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 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/dma.h>
  56 #include <linux/uaccess.h>
  57 #include <linux/errno.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/inetdevice.h>
  60 #include <linux/igmp.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/skbuff.h>
  63 #include <net/sock.h>
  64 #include <linux/rtnetlink.h>
  65 #include <linux/smp.h>
  66 #include <linux/if_ether.h>
  67 #include <net/arp.h>
  68 #include <linux/mii.h>
  69 #include <linux/ethtool.h>
  70 #include <linux/if_vlan.h>
  71 #include <linux/if_bonding.h>
  72 #include <linux/jiffies.h>
  73 #include <linux/preempt.h>
  74 #include <net/route.h>
  75 #include <net/net_namespace.h>
  76 #include <net/netns/generic.h>
  77 #include <net/pkt_sched.h>
  78 #include <linux/rculist.h>
  79 #include <net/flow_dissector.h>
  80 #include <net/switchdev.h>
  81 #include <net/bonding.h>
  82 #include <net/bond_3ad.h>
  83 #include <net/bond_alb.h>
  84
  85 #include "bonding_priv.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113 static int packets_per_slave = 1;
 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 115
 116 module_param(max_bonds, int, 0);
 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 118 module_param(tx_queues, int, 0);
 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 122                                "failover event (alias of num_unsol_na)");
 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 125                                "failover event (alias of num_grat_arp)");
 126 module_param(miimon, int, 0);
 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 128 module_param(updelay, int, 0);
 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 130 module_param(downdelay, int, 0);
 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 132                             "in milliseconds");
 133 module_param(use_carrier, int, 0);
 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 135                               "0 for off, 1 for on (default)");
 136 module_param(mode, charp, 0);
 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 138                        "1 for active-backup, 2 for balance-xor, "
 139                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 140                        "6 for balance-alb");
 141 module_param(primary, charp, 0);
 142 MODULE_PARM_DESC(primary, "Primary network device to use");
 143 module_param(primary_reselect, charp, 0);
 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 145                                    "once it comes up; "
 146                                    "0 for always (default), "
 147                                    "1 for only if speed of primary is "
 148                                    "better, "
 149                                    "2 for only on active slave "
 150                                    "failure");
 151 module_param(lacp_rate, charp, 0);
 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 153                             "0 for slow, 1 for fast");
 154 module_param(ad_select, charp, 0);
 155 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 int bond_net_id __read_mostly;
 203
 204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 205 static int arp_ip_count;
 206 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 208 static int lacp_fast;
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static int bond_init(struct net_device *bond_dev);
 213 static void bond_uninit(struct net_device *bond_dev);
 214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 215                                                 struct rtnl_link_stats64 *stats);
 216 static void bond_slave_arr_handler(struct work_struct *work);
 217
 218 /*---------------------------- General routines -----------------------------*/
 219
 220 const char *bond_mode_name(int mode)
 221 {
 222         static const char *names[] = {
 223                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 224                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 225                 [BOND_MODE_XOR] = "load balancing (xor)",
 226                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 227                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 228                 [BOND_MODE_TLB] = "transmit load balancing",
 229                 [BOND_MODE_ALB] = "adaptive load balancing",
 230         };
 231
 232         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 233                 return "unknown";
 234
 235         return names[mode];
 236 }
 237
 238 /*---------------------------------- VLAN -----------------------------------*/
 239
 240 /**
 241  * bond_dev_queue_xmit - Prepare skb for xmit.
 242  *
 243  * @bond: bond device that got this skb for tx.
 244  * @skb: hw accel VLAN tagged skb to transmit
 245  * @slave_dev: slave that is supposed to xmit this skbuff
 246  */
 247 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 248                         struct net_device *slave_dev)
 249 {
 250         skb->dev = slave_dev;
 251
 252         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 253                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 254         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 255
 256         if (unlikely(netpoll_tx_running(bond->dev)))
 257                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 258         else
 259                 dev_queue_xmit(skb);
 260 }
 261
 262 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 263  * We don't protect the slave list iteration with a lock because:
 264  * a. This operation is performed in IOCTL context,
 265  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 266  * c. Holding a lock with BH disabled while directly calling a base driver
 267  *    entry point is generally a BAD idea.
 268  *
 269  * The design of synchronization/protection for this operation in the 8021q
 270  * module is good for one or more VLAN devices over a single physical device
 271  * and cannot be extended for a teaming solution like bonding, so there is a
 272  * potential race condition here where a net device from the vlan group might
 273  * be referenced (either by a base driver or the 8021q code) while it is being
 274  * removed from the system. However, it turns out we're not making matters
 275  * worse, and if it works for regular VLAN usage it will work here too.
 276 */
 277
 278 /**
 279  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 280  * @bond_dev: bonding net device that got called
 281  * @vid: vlan id being added
 282  */
 283 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 284                                 __be16 proto, u16 vid)
 285 {
 286         struct bonding *bond = netdev_priv(bond_dev);
 287         struct slave *slave, *rollback_slave;
 288         struct list_head *iter;
 289         int res;
 290
 291         bond_for_each_slave(bond, slave, iter) {
 292                 res = vlan_vid_add(slave->dev, proto, vid);
 293                 if (res)
 294                         goto unwind;
 295         }
 296
 297         return 0;
 298
 299 unwind:
 300         /* unwind to the slave that failed */
 301         bond_for_each_slave(bond, rollback_slave, iter) {
 302                 if (rollback_slave == slave)
 303                         break;
 304
 305                 vlan_vid_del(rollback_slave->dev, proto, vid);
 306         }
 307
 308         return res;
 309 }
 310
 311 /**
 312  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 313  * @bond_dev: bonding net device that got called
 314  * @vid: vlan id being removed
 315  */
 316 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 317                                  __be16 proto, u16 vid)
 318 {
 319         struct bonding *bond = netdev_priv(bond_dev);
 320         struct list_head *iter;
 321         struct slave *slave;
 322
 323         bond_for_each_slave(bond, slave, iter)
 324                 vlan_vid_del(slave->dev, proto, vid);
 325
 326         if (bond_is_lb(bond))
 327                 bond_alb_clear_vlan(bond, vid);
 328
 329         return 0;
 330 }
 331
 332 /*------------------------------- Link status -------------------------------*/
 333
 334 /* Set the carrier state for the master according to the state of its
 335  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 336  * do special 802.3ad magic.
 337  *
 338  * Returns zero if carrier state does not change, nonzero if it does.
 339  */
 340 int bond_set_carrier(struct bonding *bond)
 341 {
 342         struct list_head *iter;
 343         struct slave *slave;
 344
 345         if (!bond_has_slaves(bond))
 346                 goto down;
 347
 348         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 349                 return bond_3ad_set_carrier(bond);
 350
 351         bond_for_each_slave(bond, slave, iter) {
 352                 if (slave->link == BOND_LINK_UP) {
 353                         if (!netif_carrier_ok(bond->dev)) {
 354                                 netif_carrier_on(bond->dev);
 355                                 return 1;
 356                         }
 357                         return 0;
 358                 }
 359         }
 360
 361 down:
 362         if (netif_carrier_ok(bond->dev)) {
 363                 netif_carrier_off(bond->dev);
 364                 return 1;
 365         }
 366         return 0;
 367 }
 368
 369 /* Get link speed and duplex from the slave's base driver
 370  * using ethtool. If for some reason the call fails or the
 371  * values are invalid, set speed and duplex to -1,
 372  * and return.
 373  */
 374 static void bond_update_speed_duplex(struct slave *slave)
 375 {
 376         struct net_device *slave_dev = slave->dev;
 377         struct ethtool_cmd ecmd;
 378         u32 slave_speed;
 379         int res;
 380
 381         slave->speed = SPEED_UNKNOWN;
 382         slave->duplex = DUPLEX_UNKNOWN;
 383
 384         res = __ethtool_get_settings(slave_dev, &ecmd);
 385         if (res < 0)
 386                 return;
 387
 388         slave_speed = ethtool_cmd_speed(&ecmd);
 389         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 390                 return;
 391
 392         switch (ecmd.duplex) {
 393         case DUPLEX_FULL:
 394         case DUPLEX_HALF:
 395                 break;
 396         default:
 397                 return;
 398         }
 399
 400         slave->speed = slave_speed;
 401         slave->duplex = ecmd.duplex;
 402
 403         return;
 404 }
 405
 406 const char *bond_slave_link_status(s8 link)
 407 {
 408         switch (link) {
 409         case BOND_LINK_UP:
 410                 return "up";
 411         case BOND_LINK_FAIL:
 412                 return "going down";
 413         case BOND_LINK_DOWN:
 414                 return "down";
 415         case BOND_LINK_BACK:
 416                 return "going back";
 417         default:
 418                 return "unknown";
 419         }
 420 }
 421
 422 /* if <dev> supports MII link status reporting, check its link status.
 423  *
 424  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 425  * depending upon the setting of the use_carrier parameter.
 426  *
 427  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 428  * can't tell and just pretend it is), or 0, meaning that the link is
 429  * down.
 430  *
 431  * If reporting is non-zero, instead of faking link up, return -1 if
 432  * both ETHTOOL and MII ioctls fail (meaning the device does not
 433  * support them).  If use_carrier is set, return whatever it says.
 434  * It'd be nice if there was a good way to tell if a driver supports
 435  * netif_carrier, but there really isn't.
 436  */
 437 static int bond_check_dev_link(struct bonding *bond,
 438                                struct net_device *slave_dev, int reporting)
 439 {
 440         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 441         int (*ioctl)(struct net_device *, struct ifreq *, int);
 442         struct ifreq ifr;
 443         struct mii_ioctl_data *mii;
 444
 445         if (!reporting && !netif_running(slave_dev))
 446                 return 0;
 447
 448         if (bond->params.use_carrier)
 449                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 450
 451         /* Try to get link status using Ethtool first. */
 452         if (slave_dev->ethtool_ops->get_link)
 453                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 454                         BMSR_LSTATUS : 0;
 455
 456         /* Ethtool can't be used, fallback to MII ioctls. */
 457         ioctl = slave_ops->ndo_do_ioctl;
 458         if (ioctl) {
 459                 /* TODO: set pointer to correct ioctl on a per team member
 460                  *       bases to make this more efficient. that is, once
 461                  *       we determine the correct ioctl, we will always
 462                  *       call it and not the others for that team
 463                  *       member.
 464                  */
 465
 466                 /* We cannot assume that SIOCGMIIPHY will also read a
 467                  * register; not all network drivers (e.g., e100)
 468                  * support that.
 469                  */
 470
 471                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 472                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 473                 mii = if_mii(&ifr);
 474                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 475                         mii->reg_num = MII_BMSR;
 476                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 477                                 return mii->val_out & BMSR_LSTATUS;
 478                 }
 479         }
 480
 481         /* If reporting, report that either there's no dev->do_ioctl,
 482          * or both SIOCGMIIREG and get_link failed (meaning that we
 483          * cannot report link status).  If not reporting, pretend
 484          * we're ok.
 485          */
 486         return reporting ? -1 : BMSR_LSTATUS;
 487 }
 488
 489 /*----------------------------- Multicast list ------------------------------*/
 490
 491 /* Push the promiscuity flag down to appropriate slaves */
 492 static int bond_set_promiscuity(struct bonding *bond, int inc)
 493 {
 494         struct list_head *iter;
 495         int err = 0;
 496
 497         if (bond_uses_primary(bond)) {
 498                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 499
 500                 if (curr_active)
 501                         err = dev_set_promiscuity(curr_active->dev, inc);
 502         } else {
 503                 struct slave *slave;
 504
 505                 bond_for_each_slave(bond, slave, iter) {
 506                         err = dev_set_promiscuity(slave->dev, inc);
 507                         if (err)
 508                                 return err;
 509                 }
 510         }
 511         return err;
 512 }
 513
 514 /* Push the allmulti flag down to all slaves */
 515 static int bond_set_allmulti(struct bonding *bond, int inc)
 516 {
 517         struct list_head *iter;
 518         int err = 0;
 519
 520         if (bond_uses_primary(bond)) {
 521                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 522
 523                 if (curr_active)
 524                         err = dev_set_allmulti(curr_active->dev, inc);
 525         } else {
 526                 struct slave *slave;
 527
 528                 bond_for_each_slave(bond, slave, iter) {
 529                         err = dev_set_allmulti(slave->dev, inc);
 530                         if (err)
 531                                 return err;
 532                 }
 533         }
 534         return err;
 535 }
 536
 537 /* Retrieve the list of registered multicast addresses for the bonding
 538  * device and retransmit an IGMP JOIN request to the current active
 539  * slave.
 540  */
 541 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 542 {
 543         struct bonding *bond = container_of(work, struct bonding,
 544                                             mcast_work.work);
 545
 546         if (!rtnl_trylock()) {
 547                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 548                 return;
 549         }
 550         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 551
 552         if (bond->igmp_retrans > 1) {
 553                 bond->igmp_retrans--;
 554                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 555         }
 556         rtnl_unlock();
 557 }
 558
 559 /* Flush bond's hardware addresses from slave */
 560 static void bond_hw_addr_flush(struct net_device *bond_dev,
 561                                struct net_device *slave_dev)
 562 {
 563         struct bonding *bond = netdev_priv(bond_dev);
 564
 565         dev_uc_unsync(slave_dev, bond_dev);
 566         dev_mc_unsync(slave_dev, bond_dev);
 567
 568         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 569                 /* del lacpdu mc addr from mc list */
 570                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 571
 572                 dev_mc_del(slave_dev, lacpdu_multicast);
 573         }
 574 }
 575
 576 /*--------------------------- Active slave change ---------------------------*/
 577
 578 /* Update the hardware address list and promisc/allmulti for the new and
 579  * old active slaves (if any).  Modes that are not using primary keep all
 580  * slaves up date at all times; only the modes that use primary need to call
 581  * this function to swap these settings during a failover.
 582  */
 583 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 584                               struct slave *old_active)
 585 {
 586         if (old_active) {
 587                 if (bond->dev->flags & IFF_PROMISC)
 588                         dev_set_promiscuity(old_active->dev, -1);
 589
 590                 if (bond->dev->flags & IFF_ALLMULTI)
 591                         dev_set_allmulti(old_active->dev, -1);
 592
 593                 bond_hw_addr_flush(bond->dev, old_active->dev);
 594         }
 595
 596         if (new_active) {
 597                 /* FIXME: Signal errors upstream. */
 598                 if (bond->dev->flags & IFF_PROMISC)
 599                         dev_set_promiscuity(new_active->dev, 1);
 600
 601                 if (bond->dev->flags & IFF_ALLMULTI)
 602                         dev_set_allmulti(new_active->dev, 1);
 603
 604                 netif_addr_lock_bh(bond->dev);
 605                 dev_uc_sync(new_active->dev, bond->dev);
 606                 dev_mc_sync(new_active->dev, bond->dev);
 607                 netif_addr_unlock_bh(bond->dev);
 608         }
 609 }
 610
 611 /**
 612  * bond_set_dev_addr - clone slave's address to bond
 613  * @bond_dev: bond net device
 614  * @slave_dev: slave net device
 615  *
 616  * Should be called with RTNL held.
 617  */
 618 static void bond_set_dev_addr(struct net_device *bond_dev,
 619                               struct net_device *slave_dev)
 620 {
 621         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
 622                    bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
 623         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 624         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 625         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 626 }
 627
 628 static struct slave *bond_get_old_active(struct bonding *bond,
 629                                          struct slave *new_active)
 630 {
 631         struct slave *slave;
 632         struct list_head *iter;
 633
 634         bond_for_each_slave(bond, slave, iter) {
 635                 if (slave == new_active)
 636                         continue;
 637
 638                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 639                         return slave;
 640         }
 641
 642         return NULL;
 643 }
 644
 645 /* bond_do_fail_over_mac
 646  *
 647  * Perform special MAC address swapping for fail_over_mac settings
 648  *
 649  * Called with RTNL
 650  */
 651 static void bond_do_fail_over_mac(struct bonding *bond,
 652                                   struct slave *new_active,
 653                                   struct slave *old_active)
 654 {
 655         u8 tmp_mac[ETH_ALEN];
 656         struct sockaddr saddr;
 657         int rv;
 658
 659         switch (bond->params.fail_over_mac) {
 660         case BOND_FOM_ACTIVE:
 661                 if (new_active)
 662                         bond_set_dev_addr(bond->dev, new_active->dev);
 663                 break;
 664         case BOND_FOM_FOLLOW:
 665                 /* if new_active && old_active, swap them
 666                  * if just old_active, do nothing (going to no active slave)
 667                  * if just new_active, set new_active to bond's MAC
 668                  */
 669                 if (!new_active)
 670                         return;
 671
 672                 if (!old_active)
 673                         old_active = bond_get_old_active(bond, new_active);
 674
 675                 if (old_active) {
 676                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 677                         ether_addr_copy(saddr.sa_data,
 678                                         old_active->dev->dev_addr);
 679                         saddr.sa_family = new_active->dev->type;
 680                 } else {
 681                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 682                         saddr.sa_family = bond->dev->type;
 683                 }
 684
 685                 rv = dev_set_mac_address(new_active->dev, &saddr);
 686                 if (rv) {
 687                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 688                                    -rv, new_active->dev->name);
 689                         goto out;
 690                 }
 691
 692                 if (!old_active)
 693                         goto out;
 694
 695                 ether_addr_copy(saddr.sa_data, tmp_mac);
 696                 saddr.sa_family = old_active->dev->type;
 697
 698                 rv = dev_set_mac_address(old_active->dev, &saddr);
 699                 if (rv)
 700                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 701                                    -rv, new_active->dev->name);
 702 out:
 703                 break;
 704         default:
 705                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 706                            bond->params.fail_over_mac);
 707                 break;
 708         }
 709
 710 }
 711
 712 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 713 {
 714         struct slave *prim = rtnl_dereference(bond->primary_slave);
 715         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 716
 717         if (!prim || prim->link != BOND_LINK_UP) {
 718                 if (!curr || curr->link != BOND_LINK_UP)
 719                         return NULL;
 720                 return curr;
 721         }
 722
 723         if (bond->force_primary) {
 724                 bond->force_primary = false;
 725                 return prim;
 726         }
 727
 728         if (!curr || curr->link != BOND_LINK_UP)
 729                 return prim;
 730
 731         /* At this point, prim and curr are both up */
 732         switch (bond->params.primary_reselect) {
 733         case BOND_PRI_RESELECT_ALWAYS:
 734                 return prim;
 735         case BOND_PRI_RESELECT_BETTER:
 736                 if (prim->speed < curr->speed)
 737                         return curr;
 738                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 739                         return curr;
 740                 return prim;
 741         case BOND_PRI_RESELECT_FAILURE:
 742                 return curr;
 743         default:
 744                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 745                            bond->params.primary_reselect);
 746                 return curr;
 747         }
 748 }
 749
 750 /**
 751  * bond_find_best_slave - select the best available slave to be the active one
 752  * @bond: our bonding struct
 753  */
 754 static struct slave *bond_find_best_slave(struct bonding *bond)
 755 {
 756         struct slave *slave, *bestslave = NULL;
 757         struct list_head *iter;
 758         int mintime = bond->params.updelay;
 759
 760         slave = bond_choose_primary_or_current(bond);
 761         if (slave)
 762                 return slave;
 763
 764         bond_for_each_slave(bond, slave, iter) {
 765                 if (slave->link == BOND_LINK_UP)
 766                         return slave;
 767                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 768                     slave->delay < mintime) {
 769                         mintime = slave->delay;
 770                         bestslave = slave;
 771                 }
 772         }
 773
 774         return bestslave;
 775 }
 776
 777 static bool bond_should_notify_peers(struct bonding *bond)
 778 {
 779         struct slave *slave;
 780
 781         rcu_read_lock();
 782         slave = rcu_dereference(bond->curr_active_slave);
 783         rcu_read_unlock();
 784
 785         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 786                    slave ? slave->dev->name : "NULL");
 787
 788         if (!slave || !bond->send_peer_notif ||
 789             !netif_carrier_ok(bond->dev) ||
 790             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 791                 return false;
 792
 793         return true;
 794 }
 795
 796 /**
 797  * change_active_interface - change the active slave into the specified one
 798  * @bond: our bonding struct
 799  * @new: the new slave to make the active one
 800  *
 801  * Set the new slave to the bond's settings and unset them on the old
 802  * curr_active_slave.
 803  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 804  *
 805  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 806  * because it is apparently the best available slave we have, even though its
 807  * updelay hasn't timed out yet.
 808  *
 809  * Caller must hold RTNL.
 810  */
 811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 812 {
 813         struct slave *old_active;
 814
 815         ASSERT_RTNL();
 816
 817         old_active = rtnl_dereference(bond->curr_active_slave);
 818
 819         if (old_active == new_active)
 820                 return;
 821
 822         if (new_active) {
 823                 new_active->last_link_up = jiffies;
 824
 825                 if (new_active->link == BOND_LINK_BACK) {
 826                         if (bond_uses_primary(bond)) {
 827                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 828                                             new_active->dev->name,
 829                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 830                         }
 831
 832                         new_active->delay = 0;
 833                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 834                                                   BOND_SLAVE_NOTIFY_NOW);
 835
 836                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 837                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 838
 839                         if (bond_is_lb(bond))
 840                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 841                 } else {
 842                         if (bond_uses_primary(bond)) {
 843                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 844                                             new_active->dev->name);
 845                         }
 846                 }
 847         }
 848
 849         if (bond_uses_primary(bond))
 850                 bond_hw_addr_swap(bond, new_active, old_active);
 851
 852         if (bond_is_lb(bond)) {
 853                 bond_alb_handle_active_change(bond, new_active);
 854                 if (old_active)
 855                         bond_set_slave_inactive_flags(old_active,
 856                                                       BOND_SLAVE_NOTIFY_NOW);
 857                 if (new_active)
 858                         bond_set_slave_active_flags(new_active,
 859                                                     BOND_SLAVE_NOTIFY_NOW);
 860         } else {
 861                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 862         }
 863
 864         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 865                 if (old_active)
 866                         bond_set_slave_inactive_flags(old_active,
 867                                                       BOND_SLAVE_NOTIFY_NOW);
 868
 869                 if (new_active) {
 870                         bool should_notify_peers = false;
 871
 872                         bond_set_slave_active_flags(new_active,
 873                                                     BOND_SLAVE_NOTIFY_NOW);
 874
 875                         if (bond->params.fail_over_mac)
 876                                 bond_do_fail_over_mac(bond, new_active,
 877                                                       old_active);
 878
 879                         if (netif_running(bond->dev)) {
 880                                 bond->send_peer_notif =
 881                                         bond->params.num_peer_notif;
 882                                 should_notify_peers =
 883                                         bond_should_notify_peers(bond);
 884                         }
 885
 886                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 887                         if (should_notify_peers)
 888                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 889                                                          bond->dev);
 890                 }
 891         }
 892
 893         /* resend IGMP joins since active slave has changed or
 894          * all were sent on curr_active_slave.
 895          * resend only if bond is brought up with the affected
 896          * bonding modes and the retransmission is enabled
 897          */
 898         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 899             ((bond_uses_primary(bond) && new_active) ||
 900              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 901                 bond->igmp_retrans = bond->params.resend_igmp;
 902                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 903         }
 904 }
 905
 906 /**
 907  * bond_select_active_slave - select a new active slave, if needed
 908  * @bond: our bonding struct
 909  *
 910  * This functions should be called when one of the following occurs:
 911  * - The old curr_active_slave has been released or lost its link.
 912  * - The primary_slave has got its link back.
 913  * - A slave has got its link back and there's no old curr_active_slave.
 914  *
 915  * Caller must hold RTNL.
 916  */
 917 void bond_select_active_slave(struct bonding *bond)
 918 {
 919         struct slave *best_slave;
 920         int rv;
 921
 922         ASSERT_RTNL();
 923
 924         best_slave = bond_find_best_slave(bond);
 925         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 926                 bond_change_active_slave(bond, best_slave);
 927                 rv = bond_set_carrier(bond);
 928                 if (!rv)
 929                         return;
 930
 931                 if (netif_carrier_ok(bond->dev))
 932                         netdev_info(bond->dev, "first active interface up!\n");
 933                 else
 934                         netdev_info(bond->dev, "now running without any active interface!\n");
 935         }
 936 }
 937
 938 #ifdef CONFIG_NET_POLL_CONTROLLER
 939 static inline int slave_enable_netpoll(struct slave *slave)
 940 {
 941         struct netpoll *np;
 942         int err = 0;
 943
 944         np = kzalloc(sizeof(*np), GFP_KERNEL);
 945         err = -ENOMEM;
 946         if (!np)
 947                 goto out;
 948
 949         err = __netpoll_setup(np, slave->dev);
 950         if (err) {
 951                 kfree(np);
 952                 goto out;
 953         }
 954         slave->np = np;
 955 out:
 956         return err;
 957 }
 958 static inline void slave_disable_netpoll(struct slave *slave)
 959 {
 960         struct netpoll *np = slave->np;
 961
 962         if (!np)
 963                 return;
 964
 965         slave->np = NULL;
 966         __netpoll_free_async(np);
 967 }
 968
 969 static void bond_poll_controller(struct net_device *bond_dev)
 970 {
 971         struct bonding *bond = netdev_priv(bond_dev);
 972         struct slave *slave = NULL;
 973         struct list_head *iter;
 974         struct ad_info ad_info;
 975         struct netpoll_info *ni;
 976         const struct net_device_ops *ops;
 977
 978         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 979                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 980                         return;
 981
 982         bond_for_each_slave_rcu(bond, slave, iter) {
 983                 ops = slave->dev->netdev_ops;
 984                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 985                         continue;
 986
 987                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 988                         struct aggregator *agg =
 989                             SLAVE_AD_INFO(slave)->port.aggregator;
 990
 991                         if (agg &&
 992                             agg->aggregator_identifier != ad_info.aggregator_id)
 993                                 continue;
 994                 }
 995
 996                 ni = rcu_dereference_bh(slave->dev->npinfo);
 997                 if (down_trylock(&ni->dev_lock))
 998                         continue;
 999                 ops->ndo_poll_controller(slave->dev);
1000                 up(&ni->dev_lock);
1001         }
1002 }
1003
1004 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1005 {
1006         struct bonding *bond = netdev_priv(bond_dev);
1007         struct list_head *iter;
1008         struct slave *slave;
1009
1010         bond_for_each_slave(bond, slave, iter)
1011                 if (bond_slave_is_up(slave))
1012                         slave_disable_netpoll(slave);
1013 }
1014
1015 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1016 {
1017         struct bonding *bond = netdev_priv(dev);
1018         struct list_head *iter;
1019         struct slave *slave;
1020         int err = 0;
1021
1022         bond_for_each_slave(bond, slave, iter) {
1023                 err = slave_enable_netpoll(slave);
1024                 if (err) {
1025                         bond_netpoll_cleanup(dev);
1026                         break;
1027                 }
1028         }
1029         return err;
1030 }
1031 #else
1032 static inline int slave_enable_netpoll(struct slave *slave)
1033 {
1034         return 0;
1035 }
1036 static inline void slave_disable_netpoll(struct slave *slave)
1037 {
1038 }
1039 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1040 {
1041 }
1042 #endif
1043
1044 /*---------------------------------- IOCTL ----------------------------------*/
1045
1046 static netdev_features_t bond_fix_features(struct net_device *dev,
1047                                            netdev_features_t features)
1048 {
1049         struct bonding *bond = netdev_priv(dev);
1050         struct list_head *iter;
1051         netdev_features_t mask;
1052         struct slave *slave;
1053
1054         mask = features;
1055
1056         features &= ~NETIF_F_ONE_FOR_ALL;
1057         features |= NETIF_F_ALL_FOR_ALL;
1058
1059         bond_for_each_slave(bond, slave, iter) {
1060                 features = netdev_increment_features(features,
1061                                                      slave->dev->features,
1062                                                      mask);
1063         }
1064         features = netdev_add_tso_features(features, mask);
1065
1066         return features;
1067 }
1068
1069 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1070                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1071                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1072
1073 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1074                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1075
1076 static void bond_compute_features(struct bonding *bond)
1077 {
1078         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1079                                         IFF_XMIT_DST_RELEASE_PERM;
1080         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1081         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1082         struct net_device *bond_dev = bond->dev;
1083         struct list_head *iter;
1084         struct slave *slave;
1085         unsigned short max_hard_header_len = ETH_HLEN;
1086         unsigned int gso_max_size = GSO_MAX_SIZE;
1087         u16 gso_max_segs = GSO_MAX_SEGS;
1088
1089         if (!bond_has_slaves(bond))
1090                 goto done;
1091         vlan_features &= NETIF_F_ALL_FOR_ALL;
1092
1093         bond_for_each_slave(bond, slave, iter) {
1094                 vlan_features = netdev_increment_features(vlan_features,
1095                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1096
1097                 enc_features = netdev_increment_features(enc_features,
1098                                                          slave->dev->hw_enc_features,
1099                                                          BOND_ENC_FEATURES);
1100                 dst_release_flag &= slave->dev->priv_flags;
1101                 if (slave->dev->hard_header_len > max_hard_header_len)
1102                         max_hard_header_len = slave->dev->hard_header_len;
1103
1104                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1105                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1106         }
1107
1108 done:
1109         bond_dev->vlan_features = vlan_features;
1110         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1111         bond_dev->hard_header_len = max_hard_header_len;
1112         bond_dev->gso_max_segs = gso_max_segs;
1113         netif_set_gso_max_size(bond_dev, gso_max_size);
1114
1115         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1116         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1117             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1118                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1119
1120         netdev_change_features(bond_dev);
1121 }
1122
1123 static void bond_setup_by_slave(struct net_device *bond_dev,
1124                                 struct net_device *slave_dev)
1125 {
1126         bond_dev->header_ops        = slave_dev->header_ops;
1127
1128         bond_dev->type              = slave_dev->type;
1129         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1130         bond_dev->addr_len          = slave_dev->addr_len;
1131
1132         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1133                 slave_dev->addr_len);
1134 }
1135
1136 /* On bonding slaves other than the currently active slave, suppress
1137  * duplicates except for alb non-mcast/bcast.
1138  */
1139 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1140                                             struct slave *slave,
1141                                             struct bonding *bond)
1142 {
1143         if (bond_is_slave_inactive(slave)) {
1144                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1145                     skb->pkt_type != PACKET_BROADCAST &&
1146                     skb->pkt_type != PACKET_MULTICAST)
1147                         return false;
1148                 return true;
1149         }
1150         return false;
1151 }
1152
1153 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1154 {
1155         struct sk_buff *skb = *pskb;
1156         struct slave *slave;
1157         struct bonding *bond;
1158         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1159                           struct slave *);
1160         int ret = RX_HANDLER_ANOTHER;
1161
1162         skb = skb_share_check(skb, GFP_ATOMIC);
1163         if (unlikely(!skb))
1164                 return RX_HANDLER_CONSUMED;
1165
1166         *pskb = skb;
1167
1168         slave = bond_slave_get_rcu(skb->dev);
1169         bond = slave->bond;
1170
1171         recv_probe = ACCESS_ONCE(bond->recv_probe);
1172         if (recv_probe) {
1173                 ret = recv_probe(skb, bond, slave);
1174                 if (ret == RX_HANDLER_CONSUMED) {
1175                         consume_skb(skb);
1176                         return ret;
1177                 }
1178         }
1179
1180         if (bond_should_deliver_exact_match(skb, slave, bond))
1181                 return RX_HANDLER_EXACT;
1182
1183         skb->dev = bond->dev;
1184
1185         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1186             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1187             skb->pkt_type == PACKET_HOST) {
1188
1189                 if (unlikely(skb_cow_head(skb,
1190                                           skb->data - skb_mac_header(skb)))) {
1191                         kfree_skb(skb);
1192                         return RX_HANDLER_CONSUMED;
1193                 }
1194                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1195         }
1196
1197         return ret;
1198 }
1199
1200 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1201 {
1202         switch (BOND_MODE(bond)) {
1203         case BOND_MODE_ROUNDROBIN:
1204                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1205         case BOND_MODE_ACTIVEBACKUP:
1206                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1207         case BOND_MODE_BROADCAST:
1208                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1209         case BOND_MODE_XOR:
1210         case BOND_MODE_8023AD:
1211                 return NETDEV_LAG_TX_TYPE_HASH;
1212         default:
1213                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1214         }
1215 }
1216
1217 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave)
1218 {
1219         struct netdev_lag_upper_info lag_upper_info;
1220         int err;
1221
1222         lag_upper_info.tx_type = bond_lag_tx_type(bond);
1223         err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1224                                            &lag_upper_info);
1225         if (err)
1226                 return err;
1227         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1228         return 0;
1229 }
1230
1231 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1232 {
1233         netdev_upper_dev_unlink(slave->dev, bond->dev);
1234         slave->dev->flags &= ~IFF_SLAVE;
1235         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1236 }
1237
1238 static struct slave *bond_alloc_slave(struct bonding *bond)
1239 {
1240         struct slave *slave = NULL;
1241
1242         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1243         if (!slave)
1244                 return NULL;
1245
1246         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1247                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1248                                                GFP_KERNEL);
1249                 if (!SLAVE_AD_INFO(slave)) {
1250                         kfree(slave);
1251                         return NULL;
1252                 }
1253         }
1254         return slave;
1255 }
1256
1257 static void bond_free_slave(struct slave *slave)
1258 {
1259         struct bonding *bond = bond_get_bond_by_slave(slave);
1260
1261         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1262                 kfree(SLAVE_AD_INFO(slave));
1263
1264         kfree(slave);
1265 }
1266
1267 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1268 {
1269         info->bond_mode = BOND_MODE(bond);
1270         info->miimon = bond->params.miimon;
1271         info->num_slaves = bond->slave_cnt;
1272 }
1273
1274 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1275 {
1276         strcpy(info->slave_name, slave->dev->name);
1277         info->link = slave->link;
1278         info->state = bond_slave_state(slave);
1279         info->link_failure_count = slave->link_failure_count;
1280 }
1281
1282 static void bond_netdev_notify(struct net_device *dev,
1283                                struct netdev_bonding_info *info)
1284 {
1285         rtnl_lock();
1286         netdev_bonding_info_change(dev, info);
1287         rtnl_unlock();
1288 }
1289
1290 static void bond_netdev_notify_work(struct work_struct *_work)
1291 {
1292         struct netdev_notify_work *w =
1293                 container_of(_work, struct netdev_notify_work, work.work);
1294
1295         bond_netdev_notify(w->dev, &w->bonding_info);
1296         dev_put(w->dev);
1297         kfree(w);
1298 }
1299
1300 void bond_queue_slave_event(struct slave *slave)
1301 {
1302         struct bonding *bond = slave->bond;
1303         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1304
1305         if (!nnw)
1306                 return;
1307
1308         dev_hold(slave->dev);
1309         nnw->dev = slave->dev;
1310         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1311         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1312         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1313
1314         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1315 }
1316
1317 void bond_lower_state_changed(struct slave *slave)
1318 {
1319         struct netdev_lag_lower_state_info info;
1320
1321         info.link_up = slave->link == BOND_LINK_UP ||
1322                        slave->link == BOND_LINK_FAIL;
1323         info.tx_enabled = bond_is_active_slave(slave);
1324         netdev_lower_state_changed(slave->dev, &info);
1325 }
1326
1327 /* enslave device <slave> to bond device <master> */
1328 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1329 {
1330         struct bonding *bond = netdev_priv(bond_dev);
1331         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1332         struct slave *new_slave = NULL, *prev_slave;
1333         struct sockaddr addr;
1334         int link_reporting;
1335         int res = 0, i;
1336
1337         if (!bond->params.use_carrier &&
1338             slave_dev->ethtool_ops->get_link == NULL &&
1339             slave_ops->ndo_do_ioctl == NULL) {
1340                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1341                             slave_dev->name);
1342         }
1343
1344         /* already enslaved */
1345         if (slave_dev->flags & IFF_SLAVE) {
1346                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1347                 return -EBUSY;
1348         }
1349
1350         if (bond_dev == slave_dev) {
1351                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1352                 return -EPERM;
1353         }
1354
1355         /* vlan challenged mutual exclusion */
1356         /* no need to lock since we're protected by rtnl_lock */
1357         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1358                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1359                            slave_dev->name);
1360                 if (vlan_uses_dev(bond_dev)) {
1361                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1362                                    slave_dev->name, bond_dev->name);
1363                         return -EPERM;
1364                 } else {
1365                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1366                                     slave_dev->name, slave_dev->name,
1367                                     bond_dev->name);
1368                 }
1369         } else {
1370                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1371                            slave_dev->name);
1372         }
1373
1374         /* Old ifenslave binaries are no longer supported.  These can
1375          * be identified with moderate accuracy by the state of the slave:
1376          * the current ifenslave will set the interface down prior to
1377          * enslaving it; the old ifenslave will not.
1378          */
1379         if (slave_dev->flags & IFF_UP) {
1380                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1381                            slave_dev->name);
1382                 res = -EPERM;
1383                 goto err_undo_flags;
1384         }
1385
1386         /* set bonding device ether type by slave - bonding netdevices are
1387          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1388          * there is a need to override some of the type dependent attribs/funcs.
1389          *
1390          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1391          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1392          */
1393         if (!bond_has_slaves(bond)) {
1394                 if (bond_dev->type != slave_dev->type) {
1395                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1396                                    bond_dev->type, slave_dev->type);
1397
1398                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1399                                                        bond_dev);
1400                         res = notifier_to_errno(res);
1401                         if (res) {
1402                                 netdev_err(bond_dev, "refused to change device type\n");
1403                                 res = -EBUSY;
1404                                 goto err_undo_flags;
1405                         }
1406
1407                         /* Flush unicast and multicast addresses */
1408                         dev_uc_flush(bond_dev);
1409                         dev_mc_flush(bond_dev);
1410
1411                         if (slave_dev->type != ARPHRD_ETHER)
1412                                 bond_setup_by_slave(bond_dev, slave_dev);
1413                         else {
1414                                 ether_setup(bond_dev);
1415                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1416                         }
1417
1418                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1419                                                  bond_dev);
1420                 }
1421         } else if (bond_dev->type != slave_dev->type) {
1422                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1423                            slave_dev->name, slave_dev->type, bond_dev->type);
1424                 res = -EINVAL;
1425                 goto err_undo_flags;
1426         }
1427
1428         if (slave_ops->ndo_set_mac_address == NULL) {
1429                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1430                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1431                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1432                         if (!bond_has_slaves(bond)) {
1433                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1434                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1435                         } else {
1436                                 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1437                                 res = -EOPNOTSUPP;
1438                                 goto err_undo_flags;
1439                         }
1440                 }
1441         }
1442
1443         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1444
1445         /* If this is the first slave, then we need to set the master's hardware
1446          * address to be the same as the slave's.
1447          */
1448         if (!bond_has_slaves(bond) &&
1449             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1450                 bond_set_dev_addr(bond->dev, slave_dev);
1451
1452         new_slave = bond_alloc_slave(bond);
1453         if (!new_slave) {
1454                 res = -ENOMEM;
1455                 goto err_undo_flags;
1456         }
1457
1458         new_slave->bond = bond;
1459         new_slave->dev = slave_dev;
1460         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1461          * is set via sysfs or module option if desired.
1462          */
1463         new_slave->queue_id = 0;
1464
1465         /* Save slave's original mtu and then set it to match the bond */
1466         new_slave->original_mtu = slave_dev->mtu;
1467         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1468         if (res) {
1469                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1470                 goto err_free;
1471         }
1472
1473         /* Save slave's original ("permanent") mac address for modes
1474          * that need it, and for restoring it upon release, and then
1475          * set it to the master's address
1476          */
1477         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1478
1479         if (!bond->params.fail_over_mac ||
1480             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1481                 /* Set slave to master's mac address.  The application already
1482                  * set the master's mac address to that of the first slave
1483                  */
1484                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1485                 addr.sa_family = slave_dev->type;
1486                 res = dev_set_mac_address(slave_dev, &addr);
1487                 if (res) {
1488                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1489                         goto err_restore_mtu;
1490                 }
1491         }
1492
1493         /* set slave flag before open to prevent IPv6 addrconf */
1494         slave_dev->flags |= IFF_SLAVE;
1495
1496         /* open the slave since the application closed it */
1497         res = dev_open(slave_dev);
1498         if (res) {
1499                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1500                 goto err_restore_mac;
1501         }
1502
1503         slave_dev->priv_flags |= IFF_BONDING;
1504         /* initialize slave stats */
1505         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1506
1507         if (bond_is_lb(bond)) {
1508                 /* bond_alb_init_slave() must be called before all other stages since
1509                  * it might fail and we do not want to have to undo everything
1510                  */
1511                 res = bond_alb_init_slave(bond, new_slave);
1512                 if (res)
1513                         goto err_close;
1514         }
1515
1516         /* If the mode uses primary, then the following is handled by
1517          * bond_change_active_slave().
1518          */
1519         if (!bond_uses_primary(bond)) {
1520                 /* set promiscuity level to new slave */
1521                 if (bond_dev->flags & IFF_PROMISC) {
1522                         res = dev_set_promiscuity(slave_dev, 1);
1523                         if (res)
1524                                 goto err_close;
1525                 }
1526
1527                 /* set allmulti level to new slave */
1528                 if (bond_dev->flags & IFF_ALLMULTI) {
1529                         res = dev_set_allmulti(slave_dev, 1);
1530                         if (res)
1531                                 goto err_close;
1532                 }
1533
1534                 netif_addr_lock_bh(bond_dev);
1535
1536                 dev_mc_sync_multiple(slave_dev, bond_dev);
1537                 dev_uc_sync_multiple(slave_dev, bond_dev);
1538
1539                 netif_addr_unlock_bh(bond_dev);
1540         }
1541
1542         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1543                 /* add lacpdu mc addr to mc list */
1544                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1545
1546                 dev_mc_add(slave_dev, lacpdu_multicast);
1547         }
1548
1549         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1550         if (res) {
1551                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1552                            slave_dev->name);
1553                 goto err_close;
1554         }
1555
1556         prev_slave = bond_last_slave(bond);
1557
1558         new_slave->delay = 0;
1559         new_slave->link_failure_count = 0;
1560
1561         bond_update_speed_duplex(new_slave);
1562
1563         new_slave->last_rx = jiffies -
1564                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1565         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1566                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1567
1568         if (bond->params.miimon && !bond->params.use_carrier) {
1569                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1570
1571                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1572                         /* miimon is set but a bonded network driver
1573                          * does not support ETHTOOL/MII and
1574                          * arp_interval is not set.  Note: if
1575                          * use_carrier is enabled, we will never go
1576                          * here (because netif_carrier is always
1577                          * supported); thus, we don't need to change
1578                          * the messages for netif_carrier.
1579                          */
1580                         netdev_warn(bond_dev, "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",
1581                                     slave_dev->name);
1582                 } else if (link_reporting == -1) {
1583                         /* unable get link status using mii/ethtool */
1584                         netdev_warn(bond_dev, "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",
1585                                     slave_dev->name);
1586                 }
1587         }
1588
1589         /* check for initial state */
1590         if (bond->params.miimon) {
1591                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1592                         if (bond->params.updelay) {
1593                                 bond_set_slave_link_state(new_slave,
1594                                                           BOND_LINK_BACK,
1595                                                           BOND_SLAVE_NOTIFY_NOW);
1596                                 new_slave->delay = bond->params.updelay;
1597                         } else {
1598                                 bond_set_slave_link_state(new_slave,
1599                                                           BOND_LINK_UP,
1600                                                           BOND_SLAVE_NOTIFY_NOW);
1601                         }
1602                 } else {
1603                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1604                                                   BOND_SLAVE_NOTIFY_NOW);
1605                 }
1606         } else if (bond->params.arp_interval) {
1607                 bond_set_slave_link_state(new_slave,
1608                                           (netif_carrier_ok(slave_dev) ?
1609                                           BOND_LINK_UP : BOND_LINK_DOWN),
1610                                           BOND_SLAVE_NOTIFY_NOW);
1611         } else {
1612                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1613                                           BOND_SLAVE_NOTIFY_NOW);
1614         }
1615
1616         if (new_slave->link != BOND_LINK_DOWN)
1617                 new_slave->last_link_up = jiffies;
1618         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1619                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1620                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1621
1622         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1623                 /* if there is a primary slave, remember it */
1624                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1625                         rcu_assign_pointer(bond->primary_slave, new_slave);
1626                         bond->force_primary = true;
1627                 }
1628         }
1629
1630         switch (BOND_MODE(bond)) {
1631         case BOND_MODE_ACTIVEBACKUP:
1632                 bond_set_slave_inactive_flags(new_slave,
1633                                               BOND_SLAVE_NOTIFY_NOW);
1634                 break;
1635         case BOND_MODE_8023AD:
1636                 /* in 802.3ad mode, the internal mechanism
1637                  * will activate the slaves in the selected
1638                  * aggregator
1639                  */
1640                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1641                 /* if this is the first slave */
1642                 if (!prev_slave) {
1643                         SLAVE_AD_INFO(new_slave)->id = 1;
1644                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1645                          * can be called only after the mac address of the bond is set
1646                          */
1647                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1648                 } else {
1649                         SLAVE_AD_INFO(new_slave)->id =
1650                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1651                 }
1652
1653                 bond_3ad_bind_slave(new_slave);
1654                 break;
1655         case BOND_MODE_TLB:
1656         case BOND_MODE_ALB:
1657                 bond_set_active_slave(new_slave);
1658                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1659                 break;
1660         default:
1661                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1662
1663                 /* always active in trunk mode */
1664                 bond_set_active_slave(new_slave);
1665
1666                 /* In trunking mode there is little meaning to curr_active_slave
1667                  * anyway (it holds no special properties of the bond device),
1668                  * so we can change it without calling change_active_interface()
1669                  */
1670                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1671                     new_slave->link == BOND_LINK_UP)
1672                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1673
1674                 break;
1675         } /* switch(bond_mode) */
1676
1677 #ifdef CONFIG_NET_POLL_CONTROLLER
1678         slave_dev->npinfo = bond->dev->npinfo;
1679         if (slave_dev->npinfo) {
1680                 if (slave_enable_netpoll(new_slave)) {
1681                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1682                         res = -EBUSY;
1683                         goto err_detach;
1684                 }
1685         }
1686 #endif
1687
1688         if (!(bond_dev->features & NETIF_F_LRO))
1689                 dev_disable_lro(slave_dev);
1690
1691         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1692                                          new_slave);
1693         if (res) {
1694                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1695                 goto err_detach;
1696         }
1697
1698         res = bond_master_upper_dev_link(bond, new_slave);
1699         if (res) {
1700                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1701                 goto err_unregister;
1702         }
1703
1704         res = bond_sysfs_slave_add(new_slave);
1705         if (res) {
1706                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1707                 goto err_upper_unlink;
1708         }
1709
1710         bond->slave_cnt++;
1711         bond_compute_features(bond);
1712         bond_set_carrier(bond);
1713
1714         if (bond_uses_primary(bond)) {
1715                 block_netpoll_tx();
1716                 bond_select_active_slave(bond);
1717                 unblock_netpoll_tx();
1718         }
1719
1720         if (bond_mode_uses_xmit_hash(bond))
1721                 bond_update_slave_arr(bond, NULL);
1722
1723         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1724                     slave_dev->name,
1725                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1726                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1727
1728         /* enslave is successful */
1729         bond_queue_slave_event(new_slave);
1730         return 0;
1731
1732 /* Undo stages on error */
1733 err_upper_unlink:
1734         bond_upper_dev_unlink(bond, new_slave);
1735
1736 err_unregister:
1737         netdev_rx_handler_unregister(slave_dev);
1738
1739 err_detach:
1740         if (!bond_uses_primary(bond))
1741                 bond_hw_addr_flush(bond_dev, slave_dev);
1742
1743         vlan_vids_del_by_dev(slave_dev, bond_dev);
1744         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1745                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1746         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1747                 block_netpoll_tx();
1748                 bond_change_active_slave(bond, NULL);
1749                 bond_select_active_slave(bond);
1750                 unblock_netpoll_tx();
1751         }
1752         /* either primary_slave or curr_active_slave might've changed */
1753         synchronize_rcu();
1754         slave_disable_netpoll(new_slave);
1755
1756 err_close:
1757         slave_dev->priv_flags &= ~IFF_BONDING;
1758         dev_close(slave_dev);
1759
1760 err_restore_mac:
1761         slave_dev->flags &= ~IFF_SLAVE;
1762         if (!bond->params.fail_over_mac ||
1763             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1764                 /* XXX TODO - fom follow mode needs to change master's
1765                  * MAC if this slave's MAC is in use by the bond, or at
1766                  * least print a warning.
1767                  */
1768                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1769                 addr.sa_family = slave_dev->type;
1770                 dev_set_mac_address(slave_dev, &addr);
1771         }
1772
1773 err_restore_mtu:
1774         dev_set_mtu(slave_dev, new_slave->original_mtu);
1775
1776 err_free:
1777         bond_free_slave(new_slave);
1778
1779 err_undo_flags:
1780         /* Enslave of first slave has failed and we need to fix master's mac */
1781         if (!bond_has_slaves(bond)) {
1782                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1783                                             slave_dev->dev_addr))
1784                         eth_hw_addr_random(bond_dev);
1785                 if (bond_dev->type != ARPHRD_ETHER) {
1786                         dev_close(bond_dev);
1787                         ether_setup(bond_dev);
1788                         bond_dev->flags |= IFF_MASTER;
1789                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1790                 }
1791         }
1792
1793         return res;
1794 }
1795
1796 /* Try to release the slave device <slave> from the bond device <master>
1797  * It is legal to access curr_active_slave without a lock because all the function
1798  * is RTNL-locked. If "all" is true it means that the function is being called
1799  * while destroying a bond interface and all slaves are being released.
1800  *
1801  * The rules for slave state should be:
1802  *   for Active/Backup:
1803  *     Active stays on all backups go down
1804  *   for Bonded connections:
1805  *     The first up interface should be left on and all others downed.
1806  */
1807 static int __bond_release_one(struct net_device *bond_dev,
1808                               struct net_device *slave_dev,
1809                               bool all)
1810 {
1811         struct bonding *bond = netdev_priv(bond_dev);
1812         struct slave *slave, *oldcurrent;
1813         struct sockaddr addr;
1814         int old_flags = bond_dev->flags;
1815         netdev_features_t old_features = bond_dev->features;
1816
1817         /* slave is not a slave or master is not master of this slave */
1818         if (!(slave_dev->flags & IFF_SLAVE) ||
1819             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1820                 netdev_dbg(bond_dev, "cannot release %s\n",
1821                            slave_dev->name);
1822                 return -EINVAL;
1823         }
1824
1825         block_netpoll_tx();
1826
1827         slave = bond_get_slave_by_dev(bond, slave_dev);
1828         if (!slave) {
1829                 /* not a slave of this bond */
1830                 netdev_info(bond_dev, "%s not enslaved\n",
1831                             slave_dev->name);
1832                 unblock_netpoll_tx();
1833                 return -EINVAL;
1834         }
1835
1836         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1837
1838         bond_sysfs_slave_del(slave);
1839
1840         /* recompute stats just before removing the slave */
1841         bond_get_stats(bond->dev, &bond->bond_stats);
1842
1843         bond_upper_dev_unlink(bond, slave);
1844         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1845          * for this slave anymore.
1846          */
1847         netdev_rx_handler_unregister(slave_dev);
1848
1849         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1850                 bond_3ad_unbind_slave(slave);
1851
1852         if (bond_mode_uses_xmit_hash(bond))
1853                 bond_update_slave_arr(bond, slave);
1854
1855         netdev_info(bond_dev, "Releasing %s interface %s\n",
1856                     bond_is_active_slave(slave) ? "active" : "backup",
1857                     slave_dev->name);
1858
1859         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1860
1861         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1862
1863         if (!all && (!bond->params.fail_over_mac ||
1864                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1865                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1866                     bond_has_slaves(bond))
1867                         netdev_warn(bond_dev, "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",
1868                                     slave_dev->name, slave->perm_hwaddr,
1869                                     bond_dev->name, slave_dev->name);
1870         }
1871
1872         if (rtnl_dereference(bond->primary_slave) == slave)
1873                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1874
1875         if (oldcurrent == slave)
1876                 bond_change_active_slave(bond, NULL);
1877
1878         if (bond_is_lb(bond)) {
1879                 /* Must be called only after the slave has been
1880                  * detached from the list and the curr_active_slave
1881                  * has been cleared (if our_slave == old_current),
1882                  * but before a new active slave is selected.
1883                  */
1884                 bond_alb_deinit_slave(bond, slave);
1885         }
1886
1887         if (all) {
1888                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1889         } else if (oldcurrent == slave) {
1890                 /* Note that we hold RTNL over this sequence, so there
1891                  * is no concern that another slave add/remove event
1892                  * will interfere.
1893                  */
1894                 bond_select_active_slave(bond);
1895         }
1896
1897         if (!bond_has_slaves(bond)) {
1898                 bond_set_carrier(bond);
1899                 eth_hw_addr_random(bond_dev);
1900         }
1901
1902         unblock_netpoll_tx();
1903         synchronize_rcu();
1904         bond->slave_cnt--;
1905
1906         if (!bond_has_slaves(bond)) {
1907                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1908                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1909         }
1910
1911         bond_compute_features(bond);
1912         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1913             (old_features & NETIF_F_VLAN_CHALLENGED))
1914                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1915                             slave_dev->name, bond_dev->name);
1916
1917         vlan_vids_del_by_dev(slave_dev, bond_dev);
1918
1919         /* If the mode uses primary, then this case was handled above by
1920          * bond_change_active_slave(..., NULL)
1921          */
1922         if (!bond_uses_primary(bond)) {
1923                 /* unset promiscuity level from slave
1924                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1925                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1926                  * value of that flag before that change, as that was the value
1927                  * when this slave was attached, so we cache at the start of the
1928                  * function and use it here. Same goes for ALLMULTI below
1929                  */
1930                 if (old_flags & IFF_PROMISC)
1931                         dev_set_promiscuity(slave_dev, -1);
1932
1933                 /* unset allmulti level from slave */
1934                 if (old_flags & IFF_ALLMULTI)
1935                         dev_set_allmulti(slave_dev, -1);
1936
1937                 bond_hw_addr_flush(bond_dev, slave_dev);
1938         }
1939
1940         slave_disable_netpoll(slave);
1941
1942         /* close slave before restoring its mac address */
1943         dev_close(slave_dev);
1944
1945         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1946             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1947                 /* restore original ("permanent") mac address */
1948                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1949                 addr.sa_family = slave_dev->type;
1950                 dev_set_mac_address(slave_dev, &addr);
1951         }
1952
1953         dev_set_mtu(slave_dev, slave->original_mtu);
1954
1955         slave_dev->priv_flags &= ~IFF_BONDING;
1956
1957         bond_free_slave(slave);
1958
1959         return 0;
1960 }
1961
1962 /* A wrapper used because of ndo_del_link */
1963 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1964 {
1965         return __bond_release_one(bond_dev, slave_dev, false);
1966 }
1967
1968 /* First release a slave and then destroy the bond if no more slaves are left.
1969  * Must be under rtnl_lock when this function is called.
1970  */
1971 static int  bond_release_and_destroy(struct net_device *bond_dev,
1972                                      struct net_device *slave_dev)
1973 {
1974         struct bonding *bond = netdev_priv(bond_dev);
1975         int ret;
1976
1977         ret = bond_release(bond_dev, slave_dev);
1978         if (ret == 0 && !bond_has_slaves(bond)) {
1979                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1980                 netdev_info(bond_dev, "Destroying bond %s\n",
1981                             bond_dev->name);
1982                 bond_remove_proc_entry(bond);
1983                 unregister_netdevice(bond_dev);
1984         }
1985         return ret;
1986 }
1987
1988 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1989 {
1990         struct bonding *bond = netdev_priv(bond_dev);
1991         bond_fill_ifbond(bond, info);
1992         return 0;
1993 }
1994
1995 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1996 {
1997         struct bonding *bond = netdev_priv(bond_dev);
1998         struct list_head *iter;
1999         int i = 0, res = -ENODEV;
2000         struct slave *slave;
2001
2002         bond_for_each_slave(bond, slave, iter) {
2003                 if (i++ == (int)info->slave_id) {
2004                         res = 0;
2005                         bond_fill_ifslave(slave, info);
2006                         break;
2007                 }
2008         }
2009
2010         return res;
2011 }
2012
2013 /*-------------------------------- Monitoring -------------------------------*/
2014
2015 /* called with rcu_read_lock() */
2016 static int bond_miimon_inspect(struct bonding *bond)
2017 {
2018         int link_state, commit = 0;
2019         struct list_head *iter;
2020         struct slave *slave;
2021         bool ignore_updelay;
2022
2023         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2024
2025         bond_for_each_slave_rcu(bond, slave, iter) {
2026                 slave->new_link = BOND_LINK_NOCHANGE;
2027
2028                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2029
2030                 switch (slave->link) {
2031                 case BOND_LINK_UP:
2032                         if (link_state)
2033                                 continue;
2034
2035                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2036                                                   BOND_SLAVE_NOTIFY_LATER);
2037                         slave->delay = bond->params.downdelay;
2038                         if (slave->delay) {
2039                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2040                                             (BOND_MODE(bond) ==
2041                                              BOND_MODE_ACTIVEBACKUP) ?
2042                                              (bond_is_active_slave(slave) ?
2043                                               "active " : "backup ") : "",
2044                                             slave->dev->name,
2045                                             bond->params.downdelay * bond->params.miimon);
2046                         }
2047                         /*FALLTHRU*/
2048                 case BOND_LINK_FAIL:
2049                         if (link_state) {
2050                                 /* recovered before downdelay expired */
2051                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2052                                                           BOND_SLAVE_NOTIFY_LATER);
2053                                 slave->last_link_up = jiffies;
2054                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2055                                             (bond->params.downdelay - slave->delay) *
2056                                             bond->params.miimon,
2057                                             slave->dev->name);
2058                                 continue;
2059                         }
2060
2061                         if (slave->delay <= 0) {
2062                                 slave->new_link = BOND_LINK_DOWN;
2063                                 commit++;
2064                                 continue;
2065                         }
2066
2067                         slave->delay--;
2068                         break;
2069
2070                 case BOND_LINK_DOWN:
2071                         if (!link_state)
2072                                 continue;
2073
2074                         bond_set_slave_link_state(slave, BOND_LINK_BACK,
2075                                                   BOND_SLAVE_NOTIFY_LATER);
2076                         slave->delay = bond->params.updelay;
2077
2078                         if (slave->delay) {
2079                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2080                                             slave->dev->name,
2081                                             ignore_updelay ? 0 :
2082                                             bond->params.updelay *
2083                                             bond->params.miimon);
2084                         }
2085                         /*FALLTHRU*/
2086                 case BOND_LINK_BACK:
2087                         if (!link_state) {
2088                                 bond_set_slave_link_state(slave,
2089                                                           BOND_LINK_DOWN,
2090                                                           BOND_SLAVE_NOTIFY_LATER);
2091                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2092                                             (bond->params.updelay - slave->delay) *
2093                                             bond->params.miimon,
2094                                             slave->dev->name);
2095
2096                                 continue;
2097                         }
2098
2099                         if (ignore_updelay)
2100                                 slave->delay = 0;
2101
2102                         if (slave->delay <= 0) {
2103                                 slave->new_link = BOND_LINK_UP;
2104                                 commit++;
2105                                 ignore_updelay = false;
2106                                 continue;
2107                         }
2108
2109                         slave->delay--;
2110                         break;
2111                 }
2112         }
2113
2114         return commit;
2115 }
2116
2117 static void bond_miimon_commit(struct bonding *bond)
2118 {
2119         struct list_head *iter;
2120         struct slave *slave, *primary;
2121
2122         bond_for_each_slave(bond, slave, iter) {
2123                 switch (slave->new_link) {
2124                 case BOND_LINK_NOCHANGE:
2125                         continue;
2126
2127                 case BOND_LINK_UP:
2128                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2129                                                   BOND_SLAVE_NOTIFY_NOW);
2130                         slave->last_link_up = jiffies;
2131
2132                         primary = rtnl_dereference(bond->primary_slave);
2133                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2134                                 /* prevent it from being the active one */
2135                                 bond_set_backup_slave(slave);
2136                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2137                                 /* make it immediately active */
2138                                 bond_set_active_slave(slave);
2139                         } else if (slave != primary) {
2140                                 /* prevent it from being the active one */
2141                                 bond_set_backup_slave(slave);
2142                         }
2143
2144                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2145                                     slave->dev->name,
2146                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2147                                     slave->duplex ? "full" : "half");
2148
2149                         /* notify ad that the link status has changed */
2150                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2151                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2152
2153                         if (bond_is_lb(bond))
2154                                 bond_alb_handle_link_change(bond, slave,
2155                                                             BOND_LINK_UP);
2156
2157                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2158                                 bond_update_slave_arr(bond, NULL);
2159
2160                         if (!bond->curr_active_slave || slave == primary)
2161                                 goto do_failover;
2162
2163                         continue;
2164
2165                 case BOND_LINK_DOWN:
2166                         if (slave->link_failure_count < UINT_MAX)
2167                                 slave->link_failure_count++;
2168
2169                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2170                                                   BOND_SLAVE_NOTIFY_NOW);
2171
2172                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2173                             BOND_MODE(bond) == BOND_MODE_8023AD)
2174                                 bond_set_slave_inactive_flags(slave,
2175                                                               BOND_SLAVE_NOTIFY_NOW);
2176
2177                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2178                                     slave->dev->name);
2179
2180                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2181                                 bond_3ad_handle_link_change(slave,
2182                                                             BOND_LINK_DOWN);
2183
2184                         if (bond_is_lb(bond))
2185                                 bond_alb_handle_link_change(bond, slave,
2186                                                             BOND_LINK_DOWN);
2187
2188                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2189                                 bond_update_slave_arr(bond, NULL);
2190
2191                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2192                                 goto do_failover;
2193
2194                         continue;
2195
2196                 default:
2197                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2198                                    slave->new_link, slave->dev->name);
2199                         slave->new_link = BOND_LINK_NOCHANGE;
2200
2201                         continue;
2202                 }
2203
2204 do_failover:
2205                 block_netpoll_tx();
2206                 bond_select_active_slave(bond);
2207                 unblock_netpoll_tx();
2208         }
2209
2210         bond_set_carrier(bond);
2211 }
2212
2213 /* bond_mii_monitor
2214  *
2215  * Really a wrapper that splits the mii monitor into two phases: an
2216  * inspection, then (if inspection indicates something needs to be done)
2217  * an acquisition of appropriate locks followed by a commit phase to
2218  * implement whatever link state changes are indicated.
2219  */
2220 static void bond_mii_monitor(struct work_struct *work)
2221 {
2222         struct bonding *bond = container_of(work, struct bonding,
2223                                             mii_work.work);
2224         bool should_notify_peers = false;
2225         unsigned long delay;
2226
2227         delay = msecs_to_jiffies(bond->params.miimon);
2228
2229         if (!bond_has_slaves(bond))
2230                 goto re_arm;
2231
2232         rcu_read_lock();
2233
2234         should_notify_peers = bond_should_notify_peers(bond);
2235
2236         if (bond_miimon_inspect(bond)) {
2237                 rcu_read_unlock();
2238
2239                 /* Race avoidance with bond_close cancel of workqueue */
2240                 if (!rtnl_trylock()) {
2241                         delay = 1;
2242                         should_notify_peers = false;
2243                         goto re_arm;
2244                 }
2245
2246                 bond_miimon_commit(bond);
2247
2248                 rtnl_unlock();  /* might sleep, hold no other locks */
2249         } else
2250                 rcu_read_unlock();
2251
2252 re_arm:
2253         if (bond->params.miimon)
2254                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2255
2256         if (should_notify_peers) {
2257                 if (!rtnl_trylock())
2258                         return;
2259                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2260                 rtnl_unlock();
2261         }
2262 }
2263
2264 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2265 {
2266         struct net_device *upper;
2267         struct list_head *iter;
2268         bool ret = false;
2269
2270         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2271                 return true;
2272
2273         rcu_read_lock();
2274         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2275                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2276                         ret = true;
2277                         break;
2278                 }
2279         }
2280         rcu_read_unlock();
2281
2282         return ret;
2283 }
2284
2285 /* We go to the (large) trouble of VLAN tagging ARP frames because
2286  * switches in VLAN mode (especially if ports are configured as
2287  * "native" to a VLAN) might not pass non-tagged frames.
2288  */
2289 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2290                           __be32 dest_ip, __be32 src_ip,
2291                           struct bond_vlan_tag *tags)
2292 {
2293         struct sk_buff *skb;
2294         struct bond_vlan_tag *outer_tag = tags;
2295
2296         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2297                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2298
2299         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2300                          NULL, slave_dev->dev_addr, NULL);
2301
2302         if (!skb) {
2303                 net_err_ratelimited("ARP packet allocation failed\n");
2304                 return;
2305         }
2306
2307         if (!tags || tags->vlan_proto == VLAN_N_VID)
2308                 goto xmit;
2309
2310         tags++;
2311
2312         /* Go through all the tags backwards and add them to the packet */
2313         while (tags->vlan_proto != VLAN_N_VID) {
2314                 if (!tags->vlan_id) {
2315                         tags++;
2316                         continue;
2317                 }
2318
2319                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2320                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2321                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2322                                                 tags->vlan_id);
2323                 if (!skb) {
2324                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2325                         return;
2326                 }
2327
2328                 tags++;
2329         }
2330         /* Set the outer tag */
2331         if (outer_tag->vlan_id) {
2332                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2333                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2334                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2335                                        outer_tag->vlan_id);
2336         }
2337
2338 xmit:
2339         arp_xmit(skb);
2340 }
2341
2342 /* Validate the device path between the @start_dev and the @end_dev.
2343  * The path is valid if the @end_dev is reachable through device
2344  * stacking.
2345  * When the path is validated, collect any vlan information in the
2346  * path.
2347  */
2348 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2349                                               struct net_device *end_dev,
2350                                               int level)
2351 {
2352         struct bond_vlan_tag *tags;
2353         struct net_device *upper;
2354         struct list_head  *iter;
2355
2356         if (start_dev == end_dev) {
2357                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2358                 if (!tags)
2359                         return ERR_PTR(-ENOMEM);
2360                 tags[level].vlan_proto = VLAN_N_VID;
2361                 return tags;
2362         }
2363
2364         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2365                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2366                 if (IS_ERR_OR_NULL(tags)) {
2367                         if (IS_ERR(tags))
2368                                 return tags;
2369                         continue;
2370                 }
2371                 if (is_vlan_dev(upper)) {
2372                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2373                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2374                 }
2375
2376                 return tags;
2377         }
2378
2379         return NULL;
2380 }
2381
2382 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2383 {
2384         struct rtable *rt;
2385         struct bond_vlan_tag *tags;
2386         __be32 *targets = bond->params.arp_targets, addr;
2387         int i;
2388
2389         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2390                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2391                 tags = NULL;
2392
2393                 /* Find out through which dev should the packet go */
2394                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2395                                      RTO_ONLINK, 0);
2396                 if (IS_ERR(rt)) {
2397                         /* there's no route to target - try to send arp
2398                          * probe to generate any traffic (arp_validate=0)
2399                          */
2400                         if (bond->params.arp_validate)
2401                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2402                                                      bond->dev->name,
2403                                                      &targets[i]);
2404                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2405                                       0, tags);
2406                         continue;
2407                 }
2408
2409                 /* bond device itself */
2410                 if (rt->dst.dev == bond->dev)
2411                         goto found;
2412
2413                 rcu_read_lock();
2414                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2415                 rcu_read_unlock();
2416
2417                 if (!IS_ERR_OR_NULL(tags))
2418                         goto found;
2419
2420                 /* Not our device - skip */
2421                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2422                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2423
2424                 ip_rt_put(rt);
2425                 continue;
2426
2427 found:
2428                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2429                 ip_rt_put(rt);
2430                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2431                               addr, tags);
2432                 kfree(tags);
2433         }
2434 }
2435
2436 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2437 {
2438         int i;
2439
2440         if (!sip || !bond_has_this_ip(bond, tip)) {
2441                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2442                            &sip, &tip);
2443                 return;
2444         }
2445
2446         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2447         if (i == -1) {
2448                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2449                            &sip);
2450                 return;
2451         }
2452         slave->last_rx = jiffies;
2453         slave->target_last_arp_rx[i] = jiffies;
2454 }
2455
2456 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2457                  struct slave *slave)
2458 {
2459         struct arphdr *arp = (struct arphdr *)skb->data;
2460         struct slave *curr_active_slave;
2461         unsigned char *arp_ptr;
2462         __be32 sip, tip;
2463         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2464
2465         if (!slave_do_arp_validate(bond, slave)) {
2466                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2467                     !slave_do_arp_validate_only(bond))
2468                         slave->last_rx = jiffies;
2469                 return RX_HANDLER_ANOTHER;
2470         } else if (!is_arp) {
2471                 return RX_HANDLER_ANOTHER;
2472         }
2473
2474         alen = arp_hdr_len(bond->dev);
2475
2476         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2477                    skb->dev->name);
2478
2479         if (alen > skb_headlen(skb)) {
2480                 arp = kmalloc(alen, GFP_ATOMIC);
2481                 if (!arp)
2482                         goto out_unlock;
2483                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2484                         goto out_unlock;
2485         }
2486
2487         if (arp->ar_hln != bond->dev->addr_len ||
2488             skb->pkt_type == PACKET_OTHERHOST ||
2489             skb->pkt_type == PACKET_LOOPBACK ||
2490             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2491             arp->ar_pro != htons(ETH_P_IP) ||
2492             arp->ar_pln != 4)
2493                 goto out_unlock;
2494
2495         arp_ptr = (unsigned char *)(arp + 1);
2496         arp_ptr += bond->dev->addr_len;
2497         memcpy(&sip, arp_ptr, 4);
2498         arp_ptr += 4 + bond->dev->addr_len;
2499         memcpy(&tip, arp_ptr, 4);
2500
2501         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2502                    slave->dev->name, bond_slave_state(slave),
2503                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2504                      &sip, &tip);
2505
2506         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2507
2508         /* Backup slaves won't see the ARP reply, but do come through
2509          * here for each ARP probe (so we swap the sip/tip to validate
2510          * the probe).  In a "redundant switch, common router" type of
2511          * configuration, the ARP probe will (hopefully) travel from
2512          * the active, through one switch, the router, then the other
2513          * switch before reaching the backup.
2514          *
2515          * We 'trust' the arp requests if there is an active slave and
2516          * it received valid arp reply(s) after it became active. This
2517          * is done to avoid endless looping when we can't reach the
2518          * arp_ip_target and fool ourselves with our own arp requests.
2519          */
2520
2521         if (bond_is_active_slave(slave))
2522                 bond_validate_arp(bond, slave, sip, tip);
2523         else if (curr_active_slave &&
2524                  time_after(slave_last_rx(bond, curr_active_slave),
2525                             curr_active_slave->last_link_up))
2526                 bond_validate_arp(bond, slave, tip, sip);
2527
2528 out_unlock:
2529         if (arp != (struct arphdr *)skb->data)
2530                 kfree(arp);
2531         return RX_HANDLER_ANOTHER;
2532 }
2533
2534 /* function to verify if we're in the arp_interval timeslice, returns true if
2535  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2536  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2537  */
2538 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2539                                   int mod)
2540 {
2541         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2542
2543         return time_in_range(jiffies,
2544                              last_act - delta_in_ticks,
2545                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2546 }
2547
2548 /* This function is called regularly to monitor each slave's link
2549  * ensuring that traffic is being sent and received when arp monitoring
2550  * is used in load-balancing mode. if the adapter has been dormant, then an
2551  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2552  * arp monitoring in active backup mode.
2553  */
2554 static void bond_loadbalance_arp_mon(struct work_struct *work)
2555 {
2556         struct bonding *bond = container_of(work, struct bonding,
2557                                             arp_work.work);
2558         struct slave *slave, *oldcurrent;
2559         struct list_head *iter;
2560         int do_failover = 0, slave_state_changed = 0;
2561
2562         if (!bond_has_slaves(bond))
2563                 goto re_arm;
2564
2565         rcu_read_lock();
2566
2567         oldcurrent = rcu_dereference(bond->curr_active_slave);
2568         /* see if any of the previous devices are up now (i.e. they have
2569          * xmt and rcv traffic). the curr_active_slave does not come into
2570          * the picture unless it is null. also, slave->last_link_up is not
2571          * needed here because we send an arp on each slave and give a slave
2572          * as long as it needs to get the tx/rx within the delta.
2573          * TODO: what about up/down delay in arp mode? it wasn't here before
2574          *       so it can wait
2575          */
2576         bond_for_each_slave_rcu(bond, slave, iter) {
2577                 unsigned long trans_start = dev_trans_start(slave->dev);
2578
2579                 if (slave->link != BOND_LINK_UP) {
2580                         if (bond_time_in_interval(bond, trans_start, 1) &&
2581                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2582
2583                                 slave->link  = BOND_LINK_UP;
2584                                 slave_state_changed = 1;
2585
2586                                 /* primary_slave has no meaning in round-robin
2587                                  * mode. the window of a slave being up and
2588                                  * curr_active_slave being null after enslaving
2589                                  * is closed.
2590                                  */
2591                                 if (!oldcurrent) {
2592                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2593                                                     slave->dev->name);
2594                                         do_failover = 1;
2595                                 } else {
2596                                         netdev_info(bond->dev, "interface %s is now up\n",
2597                                                     slave->dev->name);
2598                                 }
2599                         }
2600                 } else {
2601                         /* slave->link == BOND_LINK_UP */
2602
2603                         /* not all switches will respond to an arp request
2604                          * when the source ip is 0, so don't take the link down
2605                          * if we don't know our ip yet
2606                          */
2607                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2608                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2609
2610                                 slave->link  = BOND_LINK_DOWN;
2611                                 slave_state_changed = 1;
2612
2613                                 if (slave->link_failure_count < UINT_MAX)
2614                                         slave->link_failure_count++;
2615
2616                                 netdev_info(bond->dev, "interface %s is now down\n",
2617                                             slave->dev->name);
2618
2619                                 if (slave == oldcurrent)
2620                                         do_failover = 1;
2621                         }
2622                 }
2623
2624                 /* note: if switch is in round-robin mode, all links
2625                  * must tx arp to ensure all links rx an arp - otherwise
2626                  * links may oscillate or not come up at all; if switch is
2627                  * in something like xor mode, there is nothing we can
2628                  * do - all replies will be rx'ed on same link causing slaves
2629                  * to be unstable during low/no traffic periods
2630                  */
2631                 if (bond_slave_is_up(slave))
2632                         bond_arp_send_all(bond, slave);
2633         }
2634
2635         rcu_read_unlock();
2636
2637         if (do_failover || slave_state_changed) {
2638                 if (!rtnl_trylock())
2639                         goto re_arm;
2640
2641                 if (slave_state_changed) {
2642                         bond_slave_state_change(bond);
2643                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2644                                 bond_update_slave_arr(bond, NULL);
2645                 }
2646                 if (do_failover) {
2647                         block_netpoll_tx();
2648                         bond_select_active_slave(bond);
2649                         unblock_netpoll_tx();
2650                 }
2651                 rtnl_unlock();
2652         }
2653
2654 re_arm:
2655         if (bond->params.arp_interval)
2656                 queue_delayed_work(bond->wq, &bond->arp_work,
2657                                    msecs_to_jiffies(bond->params.arp_interval));
2658 }
2659
2660 /* Called to inspect slaves for active-backup mode ARP monitor link state
2661  * changes.  Sets new_link in slaves to specify what action should take
2662  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2663  * to link states must be committed.
2664  *
2665  * Called with rcu_read_lock held.
2666  */
2667 static int bond_ab_arp_inspect(struct bonding *bond)
2668 {
2669         unsigned long trans_start, last_rx;
2670         struct list_head *iter;
2671         struct slave *slave;
2672         int commit = 0;
2673
2674         bond_for_each_slave_rcu(bond, slave, iter) {
2675                 slave->new_link = BOND_LINK_NOCHANGE;
2676                 last_rx = slave_last_rx(bond, slave);
2677
2678                 if (slave->link != BOND_LINK_UP) {
2679                         if (bond_time_in_interval(bond, last_rx, 1)) {
2680                                 slave->new_link = BOND_LINK_UP;
2681                                 commit++;
2682                         }
2683                         continue;
2684                 }
2685
2686                 /* Give slaves 2*delta after being enslaved or made
2687                  * active.  This avoids bouncing, as the last receive
2688                  * times need a full ARP monitor cycle to be updated.
2689                  */
2690                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2691                         continue;
2692
2693                 /* Backup slave is down if:
2694                  * - No current_arp_slave AND
2695                  * - more than 3*delta since last receive AND
2696                  * - the bond has an IP address
2697                  *
2698                  * Note: a non-null current_arp_slave indicates
2699                  * the curr_active_slave went down and we are
2700                  * searching for a new one; under this condition
2701                  * we only take the curr_active_slave down - this
2702                  * gives each slave a chance to tx/rx traffic
2703                  * before being taken out
2704                  */
2705                 if (!bond_is_active_slave(slave) &&
2706                     !rcu_access_pointer(bond->current_arp_slave) &&
2707                     !bond_time_in_interval(bond, last_rx, 3)) {
2708                         slave->new_link = BOND_LINK_DOWN;
2709                         commit++;
2710                 }
2711
2712                 /* Active slave is down if:
2713                  * - more than 2*delta since transmitting OR
2714                  * - (more than 2*delta since receive AND
2715                  *    the bond has an IP address)
2716                  */
2717                 trans_start = dev_trans_start(slave->dev);
2718                 if (bond_is_active_slave(slave) &&
2719                     (!bond_time_in_interval(bond, trans_start, 2) ||
2720                      !bond_time_in_interval(bond, last_rx, 2))) {
2721                         slave->new_link = BOND_LINK_DOWN;
2722                         commit++;
2723                 }
2724         }
2725
2726         return commit;
2727 }
2728
2729 /* Called to commit link state changes noted by inspection step of
2730  * active-backup mode ARP monitor.
2731  *
2732  * Called with RTNL hold.
2733  */
2734 static void bond_ab_arp_commit(struct bonding *bond)
2735 {
2736         unsigned long trans_start;
2737         struct list_head *iter;
2738         struct slave *slave;
2739
2740         bond_for_each_slave(bond, slave, iter) {
2741                 switch (slave->new_link) {
2742                 case BOND_LINK_NOCHANGE:
2743                         continue;
2744
2745                 case BOND_LINK_UP:
2746                         trans_start = dev_trans_start(slave->dev);
2747                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2748                             (!rtnl_dereference(bond->curr_active_slave) &&
2749                              bond_time_in_interval(bond, trans_start, 1))) {
2750                                 struct slave *current_arp_slave;
2751
2752                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2753                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2754                                                           BOND_SLAVE_NOTIFY_NOW);
2755                                 if (current_arp_slave) {
2756                                         bond_set_slave_inactive_flags(
2757                                                 current_arp_slave,
2758                                                 BOND_SLAVE_NOTIFY_NOW);
2759                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2760                                 }
2761
2762                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2763                                             slave->dev->name);
2764
2765                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2766                                     slave == rtnl_dereference(bond->primary_slave))
2767                                         goto do_failover;
2768
2769                         }
2770
2771                         continue;
2772
2773                 case BOND_LINK_DOWN:
2774                         if (slave->link_failure_count < UINT_MAX)
2775                                 slave->link_failure_count++;
2776
2777                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2778                                                   BOND_SLAVE_NOTIFY_NOW);
2779                         bond_set_slave_inactive_flags(slave,
2780                                                       BOND_SLAVE_NOTIFY_NOW);
2781
2782                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2783                                     slave->dev->name);
2784
2785                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2786                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2787                                 goto do_failover;
2788                         }
2789
2790                         continue;
2791
2792                 default:
2793                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2794                                    slave->new_link, slave->dev->name);
2795                         continue;
2796                 }
2797
2798 do_failover:
2799                 block_netpoll_tx();
2800                 bond_select_active_slave(bond);
2801                 unblock_netpoll_tx();
2802         }
2803
2804         bond_set_carrier(bond);
2805 }
2806
2807 /* Send ARP probes for active-backup mode ARP monitor.
2808  *
2809  * Called with rcu_read_lock held.
2810  */
2811 static bool bond_ab_arp_probe(struct bonding *bond)
2812 {
2813         struct slave *slave, *before = NULL, *new_slave = NULL,
2814                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2815                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2816         struct list_head *iter;
2817         bool found = false;
2818         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2819
2820         if (curr_arp_slave && curr_active_slave)
2821                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2822                             curr_arp_slave->dev->name,
2823                             curr_active_slave->dev->name);
2824
2825         if (curr_active_slave) {
2826                 bond_arp_send_all(bond, curr_active_slave);
2827                 return should_notify_rtnl;
2828         }
2829
2830         /* if we don't have a curr_active_slave, search for the next available
2831          * backup slave from the current_arp_slave and make it the candidate
2832          * for becoming the curr_active_slave
2833          */
2834
2835         if (!curr_arp_slave) {
2836                 curr_arp_slave = bond_first_slave_rcu(bond);
2837                 if (!curr_arp_slave)
2838                         return should_notify_rtnl;
2839         }
2840
2841         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2842
2843         bond_for_each_slave_rcu(bond, slave, iter) {
2844                 if (!found && !before && bond_slave_is_up(slave))
2845                         before = slave;
2846
2847                 if (found && !new_slave && bond_slave_is_up(slave))
2848                         new_slave = slave;
2849                 /* if the link state is up at this point, we
2850                  * mark it down - this can happen if we have
2851                  * simultaneous link failures and
2852                  * reselect_active_interface doesn't make this
2853                  * one the current slave so it is still marked
2854                  * up when it is actually down
2855                  */
2856                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2857                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2858                                                   BOND_SLAVE_NOTIFY_LATER);
2859                         if (slave->link_failure_count < UINT_MAX)
2860                                 slave->link_failure_count++;
2861
2862                         bond_set_slave_inactive_flags(slave,
2863                                                       BOND_SLAVE_NOTIFY_LATER);
2864
2865                         netdev_info(bond->dev, "backup interface %s is now down\n",
2866                                     slave->dev->name);
2867                 }
2868                 if (slave == curr_arp_slave)
2869                         found = true;
2870         }
2871
2872         if (!new_slave && before)
2873                 new_slave = before;
2874
2875         if (!new_slave)
2876                 goto check_state;
2877
2878         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2879                                   BOND_SLAVE_NOTIFY_LATER);
2880         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2881         bond_arp_send_all(bond, new_slave);
2882         new_slave->last_link_up = jiffies;
2883         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2884
2885 check_state:
2886         bond_for_each_slave_rcu(bond, slave, iter) {
2887                 if (slave->should_notify || slave->should_notify_link) {
2888                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2889                         break;
2890                 }
2891         }
2892         return should_notify_rtnl;
2893 }
2894
2895 static void bond_activebackup_arp_mon(struct work_struct *work)
2896 {
2897         struct bonding *bond = container_of(work, struct bonding,
2898                                             arp_work.work);
2899         bool should_notify_peers = false;
2900         bool should_notify_rtnl = false;
2901         int delta_in_ticks;
2902
2903         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2904
2905         if (!bond_has_slaves(bond))
2906                 goto re_arm;
2907
2908         rcu_read_lock();
2909
2910         should_notify_peers = bond_should_notify_peers(bond);
2911
2912         if (bond_ab_arp_inspect(bond)) {
2913                 rcu_read_unlock();
2914
2915                 /* Race avoidance with bond_close flush of workqueue */
2916                 if (!rtnl_trylock()) {
2917                         delta_in_ticks = 1;
2918                         should_notify_peers = false;
2919                         goto re_arm;
2920                 }
2921
2922                 bond_ab_arp_commit(bond);
2923
2924                 rtnl_unlock();
2925                 rcu_read_lock();
2926         }
2927
2928         should_notify_rtnl = bond_ab_arp_probe(bond);
2929         rcu_read_unlock();
2930
2931 re_arm:
2932         if (bond->params.arp_interval)
2933                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2934
2935         if (should_notify_peers || should_notify_rtnl) {
2936                 if (!rtnl_trylock())
2937                         return;
2938
2939                 if (should_notify_peers)
2940                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2941                                                  bond->dev);
2942                 if (should_notify_rtnl) {
2943                         bond_slave_state_notify(bond);
2944                         bond_slave_link_notify(bond);
2945                 }
2946
2947                 rtnl_unlock();
2948         }
2949 }
2950
2951 /*-------------------------- netdev event handling --------------------------*/
2952
2953 /* Change device name */
2954 static int bond_event_changename(struct bonding *bond)
2955 {
2956         bond_remove_proc_entry(bond);
2957         bond_create_proc_entry(bond);
2958
2959         bond_debug_reregister(bond);
2960
2961         return NOTIFY_DONE;
2962 }
2963
2964 static int bond_master_netdev_event(unsigned long event,
2965                                     struct net_device *bond_dev)
2966 {
2967         struct bonding *event_bond = netdev_priv(bond_dev);
2968
2969         switch (event) {
2970         case NETDEV_CHANGENAME:
2971                 return bond_event_changename(event_bond);
2972         case NETDEV_UNREGISTER:
2973                 bond_remove_proc_entry(event_bond);
2974                 break;
2975         case NETDEV_REGISTER:
2976                 bond_create_proc_entry(event_bond);
2977                 break;
2978         case NETDEV_NOTIFY_PEERS:
2979                 if (event_bond->send_peer_notif)
2980                         event_bond->send_peer_notif--;
2981                 break;
2982         default:
2983                 break;
2984         }
2985
2986         return NOTIFY_DONE;
2987 }
2988
2989 static int bond_slave_netdev_event(unsigned long event,
2990                                    struct net_device *slave_dev)
2991 {
2992         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2993         struct bonding *bond;
2994         struct net_device *bond_dev;
2995
2996         /* A netdev event can be generated while enslaving a device
2997          * before netdev_rx_handler_register is called in which case
2998          * slave will be NULL
2999          */
3000         if (!slave)
3001                 return NOTIFY_DONE;
3002         bond_dev = slave->bond->dev;
3003         bond = slave->bond;
3004         primary = rtnl_dereference(bond->primary_slave);
3005
3006         switch (event) {
3007         case NETDEV_UNREGISTER:
3008                 if (bond_dev->type != ARPHRD_ETHER)
3009                         bond_release_and_destroy(bond_dev, slave_dev);
3010                 else
3011                         bond_release(bond_dev, slave_dev);
3012                 break;
3013         case NETDEV_UP:
3014         case NETDEV_CHANGE:
3015                 bond_update_speed_duplex(slave);
3016                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3017                         bond_3ad_adapter_speed_duplex_changed(slave);
3018                 /* Fallthrough */
3019         case NETDEV_DOWN:
3020                 /* Refresh slave-array if applicable!
3021                  * If the setup does not use miimon or arpmon (mode-specific!),
3022                  * then these events will not cause the slave-array to be
3023                  * refreshed. This will cause xmit to use a slave that is not
3024                  * usable. Avoid such situation by refeshing the array at these
3025                  * events. If these (miimon/arpmon) parameters are configured
3026                  * then array gets refreshed twice and that should be fine!
3027                  */
3028                 if (bond_mode_uses_xmit_hash(bond))
3029                         bond_update_slave_arr(bond, NULL);
3030                 break;
3031         case NETDEV_CHANGEMTU:
3032                 /* TODO: Should slaves be allowed to
3033                  * independently alter their MTU?  For
3034                  * an active-backup bond, slaves need
3035                  * not be the same type of device, so
3036                  * MTUs may vary.  For other modes,
3037                  * slaves arguably should have the
3038                  * same MTUs. To do this, we'd need to
3039                  * take over the slave's change_mtu
3040                  * function for the duration of their
3041                  * servitude.
3042                  */
3043                 break;
3044         case NETDEV_CHANGENAME:
3045                 /* we don't care if we don't have primary set */
3046                 if (!bond_uses_primary(bond) ||
3047                     !bond->params.primary[0])
3048                         break;
3049
3050                 if (slave == primary) {
3051                         /* slave's name changed - he's no longer primary */
3052                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3053                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3054                         /* we have a new primary slave */
3055                         rcu_assign_pointer(bond->primary_slave, slave);
3056                 } else { /* we didn't change primary - exit */
3057                         break;
3058                 }
3059
3060                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3061                             primary ? slave_dev->name : "none");
3062
3063                 block_netpoll_tx();
3064                 bond_select_active_slave(bond);
3065                 unblock_netpoll_tx();
3066                 break;
3067         case NETDEV_FEAT_CHANGE:
3068                 bond_compute_features(bond);
3069                 break;
3070         case NETDEV_RESEND_IGMP:
3071                 /* Propagate to master device */
3072                 call_netdevice_notifiers(event, slave->bond->dev);
3073                 break;
3074         default:
3075                 break;
3076         }
3077
3078         return NOTIFY_DONE;
3079 }
3080
3081 /* bond_netdev_event: handle netdev notifier chain events.
3082  *
3083  * This function receives events for the netdev chain.  The caller (an
3084  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3085  * locks for us to safely manipulate the slave devices (RTNL lock,
3086  * dev_probe_lock).
3087  */
3088 static int bond_netdev_event(struct notifier_block *this,
3089                              unsigned long event, void *ptr)
3090 {
3091         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3092
3093         netdev_dbg(event_dev, "event: %lx\n", event);
3094
3095         if (!(event_dev->priv_flags & IFF_BONDING))
3096                 return NOTIFY_DONE;
3097
3098         if (event_dev->flags & IFF_MASTER) {
3099                 netdev_dbg(event_dev, "IFF_MASTER\n");
3100                 return bond_master_netdev_event(event, event_dev);
3101         }
3102
3103         if (event_dev->flags & IFF_SLAVE) {
3104                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3105                 return bond_slave_netdev_event(event, event_dev);
3106         }
3107
3108         return NOTIFY_DONE;
3109 }
3110
3111 static struct notifier_block bond_netdev_notifier = {
3112         .notifier_call = bond_netdev_event,
3113 };
3114
3115 /*---------------------------- Hashing Policies -----------------------------*/
3116
3117 /* L2 hash helper */
3118 static inline u32 bond_eth_hash(struct sk_buff *skb)
3119 {
3120         struct ethhdr *ep, hdr_tmp;
3121
3122         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3123         if (ep)
3124                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3125         return 0;
3126 }
3127
3128 /* Extract the appropriate headers based on bond's xmit policy */
3129 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3130                               struct flow_keys *fk)
3131 {
3132         const struct ipv6hdr *iph6;
3133         const struct iphdr *iph;
3134         int noff, proto = -1;
3135
3136         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3137                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3138
3139         fk->ports.ports = 0;
3140         noff = skb_network_offset(skb);
3141         if (skb->protocol == htons(ETH_P_IP)) {
3142                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3143                         return false;
3144                 iph = ip_hdr(skb);
3145                 iph_to_flow_copy_v4addrs(fk, iph);
3146                 noff += iph->ihl << 2;
3147                 if (!ip_is_fragment(iph))
3148                         proto = iph->protocol;
3149         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3150                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3151                         return false;
3152                 iph6 = ipv6_hdr(skb);
3153                 iph_to_flow_copy_v6addrs(fk, iph6);
3154                 noff += sizeof(*iph6);
3155                 proto = iph6->nexthdr;
3156         } else {
3157                 return false;
3158         }
3159         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3160                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3161
3162         return true;
3163 }
3164
3165 /**
3166  * bond_xmit_hash - generate a hash value based on the xmit policy
3167  * @bond: bonding device
3168  * @skb: buffer to use for headers
3169  *
3170  * This function will extract the necessary headers from the skb buffer and use
3171  * them to generate a hash based on the xmit_policy set in the bonding device
3172  */
3173 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3174 {
3175         struct flow_keys flow;
3176         u32 hash;
3177
3178         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3179             skb->l4_hash)
3180                 return skb->hash;
3181
3182         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3183             !bond_flow_dissect(bond, skb, &flow))
3184                 return bond_eth_hash(skb);
3185
3186         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3187             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3188                 hash = bond_eth_hash(skb);
3189         else
3190                 hash = (__force u32)flow.ports.ports;
3191         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3192                 (__force u32)flow_get_u32_src(&flow);
3193         hash ^= (hash >> 16);
3194         hash ^= (hash >> 8);
3195
3196         return hash;
3197 }
3198
3199 /*-------------------------- Device entry points ----------------------------*/
3200
3201 static void bond_work_init_all(struct bonding *bond)
3202 {
3203         INIT_DELAYED_WORK(&bond->mcast_work,
3204                           bond_resend_igmp_join_requests_delayed);
3205         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3206         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3207         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3208                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3209         else
3210                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3211         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3212         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3213 }
3214
3215 static void bond_work_cancel_all(struct bonding *bond)
3216 {
3217         cancel_delayed_work_sync(&bond->mii_work);
3218         cancel_delayed_work_sync(&bond->arp_work);
3219         cancel_delayed_work_sync(&bond->alb_work);
3220         cancel_delayed_work_sync(&bond->ad_work);
3221         cancel_delayed_work_sync(&bond->mcast_work);
3222         cancel_delayed_work_sync(&bond->slave_arr_work);
3223 }
3224
3225 static int bond_open(struct net_device *bond_dev)
3226 {
3227         struct bonding *bond = netdev_priv(bond_dev);
3228         struct list_head *iter;
3229         struct slave *slave;
3230
3231         /* reset slave->backup and slave->inactive */
3232         if (bond_has_slaves(bond)) {
3233                 bond_for_each_slave(bond, slave, iter) {
3234                         if (bond_uses_primary(bond) &&
3235                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3236                                 bond_set_slave_inactive_flags(slave,
3237                                                               BOND_SLAVE_NOTIFY_NOW);
3238                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3239                                 bond_set_slave_active_flags(slave,
3240                                                             BOND_SLAVE_NOTIFY_NOW);
3241                         }
3242                 }
3243         }
3244
3245         bond_work_init_all(bond);
3246
3247         if (bond_is_lb(bond)) {
3248                 /* bond_alb_initialize must be called before the timer
3249                  * is started.
3250                  */
3251                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3252                         return -ENOMEM;
3253                 if (bond->params.tlb_dynamic_lb)
3254                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3255         }
3256
3257         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3258                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3259
3260         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3261                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3262                 bond->recv_probe = bond_arp_rcv;
3263         }
3264
3265         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3266                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3267                 /* register to receive LACPDUs */
3268                 bond->recv_probe = bond_3ad_lacpdu_recv;
3269                 bond_3ad_initiate_agg_selection(bond, 1);
3270         }
3271
3272         if (bond_mode_uses_xmit_hash(bond))
3273                 bond_update_slave_arr(bond, NULL);
3274
3275         return 0;
3276 }
3277
3278 static int bond_close(struct net_device *bond_dev)
3279 {
3280         struct bonding *bond = netdev_priv(bond_dev);
3281
3282         bond_work_cancel_all(bond);
3283         bond->send_peer_notif = 0;
3284         if (bond_is_lb(bond))
3285                 bond_alb_deinitialize(bond);
3286         bond->recv_probe = NULL;
3287
3288         return 0;
3289 }
3290
3291 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3292                                                 struct rtnl_link_stats64 *stats)
3293 {
3294         struct bonding *bond = netdev_priv(bond_dev);
3295         struct rtnl_link_stats64 temp;
3296         struct list_head *iter;
3297         struct slave *slave;
3298
3299         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3300
3301         bond_for_each_slave(bond, slave, iter) {
3302                 const struct rtnl_link_stats64 *sstats =
3303                         dev_get_stats(slave->dev, &temp);
3304                 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3305
3306                 stats->rx_packets +=  sstats->rx_packets - pstats->rx_packets;
3307                 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3308                 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3309                 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3310                 stats->rx_nohandler += sstats->rx_nohandler - pstats->rx_nohandler;
3311
3312                 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3313                 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3314                 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3315                 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3316
3317                 stats->multicast += sstats->multicast - pstats->multicast;
3318                 stats->collisions += sstats->collisions - pstats->collisions;
3319
3320                 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3321                 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3322                 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3323                 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3324                 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3325                 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3326
3327                 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3328                 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3329                 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3330                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3331                 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3332
3333                 /* save off the slave stats for the next run */
3334                 memcpy(pstats, sstats, sizeof(*sstats));
3335         }
3336         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3337
3338         return stats;
3339 }
3340
3341 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3342 {
3343         struct bonding *bond = netdev_priv(bond_dev);
3344         struct net_device *slave_dev = NULL;
3345         struct ifbond k_binfo;
3346         struct ifbond __user *u_binfo = NULL;
3347         struct ifslave k_sinfo;
3348         struct ifslave __user *u_sinfo = NULL;
3349         struct mii_ioctl_data *mii = NULL;
3350         struct bond_opt_value newval;
3351         struct net *net;
3352         int res = 0;
3353
3354         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3355
3356         switch (cmd) {
3357         case SIOCGMIIPHY:
3358                 mii = if_mii(ifr);
3359                 if (!mii)
3360                         return -EINVAL;
3361
3362                 mii->phy_id = 0;
3363                 /* Fall Through */
3364         case SIOCGMIIREG:
3365                 /* We do this again just in case we were called by SIOCGMIIREG
3366                  * instead of SIOCGMIIPHY.
3367                  */
3368                 mii = if_mii(ifr);
3369                 if (!mii)
3370                         return -EINVAL;
3371
3372                 if (mii->reg_num == 1) {
3373                         mii->val_out = 0;
3374                         if (netif_carrier_ok(bond->dev))
3375                                 mii->val_out = BMSR_LSTATUS;
3376                 }
3377
3378                 return 0;
3379         case BOND_INFO_QUERY_OLD:
3380         case SIOCBONDINFOQUERY:
3381                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3382
3383                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3384                         return -EFAULT;
3385
3386                 res = bond_info_query(bond_dev, &k_binfo);
3387                 if (res == 0 &&
3388                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3389                         return -EFAULT;
3390
3391                 return res;
3392         case BOND_SLAVE_INFO_QUERY_OLD:
3393         case SIOCBONDSLAVEINFOQUERY:
3394                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3395
3396                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3397                         return -EFAULT;
3398
3399                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3400                 if (res == 0 &&
3401                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3402                         return -EFAULT;
3403
3404                 return res;
3405         default:
3406                 break;
3407         }
3408
3409         net = dev_net(bond_dev);
3410
3411         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3412                 return -EPERM;
3413
3414         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3415
3416         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3417
3418         if (!slave_dev)
3419                 return -ENODEV;
3420
3421         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3422         switch (cmd) {
3423         case BOND_ENSLAVE_OLD:
3424         case SIOCBONDENSLAVE:
3425                 res = bond_enslave(bond_dev, slave_dev);
3426                 break;
3427         case BOND_RELEASE_OLD:
3428         case SIOCBONDRELEASE:
3429                 res = bond_release(bond_dev, slave_dev);
3430                 break;
3431         case BOND_SETHWADDR_OLD:
3432         case SIOCBONDSETHWADDR:
3433                 bond_set_dev_addr(bond_dev, slave_dev);
3434                 res = 0;
3435                 break;
3436         case BOND_CHANGE_ACTIVE_OLD:
3437         case SIOCBONDCHANGEACTIVE:
3438                 bond_opt_initstr(&newval, slave_dev->name);
3439                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3440                 break;
3441         default:
3442                 res = -EOPNOTSUPP;
3443         }
3444
3445         return res;
3446 }
3447
3448 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3449 {
3450         struct bonding *bond = netdev_priv(bond_dev);
3451
3452         if (change & IFF_PROMISC)
3453                 bond_set_promiscuity(bond,
3454                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3455
3456         if (change & IFF_ALLMULTI)
3457                 bond_set_allmulti(bond,
3458                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3459 }
3460
3461 static void bond_set_rx_mode(struct net_device *bond_dev)
3462 {
3463         struct bonding *bond = netdev_priv(bond_dev);
3464         struct list_head *iter;
3465         struct slave *slave;
3466
3467         rcu_read_lock();
3468         if (bond_uses_primary(bond)) {
3469                 slave = rcu_dereference(bond->curr_active_slave);
3470                 if (slave) {
3471                         dev_uc_sync(slave->dev, bond_dev);
3472                         dev_mc_sync(slave->dev, bond_dev);
3473                 }
3474         } else {
3475                 bond_for_each_slave_rcu(bond, slave, iter) {
3476                         dev_uc_sync_multiple(slave->dev, bond_dev);
3477                         dev_mc_sync_multiple(slave->dev, bond_dev);
3478                 }
3479         }
3480         rcu_read_unlock();
3481 }
3482
3483 static int bond_neigh_init(struct neighbour *n)
3484 {
3485         struct bonding *bond = netdev_priv(n->dev);
3486         const struct net_device_ops *slave_ops;
3487         struct neigh_parms parms;
3488         struct slave *slave;
3489         int ret;
3490
3491         slave = bond_first_slave(bond);
3492         if (!slave)
3493                 return 0;
3494         slave_ops = slave->dev->netdev_ops;
3495         if (!slave_ops->ndo_neigh_setup)
3496                 return 0;
3497
3498         parms.neigh_setup = NULL;
3499         parms.neigh_cleanup = NULL;
3500         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3501         if (ret)
3502                 return ret;
3503
3504         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3505          * after the last slave has been detached.  Assumes that all slaves
3506          * utilize the same neigh_cleanup (true at this writing as only user
3507          * is ipoib).
3508          */
3509         n->parms->neigh_cleanup = parms.neigh_cleanup;
3510
3511         if (!parms.neigh_setup)
3512                 return 0;
3513
3514         return parms.neigh_setup(n);
3515 }
3516
3517 /* The bonding ndo_neigh_setup is called at init time beofre any
3518  * slave exists. So we must declare proxy setup function which will
3519  * be used at run time to resolve the actual slave neigh param setup.
3520  *
3521  * It's also called by master devices (such as vlans) to setup their
3522  * underlying devices. In that case - do nothing, we're already set up from
3523  * our init.
3524  */
3525 static int bond_neigh_setup(struct net_device *dev,
3526                             struct neigh_parms *parms)
3527 {
3528         /* modify only our neigh_parms */
3529         if (parms->dev == dev)
3530                 parms->neigh_setup = bond_neigh_init;
3531
3532         return 0;
3533 }
3534
3535 /* Change the MTU of all of a master's slaves to match the master */
3536 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3537 {
3538         struct bonding *bond = netdev_priv(bond_dev);
3539         struct slave *slave, *rollback_slave;
3540         struct list_head *iter;
3541         int res = 0;
3542
3543         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3544
3545         bond_for_each_slave(bond, slave, iter) {
3546                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3547                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3548
3549                 res = dev_set_mtu(slave->dev, new_mtu);
3550
3551                 if (res) {
3552                         /* If we failed to set the slave's mtu to the new value
3553                          * we must abort the operation even in ACTIVE_BACKUP
3554                          * mode, because if we allow the backup slaves to have
3555                          * different mtu values than the active slave we'll
3556                          * need to change their mtu when doing a failover. That
3557                          * means changing their mtu from timer context, which
3558                          * is probably not a good idea.
3559                          */
3560                         netdev_dbg(bond_dev, "err %d %s\n", res,
3561                                    slave->dev->name);
3562                         goto unwind;
3563                 }
3564         }
3565
3566         bond_dev->mtu = new_mtu;
3567
3568         return 0;
3569
3570 unwind:
3571         /* unwind from head to the slave that failed */
3572         bond_for_each_slave(bond, rollback_slave, iter) {
3573                 int tmp_res;
3574
3575                 if (rollback_slave == slave)
3576                         break;
3577
3578                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3579                 if (tmp_res) {
3580                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3581                                    tmp_res, rollback_slave->dev->name);
3582                 }
3583         }
3584
3585         return res;
3586 }
3587
3588 /* Change HW address
3589  *
3590  * Note that many devices must be down to change the HW address, and
3591  * downing the master releases all slaves.  We can make bonds full of
3592  * bonding devices to test this, however.
3593  */
3594 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3595 {
3596         struct bonding *bond = netdev_priv(bond_dev);
3597         struct slave *slave, *rollback_slave;
3598         struct sockaddr *sa = addr, tmp_sa;
3599         struct list_head *iter;
3600         int res = 0;
3601
3602         if (BOND_MODE(bond) == BOND_MODE_ALB)
3603                 return bond_alb_set_mac_address(bond_dev, addr);
3604
3605
3606         netdev_dbg(bond_dev, "bond=%p\n", bond);
3607
3608         /* If fail_over_mac is enabled, do nothing and return success.
3609          * Returning an error causes ifenslave to fail.
3610          */
3611         if (bond->params.fail_over_mac &&
3612             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3613                 return 0;
3614
3615         if (!is_valid_ether_addr(sa->sa_data))
3616                 return -EADDRNOTAVAIL;
3617
3618         bond_for_each_slave(bond, slave, iter) {
3619                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3620                 res = dev_set_mac_address(slave->dev, addr);
3621                 if (res) {
3622                         /* TODO: consider downing the slave
3623                          * and retry ?
3624                          * User should expect communications
3625                          * breakage anyway until ARP finish
3626                          * updating, so...
3627                          */
3628                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3629                         goto unwind;
3630                 }
3631         }
3632
3633         /* success */
3634         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3635         return 0;
3636
3637 unwind:
3638         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3639         tmp_sa.sa_family = bond_dev->type;
3640
3641         /* unwind from head to the slave that failed */
3642         bond_for_each_slave(bond, rollback_slave, iter) {
3643                 int tmp_res;
3644
3645                 if (rollback_slave == slave)
3646                         break;
3647
3648                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3649                 if (tmp_res) {
3650                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3651                                    tmp_res, rollback_slave->dev->name);
3652                 }
3653         }
3654
3655         return res;
3656 }
3657
3658 /**
3659  * bond_xmit_slave_id - transmit skb through slave with slave_id
3660  * @bond: bonding device that is transmitting
3661  * @skb: buffer to transmit
3662  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3663  *
3664  * This function tries to transmit through slave with slave_id but in case
3665  * it fails, it tries to find the first available slave for transmission.
3666  * The skb is consumed in all cases, thus the function is void.
3667  */
3668 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3669 {
3670         struct list_head *iter;
3671         struct slave *slave;
3672         int i = slave_id;
3673
3674         /* Here we start from the slave with slave_id */
3675         bond_for_each_slave_rcu(bond, slave, iter) {
3676                 if (--i < 0) {
3677                         if (bond_slave_can_tx(slave)) {
3678                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3679                                 return;
3680                         }
3681                 }
3682         }
3683
3684         /* Here we start from the first slave up to slave_id */
3685         i = slave_id;
3686         bond_for_each_slave_rcu(bond, slave, iter) {
3687                 if (--i < 0)
3688                         break;
3689                 if (bond_slave_can_tx(slave)) {
3690                         bond_dev_queue_xmit(bond, skb, slave->dev);
3691                         return;
3692                 }
3693         }
3694         /* no slave that can tx has been found */
3695         bond_tx_drop(bond->dev, skb);
3696 }
3697
3698 /**
3699  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3700  * @bond: bonding device to use
3701  *
3702  * Based on the value of the bonding device's packets_per_slave parameter
3703  * this function generates a slave id, which is usually used as the next
3704  * slave to transmit through.
3705  */
3706 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3707 {
3708         u32 slave_id;
3709         struct reciprocal_value reciprocal_packets_per_slave;
3710         int packets_per_slave = bond->params.packets_per_slave;
3711
3712         switch (packets_per_slave) {
3713         case 0:
3714                 slave_id = prandom_u32();
3715                 break;
3716         case 1:
3717                 slave_id = bond->rr_tx_counter;
3718                 break;
3719         default:
3720                 reciprocal_packets_per_slave =
3721                         bond->params.reciprocal_packets_per_slave;
3722                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3723                                              reciprocal_packets_per_slave);
3724                 break;
3725         }
3726         bond->rr_tx_counter++;
3727
3728         return slave_id;
3729 }
3730
3731 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3732 {
3733         struct bonding *bond = netdev_priv(bond_dev);
3734         struct iphdr *iph = ip_hdr(skb);
3735         struct slave *slave;
3736         u32 slave_id;
3737
3738         /* Start with the curr_active_slave that joined the bond as the
3739          * default for sending IGMP traffic.  For failover purposes one
3740          * needs to maintain some consistency for the interface that will
3741          * send the join/membership reports.  The curr_active_slave found
3742          * will send all of this type of traffic.
3743          */
3744         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3745                 slave = rcu_dereference(bond->curr_active_slave);
3746                 if (slave)
3747                         bond_dev_queue_xmit(bond, skb, slave->dev);
3748                 else
3749                         bond_xmit_slave_id(bond, skb, 0);
3750         } else {
3751                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3752
3753                 if (likely(slave_cnt)) {
3754                         slave_id = bond_rr_gen_slave_id(bond);
3755                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3756                 } else {
3757                         bond_tx_drop(bond_dev, skb);
3758                 }
3759         }
3760
3761         return NETDEV_TX_OK;
3762 }
3763
3764 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3765  * the bond has a usable interface.
3766  */
3767 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3768 {
3769         struct bonding *bond = netdev_priv(bond_dev);
3770         struct slave *slave;
3771
3772         slave = rcu_dereference(bond->curr_active_slave);
3773         if (slave)
3774                 bond_dev_queue_xmit(bond, skb, slave->dev);
3775         else
3776                 bond_tx_drop(bond_dev, skb);
3777
3778         return NETDEV_TX_OK;
3779 }
3780
3781 /* Use this to update slave_array when (a) it's not appropriate to update
3782  * slave_array right away (note that update_slave_array() may sleep)
3783  * and / or (b) RTNL is not held.
3784  */
3785 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3786 {
3787         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3788 }
3789
3790 /* Slave array work handler. Holds only RTNL */
3791 static void bond_slave_arr_handler(struct work_struct *work)
3792 {
3793         struct bonding *bond = container_of(work, struct bonding,
3794                                             slave_arr_work.work);
3795         int ret;
3796
3797         if (!rtnl_trylock())
3798                 goto err;
3799
3800         ret = bond_update_slave_arr(bond, NULL);
3801         rtnl_unlock();
3802         if (ret) {
3803                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3804                 goto err;
3805         }
3806         return;
3807
3808 err:
3809         bond_slave_arr_work_rearm(bond, 1);
3810 }
3811
3812 /* Build the usable slaves array in control path for modes that use xmit-hash
3813  * to determine the slave interface -
3814  * (a) BOND_MODE_8023AD
3815  * (b) BOND_MODE_XOR
3816  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3817  *
3818  * The caller is expected to hold RTNL only and NO other lock!
3819  */
3820 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3821 {
3822         struct slave *slave;
3823         struct list_head *iter;
3824         struct bond_up_slave *new_arr, *old_arr;
3825         int agg_id = 0;
3826         int ret = 0;
3827
3828 #ifdef CONFIG_LOCKDEP
3829         WARN_ON(lockdep_is_held(&bond->mode_lock));
3830 #endif
3831
3832         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3833                           GFP_KERNEL);
3834         if (!new_arr) {
3835                 ret = -ENOMEM;
3836                 pr_err("Failed to build slave-array.\n");
3837                 goto out;
3838         }
3839         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3840                 struct ad_info ad_info;
3841
3842                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3843                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3844                         kfree_rcu(new_arr, rcu);
3845                         /* No active aggragator means it's not safe to use
3846                          * the previous array.
3847                          */
3848                         old_arr = rtnl_dereference(bond->slave_arr);
3849                         if (old_arr) {
3850                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3851                                 kfree_rcu(old_arr, rcu);
3852                         }
3853                         goto out;
3854                 }
3855                 agg_id = ad_info.aggregator_id;
3856         }
3857         bond_for_each_slave(bond, slave, iter) {
3858                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3859                         struct aggregator *agg;
3860
3861                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3862                         if (!agg || agg->aggregator_identifier != agg_id)
3863                                 continue;
3864                 }
3865                 if (!bond_slave_can_tx(slave))
3866                         continue;
3867                 if (skipslave == slave)
3868                         continue;
3869                 new_arr->arr[new_arr->count++] = slave;
3870         }
3871
3872         old_arr = rtnl_dereference(bond->slave_arr);
3873         rcu_assign_pointer(bond->slave_arr, new_arr);
3874         if (old_arr)
3875                 kfree_rcu(old_arr, rcu);
3876 out:
3877         if (ret != 0 && skipslave) {
3878                 int idx;
3879
3880                 /* Rare situation where caller has asked to skip a specific
3881                  * slave but allocation failed (most likely!). BTW this is
3882                  * only possible when the call is initiated from
3883                  * __bond_release_one(). In this situation; overwrite the
3884                  * skipslave entry in the array with the last entry from the
3885                  * array to avoid a situation where the xmit path may choose
3886                  * this to-be-skipped slave to send a packet out.
3887                  */
3888                 old_arr = rtnl_dereference(bond->slave_arr);
3889                 for (idx = 0; idx < old_arr->count; idx++) {
3890                         if (skipslave == old_arr->arr[idx]) {
3891                                 old_arr->arr[idx] =
3892                                     old_arr->arr[old_arr->count-1];
3893                                 old_arr->count--;
3894                                 break;
3895                         }
3896                 }
3897         }
3898         return ret;
3899 }
3900
3901 /* Use this Xmit function for 3AD as well as XOR modes. The current
3902  * usable slave array is formed in the control path. The xmit function
3903  * just calculates hash and sends the packet out.
3904  */
3905 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3906 {
3907         struct bonding *bond = netdev_priv(dev);
3908         struct slave *slave;
3909         struct bond_up_slave *slaves;
3910         unsigned int count;
3911
3912         slaves = rcu_dereference(bond->slave_arr);
3913         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3914         if (likely(count)) {
3915                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3916                 bond_dev_queue_xmit(bond, skb, slave->dev);
3917         } else {
3918                 bond_tx_drop(dev, skb);
3919         }
3920
3921         return NETDEV_TX_OK;
3922 }
3923
3924 /* in broadcast mode, we send everything to all usable interfaces. */
3925 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3926 {
3927         struct bonding *bond = netdev_priv(bond_dev);
3928         struct slave *slave = NULL;
3929         struct list_head *iter;
3930
3931         bond_for_each_slave_rcu(bond, slave, iter) {
3932                 if (bond_is_last_slave(bond, slave))
3933                         break;
3934                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3935                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3936
3937                         if (!skb2) {
3938                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3939                                                     bond_dev->name, __func__);
3940                                 continue;
3941                         }
3942                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3943                 }
3944         }
3945         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3946                 bond_dev_queue_xmit(bond, skb, slave->dev);
3947         else
3948                 bond_tx_drop(bond_dev, skb);
3949
3950         return NETDEV_TX_OK;
3951 }
3952
3953 /*------------------------- Device initialization ---------------------------*/
3954
3955 /* Lookup the slave that corresponds to a qid */
3956 static inline int bond_slave_override(struct bonding *bond,
3957                                       struct sk_buff *skb)
3958 {
3959         struct slave *slave = NULL;
3960         struct list_head *iter;
3961
3962         if (!skb->queue_mapping)
3963                 return 1;
3964
3965         /* Find out if any slaves have the same mapping as this skb. */
3966         bond_for_each_slave_rcu(bond, slave, iter) {
3967                 if (slave->queue_id == skb->queue_mapping) {
3968                         if (bond_slave_is_up(slave) &&
3969                             slave->link == BOND_LINK_UP) {
3970                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3971                                 return 0;
3972                         }
3973                         /* If the slave isn't UP, use default transmit policy. */
3974                         break;
3975                 }
3976         }
3977
3978         return 1;
3979 }
3980
3981
3982 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3983                              void *accel_priv, select_queue_fallback_t fallback)
3984 {
3985         /* This helper function exists to help dev_pick_tx get the correct
3986          * destination queue.  Using a helper function skips a call to
3987          * skb_tx_hash and will put the skbs in the queue we expect on their
3988          * way down to the bonding driver.
3989          */
3990         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3991
3992         /* Save the original txq to restore before passing to the driver */
3993         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3994
3995         if (unlikely(txq >= dev->real_num_tx_queues)) {
3996                 do {
3997                         txq -= dev->real_num_tx_queues;
3998                 } while (txq >= dev->real_num_tx_queues);
3999         }
4000         return txq;
4001 }
4002
4003 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4004 {
4005         struct bonding *bond = netdev_priv(dev);
4006
4007         if (bond_should_override_tx_queue(bond) &&
4008             !bond_slave_override(bond, skb))
4009                 return NETDEV_TX_OK;
4010
4011         switch (BOND_MODE(bond)) {
4012         case BOND_MODE_ROUNDROBIN:
4013                 return bond_xmit_roundrobin(skb, dev);
4014         case BOND_MODE_ACTIVEBACKUP:
4015                 return bond_xmit_activebackup(skb, dev);
4016         case BOND_MODE_8023AD:
4017         case BOND_MODE_XOR:
4018                 return bond_3ad_xor_xmit(skb, dev);
4019         case BOND_MODE_BROADCAST:
4020                 return bond_xmit_broadcast(skb, dev);
4021         case BOND_MODE_ALB:
4022                 return bond_alb_xmit(skb, dev);
4023         case BOND_MODE_TLB:
4024                 return bond_tlb_xmit(skb, dev);
4025         default:
4026                 /* Should never happen, mode already checked */
4027                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4028                 WARN_ON_ONCE(1);
4029                 bond_tx_drop(dev, skb);
4030                 return NETDEV_TX_OK;
4031         }
4032 }
4033
4034 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4035 {
4036         struct bonding *bond = netdev_priv(dev);
4037         netdev_tx_t ret = NETDEV_TX_OK;
4038
4039         /* If we risk deadlock from transmitting this in the
4040          * netpoll path, tell netpoll to queue the frame for later tx
4041          */
4042         if (unlikely(is_netpoll_tx_blocked(dev)))
4043                 return NETDEV_TX_BUSY;
4044
4045         rcu_read_lock();
4046         if (bond_has_slaves(bond))
4047                 ret = __bond_start_xmit(skb, dev);
4048         else
4049                 bond_tx_drop(dev, skb);
4050         rcu_read_unlock();
4051
4052         return ret;
4053 }
4054
4055 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4056                                      struct ethtool_cmd *ecmd)
4057 {
4058         struct bonding *bond = netdev_priv(bond_dev);
4059         unsigned long speed = 0;
4060         struct list_head *iter;
4061         struct slave *slave;
4062
4063         ecmd->duplex = DUPLEX_UNKNOWN;
4064         ecmd->port = PORT_OTHER;
4065
4066         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4067          * do not need to check mode.  Though link speed might not represent
4068          * the true receive or transmit bandwidth (not all modes are symmetric)
4069          * this is an accurate maximum.
4070          */
4071         bond_for_each_slave(bond, slave, iter) {
4072                 if (bond_slave_can_tx(slave)) {
4073                         if (slave->speed != SPEED_UNKNOWN)
4074                                 speed += slave->speed;
4075                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4076                             slave->duplex != DUPLEX_UNKNOWN)
4077                                 ecmd->duplex = slave->duplex;
4078                 }
4079         }
4080         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4081
4082         return 0;
4083 }
4084
4085 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4086                                      struct ethtool_drvinfo *drvinfo)
4087 {
4088         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4089         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4090         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4091                  BOND_ABI_VERSION);
4092 }
4093
4094 static const struct ethtool_ops bond_ethtool_ops = {
4095         .get_drvinfo            = bond_ethtool_get_drvinfo,
4096         .get_settings           = bond_ethtool_get_settings,
4097         .get_link               = ethtool_op_get_link,
4098 };
4099
4100 static const struct net_device_ops bond_netdev_ops = {
4101         .ndo_init               = bond_init,
4102         .ndo_uninit             = bond_uninit,
4103         .ndo_open               = bond_open,
4104         .ndo_stop               = bond_close,
4105         .ndo_start_xmit         = bond_start_xmit,
4106         .ndo_select_queue       = bond_select_queue,
4107         .ndo_get_stats64        = bond_get_stats,
4108         .ndo_do_ioctl           = bond_do_ioctl,
4109         .ndo_change_rx_flags    = bond_change_rx_flags,
4110         .ndo_set_rx_mode        = bond_set_rx_mode,
4111         .ndo_change_mtu         = bond_change_mtu,
4112         .ndo_set_mac_address    = bond_set_mac_address,
4113         .ndo_neigh_setup        = bond_neigh_setup,
4114         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4115         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4116 #ifdef CONFIG_NET_POLL_CONTROLLER
4117         .ndo_netpoll_setup      = bond_netpoll_setup,
4118         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4119         .ndo_poll_controller    = bond_poll_controller,
4120 #endif
4121         .ndo_add_slave          = bond_enslave,
4122         .ndo_del_slave          = bond_release,
4123         .ndo_fix_features       = bond_fix_features,
4124         .ndo_bridge_setlink     = switchdev_port_bridge_setlink,
4125         .ndo_bridge_getlink     = switchdev_port_bridge_getlink,
4126         .ndo_bridge_dellink     = switchdev_port_bridge_dellink,
4127         .ndo_fdb_add            = switchdev_port_fdb_add,
4128         .ndo_fdb_del            = switchdev_port_fdb_del,
4129         .ndo_fdb_dump           = switchdev_port_fdb_dump,
4130         .ndo_features_check     = passthru_features_check,
4131 };
4132
4133 static const struct device_type bond_type = {
4134         .name = "bond",
4135 };
4136
4137 static void bond_destructor(struct net_device *bond_dev)
4138 {
4139         struct bonding *bond = netdev_priv(bond_dev);
4140         if (bond->wq)
4141                 destroy_workqueue(bond->wq);
4142         free_netdev(bond_dev);
4143 }
4144
4145 void bond_setup(struct net_device *bond_dev)
4146 {
4147         struct bonding *bond = netdev_priv(bond_dev);
4148
4149         spin_lock_init(&bond->mode_lock);
4150         bond->params = bonding_defaults;
4151
4152         /* Initialize pointers */
4153         bond->dev = bond_dev;
4154
4155         /* Initialize the device entry points */
4156         ether_setup(bond_dev);
4157         bond_dev->netdev_ops = &bond_netdev_ops;
4158         bond_dev->ethtool_ops = &bond_ethtool_ops;
4159
4160         bond_dev->destructor = bond_destructor;
4161
4162         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4163
4164         /* Initialize the device options */
4165         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4166         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4167         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4168
4169         /* don't acquire bond device's netif_tx_lock when transmitting */
4170         bond_dev->features |= NETIF_F_LLTX;
4171
4172         /* By default, we declare the bond to be fully
4173          * VLAN hardware accelerated capable. Special
4174          * care is taken in the various xmit functions
4175          * when there are slaves that are not hw accel
4176          * capable
4177          */
4178
4179         /* Don't allow bond devices to change network namespaces. */
4180         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4181
4182         bond_dev->hw_features = BOND_VLAN_FEATURES |
4183                                 NETIF_F_HW_VLAN_CTAG_TX |
4184                                 NETIF_F_HW_VLAN_CTAG_RX |
4185                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4186
4187         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4188         bond_dev->features |= bond_dev->hw_features;
4189 }
4190
4191 /* Destroy a bonding device.
4192  * Must be under rtnl_lock when this function is called.
4193  */
4194 static void bond_uninit(struct net_device *bond_dev)
4195 {
4196         struct bonding *bond = netdev_priv(bond_dev);
4197         struct list_head *iter;
4198         struct slave *slave;
4199         struct bond_up_slave *arr;
4200
4201         bond_netpoll_cleanup(bond_dev);
4202
4203         /* Release the bonded slaves */
4204         bond_for_each_slave(bond, slave, iter)
4205                 __bond_release_one(bond_dev, slave->dev, true);
4206         netdev_info(bond_dev, "Released all slaves\n");
4207
4208         arr = rtnl_dereference(bond->slave_arr);
4209         if (arr) {
4210                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4211                 kfree_rcu(arr, rcu);
4212         }
4213
4214         list_del(&bond->bond_list);
4215
4216         bond_debug_unregister(bond);
4217 }
4218
4219 /*------------------------- Module initialization ---------------------------*/
4220
4221 static int bond_check_params(struct bond_params *params)
4222 {
4223         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4224         struct bond_opt_value newval;
4225         const struct bond_opt_value *valptr;
4226         int arp_all_targets_value;
4227         u16 ad_actor_sys_prio = 0;
4228         u16 ad_user_port_key = 0;
4229
4230         /* Convert string parameters. */
4231         if (mode) {
4232                 bond_opt_initstr(&newval, mode);
4233                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4234                 if (!valptr) {
4235                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4236                         return -EINVAL;
4237                 }
4238                 bond_mode = valptr->value;
4239         }
4240
4241         if (xmit_hash_policy) {
4242                 if ((bond_mode != BOND_MODE_XOR) &&
4243                     (bond_mode != BOND_MODE_8023AD) &&
4244                     (bond_mode != BOND_MODE_TLB)) {
4245                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4246                                 bond_mode_name(bond_mode));
4247                 } else {
4248                         bond_opt_initstr(&newval, xmit_hash_policy);
4249                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4250                                                 &newval);
4251                         if (!valptr) {
4252                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4253                                        xmit_hash_policy);
4254                                 return -EINVAL;
4255                         }
4256                         xmit_hashtype = valptr->value;
4257                 }
4258         }
4259
4260         if (lacp_rate) {
4261                 if (bond_mode != BOND_MODE_8023AD) {
4262                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4263                                 bond_mode_name(bond_mode));
4264                 } else {
4265                         bond_opt_initstr(&newval, lacp_rate);
4266                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4267                                                 &newval);
4268                         if (!valptr) {
4269                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4270                                        lacp_rate);
4271                                 return -EINVAL;
4272                         }
4273                         lacp_fast = valptr->value;
4274                 }
4275         }
4276
4277         if (ad_select) {
4278                 bond_opt_initstr(&newval, ad_select);
4279                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4280                                         &newval);
4281                 if (!valptr) {
4282                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4283                         return -EINVAL;
4284                 }
4285                 params->ad_select = valptr->value;
4286                 if (bond_mode != BOND_MODE_8023AD)
4287                         pr_warn("ad_select param only affects 802.3ad mode\n");
4288         } else {
4289                 params->ad_select = BOND_AD_STABLE;
4290         }
4291
4292         if (max_bonds < 0) {
4293                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4294                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4295                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4296         }
4297
4298         if (miimon < 0) {
4299                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4300                         miimon, INT_MAX);
4301                 miimon = 0;
4302         }
4303
4304         if (updelay < 0) {
4305                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4306                         updelay, INT_MAX);
4307                 updelay = 0;
4308         }
4309
4310         if (downdelay < 0) {
4311                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4312                         downdelay, INT_MAX);
4313                 downdelay = 0;
4314         }
4315
4316         if ((use_carrier != 0) && (use_carrier != 1)) {
4317                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4318                         use_carrier);
4319                 use_carrier = 1;
4320         }
4321
4322         if (num_peer_notif < 0 || num_peer_notif > 255) {
4323                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4324                         num_peer_notif);
4325                 num_peer_notif = 1;
4326         }
4327
4328         /* reset values for 802.3ad/TLB/ALB */
4329         if (!bond_mode_uses_arp(bond_mode)) {
4330                 if (!miimon) {
4331                         pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4332                         pr_warn("Forcing miimon to 100msec\n");
4333                         miimon = BOND_DEFAULT_MIIMON;
4334                 }
4335         }
4336
4337         if (tx_queues < 1 || tx_queues > 255) {
4338                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4339                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4340                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4341         }
4342
4343         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4344                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4345                         all_slaves_active);
4346                 all_slaves_active = 0;
4347         }
4348
4349         if (resend_igmp < 0 || resend_igmp > 255) {
4350                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4351                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4352                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4353         }
4354
4355         bond_opt_initval(&newval, packets_per_slave);
4356         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4357                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4358                         packets_per_slave, USHRT_MAX);
4359                 packets_per_slave = 1;
4360         }
4361
4362         if (bond_mode == BOND_MODE_ALB) {
4363                 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",
4364                           updelay);
4365         }
4366
4367         if (!miimon) {
4368                 if (updelay || downdelay) {
4369                         /* just warn the user the up/down delay will have
4370                          * no effect since miimon is zero...
4371                          */
4372                         pr_warn("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",
4373                                 updelay, downdelay);
4374                 }
4375         } else {
4376                 /* don't allow arp monitoring */
4377                 if (arp_interval) {
4378                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4379                                 miimon, arp_interval);
4380                         arp_interval = 0;
4381                 }
4382
4383                 if ((updelay % miimon) != 0) {
4384                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4385                                 updelay, miimon, (updelay / miimon) * miimon);
4386                 }
4387
4388                 updelay /= miimon;
4389
4390                 if ((downdelay % miimon) != 0) {
4391                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4392                                 downdelay, miimon,
4393                                 (downdelay / miimon) * miimon);
4394                 }
4395
4396                 downdelay /= miimon;
4397         }
4398
4399         if (arp_interval < 0) {
4400                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4401                         arp_interval, INT_MAX);
4402                 arp_interval = 0;
4403         }
4404
4405         for (arp_ip_count = 0, i = 0;
4406              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4407                 __be32 ip;
4408
4409                 /* not a complete check, but good enough to catch mistakes */
4410                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4411                     !bond_is_ip_target_ok(ip)) {
4412                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4413                                 arp_ip_target[i]);
4414                         arp_interval = 0;
4415                 } else {
4416                         if (bond_get_targets_ip(arp_target, ip) == -1)
4417                                 arp_target[arp_ip_count++] = ip;
4418                         else
4419                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4420                                         &ip);
4421                 }
4422         }
4423
4424         if (arp_interval && !arp_ip_count) {
4425                 /* don't allow arping if no arp_ip_target given... */
4426                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4427                         arp_interval);
4428                 arp_interval = 0;
4429         }
4430
4431         if (arp_validate) {
4432                 if (!arp_interval) {
4433                         pr_err("arp_validate requires arp_interval\n");
4434                         return -EINVAL;
4435                 }
4436
4437                 bond_opt_initstr(&newval, arp_validate);
4438                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4439                                         &newval);
4440                 if (!valptr) {
4441                         pr_err("Error: invalid arp_validate \"%s\"\n",
4442                                arp_validate);
4443                         return -EINVAL;
4444                 }
4445                 arp_validate_value = valptr->value;
4446         } else {
4447                 arp_validate_value = 0;
4448         }
4449
4450         arp_all_targets_value = 0;
4451         if (arp_all_targets) {
4452                 bond_opt_initstr(&newval, arp_all_targets);
4453                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4454                                         &newval);
4455                 if (!valptr) {
4456                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4457                                arp_all_targets);
4458                         arp_all_targets_value = 0;
4459                 } else {
4460                         arp_all_targets_value = valptr->value;
4461                 }
4462         }
4463
4464         if (miimon) {
4465                 pr_info("MII link monitoring set to %d ms\n", miimon);
4466         } else if (arp_interval) {
4467                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4468                                           arp_validate_value);
4469                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4470                         arp_interval, valptr->string, arp_ip_count);
4471
4472                 for (i = 0; i < arp_ip_count; i++)
4473                         pr_cont(" %s", arp_ip_target[i]);
4474
4475                 pr_cont("\n");
4476
4477         } else if (max_bonds) {
4478                 /* miimon and arp_interval not set, we need one so things
4479                  * work as expected, see bonding.txt for details
4480                  */
4481                 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");
4482         }
4483
4484         if (primary && !bond_mode_uses_primary(bond_mode)) {
4485                 /* currently, using a primary only makes sense
4486                  * in active backup, TLB or ALB modes
4487                  */
4488                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4489                         primary, bond_mode_name(bond_mode));
4490                 primary = NULL;
4491         }
4492
4493         if (primary && primary_reselect) {
4494                 bond_opt_initstr(&newval, primary_reselect);
4495                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4496                                         &newval);
4497                 if (!valptr) {
4498                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4499                                primary_reselect);
4500                         return -EINVAL;
4501                 }
4502                 primary_reselect_value = valptr->value;
4503         } else {
4504                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4505         }
4506
4507         if (fail_over_mac) {
4508                 bond_opt_initstr(&newval, fail_over_mac);
4509                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4510                                         &newval);
4511                 if (!valptr) {
4512                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4513                                fail_over_mac);
4514                         return -EINVAL;
4515                 }
4516                 fail_over_mac_value = valptr->value;
4517                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4518                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4519         } else {
4520                 fail_over_mac_value = BOND_FOM_NONE;
4521         }
4522
4523         bond_opt_initstr(&newval, "default");
4524         valptr = bond_opt_parse(
4525                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4526                                      &newval);
4527         if (!valptr) {
4528                 pr_err("Error: No ad_actor_sys_prio default value");
4529                 return -EINVAL;
4530         }
4531         ad_actor_sys_prio = valptr->value;
4532
4533         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4534                                 &newval);
4535         if (!valptr) {
4536                 pr_err("Error: No ad_user_port_key default value");
4537                 return -EINVAL;
4538         }
4539         ad_user_port_key = valptr->value;
4540
4541         if (lp_interval == 0) {
4542                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4543                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4544                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4545         }
4546
4547         /* fill params struct with the proper values */
4548         params->mode = bond_mode;
4549         params->xmit_policy = xmit_hashtype;
4550         params->miimon = miimon;
4551         params->num_peer_notif = num_peer_notif;
4552         params->arp_interval = arp_interval;
4553         params->arp_validate = arp_validate_value;
4554         params->arp_all_targets = arp_all_targets_value;
4555         params->updelay = updelay;
4556         params->downdelay = downdelay;
4557         params->use_carrier = use_carrier;
4558         params->lacp_fast = lacp_fast;
4559         params->primary[0] = 0;
4560         params->primary_reselect = primary_reselect_value;
4561         params->fail_over_mac = fail_over_mac_value;
4562         params->tx_queues = tx_queues;
4563         params->all_slaves_active = all_slaves_active;
4564         params->resend_igmp = resend_igmp;
4565         params->min_links = min_links;
4566         params->lp_interval = lp_interval;
4567         params->packets_per_slave = packets_per_slave;
4568         params->tlb_dynamic_lb = 1; /* Default value */
4569         params->ad_actor_sys_prio = ad_actor_sys_prio;
4570         eth_zero_addr(params->ad_actor_system);
4571         params->ad_user_port_key = ad_user_port_key;
4572         if (packets_per_slave > 0) {
4573                 params->reciprocal_packets_per_slave =
4574                         reciprocal_value(packets_per_slave);
4575         } else {
4576                 /* reciprocal_packets_per_slave is unused if
4577                  * packets_per_slave is 0 or 1, just initialize it
4578                  */
4579                 params->reciprocal_packets_per_slave =
4580                         (struct reciprocal_value) { 0 };
4581         }
4582
4583         if (primary) {
4584                 strncpy(params->primary, primary, IFNAMSIZ);
4585                 params->primary[IFNAMSIZ - 1] = 0;
4586         }
4587
4588         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4589
4590         return 0;
4591 }
4592
4593 static struct lock_class_key bonding_netdev_xmit_lock_key;
4594 static struct lock_class_key bonding_netdev_addr_lock_key;
4595 static struct lock_class_key bonding_tx_busylock_key;
4596
4597 static void bond_set_lockdep_class_one(struct net_device *dev,
4598                                        struct netdev_queue *txq,
4599                                        void *_unused)
4600 {
4601         lockdep_set_class(&txq->_xmit_lock,
4602                           &bonding_netdev_xmit_lock_key);
4603 }
4604
4605 static void bond_set_lockdep_class(struct net_device *dev)
4606 {
4607         lockdep_set_class(&dev->addr_list_lock,
4608                           &bonding_netdev_addr_lock_key);
4609         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4610         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4611 }
4612
4613 /* Called from registration process */
4614 static int bond_init(struct net_device *bond_dev)
4615 {
4616         struct bonding *bond = netdev_priv(bond_dev);
4617         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4618
4619         netdev_dbg(bond_dev, "Begin bond_init\n");
4620
4621         bond->wq = create_singlethread_workqueue(bond_dev->name);
4622         if (!bond->wq)
4623                 return -ENOMEM;
4624
4625         bond_set_lockdep_class(bond_dev);
4626
4627         list_add_tail(&bond->bond_list, &bn->dev_list);
4628
4629         bond_prepare_sysfs_group(bond);
4630
4631         bond_debug_register(bond);
4632
4633         /* Ensure valid dev_addr */
4634         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4635             bond_dev->addr_assign_type == NET_ADDR_PERM)
4636                 eth_hw_addr_random(bond_dev);
4637
4638         return 0;
4639 }
4640
4641 unsigned int bond_get_num_tx_queues(void)
4642 {
4643         return tx_queues;
4644 }
4645
4646 /* Create a new bond based on the specified name and bonding parameters.
4647  * If name is NULL, obtain a suitable "bond%d" name for us.
4648  * Caller must NOT hold rtnl_lock; we need to release it here before we
4649  * set up our sysfs entries.
4650  */
4651 int bond_create(struct net *net, const char *name)
4652 {
4653         struct net_device *bond_dev;
4654         struct bonding *bond;
4655         struct alb_bond_info *bond_info;
4656         int res;
4657
4658         rtnl_lock();
4659
4660         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4661                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4662                                    bond_setup, tx_queues);
4663         if (!bond_dev) {
4664                 pr_err("%s: eek! can't alloc netdev!\n", name);
4665                 rtnl_unlock();
4666                 return -ENOMEM;
4667         }
4668
4669         /*
4670          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4671          * It is set to 0 by default which is wrong.
4672          */
4673         bond = netdev_priv(bond_dev);
4674         bond_info = &(BOND_ALB_INFO(bond));
4675         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4676
4677         dev_net_set(bond_dev, net);
4678         bond_dev->rtnl_link_ops = &bond_link_ops;
4679
4680         res = register_netdevice(bond_dev);
4681
4682         netif_carrier_off(bond_dev);
4683
4684         rtnl_unlock();
4685         if (res < 0)
4686                 bond_destructor(bond_dev);
4687         return res;
4688 }
4689
4690 static int __net_init bond_net_init(struct net *net)
4691 {
4692         struct bond_net *bn = net_generic(net, bond_net_id);
4693
4694         bn->net = net;
4695         INIT_LIST_HEAD(&bn->dev_list);
4696
4697         bond_create_proc_dir(bn);
4698         bond_create_sysfs(bn);
4699
4700         return 0;
4701 }
4702
4703 static void __net_exit bond_net_exit(struct net *net)
4704 {
4705         struct bond_net *bn = net_generic(net, bond_net_id);
4706         struct bonding *bond, *tmp_bond;
4707         LIST_HEAD(list);
4708
4709         bond_destroy_sysfs(bn);
4710
4711         /* Kill off any bonds created after unregistering bond rtnl ops */
4712         rtnl_lock();
4713         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4714                 unregister_netdevice_queue(bond->dev, &list);
4715         unregister_netdevice_many(&list);
4716         rtnl_unlock();
4717
4718         bond_destroy_proc_dir(bn);
4719 }
4720
4721 static struct pernet_operations bond_net_ops = {
4722         .init = bond_net_init,
4723         .exit = bond_net_exit,
4724         .id   = &bond_net_id,
4725         .size = sizeof(struct bond_net),
4726 };
4727
4728 static int __init bonding_init(void)
4729 {
4730         int i;
4731         int res;
4732
4733         pr_info("%s", bond_version);
4734
4735         res = bond_check_params(&bonding_defaults);
4736         if (res)
4737                 goto out;
4738
4739         res = register_pernet_subsys(&bond_net_ops);
4740         if (res)
4741                 goto out;
4742
4743         res = bond_netlink_init();
4744         if (res)
4745                 goto err_link;
4746
4747         bond_create_debugfs();
4748
4749         for (i = 0; i < max_bonds; i++) {
4750                 res = bond_create(&init_net, NULL);
4751                 if (res)
4752                         goto err;
4753         }
4754
4755         register_netdevice_notifier(&bond_netdev_notifier);
4756 out:
4757         return res;
4758 err:
4759         bond_destroy_debugfs();
4760         bond_netlink_fini();
4761 err_link:
4762         unregister_pernet_subsys(&bond_net_ops);
4763         goto out;
4764
4765 }
4766
4767 static void __exit bonding_exit(void)
4768 {
4769         unregister_netdevice_notifier(&bond_netdev_notifier);
4770
4771         bond_destroy_debugfs();
4772
4773         bond_netlink_fini();
4774         unregister_pernet_subsys(&bond_net_ops);
4775
4776 #ifdef CONFIG_NET_POLL_CONTROLLER
4777         /* Make sure we don't have an imbalance on our netpoll blocking */
4778         WARN_ON(atomic_read(&netpoll_block_tx));
4779 #endif
4780 }
4781
4782 module_init(bonding_init);
4783 module_exit(bonding_exit);
4784 MODULE_LICENSE("GPL");
4785 MODULE_VERSION(DRV_VERSION);
4786 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4787 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");