
4980                 goto out;
4981
4982         res = register_pernet_subsys(&bond_net_ops);
4983         if (res)
4984                 goto out;
4985
4986         res = rtnl_link_register(&bond_link_ops);
4987         if (res)
4988                 goto err_link;
4989
4990         bond_create_debugfs();
4991
4992         for (i = 0; i < max_bonds; i++) {
4993                 res = bond_create(&init_net, NULL);
4994                 if (res)
4995                         goto err;
4996         }
4997
4998         register_netdevice_notifier(&bond_netdev_notifier);
4999 out:
5000         return res;
5001 err:
5002         rtnl_link_unregister(&bond_link_ops);
5003 err_link:
5004         unregister_pernet_subsys(&bond_net_ops);
5005         goto out;
5006
5007 }
5008
5009 static void __exit bonding_exit(void)
5010 {
5011         unregister_netdevice_notifier(&bond_netdev_notifier);
5012
5013         bond_destroy_debugfs();
5014
5015         rtnl_link_unregister(&bond_link_ops);
5016         unregister_pernet_subsys(&bond_net_ops);
5017
5018 #ifdef CONFIG_NET_POLL_CONTROLLER
5019         /*
5020          * Make sure we don't have an imbalance on our netpoll blocking
5021          */
5022         WARN_ON(atomic_read(&netpoll_block_tx));
5023 #endif
5024 }
5025
5026 module_init(bonding_init);
5027 module_exit(bonding_exit);
5028 MODULE_LICENSE("GPL");
5029 MODULE_VERSION(DRV_VERSION);
5030 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5031 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5032 MODULE_ALIAS_RTNL_LINK("bond");