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, "802.3ad 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-alb, balance-tlb, balance-xor, 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 unsigned int bond_net_id __read_mostly;
 203
 204 /*-------------------------- Forward declarations ---------------------------*/
 205
 206 static int bond_init(struct net_device *bond_dev);
 207 static void bond_uninit(struct net_device *bond_dev);
 208 static void bond_get_stats(struct net_device *bond_dev,
 209                            struct rtnl_link_stats64 *stats);
 210 static void bond_slave_arr_handler(struct work_struct *work);
 211 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 212                                   int mod);
 213
 214 /*---------------------------- General routines -----------------------------*/
 215
 216 const char *bond_mode_name(int mode)
 217 {
 218         static const char *names[] = {
 219                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 220                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 221                 [BOND_MODE_XOR] = "load balancing (xor)",
 222                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 223                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 224                 [BOND_MODE_TLB] = "transmit load balancing",
 225                 [BOND_MODE_ALB] = "adaptive load balancing",
 226         };
 227
 228         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 229                 return "unknown";
 230
 231         return names[mode];
 232 }
 233
 234 /*---------------------------------- VLAN -----------------------------------*/
 235
 236 /**
 237  * bond_dev_queue_xmit - Prepare skb for xmit.
 238  *
 239  * @bond: bond device that got this skb for tx.
 240  * @skb: hw accel VLAN tagged skb to transmit
 241  * @slave_dev: slave that is supposed to xmit this skbuff
 242  */
 243 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 244                         struct net_device *slave_dev)
 245 {
 246         skb->dev = slave_dev;
 247
 248         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 249                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 250         skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 251
 252         if (unlikely(netpoll_tx_running(bond->dev)))
 253                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 254         else
 255                 dev_queue_xmit(skb);
 256 }
 257
 258 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 259  * We don't protect the slave list iteration with a lock because:
 260  * a. This operation is performed in IOCTL context,
 261  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 262  * c. Holding a lock with BH disabled while directly calling a base driver
 263  *    entry point is generally a BAD idea.
 264  *
 265  * The design of synchronization/protection for this operation in the 8021q
 266  * module is good for one or more VLAN devices over a single physical device
 267  * and cannot be extended for a teaming solution like bonding, so there is a
 268  * potential race condition here where a net device from the vlan group might
 269  * be referenced (either by a base driver or the 8021q code) while it is being
 270  * removed from the system. However, it turns out we're not making matters
 271  * worse, and if it works for regular VLAN usage it will work here too.
 272 */
 273
 274 /**
 275  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 276  * @bond_dev: bonding net device that got called
 277  * @vid: vlan id being added
 278  */
 279 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 280                                 __be16 proto, u16 vid)
 281 {
 282         struct bonding *bond = netdev_priv(bond_dev);
 283         struct slave *slave, *rollback_slave;
 284         struct list_head *iter;
 285         int res;
 286
 287         bond_for_each_slave(bond, slave, iter) {
 288                 res = vlan_vid_add(slave->dev, proto, vid);
 289                 if (res)
 290                         goto unwind;
 291         }
 292
 293         return 0;
 294
 295 unwind:
 296         /* unwind to the slave that failed */
 297         bond_for_each_slave(bond, rollback_slave, iter) {
 298                 if (rollback_slave == slave)
 299                         break;
 300
 301                 vlan_vid_del(rollback_slave->dev, proto, vid);
 302         }
 303
 304         return res;
 305 }
 306
 307 /**
 308  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 309  * @bond_dev: bonding net device that got called
 310  * @vid: vlan id being removed
 311  */
 312 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 313                                  __be16 proto, u16 vid)
 314 {
 315         struct bonding *bond = netdev_priv(bond_dev);
 316         struct list_head *iter;
 317         struct slave *slave;
 318
 319         bond_for_each_slave(bond, slave, iter)
 320                 vlan_vid_del(slave->dev, proto, vid);
 321
 322         if (bond_is_lb(bond))
 323                 bond_alb_clear_vlan(bond, vid);
 324
 325         return 0;
 326 }
 327
 328 /*------------------------------- Link status -------------------------------*/
 329
 330 /* Set the carrier state for the master according to the state of its
 331  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 332  * do special 802.3ad magic.
 333  *
 334  * Returns zero if carrier state does not change, nonzero if it does.
 335  */
 336 int bond_set_carrier(struct bonding *bond)
 337 {
 338         struct list_head *iter;
 339         struct slave *slave;
 340
 341         if (!bond_has_slaves(bond))
 342                 goto down;
 343
 344         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 345                 return bond_3ad_set_carrier(bond);
 346
 347         bond_for_each_slave(bond, slave, iter) {
 348                 if (slave->link == BOND_LINK_UP) {
 349                         if (!netif_carrier_ok(bond->dev)) {
 350                                 netif_carrier_on(bond->dev);
 351                                 return 1;
 352                         }
 353                         return 0;
 354                 }
 355         }
 356
 357 down:
 358         if (netif_carrier_ok(bond->dev)) {
 359                 netif_carrier_off(bond->dev);
 360                 return 1;
 361         }
 362         return 0;
 363 }
 364
 365 /* Get link speed and duplex from the slave's base driver
 366  * using ethtool. If for some reason the call fails or the
 367  * values are invalid, set speed and duplex to -1,
 368  * and return. Return 1 if speed or duplex settings are
 369  * UNKNOWN; 0 otherwise.
 370  */
 371 static int bond_update_speed_duplex(struct slave *slave)
 372 {
 373         struct net_device *slave_dev = slave->dev;
 374         struct ethtool_link_ksettings ecmd;
 375         int res;
 376
 377         slave->speed = SPEED_UNKNOWN;
 378         slave->duplex = DUPLEX_UNKNOWN;
 379
 380         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 381         if (res < 0)
 382                 return 1;
 383         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 384                 return 1;
 385         switch (ecmd.base.duplex) {
 386         case DUPLEX_FULL:
 387         case DUPLEX_HALF:
 388                 break;
 389         default:
 390                 return 1;
 391         }
 392
 393         slave->speed = ecmd.base.speed;
 394         slave->duplex = ecmd.base.duplex;
 395
 396         return 0;
 397 }
 398
 399 const char *bond_slave_link_status(s8 link)
 400 {
 401         switch (link) {
 402         case BOND_LINK_UP:
 403                 return "up";
 404         case BOND_LINK_FAIL:
 405                 return "going down";
 406         case BOND_LINK_DOWN:
 407                 return "down";
 408         case BOND_LINK_BACK:
 409                 return "going back";
 410         default:
 411                 return "unknown";
 412         }
 413 }
 414
 415 /* if <dev> supports MII link status reporting, check its link status.
 416  *
 417  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 418  * depending upon the setting of the use_carrier parameter.
 419  *
 420  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 421  * can't tell and just pretend it is), or 0, meaning that the link is
 422  * down.
 423  *
 424  * If reporting is non-zero, instead of faking link up, return -1 if
 425  * both ETHTOOL and MII ioctls fail (meaning the device does not
 426  * support them).  If use_carrier is set, return whatever it says.
 427  * It'd be nice if there was a good way to tell if a driver supports
 428  * netif_carrier, but there really isn't.
 429  */
 430 static int bond_check_dev_link(struct bonding *bond,
 431                                struct net_device *slave_dev, int reporting)
 432 {
 433         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 434         int (*ioctl)(struct net_device *, struct ifreq *, int);
 435         struct ifreq ifr;
 436         struct mii_ioctl_data *mii;
 437
 438         if (!reporting && !netif_running(slave_dev))
 439                 return 0;
 440
 441         if (bond->params.use_carrier)
 442                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 443
 444         /* Try to get link status using Ethtool first. */
 445         if (slave_dev->ethtool_ops->get_link)
 446                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 447                         BMSR_LSTATUS : 0;
 448
 449         /* Ethtool can't be used, fallback to MII ioctls. */
 450         ioctl = slave_ops->ndo_do_ioctl;
 451         if (ioctl) {
 452                 /* TODO: set pointer to correct ioctl on a per team member
 453                  *       bases to make this more efficient. that is, once
 454                  *       we determine the correct ioctl, we will always
 455                  *       call it and not the others for that team
 456                  *       member.
 457                  */
 458
 459                 /* We cannot assume that SIOCGMIIPHY will also read a
 460                  * register; not all network drivers (e.g., e100)
 461                  * support that.
 462                  */
 463
 464                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 465                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 466                 mii = if_mii(&ifr);
 467                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 468                         mii->reg_num = MII_BMSR;
 469                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 470                                 return mii->val_out & BMSR_LSTATUS;
 471                 }
 472         }
 473
 474         /* If reporting, report that either there's no dev->do_ioctl,
 475          * or both SIOCGMIIREG and get_link failed (meaning that we
 476          * cannot report link status).  If not reporting, pretend
 477          * we're ok.
 478          */
 479         return reporting ? -1 : BMSR_LSTATUS;
 480 }
 481
 482 /*----------------------------- Multicast list ------------------------------*/
 483
 484 /* Push the promiscuity flag down to appropriate slaves */
 485 static int bond_set_promiscuity(struct bonding *bond, int inc)
 486 {
 487         struct list_head *iter;
 488         int err = 0;
 489
 490         if (bond_uses_primary(bond)) {
 491                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 492
 493                 if (curr_active)
 494                         err = dev_set_promiscuity(curr_active->dev, inc);
 495         } else {
 496                 struct slave *slave;
 497
 498                 bond_for_each_slave(bond, slave, iter) {
 499                         err = dev_set_promiscuity(slave->dev, inc);
 500                         if (err)
 501                                 return err;
 502                 }
 503         }
 504         return err;
 505 }
 506
 507 /* Push the allmulti flag down to all slaves */
 508 static int bond_set_allmulti(struct bonding *bond, int inc)
 509 {
 510         struct list_head *iter;
 511         int err = 0;
 512
 513         if (bond_uses_primary(bond)) {
 514                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 515
 516                 if (curr_active)
 517                         err = dev_set_allmulti(curr_active->dev, inc);
 518         } else {
 519                 struct slave *slave;
 520
 521                 bond_for_each_slave(bond, slave, iter) {
 522                         err = dev_set_allmulti(slave->dev, inc);
 523                         if (err)
 524                                 return err;
 525                 }
 526         }
 527         return err;
 528 }
 529
 530 /* Retrieve the list of registered multicast addresses for the bonding
 531  * device and retransmit an IGMP JOIN request to the current active
 532  * slave.
 533  */
 534 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 535 {
 536         struct bonding *bond = container_of(work, struct bonding,
 537                                             mcast_work.work);
 538
 539         if (!rtnl_trylock()) {
 540                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 541                 return;
 542         }
 543         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 544
 545         if (bond->igmp_retrans > 1) {
 546                 bond->igmp_retrans--;
 547                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 548         }
 549         rtnl_unlock();
 550 }
 551
 552 /* Flush bond's hardware addresses from slave */
 553 static void bond_hw_addr_flush(struct net_device *bond_dev,
 554                                struct net_device *slave_dev)
 555 {
 556         struct bonding *bond = netdev_priv(bond_dev);
 557
 558         dev_uc_unsync(slave_dev, bond_dev);
 559         dev_mc_unsync(slave_dev, bond_dev);
 560
 561         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 562                 /* del lacpdu mc addr from mc list */
 563                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 564
 565                 dev_mc_del(slave_dev, lacpdu_multicast);
 566         }
 567 }
 568
 569 /*--------------------------- Active slave change ---------------------------*/
 570
 571 /* Update the hardware address list and promisc/allmulti for the new and
 572  * old active slaves (if any).  Modes that are not using primary keep all
 573  * slaves up date at all times; only the modes that use primary need to call
 574  * this function to swap these settings during a failover.
 575  */
 576 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 577                               struct slave *old_active)
 578 {
 579         if (old_active) {
 580                 if (bond->dev->flags & IFF_PROMISC)
 581                         dev_set_promiscuity(old_active->dev, -1);
 582
 583                 if (bond->dev->flags & IFF_ALLMULTI)
 584                         dev_set_allmulti(old_active->dev, -1);
 585
 586                 bond_hw_addr_flush(bond->dev, old_active->dev);
 587         }
 588
 589         if (new_active) {
 590                 /* FIXME: Signal errors upstream. */
 591                 if (bond->dev->flags & IFF_PROMISC)
 592                         dev_set_promiscuity(new_active->dev, 1);
 593
 594                 if (bond->dev->flags & IFF_ALLMULTI)
 595                         dev_set_allmulti(new_active->dev, 1);
 596
 597                 netif_addr_lock_bh(bond->dev);
 598                 dev_uc_sync(new_active->dev, bond->dev);
 599                 dev_mc_sync(new_active->dev, bond->dev);
 600                 netif_addr_unlock_bh(bond->dev);
 601         }
 602 }
 603
 604 /**
 605  * bond_set_dev_addr - clone slave's address to bond
 606  * @bond_dev: bond net device
 607  * @slave_dev: slave net device
 608  *
 609  * Should be called with RTNL held.
 610  */
 611 static void bond_set_dev_addr(struct net_device *bond_dev,
 612                               struct net_device *slave_dev)
 613 {
 614         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
 615                    bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
 616         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 617         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 618         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 619 }
 620
 621 static struct slave *bond_get_old_active(struct bonding *bond,
 622                                          struct slave *new_active)
 623 {
 624         struct slave *slave;
 625         struct list_head *iter;
 626
 627         bond_for_each_slave(bond, slave, iter) {
 628                 if (slave == new_active)
 629                         continue;
 630
 631                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 632                         return slave;
 633         }
 634
 635         return NULL;
 636 }
 637
 638 /* bond_do_fail_over_mac
 639  *
 640  * Perform special MAC address swapping for fail_over_mac settings
 641  *
 642  * Called with RTNL
 643  */
 644 static void bond_do_fail_over_mac(struct bonding *bond,
 645                                   struct slave *new_active,
 646                                   struct slave *old_active)
 647 {
 648         u8 tmp_mac[MAX_ADDR_LEN];
 649         struct sockaddr_storage ss;
 650         int rv;
 651
 652         switch (bond->params.fail_over_mac) {
 653         case BOND_FOM_ACTIVE:
 654                 if (new_active)
 655                         bond_set_dev_addr(bond->dev, new_active->dev);
 656                 break;
 657         case BOND_FOM_FOLLOW:
 658                 /* if new_active && old_active, swap them
 659                  * if just old_active, do nothing (going to no active slave)
 660                  * if just new_active, set new_active to bond's MAC
 661                  */
 662                 if (!new_active)
 663                         return;
 664
 665                 if (!old_active)
 666                         old_active = bond_get_old_active(bond, new_active);
 667
 668                 if (old_active) {
 669                         bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 670                                           new_active->dev->addr_len);
 671                         bond_hw_addr_copy(ss.__data,
 672                                           old_active->dev->dev_addr,
 673                                           old_active->dev->addr_len);
 674                         ss.ss_family = new_active->dev->type;
 675                 } else {
 676                         bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 677                                           bond->dev->addr_len);
 678                         ss.ss_family = bond->dev->type;
 679                 }
 680
 681                 rv = dev_set_mac_address(new_active->dev,
 682                                          (struct sockaddr *)&ss);
 683                 if (rv) {
 684                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 685                                    -rv, new_active->dev->name);
 686                         goto out;
 687                 }
 688
 689                 if (!old_active)
 690                         goto out;
 691
 692                 bond_hw_addr_copy(ss.__data, tmp_mac,
 693                                   new_active->dev->addr_len);
 694                 ss.ss_family = old_active->dev->type;
 695
 696                 rv = dev_set_mac_address(old_active->dev,
 697                                          (struct sockaddr *)&ss);
 698                 if (rv)
 699                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 700                                    -rv, new_active->dev->name);
 701 out:
 702                 break;
 703         default:
 704                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 705                            bond->params.fail_over_mac);
 706                 break;
 707         }
 708
 709 }
 710
 711 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 712 {
 713         struct slave *prim = rtnl_dereference(bond->primary_slave);
 714         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 715
 716         if (!prim || prim->link != BOND_LINK_UP) {
 717                 if (!curr || curr->link != BOND_LINK_UP)
 718                         return NULL;
 719                 return curr;
 720         }
 721
 722         if (bond->force_primary) {
 723                 bond->force_primary = false;
 724                 return prim;
 725         }
 726
 727         if (!curr || curr->link != BOND_LINK_UP)
 728                 return prim;
 729
 730         /* At this point, prim and curr are both up */
 731         switch (bond->params.primary_reselect) {
 732         case BOND_PRI_RESELECT_ALWAYS:
 733                 return prim;
 734         case BOND_PRI_RESELECT_BETTER:
 735                 if (prim->speed < curr->speed)
 736                         return curr;
 737                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 738                         return curr;
 739                 return prim;
 740         case BOND_PRI_RESELECT_FAILURE:
 741                 return curr;
 742         default:
 743                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 744                            bond->params.primary_reselect);
 745                 return curr;
 746         }
 747 }
 748
 749 /**
 750  * bond_find_best_slave - select the best available slave to be the active one
 751  * @bond: our bonding struct
 752  */
 753 static struct slave *bond_find_best_slave(struct bonding *bond)
 754 {
 755         struct slave *slave, *bestslave = NULL;
 756         struct list_head *iter;
 757         int mintime = bond->params.updelay;
 758
 759         slave = bond_choose_primary_or_current(bond);
 760         if (slave)
 761                 return slave;
 762
 763         bond_for_each_slave(bond, slave, iter) {
 764                 if (slave->link == BOND_LINK_UP)
 765                         return slave;
 766                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 767                     slave->delay < mintime) {
 768                         mintime = slave->delay;
 769                         bestslave = slave;
 770                 }
 771         }
 772
 773         return bestslave;
 774 }
 775
 776 static bool bond_should_notify_peers(struct bonding *bond)
 777 {
 778         struct slave *slave;
 779
 780         rcu_read_lock();
 781         slave = rcu_dereference(bond->curr_active_slave);
 782         rcu_read_unlock();
 783
 784         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 785                    slave ? slave->dev->name : "NULL");
 786
 787         if (!slave || !bond->send_peer_notif ||
 788             !netif_carrier_ok(bond->dev) ||
 789             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 790                 return false;
 791
 792         return true;
 793 }
 794
 795 /**
 796  * change_active_interface - change the active slave into the specified one
 797  * @bond: our bonding struct
 798  * @new: the new slave to make the active one
 799  *
 800  * Set the new slave to the bond's settings and unset them on the old
 801  * curr_active_slave.
 802  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 803  *
 804  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 805  * because it is apparently the best available slave we have, even though its
 806  * updelay hasn't timed out yet.
 807  *
 808  * Caller must hold RTNL.
 809  */
 810 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 811 {
 812         struct slave *old_active;
 813
 814         ASSERT_RTNL();
 815
 816         old_active = rtnl_dereference(bond->curr_active_slave);
 817
 818         if (old_active == new_active)
 819                 return;
 820
 821         if (new_active) {
 822                 new_active->last_link_up = jiffies;
 823
 824                 if (new_active->link == BOND_LINK_BACK) {
 825                         if (bond_uses_primary(bond)) {
 826                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 827                                             new_active->dev->name,
 828                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 829                         }
 830
 831                         new_active->delay = 0;
 832                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 833                                                   BOND_SLAVE_NOTIFY_NOW);
 834
 835                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 836                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 837
 838                         if (bond_is_lb(bond))
 839                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 840                 } else {
 841                         if (bond_uses_primary(bond)) {
 842                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 843                                             new_active->dev->name);
 844                         }
 845                 }
 846         }
 847
 848         if (bond_uses_primary(bond))
 849                 bond_hw_addr_swap(bond, new_active, old_active);
 850
 851         if (bond_is_lb(bond)) {
 852                 bond_alb_handle_active_change(bond, new_active);
 853                 if (old_active)
 854                         bond_set_slave_inactive_flags(old_active,
 855                                                       BOND_SLAVE_NOTIFY_NOW);
 856                 if (new_active)
 857                         bond_set_slave_active_flags(new_active,
 858                                                     BOND_SLAVE_NOTIFY_NOW);
 859         } else {
 860                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 861         }
 862
 863         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 864                 if (old_active)
 865                         bond_set_slave_inactive_flags(old_active,
 866                                                       BOND_SLAVE_NOTIFY_NOW);
 867
 868                 if (new_active) {
 869                         bool should_notify_peers = false;
 870
 871                         bond_set_slave_active_flags(new_active,
 872                                                     BOND_SLAVE_NOTIFY_NOW);
 873
 874                         if (bond->params.fail_over_mac)
 875                                 bond_do_fail_over_mac(bond, new_active,
 876                                                       old_active);
 877
 878                         if (netif_running(bond->dev)) {
 879                                 bond->send_peer_notif =
 880                                         bond->params.num_peer_notif;
 881                                 should_notify_peers =
 882                                         bond_should_notify_peers(bond);
 883                         }
 884
 885                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 886                         if (should_notify_peers)
 887                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 888                                                          bond->dev);
 889                 }
 890         }
 891
 892         /* resend IGMP joins since active slave has changed or
 893          * all were sent on curr_active_slave.
 894          * resend only if bond is brought up with the affected
 895          * bonding modes and the retransmission is enabled
 896          */
 897         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 898             ((bond_uses_primary(bond) && new_active) ||
 899              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 900                 bond->igmp_retrans = bond->params.resend_igmp;
 901                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 902         }
 903 }
 904
 905 /**
 906  * bond_select_active_slave - select a new active slave, if needed
 907  * @bond: our bonding struct
 908  *
 909  * This functions should be called when one of the following occurs:
 910  * - The old curr_active_slave has been released or lost its link.
 911  * - The primary_slave has got its link back.
 912  * - A slave has got its link back and there's no old curr_active_slave.
 913  *
 914  * Caller must hold RTNL.
 915  */
 916 void bond_select_active_slave(struct bonding *bond)
 917 {
 918         struct slave *best_slave;
 919         int rv;
 920
 921         ASSERT_RTNL();
 922
 923         best_slave = bond_find_best_slave(bond);
 924         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 925                 bond_change_active_slave(bond, best_slave);
 926                 rv = bond_set_carrier(bond);
 927                 if (!rv)
 928                         return;
 929
 930                 if (netif_carrier_ok(bond->dev))
 931                         netdev_info(bond->dev, "first active interface up!\n");
 932                 else
 933                         netdev_info(bond->dev, "now running without any active interface!\n");
 934         }
 935 }
 936
 937 #ifdef CONFIG_NET_POLL_CONTROLLER
 938 static inline int slave_enable_netpoll(struct slave *slave)
 939 {
 940         struct netpoll *np;
 941         int err = 0;
 942
 943         np = kzalloc(sizeof(*np), GFP_KERNEL);
 944         err = -ENOMEM;
 945         if (!np)
 946                 goto out;
 947
 948         err = __netpoll_setup(np, slave->dev);
 949         if (err) {
 950                 kfree(np);
 951                 goto out;
 952         }
 953         slave->np = np;
 954 out:
 955         return err;
 956 }
 957 static inline void slave_disable_netpoll(struct slave *slave)
 958 {
 959         struct netpoll *np = slave->np;
 960
 961         if (!np)
 962                 return;
 963
 964         slave->np = NULL;
 965         __netpoll_free_async(np);
 966 }
 967
 968 static void bond_poll_controller(struct net_device *bond_dev)
 969 {
 970         struct bonding *bond = netdev_priv(bond_dev);
 971         struct slave *slave = NULL;
 972         struct list_head *iter;
 973         struct ad_info ad_info;
 974         struct netpoll_info *ni;
 975         const struct net_device_ops *ops;
 976
 977         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 978                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 979                         return;
 980
 981         bond_for_each_slave_rcu(bond, slave, iter) {
 982                 ops = slave->dev->netdev_ops;
 983                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 984                         continue;
 985
 986                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 987                         struct aggregator *agg =
 988                             SLAVE_AD_INFO(slave)->port.aggregator;
 989
 990                         if (agg &&
 991                             agg->aggregator_identifier != ad_info.aggregator_id)
 992                                 continue;
 993                 }
 994
 995                 ni = rcu_dereference_bh(slave->dev->npinfo);
 996                 if (down_trylock(&ni->dev_lock))
 997                         continue;
 998                 ops->ndo_poll_controller(slave->dev);
 999                 up(&ni->dev_lock);
1000         }
1001 }
1002
1003 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1004 {
1005         struct bonding *bond = netdev_priv(bond_dev);
1006         struct list_head *iter;
1007         struct slave *slave;
1008
1009         bond_for_each_slave(bond, slave, iter)
1010                 if (bond_slave_is_up(slave))
1011                         slave_disable_netpoll(slave);
1012 }
1013
1014 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1015 {
1016         struct bonding *bond = netdev_priv(dev);
1017         struct list_head *iter;
1018         struct slave *slave;
1019         int err = 0;
1020
1021         bond_for_each_slave(bond, slave, iter) {
1022                 err = slave_enable_netpoll(slave);
1023                 if (err) {
1024                         bond_netpoll_cleanup(dev);
1025                         break;
1026                 }
1027         }
1028         return err;
1029 }
1030 #else
1031 static inline int slave_enable_netpoll(struct slave *slave)
1032 {
1033         return 0;
1034 }
1035 static inline void slave_disable_netpoll(struct slave *slave)
1036 {
1037 }
1038 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1039 {
1040 }
1041 #endif
1042
1043 /*---------------------------------- IOCTL ----------------------------------*/
1044
1045 static netdev_features_t bond_fix_features(struct net_device *dev,
1046                                            netdev_features_t features)
1047 {
1048         struct bonding *bond = netdev_priv(dev);
1049         struct list_head *iter;
1050         netdev_features_t mask;
1051         struct slave *slave;
1052
1053         mask = features;
1054
1055         features &= ~NETIF_F_ONE_FOR_ALL;
1056         features |= NETIF_F_ALL_FOR_ALL;
1057
1058         bond_for_each_slave(bond, slave, iter) {
1059                 features = netdev_increment_features(features,
1060                                                      slave->dev->features,
1061                                                      mask);
1062         }
1063         features = netdev_add_tso_features(features, mask);
1064
1065         return features;
1066 }
1067
1068 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1069                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1070                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1071
1072 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1073                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1074
1075 static void bond_compute_features(struct bonding *bond)
1076 {
1077         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1078                                         IFF_XMIT_DST_RELEASE_PERM;
1079         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1080         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1081         struct net_device *bond_dev = bond->dev;
1082         struct list_head *iter;
1083         struct slave *slave;
1084         unsigned short max_hard_header_len = ETH_HLEN;
1085         unsigned int gso_max_size = GSO_MAX_SIZE;
1086         u16 gso_max_segs = GSO_MAX_SEGS;
1087
1088         if (!bond_has_slaves(bond))
1089                 goto done;
1090         vlan_features &= NETIF_F_ALL_FOR_ALL;
1091
1092         bond_for_each_slave(bond, slave, iter) {
1093                 vlan_features = netdev_increment_features(vlan_features,
1094                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1095
1096                 enc_features = netdev_increment_features(enc_features,
1097                                                          slave->dev->hw_enc_features,
1098                                                          BOND_ENC_FEATURES);
1099                 dst_release_flag &= slave->dev->priv_flags;
1100                 if (slave->dev->hard_header_len > max_hard_header_len)
1101                         max_hard_header_len = slave->dev->hard_header_len;
1102
1103                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1104                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1105         }
1106         bond_dev->hard_header_len = max_hard_header_len;
1107
1108 done:
1109         bond_dev->vlan_features = vlan_features;
1110         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1111                                     NETIF_F_GSO_UDP_L4;
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 = READ_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         /* don't change skb->dev for link-local packets */
1181         if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1182                 return RX_HANDLER_PASS;
1183         if (bond_should_deliver_exact_match(skb, slave, bond))
1184                 return RX_HANDLER_EXACT;
1185
1186         skb->dev = bond->dev;
1187
1188         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1189             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1190             skb->pkt_type == PACKET_HOST) {
1191
1192                 if (unlikely(skb_cow_head(skb,
1193                                           skb->data - skb_mac_header(skb)))) {
1194                         kfree_skb(skb);
1195                         return RX_HANDLER_CONSUMED;
1196                 }
1197                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1198                                   bond->dev->addr_len);
1199         }
1200
1201         return ret;
1202 }
1203
1204 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1205 {
1206         switch (BOND_MODE(bond)) {
1207         case BOND_MODE_ROUNDROBIN:
1208                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1209         case BOND_MODE_ACTIVEBACKUP:
1210                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1211         case BOND_MODE_BROADCAST:
1212                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1213         case BOND_MODE_XOR:
1214         case BOND_MODE_8023AD:
1215                 return NETDEV_LAG_TX_TYPE_HASH;
1216         default:
1217                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1218         }
1219 }
1220
1221 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1222                                                enum netdev_lag_tx_type type)
1223 {
1224         if (type != NETDEV_LAG_TX_TYPE_HASH)
1225                 return NETDEV_LAG_HASH_NONE;
1226
1227         switch (bond->params.xmit_policy) {
1228         case BOND_XMIT_POLICY_LAYER2:
1229                 return NETDEV_LAG_HASH_L2;
1230         case BOND_XMIT_POLICY_LAYER34:
1231                 return NETDEV_LAG_HASH_L34;
1232         case BOND_XMIT_POLICY_LAYER23:
1233                 return NETDEV_LAG_HASH_L23;
1234         case BOND_XMIT_POLICY_ENCAP23:
1235                 return NETDEV_LAG_HASH_E23;
1236         case BOND_XMIT_POLICY_ENCAP34:
1237                 return NETDEV_LAG_HASH_E34;
1238         default:
1239                 return NETDEV_LAG_HASH_UNKNOWN;
1240         }
1241 }
1242
1243 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1244                                       struct netlink_ext_ack *extack)
1245 {
1246         struct netdev_lag_upper_info lag_upper_info;
1247         enum netdev_lag_tx_type type;
1248
1249         type = bond_lag_tx_type(bond);
1250         lag_upper_info.tx_type = type;
1251         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1252
1253         return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1254                                             &lag_upper_info, extack);
1255 }
1256
1257 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1258 {
1259         netdev_upper_dev_unlink(slave->dev, bond->dev);
1260         slave->dev->flags &= ~IFF_SLAVE;
1261 }
1262
1263 static struct slave *bond_alloc_slave(struct bonding *bond)
1264 {
1265         struct slave *slave = NULL;
1266
1267         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1268         if (!slave)
1269                 return NULL;
1270
1271         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1272                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1273                                                GFP_KERNEL);
1274                 if (!SLAVE_AD_INFO(slave)) {
1275                         kfree(slave);
1276                         return NULL;
1277                 }
1278         }
1279         return slave;
1280 }
1281
1282 static void bond_free_slave(struct slave *slave)
1283 {
1284         struct bonding *bond = bond_get_bond_by_slave(slave);
1285
1286         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1287                 kfree(SLAVE_AD_INFO(slave));
1288
1289         kfree(slave);
1290 }
1291
1292 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1293 {
1294         info->bond_mode = BOND_MODE(bond);
1295         info->miimon = bond->params.miimon;
1296         info->num_slaves = bond->slave_cnt;
1297 }
1298
1299 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1300 {
1301         strcpy(info->slave_name, slave->dev->name);
1302         info->link = slave->link;
1303         info->state = bond_slave_state(slave);
1304         info->link_failure_count = slave->link_failure_count;
1305 }
1306
1307 static void bond_netdev_notify(struct net_device *dev,
1308                                struct netdev_bonding_info *info)
1309 {
1310         rtnl_lock();
1311         netdev_bonding_info_change(dev, info);
1312         rtnl_unlock();
1313 }
1314
1315 static void bond_netdev_notify_work(struct work_struct *_work)
1316 {
1317         struct netdev_notify_work *w =
1318                 container_of(_work, struct netdev_notify_work, work.work);
1319
1320         bond_netdev_notify(w->dev, &w->bonding_info);
1321         dev_put(w->dev);
1322         kfree(w);
1323 }
1324
1325 void bond_queue_slave_event(struct slave *slave)
1326 {
1327         struct bonding *bond = slave->bond;
1328         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1329
1330         if (!nnw)
1331                 return;
1332
1333         dev_hold(slave->dev);
1334         nnw->dev = slave->dev;
1335         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1336         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1337         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1338
1339         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1340 }
1341
1342 void bond_lower_state_changed(struct slave *slave)
1343 {
1344         struct netdev_lag_lower_state_info info;
1345
1346         info.link_up = slave->link == BOND_LINK_UP ||
1347                        slave->link == BOND_LINK_FAIL;
1348         info.tx_enabled = bond_is_active_slave(slave);
1349         netdev_lower_state_changed(slave->dev, &info);
1350 }
1351
1352 /* enslave device <slave> to bond device <master> */
1353 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1354                  struct netlink_ext_ack *extack)
1355 {
1356         struct bonding *bond = netdev_priv(bond_dev);
1357         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1358         struct slave *new_slave = NULL, *prev_slave;
1359         struct sockaddr_storage ss;
1360         int link_reporting;
1361         int res = 0, i;
1362
1363         if (!bond->params.use_carrier &&
1364             slave_dev->ethtool_ops->get_link == NULL &&
1365             slave_ops->ndo_do_ioctl == NULL) {
1366                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1367                             slave_dev->name);
1368         }
1369
1370         /* already in-use? */
1371         if (netdev_is_rx_handler_busy(slave_dev)) {
1372                 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1373                 netdev_err(bond_dev,
1374                            "Error: Device is in use and cannot be enslaved\n");
1375                 return -EBUSY;
1376         }
1377
1378         if (bond_dev == slave_dev) {
1379                 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1380                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1381                 return -EPERM;
1382         }
1383
1384         /* vlan challenged mutual exclusion */
1385         /* no need to lock since we're protected by rtnl_lock */
1386         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1387                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1388                            slave_dev->name);
1389                 if (vlan_uses_dev(bond_dev)) {
1390                         NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1391                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1392                                    slave_dev->name, bond_dev->name);
1393                         return -EPERM;
1394                 } else {
1395                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1396                                     slave_dev->name, slave_dev->name,
1397                                     bond_dev->name);
1398                 }
1399         } else {
1400                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1401                            slave_dev->name);
1402         }
1403
1404         /* Old ifenslave binaries are no longer supported.  These can
1405          * be identified with moderate accuracy by the state of the slave:
1406          * the current ifenslave will set the interface down prior to
1407          * enslaving it; the old ifenslave will not.
1408          */
1409         if (slave_dev->flags & IFF_UP) {
1410                 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1411                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1412                            slave_dev->name);
1413                 return -EPERM;
1414         }
1415
1416         /* set bonding device ether type by slave - bonding netdevices are
1417          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1418          * there is a need to override some of the type dependent attribs/funcs.
1419          *
1420          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1421          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1422          */
1423         if (!bond_has_slaves(bond)) {
1424                 if (bond_dev->type != slave_dev->type) {
1425                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1426                                    bond_dev->type, slave_dev->type);
1427
1428                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1429                                                        bond_dev);
1430                         res = notifier_to_errno(res);
1431                         if (res) {
1432                                 netdev_err(bond_dev, "refused to change device type\n");
1433                                 return -EBUSY;
1434                         }
1435
1436                         /* Flush unicast and multicast addresses */
1437                         dev_uc_flush(bond_dev);
1438                         dev_mc_flush(bond_dev);
1439
1440                         if (slave_dev->type != ARPHRD_ETHER)
1441                                 bond_setup_by_slave(bond_dev, slave_dev);
1442                         else {
1443                                 ether_setup(bond_dev);
1444                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1445                         }
1446
1447                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1448                                                  bond_dev);
1449                 }
1450         } else if (bond_dev->type != slave_dev->type) {
1451                 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1452                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1453                            slave_dev->name, slave_dev->type, bond_dev->type);
1454                 return -EINVAL;
1455         }
1456
1457         if (slave_dev->type == ARPHRD_INFINIBAND &&
1458             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1459                 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1460                 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1461                             slave_dev->type);
1462                 res = -EOPNOTSUPP;
1463                 goto err_undo_flags;
1464         }
1465
1466         if (!slave_ops->ndo_set_mac_address ||
1467             slave_dev->type == ARPHRD_INFINIBAND) {
1468                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1469                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1470                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1471                         if (!bond_has_slaves(bond)) {
1472                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1473                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1474                         } else {
1475                                 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1476                                 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");
1477                                 res = -EOPNOTSUPP;
1478                                 goto err_undo_flags;
1479                         }
1480                 }
1481         }
1482
1483         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1484
1485         /* If this is the first slave, then we need to set the master's hardware
1486          * address to be the same as the slave's.
1487          */
1488         if (!bond_has_slaves(bond) &&
1489             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1490                 bond_set_dev_addr(bond->dev, slave_dev);
1491
1492         new_slave = bond_alloc_slave(bond);
1493         if (!new_slave) {
1494                 res = -ENOMEM;
1495                 goto err_undo_flags;
1496         }
1497
1498         new_slave->bond = bond;
1499         new_slave->dev = slave_dev;
1500         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1501          * is set via sysfs or module option if desired.
1502          */
1503         new_slave->queue_id = 0;
1504
1505         /* Save slave's original mtu and then set it to match the bond */
1506         new_slave->original_mtu = slave_dev->mtu;
1507         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1508         if (res) {
1509                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1510                 goto err_free;
1511         }
1512
1513         /* Save slave's original ("permanent") mac address for modes
1514          * that need it, and for restoring it upon release, and then
1515          * set it to the master's address
1516          */
1517         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1518                           slave_dev->addr_len);
1519
1520         if (!bond->params.fail_over_mac ||
1521             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1522                 /* Set slave to master's mac address.  The application already
1523                  * set the master's mac address to that of the first slave
1524                  */
1525                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1526                 ss.ss_family = slave_dev->type;
1527                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1528                 if (res) {
1529                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1530                         goto err_restore_mtu;
1531                 }
1532         }
1533
1534         /* set slave flag before open to prevent IPv6 addrconf */
1535         slave_dev->flags |= IFF_SLAVE;
1536
1537         /* open the slave since the application closed it */
1538         res = dev_open(slave_dev);
1539         if (res) {
1540                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1541                 goto err_restore_mac;
1542         }
1543
1544         slave_dev->priv_flags |= IFF_BONDING;
1545         /* initialize slave stats */
1546         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1547
1548         if (bond_is_lb(bond)) {
1549                 /* bond_alb_init_slave() must be called before all other stages since
1550                  * it might fail and we do not want to have to undo everything
1551                  */
1552                 res = bond_alb_init_slave(bond, new_slave);
1553                 if (res)
1554                         goto err_close;
1555         }
1556
1557         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1558         if (res) {
1559                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1560                            slave_dev->name);
1561                 goto err_close;
1562         }
1563
1564         prev_slave = bond_last_slave(bond);
1565
1566         new_slave->delay = 0;
1567         new_slave->link_failure_count = 0;
1568
1569         if (bond_update_speed_duplex(new_slave) &&
1570             bond_needs_speed_duplex(bond))
1571                 new_slave->link = BOND_LINK_DOWN;
1572
1573         new_slave->last_rx = jiffies -
1574                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1575         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1576                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1577
1578         if (bond->params.miimon && !bond->params.use_carrier) {
1579                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1580
1581                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1582                         /* miimon is set but a bonded network driver
1583                          * does not support ETHTOOL/MII and
1584                          * arp_interval is not set.  Note: if
1585                          * use_carrier is enabled, we will never go
1586                          * here (because netif_carrier is always
1587                          * supported); thus, we don't need to change
1588                          * the messages for netif_carrier.
1589                          */
1590                         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",
1591                                     slave_dev->name);
1592                 } else if (link_reporting == -1) {
1593                         /* unable get link status using mii/ethtool */
1594                         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",
1595                                     slave_dev->name);
1596                 }
1597         }
1598
1599         /* check for initial state */
1600         new_slave->link = BOND_LINK_NOCHANGE;
1601         if (bond->params.miimon) {
1602                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1603                         if (bond->params.updelay) {
1604                                 bond_set_slave_link_state(new_slave,
1605                                                           BOND_LINK_BACK,
1606                                                           BOND_SLAVE_NOTIFY_NOW);
1607                                 new_slave->delay = bond->params.updelay;
1608                         } else {
1609                                 bond_set_slave_link_state(new_slave,
1610                                                           BOND_LINK_UP,
1611                                                           BOND_SLAVE_NOTIFY_NOW);
1612                         }
1613                 } else {
1614                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1615                                                   BOND_SLAVE_NOTIFY_NOW);
1616                 }
1617         } else if (bond->params.arp_interval) {
1618                 bond_set_slave_link_state(new_slave,
1619                                           (netif_carrier_ok(slave_dev) ?
1620                                           BOND_LINK_UP : BOND_LINK_DOWN),
1621                                           BOND_SLAVE_NOTIFY_NOW);
1622         } else {
1623                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1624                                           BOND_SLAVE_NOTIFY_NOW);
1625         }
1626
1627         if (new_slave->link != BOND_LINK_DOWN)
1628                 new_slave->last_link_up = jiffies;
1629         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1630                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1631                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1632
1633         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1634                 /* if there is a primary slave, remember it */
1635                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1636                         rcu_assign_pointer(bond->primary_slave, new_slave);
1637                         bond->force_primary = true;
1638                 }
1639         }
1640
1641         switch (BOND_MODE(bond)) {
1642         case BOND_MODE_ACTIVEBACKUP:
1643                 bond_set_slave_inactive_flags(new_slave,
1644                                               BOND_SLAVE_NOTIFY_NOW);
1645                 break;
1646         case BOND_MODE_8023AD:
1647                 /* in 802.3ad mode, the internal mechanism
1648                  * will activate the slaves in the selected
1649                  * aggregator
1650                  */
1651                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1652                 /* if this is the first slave */
1653                 if (!prev_slave) {
1654                         SLAVE_AD_INFO(new_slave)->id = 1;
1655                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1656                          * can be called only after the mac address of the bond is set
1657                          */
1658                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1659                 } else {
1660                         SLAVE_AD_INFO(new_slave)->id =
1661                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1662                 }
1663
1664                 bond_3ad_bind_slave(new_slave);
1665                 break;
1666         case BOND_MODE_TLB:
1667         case BOND_MODE_ALB:
1668                 bond_set_active_slave(new_slave);
1669                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1670                 break;
1671         default:
1672                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1673
1674                 /* always active in trunk mode */
1675                 bond_set_active_slave(new_slave);
1676
1677                 /* In trunking mode there is little meaning to curr_active_slave
1678                  * anyway (it holds no special properties of the bond device),
1679                  * so we can change it without calling change_active_interface()
1680                  */
1681                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1682                     new_slave->link == BOND_LINK_UP)
1683                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1684
1685                 break;
1686         } /* switch(bond_mode) */
1687
1688 #ifdef CONFIG_NET_POLL_CONTROLLER
1689         if (bond->dev->npinfo) {
1690                 if (slave_enable_netpoll(new_slave)) {
1691                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1692                         res = -EBUSY;
1693                         goto err_detach;
1694                 }
1695         }
1696 #endif
1697
1698         if (!(bond_dev->features & NETIF_F_LRO))
1699                 dev_disable_lro(slave_dev);
1700
1701         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1702                                          new_slave);
1703         if (res) {
1704                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1705                 goto err_detach;
1706         }
1707
1708         res = bond_master_upper_dev_link(bond, new_slave, extack);
1709         if (res) {
1710                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1711                 goto err_unregister;
1712         }
1713
1714         res = bond_sysfs_slave_add(new_slave);
1715         if (res) {
1716                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1717                 goto err_upper_unlink;
1718         }
1719
1720         bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1721
1722         /* If the mode uses primary, then the following is handled by
1723          * bond_change_active_slave().
1724          */
1725         if (!bond_uses_primary(bond)) {
1726                 /* set promiscuity level to new slave */
1727                 if (bond_dev->flags & IFF_PROMISC) {
1728                         res = dev_set_promiscuity(slave_dev, 1);
1729                         if (res)
1730                                 goto err_sysfs_del;
1731                 }
1732
1733                 /* set allmulti level to new slave */
1734                 if (bond_dev->flags & IFF_ALLMULTI) {
1735                         res = dev_set_allmulti(slave_dev, 1);
1736                         if (res) {
1737                                 if (bond_dev->flags & IFF_PROMISC)
1738                                         dev_set_promiscuity(slave_dev, -1);
1739                                 goto err_sysfs_del;
1740                         }
1741                 }
1742
1743                 netif_addr_lock_bh(bond_dev);
1744                 dev_mc_sync_multiple(slave_dev, bond_dev);
1745                 dev_uc_sync_multiple(slave_dev, bond_dev);
1746                 netif_addr_unlock_bh(bond_dev);
1747
1748                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1749                         /* add lacpdu mc addr to mc list */
1750                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1751
1752                         dev_mc_add(slave_dev, lacpdu_multicast);
1753                 }
1754         }
1755
1756         bond->slave_cnt++;
1757         bond_compute_features(bond);
1758         bond_set_carrier(bond);
1759
1760         if (bond_uses_primary(bond)) {
1761                 block_netpoll_tx();
1762                 bond_select_active_slave(bond);
1763                 unblock_netpoll_tx();
1764         }
1765
1766         if (bond_mode_can_use_xmit_hash(bond))
1767                 bond_update_slave_arr(bond, NULL);
1768
1769
1770         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1771                     slave_dev->name,
1772                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1773                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1774
1775         /* enslave is successful */
1776         bond_queue_slave_event(new_slave);
1777         return 0;
1778
1779 /* Undo stages on error */
1780 err_sysfs_del:
1781         bond_sysfs_slave_del(new_slave);
1782
1783 err_upper_unlink:
1784         bond_upper_dev_unlink(bond, new_slave);
1785
1786 err_unregister:
1787         netdev_rx_handler_unregister(slave_dev);
1788
1789 err_detach:
1790         vlan_vids_del_by_dev(slave_dev, bond_dev);
1791         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1792                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1793         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1794                 block_netpoll_tx();
1795                 bond_change_active_slave(bond, NULL);
1796                 bond_select_active_slave(bond);
1797                 unblock_netpoll_tx();
1798         }
1799         /* either primary_slave or curr_active_slave might've changed */
1800         synchronize_rcu();
1801         slave_disable_netpoll(new_slave);
1802
1803 err_close:
1804         slave_dev->priv_flags &= ~IFF_BONDING;
1805         dev_close(slave_dev);
1806
1807 err_restore_mac:
1808         slave_dev->flags &= ~IFF_SLAVE;
1809         if (!bond->params.fail_over_mac ||
1810             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1811                 /* XXX TODO - fom follow mode needs to change master's
1812                  * MAC if this slave's MAC is in use by the bond, or at
1813                  * least print a warning.
1814                  */
1815                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1816                                   new_slave->dev->addr_len);
1817                 ss.ss_family = slave_dev->type;
1818                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1819         }
1820
1821 err_restore_mtu:
1822         dev_set_mtu(slave_dev, new_slave->original_mtu);
1823
1824 err_free:
1825         bond_free_slave(new_slave);
1826
1827 err_undo_flags:
1828         /* Enslave of first slave has failed and we need to fix master's mac */
1829         if (!bond_has_slaves(bond)) {
1830                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1831                                             slave_dev->dev_addr))
1832                         eth_hw_addr_random(bond_dev);
1833                 if (bond_dev->type != ARPHRD_ETHER) {
1834                         dev_close(bond_dev);
1835                         ether_setup(bond_dev);
1836                         bond_dev->flags |= IFF_MASTER;
1837                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1838                 }
1839         }
1840
1841         return res;
1842 }
1843
1844 /* Try to release the slave device <slave> from the bond device <master>
1845  * It is legal to access curr_active_slave without a lock because all the function
1846  * is RTNL-locked. If "all" is true it means that the function is being called
1847  * while destroying a bond interface and all slaves are being released.
1848  *
1849  * The rules for slave state should be:
1850  *   for Active/Backup:
1851  *     Active stays on all backups go down
1852  *   for Bonded connections:
1853  *     The first up interface should be left on and all others downed.
1854  */
1855 static int __bond_release_one(struct net_device *bond_dev,
1856                               struct net_device *slave_dev,
1857                               bool all, bool unregister)
1858 {
1859         struct bonding *bond = netdev_priv(bond_dev);
1860         struct slave *slave, *oldcurrent;
1861         struct sockaddr_storage ss;
1862         int old_flags = bond_dev->flags;
1863         netdev_features_t old_features = bond_dev->features;
1864
1865         /* slave is not a slave or master is not master of this slave */
1866         if (!(slave_dev->flags & IFF_SLAVE) ||
1867             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1868                 netdev_dbg(bond_dev, "cannot release %s\n",
1869                            slave_dev->name);
1870                 return -EINVAL;
1871         }
1872
1873         block_netpoll_tx();
1874
1875         slave = bond_get_slave_by_dev(bond, slave_dev);
1876         if (!slave) {
1877                 /* not a slave of this bond */
1878                 netdev_info(bond_dev, "%s not enslaved\n",
1879                             slave_dev->name);
1880                 unblock_netpoll_tx();
1881                 return -EINVAL;
1882         }
1883
1884         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1885
1886         bond_sysfs_slave_del(slave);
1887
1888         /* recompute stats just before removing the slave */
1889         bond_get_stats(bond->dev, &bond->bond_stats);
1890
1891         bond_upper_dev_unlink(bond, slave);
1892         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1893          * for this slave anymore.
1894          */
1895         netdev_rx_handler_unregister(slave_dev);
1896
1897         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1898                 bond_3ad_unbind_slave(slave);
1899
1900         if (bond_mode_can_use_xmit_hash(bond))
1901                 bond_update_slave_arr(bond, slave);
1902
1903         netdev_info(bond_dev, "Releasing %s interface %s\n",
1904                     bond_is_active_slave(slave) ? "active" : "backup",
1905                     slave_dev->name);
1906
1907         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1908
1909         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1910
1911         if (!all && (!bond->params.fail_over_mac ||
1912                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1913                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1914                     bond_has_slaves(bond))
1915                         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",
1916                                     slave_dev->name, slave->perm_hwaddr,
1917                                     bond_dev->name, slave_dev->name);
1918         }
1919
1920         if (rtnl_dereference(bond->primary_slave) == slave)
1921                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1922
1923         if (oldcurrent == slave)
1924                 bond_change_active_slave(bond, NULL);
1925
1926         if (bond_is_lb(bond)) {
1927                 /* Must be called only after the slave has been
1928                  * detached from the list and the curr_active_slave
1929                  * has been cleared (if our_slave == old_current),
1930                  * but before a new active slave is selected.
1931                  */
1932                 bond_alb_deinit_slave(bond, slave);
1933         }
1934
1935         if (all) {
1936                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1937         } else if (oldcurrent == slave) {
1938                 /* Note that we hold RTNL over this sequence, so there
1939                  * is no concern that another slave add/remove event
1940                  * will interfere.
1941                  */
1942                 bond_select_active_slave(bond);
1943         }
1944
1945         if (!bond_has_slaves(bond)) {
1946                 bond_set_carrier(bond);
1947                 eth_hw_addr_random(bond_dev);
1948         }
1949
1950         unblock_netpoll_tx();
1951         synchronize_rcu();
1952         bond->slave_cnt--;
1953
1954         if (!bond_has_slaves(bond)) {
1955                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1956                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1957         }
1958
1959         bond_compute_features(bond);
1960         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1961             (old_features & NETIF_F_VLAN_CHALLENGED))
1962                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1963                             slave_dev->name, bond_dev->name);
1964
1965         vlan_vids_del_by_dev(slave_dev, bond_dev);
1966
1967         /* If the mode uses primary, then this case was handled above by
1968          * bond_change_active_slave(..., NULL)
1969          */
1970         if (!bond_uses_primary(bond)) {
1971                 /* unset promiscuity level from slave
1972                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1973                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1974                  * value of that flag before that change, as that was the value
1975                  * when this slave was attached, so we cache at the start of the
1976                  * function and use it here. Same goes for ALLMULTI below
1977                  */
1978                 if (old_flags & IFF_PROMISC)
1979                         dev_set_promiscuity(slave_dev, -1);
1980
1981                 /* unset allmulti level from slave */
1982                 if (old_flags & IFF_ALLMULTI)
1983                         dev_set_allmulti(slave_dev, -1);
1984
1985                 bond_hw_addr_flush(bond_dev, slave_dev);
1986         }
1987
1988         slave_disable_netpoll(slave);
1989
1990         /* close slave before restoring its mac address */
1991         dev_close(slave_dev);
1992
1993         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1994             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1995                 /* restore original ("permanent") mac address */
1996                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
1997                                   slave->dev->addr_len);
1998                 ss.ss_family = slave_dev->type;
1999                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
2000         }
2001
2002         if (unregister)
2003                 __dev_set_mtu(slave_dev, slave->original_mtu);
2004         else
2005                 dev_set_mtu(slave_dev, slave->original_mtu);
2006
2007         slave_dev->priv_flags &= ~IFF_BONDING;
2008
2009         bond_free_slave(slave);
2010
2011         return 0;
2012 }
2013
2014 /* A wrapper used because of ndo_del_link */
2015 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2016 {
2017         return __bond_release_one(bond_dev, slave_dev, false, false);
2018 }
2019
2020 /* First release a slave and then destroy the bond if no more slaves are left.
2021  * Must be under rtnl_lock when this function is called.
2022  */
2023 static int  bond_release_and_destroy(struct net_device *bond_dev,
2024                                      struct net_device *slave_dev)
2025 {
2026         struct bonding *bond = netdev_priv(bond_dev);
2027         int ret;
2028
2029         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2030         if (ret == 0 && !bond_has_slaves(bond)) {
2031                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2032                 netdev_info(bond_dev, "Destroying bond %s\n",
2033                             bond_dev->name);
2034                 bond_remove_proc_entry(bond);
2035                 unregister_netdevice(bond_dev);
2036         }
2037         return ret;
2038 }
2039
2040 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2041 {
2042         struct bonding *bond = netdev_priv(bond_dev);
2043         bond_fill_ifbond(bond, info);
2044 }
2045
2046 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2047 {
2048         struct bonding *bond = netdev_priv(bond_dev);
2049         struct list_head *iter;
2050         int i = 0, res = -ENODEV;
2051         struct slave *slave;
2052
2053         bond_for_each_slave(bond, slave, iter) {
2054                 if (i++ == (int)info->slave_id) {
2055                         res = 0;
2056                         bond_fill_ifslave(slave, info);
2057                         break;
2058                 }
2059         }
2060
2061         return res;
2062 }
2063
2064 /*-------------------------------- Monitoring -------------------------------*/
2065
2066 /* called with rcu_read_lock() */
2067 static int bond_miimon_inspect(struct bonding *bond)
2068 {
2069         int link_state, commit = 0;
2070         struct list_head *iter;
2071         struct slave *slave;
2072         bool ignore_updelay;
2073
2074         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2075
2076         bond_for_each_slave_rcu(bond, slave, iter) {
2077                 slave->new_link = BOND_LINK_NOCHANGE;
2078                 slave->link_new_state = slave->link;
2079
2080                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2081
2082                 switch (slave->link) {
2083                 case BOND_LINK_UP:
2084                         if (link_state)
2085                                 continue;
2086
2087                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2088                         commit++;
2089                         slave->delay = bond->params.downdelay;
2090                         if (slave->delay) {
2091                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2092                                             (BOND_MODE(bond) ==
2093                                              BOND_MODE_ACTIVEBACKUP) ?
2094                                              (bond_is_active_slave(slave) ?
2095                                               "active " : "backup ") : "",
2096                                             slave->dev->name,
2097                                             bond->params.downdelay * bond->params.miimon);
2098                         }
2099                         /*FALLTHRU*/
2100                 case BOND_LINK_FAIL:
2101                         if (link_state) {
2102                                 /* recovered before downdelay expired */
2103                                 bond_propose_link_state(slave, BOND_LINK_UP);
2104                                 slave->last_link_up = jiffies;
2105                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2106                                             (bond->params.downdelay - slave->delay) *
2107                                             bond->params.miimon,
2108                                             slave->dev->name);
2109                                 commit++;
2110                                 continue;
2111                         }
2112
2113                         if (slave->delay <= 0) {
2114                                 slave->new_link = BOND_LINK_DOWN;
2115                                 commit++;
2116                                 continue;
2117                         }
2118
2119                         slave->delay--;
2120                         break;
2121
2122                 case BOND_LINK_DOWN:
2123                         if (!link_state)
2124                                 continue;
2125
2126                         bond_propose_link_state(slave, BOND_LINK_BACK);
2127                         commit++;
2128                         slave->delay = bond->params.updelay;
2129
2130                         if (slave->delay) {
2131                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2132                                             slave->dev->name,
2133                                             ignore_updelay ? 0 :
2134                                             bond->params.updelay *
2135                                             bond->params.miimon);
2136                         }
2137                         /*FALLTHRU*/
2138                 case BOND_LINK_BACK:
2139                         if (!link_state) {
2140                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2141                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2142                                             (bond->params.updelay - slave->delay) *
2143                                             bond->params.miimon,
2144                                             slave->dev->name);
2145                                 commit++;
2146                                 continue;
2147                         }
2148
2149                         if (ignore_updelay)
2150                                 slave->delay = 0;
2151
2152                         if (slave->delay <= 0) {
2153                                 slave->new_link = BOND_LINK_UP;
2154                                 commit++;
2155                                 ignore_updelay = false;
2156                                 continue;
2157                         }
2158
2159                         slave->delay--;
2160                         break;
2161                 }
2162         }
2163
2164         return commit;
2165 }
2166
2167 static void bond_miimon_link_change(struct bonding *bond,
2168                                     struct slave *slave,
2169                                     char link)
2170 {
2171         switch (BOND_MODE(bond)) {
2172         case BOND_MODE_8023AD:
2173                 bond_3ad_handle_link_change(slave, link);
2174                 break;
2175         case BOND_MODE_TLB:
2176         case BOND_MODE_ALB:
2177                 bond_alb_handle_link_change(bond, slave, link);
2178                 break;
2179         case BOND_MODE_XOR:
2180                 bond_update_slave_arr(bond, NULL);
2181                 break;
2182         }
2183 }
2184
2185 static void bond_miimon_commit(struct bonding *bond)
2186 {
2187         struct list_head *iter;
2188         struct slave *slave, *primary;
2189
2190         bond_for_each_slave(bond, slave, iter) {
2191                 switch (slave->new_link) {
2192                 case BOND_LINK_NOCHANGE:
2193                         continue;
2194
2195                 case BOND_LINK_UP:
2196                         if (bond_update_speed_duplex(slave) &&
2197                             bond_needs_speed_duplex(bond)) {
2198                                 slave->link = BOND_LINK_DOWN;
2199                                 if (net_ratelimit())
2200                                         netdev_warn(bond->dev,
2201                                                     "failed to get link speed/duplex for %s\n",
2202                                                     slave->dev->name);
2203                                 continue;
2204                         }
2205                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2206                                                   BOND_SLAVE_NOTIFY_NOW);
2207                         slave->last_link_up = jiffies;
2208
2209                         primary = rtnl_dereference(bond->primary_slave);
2210                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2211                                 /* prevent it from being the active one */
2212                                 bond_set_backup_slave(slave);
2213                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2214                                 /* make it immediately active */
2215                                 bond_set_active_slave(slave);
2216                         } else if (slave != primary) {
2217                                 /* prevent it from being the active one */
2218                                 bond_set_backup_slave(slave);
2219                         }
2220
2221                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2222                                     slave->dev->name,
2223                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2224                                     slave->duplex ? "full" : "half");
2225
2226                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2227
2228                         if (!bond->curr_active_slave || slave == primary)
2229                                 goto do_failover;
2230
2231                         continue;
2232
2233                 case BOND_LINK_DOWN:
2234                         if (slave->link_failure_count < UINT_MAX)
2235                                 slave->link_failure_count++;
2236
2237                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2238                                                   BOND_SLAVE_NOTIFY_NOW);
2239
2240                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2241                             BOND_MODE(bond) == BOND_MODE_8023AD)
2242                                 bond_set_slave_inactive_flags(slave,
2243                                                               BOND_SLAVE_NOTIFY_NOW);
2244
2245                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2246                                     slave->dev->name);
2247
2248                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2249
2250                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2251                                 goto do_failover;
2252
2253                         continue;
2254
2255                 default:
2256                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2257                                    slave->new_link, slave->dev->name);
2258                         slave->new_link = BOND_LINK_NOCHANGE;
2259
2260                         continue;
2261                 }
2262
2263 do_failover:
2264                 block_netpoll_tx();
2265                 bond_select_active_slave(bond);
2266                 unblock_netpoll_tx();
2267         }
2268
2269         bond_set_carrier(bond);
2270 }
2271
2272 /* bond_mii_monitor
2273  *
2274  * Really a wrapper that splits the mii monitor into two phases: an
2275  * inspection, then (if inspection indicates something needs to be done)
2276  * an acquisition of appropriate locks followed by a commit phase to
2277  * implement whatever link state changes are indicated.
2278  */
2279 static void bond_mii_monitor(struct work_struct *work)
2280 {
2281         struct bonding *bond = container_of(work, struct bonding,
2282                                             mii_work.work);
2283         bool should_notify_peers = false;
2284         unsigned long delay;
2285         struct slave *slave;
2286         struct list_head *iter;
2287
2288         delay = msecs_to_jiffies(bond->params.miimon);
2289
2290         if (!bond_has_slaves(bond))
2291                 goto re_arm;
2292
2293         rcu_read_lock();
2294
2295         should_notify_peers = bond_should_notify_peers(bond);
2296
2297         if (bond_miimon_inspect(bond)) {
2298                 rcu_read_unlock();
2299
2300                 /* Race avoidance with bond_close cancel of workqueue */
2301                 if (!rtnl_trylock()) {
2302                         delay = 1;
2303                         should_notify_peers = false;
2304                         goto re_arm;
2305                 }
2306
2307                 bond_for_each_slave(bond, slave, iter) {
2308                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2309                 }
2310                 bond_miimon_commit(bond);
2311
2312                 rtnl_unlock();  /* might sleep, hold no other locks */
2313         } else
2314                 rcu_read_unlock();
2315
2316 re_arm:
2317         if (bond->params.miimon)
2318                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2319
2320         if (should_notify_peers) {
2321                 if (!rtnl_trylock())
2322                         return;
2323                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2324                 rtnl_unlock();
2325         }
2326 }
2327
2328 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2329 {
2330         __be32 ip = *((__be32 *)data);
2331
2332         return ip == bond_confirm_addr(upper, 0, ip);
2333 }
2334
2335 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2336 {
2337         bool ret = false;
2338
2339         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2340                 return true;
2341
2342         rcu_read_lock();
2343         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2344                 ret = true;
2345         rcu_read_unlock();
2346
2347         return ret;
2348 }
2349
2350 /* We go to the (large) trouble of VLAN tagging ARP frames because
2351  * switches in VLAN mode (especially if ports are configured as
2352  * "native" to a VLAN) might not pass non-tagged frames.
2353  */
2354 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2355                           __be32 dest_ip, __be32 src_ip,
2356                           struct bond_vlan_tag *tags)
2357 {
2358         struct sk_buff *skb;
2359         struct bond_vlan_tag *outer_tag = tags;
2360
2361         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2362                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2363
2364         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2365                          NULL, slave_dev->dev_addr, NULL);
2366
2367         if (!skb) {
2368                 net_err_ratelimited("ARP packet allocation failed\n");
2369                 return;
2370         }
2371
2372         if (!tags || tags->vlan_proto == VLAN_N_VID)
2373                 goto xmit;
2374
2375         tags++;
2376
2377         /* Go through all the tags backwards and add them to the packet */
2378         while (tags->vlan_proto != VLAN_N_VID) {
2379                 if (!tags->vlan_id) {
2380                         tags++;
2381                         continue;
2382                 }
2383
2384                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2385                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2386                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2387                                                 tags->vlan_id);
2388                 if (!skb) {
2389                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2390                         return;
2391                 }
2392
2393                 tags++;
2394         }
2395         /* Set the outer tag */
2396         if (outer_tag->vlan_id) {
2397                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2398                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2399                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2400                                        outer_tag->vlan_id);
2401         }
2402
2403 xmit:
2404         arp_xmit(skb);
2405 }
2406
2407 /* Validate the device path between the @start_dev and the @end_dev.
2408  * The path is valid if the @end_dev is reachable through device
2409  * stacking.
2410  * When the path is validated, collect any vlan information in the
2411  * path.
2412  */
2413 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2414                                               struct net_device *end_dev,
2415                                               int level)
2416 {
2417         struct bond_vlan_tag *tags;
2418         struct net_device *upper;
2419         struct list_head  *iter;
2420
2421         if (start_dev == end_dev) {
2422                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2423                 if (!tags)
2424                         return ERR_PTR(-ENOMEM);
2425                 tags[level].vlan_proto = VLAN_N_VID;
2426                 return tags;
2427         }
2428
2429         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2430                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2431                 if (IS_ERR_OR_NULL(tags)) {
2432                         if (IS_ERR(tags))
2433                                 return tags;
2434                         continue;
2435                 }
2436                 if (is_vlan_dev(upper)) {
2437                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2438                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2439                 }
2440
2441                 return tags;
2442         }
2443
2444         return NULL;
2445 }
2446
2447 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2448 {
2449         struct rtable *rt;
2450         struct bond_vlan_tag *tags;
2451         __be32 *targets = bond->params.arp_targets, addr;
2452         int i;
2453
2454         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2455                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2456                 tags = NULL;
2457
2458                 /* Find out through which dev should the packet go */
2459                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2460                                      RTO_ONLINK, 0);
2461                 if (IS_ERR(rt)) {
2462                         /* there's no route to target - try to send arp
2463                          * probe to generate any traffic (arp_validate=0)
2464                          */
2465                         if (bond->params.arp_validate)
2466                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2467                                                      bond->dev->name,
2468                                                      &targets[i]);
2469                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2470                                       0, tags);
2471                         continue;
2472                 }
2473
2474                 /* bond device itself */
2475                 if (rt->dst.dev == bond->dev)
2476                         goto found;
2477
2478                 rcu_read_lock();
2479                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2480                 rcu_read_unlock();
2481
2482                 if (!IS_ERR_OR_NULL(tags))
2483                         goto found;
2484
2485                 /* Not our device - skip */
2486                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2487                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2488
2489                 ip_rt_put(rt);
2490                 continue;
2491
2492 found:
2493                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2494                 ip_rt_put(rt);
2495                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2496                               addr, tags);
2497                 kfree(tags);
2498         }
2499 }
2500
2501 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2502 {
2503         int i;
2504
2505         if (!sip || !bond_has_this_ip(bond, tip)) {
2506                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2507                            &sip, &tip);
2508                 return;
2509         }
2510
2511         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2512         if (i == -1) {
2513                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2514                            &sip);
2515                 return;
2516         }
2517         slave->last_rx = jiffies;
2518         slave->target_last_arp_rx[i] = jiffies;
2519 }
2520
2521 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2522                  struct slave *slave)
2523 {
2524         struct arphdr *arp = (struct arphdr *)skb->data;
2525         struct slave *curr_active_slave, *curr_arp_slave;
2526         unsigned char *arp_ptr;
2527         __be32 sip, tip;
2528         int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2529         unsigned int alen;
2530
2531         if (!slave_do_arp_validate(bond, slave)) {
2532                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2533                     !slave_do_arp_validate_only(bond))
2534                         slave->last_rx = jiffies;
2535                 return RX_HANDLER_ANOTHER;
2536         } else if (!is_arp) {
2537                 return RX_HANDLER_ANOTHER;
2538         }
2539
2540         alen = arp_hdr_len(bond->dev);
2541
2542         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2543                    skb->dev->name);
2544
2545         if (alen > skb_headlen(skb)) {
2546                 arp = kmalloc(alen, GFP_ATOMIC);
2547                 if (!arp)
2548                         goto out_unlock;
2549                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2550                         goto out_unlock;
2551         }
2552
2553         if (arp->ar_hln != bond->dev->addr_len ||
2554             skb->pkt_type == PACKET_OTHERHOST ||
2555             skb->pkt_type == PACKET_LOOPBACK ||
2556             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2557             arp->ar_pro != htons(ETH_P_IP) ||
2558             arp->ar_pln != 4)
2559                 goto out_unlock;
2560
2561         arp_ptr = (unsigned char *)(arp + 1);
2562         arp_ptr += bond->dev->addr_len;
2563         memcpy(&sip, arp_ptr, 4);
2564         arp_ptr += 4 + bond->dev->addr_len;
2565         memcpy(&tip, arp_ptr, 4);
2566
2567         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2568                    slave->dev->name, bond_slave_state(slave),
2569                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2570                      &sip, &tip);
2571
2572         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2573         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2574
2575         /* We 'trust' the received ARP enough to validate it if:
2576          *
2577          * (a) the slave receiving the ARP is active (which includes the
2578          * current ARP slave, if any), or
2579          *
2580          * (b) the receiving slave isn't active, but there is a currently
2581          * active slave and it received valid arp reply(s) after it became
2582          * the currently active slave, or
2583          *
2584          * (c) there is an ARP slave that sent an ARP during the prior ARP
2585          * interval, and we receive an ARP reply on any slave.  We accept
2586          * these because switch FDB update delays may deliver the ARP
2587          * reply to a slave other than the sender of the ARP request.
2588          *
2589          * Note: for (b), backup slaves are receiving the broadcast ARP
2590          * request, not a reply.  This request passes from the sending
2591          * slave through the L2 switch(es) to the receiving slave.  Since
2592          * this is checking the request, sip/tip are swapped for
2593          * validation.
2594          *
2595          * This is done to avoid endless looping when we can't reach the
2596          * arp_ip_target and fool ourselves with our own arp requests.
2597          */
2598         if (bond_is_active_slave(slave))
2599                 bond_validate_arp(bond, slave, sip, tip);
2600         else if (curr_active_slave &&
2601                  time_after(slave_last_rx(bond, curr_active_slave),
2602                             curr_active_slave->last_link_up))
2603                 bond_validate_arp(bond, slave, tip, sip);
2604         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2605                  bond_time_in_interval(bond,
2606                                        dev_trans_start(curr_arp_slave->dev), 1))
2607                 bond_validate_arp(bond, slave, sip, tip);
2608
2609 out_unlock:
2610         if (arp != (struct arphdr *)skb->data)
2611                 kfree(arp);
2612         return RX_HANDLER_ANOTHER;
2613 }
2614
2615 /* function to verify if we're in the arp_interval timeslice, returns true if
2616  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2617  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2618  */
2619 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2620                                   int mod)
2621 {
2622         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2623
2624         return time_in_range(jiffies,
2625                              last_act - delta_in_ticks,
2626                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2627 }
2628
2629 /* This function is called regularly to monitor each slave's link
2630  * ensuring that traffic is being sent and received when arp monitoring
2631  * is used in load-balancing mode. if the adapter has been dormant, then an
2632  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2633  * arp monitoring in active backup mode.
2634  */
2635 static void bond_loadbalance_arp_mon(struct bonding *bond)
2636 {
2637         struct slave *slave, *oldcurrent;
2638         struct list_head *iter;
2639         int do_failover = 0, slave_state_changed = 0;
2640
2641         if (!bond_has_slaves(bond))
2642                 goto re_arm;
2643
2644         rcu_read_lock();
2645
2646         oldcurrent = rcu_dereference(bond->curr_active_slave);
2647         /* see if any of the previous devices are up now (i.e. they have
2648          * xmt and rcv traffic). the curr_active_slave does not come into
2649          * the picture unless it is null. also, slave->last_link_up is not
2650          * needed here because we send an arp on each slave and give a slave
2651          * as long as it needs to get the tx/rx within the delta.
2652          * TODO: what about up/down delay in arp mode? it wasn't here before
2653          *       so it can wait
2654          */
2655         bond_for_each_slave_rcu(bond, slave, iter) {
2656                 unsigned long trans_start = dev_trans_start(slave->dev);
2657
2658                 slave->new_link = BOND_LINK_NOCHANGE;
2659
2660                 if (slave->link != BOND_LINK_UP) {
2661                         if (bond_time_in_interval(bond, trans_start, 1) &&
2662                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2663
2664                                 slave->new_link = BOND_LINK_UP;
2665                                 slave_state_changed = 1;
2666
2667                                 /* primary_slave has no meaning in round-robin
2668                                  * mode. the window of a slave being up and
2669                                  * curr_active_slave being null after enslaving
2670                                  * is closed.
2671                                  */
2672                                 if (!oldcurrent) {
2673                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2674                                                     slave->dev->name);
2675                                         do_failover = 1;
2676                                 } else {
2677                                         netdev_info(bond->dev, "interface %s is now up\n",
2678                                                     slave->dev->name);
2679                                 }
2680                         }
2681                 } else {
2682                         /* slave->link == BOND_LINK_UP */
2683
2684                         /* not all switches will respond to an arp request
2685                          * when the source ip is 0, so don't take the link down
2686                          * if we don't know our ip yet
2687                          */
2688                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2689                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2690
2691                                 slave->new_link = BOND_LINK_DOWN;
2692                                 slave_state_changed = 1;
2693
2694                                 if (slave->link_failure_count < UINT_MAX)
2695                                         slave->link_failure_count++;
2696
2697                                 netdev_info(bond->dev, "interface %s is now down\n",
2698                                             slave->dev->name);
2699
2700                                 if (slave == oldcurrent)
2701                                         do_failover = 1;
2702                         }
2703                 }
2704
2705                 /* note: if switch is in round-robin mode, all links
2706                  * must tx arp to ensure all links rx an arp - otherwise
2707                  * links may oscillate or not come up at all; if switch is
2708                  * in something like xor mode, there is nothing we can
2709                  * do - all replies will be rx'ed on same link causing slaves
2710                  * to be unstable during low/no traffic periods
2711                  */
2712                 if (bond_slave_is_up(slave))
2713                         bond_arp_send_all(bond, slave);
2714         }
2715
2716         rcu_read_unlock();
2717
2718         if (do_failover || slave_state_changed) {
2719                 if (!rtnl_trylock())
2720                         goto re_arm;
2721
2722                 bond_for_each_slave(bond, slave, iter) {
2723                         if (slave->new_link != BOND_LINK_NOCHANGE)
2724                                 slave->link = slave->new_link;
2725                 }
2726
2727                 if (slave_state_changed) {
2728                         bond_slave_state_change(bond);
2729                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2730                                 bond_update_slave_arr(bond, NULL);
2731                 }
2732                 if (do_failover) {
2733                         block_netpoll_tx();
2734                         bond_select_active_slave(bond);
2735                         unblock_netpoll_tx();
2736                 }
2737                 rtnl_unlock();
2738         }
2739
2740 re_arm:
2741         if (bond->params.arp_interval)
2742                 queue_delayed_work(bond->wq, &bond->arp_work,
2743                                    msecs_to_jiffies(bond->params.arp_interval));
2744 }
2745
2746 /* Called to inspect slaves for active-backup mode ARP monitor link state
2747  * changes.  Sets new_link in slaves to specify what action should take
2748  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2749  * to link states must be committed.
2750  *
2751  * Called with rcu_read_lock held.
2752  */
2753 static int bond_ab_arp_inspect(struct bonding *bond)
2754 {
2755         unsigned long trans_start, last_rx;
2756         struct list_head *iter;
2757         struct slave *slave;
2758         int commit = 0;
2759
2760         bond_for_each_slave_rcu(bond, slave, iter) {
2761                 slave->new_link = BOND_LINK_NOCHANGE;
2762                 last_rx = slave_last_rx(bond, slave);
2763
2764                 if (slave->link != BOND_LINK_UP) {
2765                         if (bond_time_in_interval(bond, last_rx, 1)) {
2766                                 slave->new_link = BOND_LINK_UP;
2767                                 commit++;
2768                         }
2769                         continue;
2770                 }
2771
2772                 /* Give slaves 2*delta after being enslaved or made
2773                  * active.  This avoids bouncing, as the last receive
2774                  * times need a full ARP monitor cycle to be updated.
2775                  */
2776                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2777                         continue;
2778
2779                 /* Backup slave is down if:
2780                  * - No current_arp_slave AND
2781                  * - more than 3*delta since last receive AND
2782                  * - the bond has an IP address
2783                  *
2784                  * Note: a non-null current_arp_slave indicates
2785                  * the curr_active_slave went down and we are
2786                  * searching for a new one; under this condition
2787                  * we only take the curr_active_slave down - this
2788                  * gives each slave a chance to tx/rx traffic
2789                  * before being taken out
2790                  */
2791                 if (!bond_is_active_slave(slave) &&
2792                     !rcu_access_pointer(bond->current_arp_slave) &&
2793                     !bond_time_in_interval(bond, last_rx, 3)) {
2794                         slave->new_link = BOND_LINK_DOWN;
2795                         commit++;
2796                 }
2797
2798                 /* Active slave is down if:
2799                  * - more than 2*delta since transmitting OR
2800                  * - (more than 2*delta since receive AND
2801                  *    the bond has an IP address)
2802                  */
2803                 trans_start = dev_trans_start(slave->dev);
2804                 if (bond_is_active_slave(slave) &&
2805                     (!bond_time_in_interval(bond, trans_start, 2) ||
2806                      !bond_time_in_interval(bond, last_rx, 2))) {
2807                         slave->new_link = BOND_LINK_DOWN;
2808                         commit++;
2809                 }
2810         }
2811
2812         return commit;
2813 }
2814
2815 /* Called to commit link state changes noted by inspection step of
2816  * active-backup mode ARP monitor.
2817  *
2818  * Called with RTNL hold.
2819  */
2820 static void bond_ab_arp_commit(struct bonding *bond)
2821 {
2822         unsigned long trans_start;
2823         struct list_head *iter;
2824         struct slave *slave;
2825
2826         bond_for_each_slave(bond, slave, iter) {
2827                 switch (slave->new_link) {
2828                 case BOND_LINK_NOCHANGE:
2829                         continue;
2830
2831                 case BOND_LINK_UP:
2832                         trans_start = dev_trans_start(slave->dev);
2833                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2834                             (!rtnl_dereference(bond->curr_active_slave) &&
2835                              bond_time_in_interval(bond, trans_start, 1))) {
2836                                 struct slave *current_arp_slave;
2837
2838                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2839                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2840                                                           BOND_SLAVE_NOTIFY_NOW);
2841                                 if (current_arp_slave) {
2842                                         bond_set_slave_inactive_flags(
2843                                                 current_arp_slave,
2844                                                 BOND_SLAVE_NOTIFY_NOW);
2845                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2846                                 }
2847
2848                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2849                                             slave->dev->name);
2850
2851                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2852                                     slave == rtnl_dereference(bond->primary_slave))
2853                                         goto do_failover;
2854
2855                         }
2856
2857                         continue;
2858
2859                 case BOND_LINK_DOWN:
2860                         if (slave->link_failure_count < UINT_MAX)
2861                                 slave->link_failure_count++;
2862
2863                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2864                                                   BOND_SLAVE_NOTIFY_NOW);
2865                         bond_set_slave_inactive_flags(slave,
2866                                                       BOND_SLAVE_NOTIFY_NOW);
2867
2868                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2869                                     slave->dev->name);
2870
2871                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2872                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2873                                 goto do_failover;
2874                         }
2875
2876                         continue;
2877
2878                 default:
2879                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2880                                    slave->new_link, slave->dev->name);
2881                         continue;
2882                 }
2883
2884 do_failover:
2885                 block_netpoll_tx();
2886                 bond_select_active_slave(bond);
2887                 unblock_netpoll_tx();
2888         }
2889
2890         bond_set_carrier(bond);
2891 }
2892
2893 /* Send ARP probes for active-backup mode ARP monitor.
2894  *
2895  * Called with rcu_read_lock held.
2896  */
2897 static bool bond_ab_arp_probe(struct bonding *bond)
2898 {
2899         struct slave *slave, *before = NULL, *new_slave = NULL,
2900                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2901                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2902         struct list_head *iter;
2903         bool found = false;
2904         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2905
2906         if (curr_arp_slave && curr_active_slave)
2907                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2908                             curr_arp_slave->dev->name,
2909                             curr_active_slave->dev->name);
2910
2911         if (curr_active_slave) {
2912                 bond_arp_send_all(bond, curr_active_slave);
2913                 return should_notify_rtnl;
2914         }
2915
2916         /* if we don't have a curr_active_slave, search for the next available
2917          * backup slave from the current_arp_slave and make it the candidate
2918          * for becoming the curr_active_slave
2919          */
2920
2921         if (!curr_arp_slave) {
2922                 curr_arp_slave = bond_first_slave_rcu(bond);
2923                 if (!curr_arp_slave)
2924                         return should_notify_rtnl;
2925         }
2926
2927         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2928
2929         bond_for_each_slave_rcu(bond, slave, iter) {
2930                 if (!found && !before && bond_slave_is_up(slave))
2931                         before = slave;
2932
2933                 if (found && !new_slave && bond_slave_is_up(slave))
2934                         new_slave = slave;
2935                 /* if the link state is up at this point, we
2936                  * mark it down - this can happen if we have
2937                  * simultaneous link failures and
2938                  * reselect_active_interface doesn't make this
2939                  * one the current slave so it is still marked
2940                  * up when it is actually down
2941                  */
2942                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2943                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2944                                                   BOND_SLAVE_NOTIFY_LATER);
2945                         if (slave->link_failure_count < UINT_MAX)
2946                                 slave->link_failure_count++;
2947
2948                         bond_set_slave_inactive_flags(slave,
2949                                                       BOND_SLAVE_NOTIFY_LATER);
2950
2951                         netdev_info(bond->dev, "backup interface %s is now down\n",
2952                                     slave->dev->name);
2953                 }
2954                 if (slave == curr_arp_slave)
2955                         found = true;
2956         }
2957
2958         if (!new_slave && before)
2959                 new_slave = before;
2960
2961         if (!new_slave)
2962                 goto check_state;
2963
2964         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2965                                   BOND_SLAVE_NOTIFY_LATER);
2966         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2967         bond_arp_send_all(bond, new_slave);
2968         new_slave->last_link_up = jiffies;
2969         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2970
2971 check_state:
2972         bond_for_each_slave_rcu(bond, slave, iter) {
2973                 if (slave->should_notify || slave->should_notify_link) {
2974                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2975                         break;
2976                 }
2977         }
2978         return should_notify_rtnl;
2979 }
2980
2981 static void bond_activebackup_arp_mon(struct bonding *bond)
2982 {
2983         bool should_notify_peers = false;
2984         bool should_notify_rtnl = false;
2985         int delta_in_ticks;
2986
2987         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2988
2989         if (!bond_has_slaves(bond))
2990                 goto re_arm;
2991
2992         rcu_read_lock();
2993
2994         should_notify_peers = bond_should_notify_peers(bond);
2995
2996         if (bond_ab_arp_inspect(bond)) {
2997                 rcu_read_unlock();
2998
2999                 /* Race avoidance with bond_close flush of workqueue */
3000                 if (!rtnl_trylock()) {
3001                         delta_in_ticks = 1;
3002                         should_notify_peers = false;
3003                         goto re_arm;
3004                 }
3005
3006                 bond_ab_arp_commit(bond);
3007
3008                 rtnl_unlock();
3009                 rcu_read_lock();
3010         }
3011
3012         should_notify_rtnl = bond_ab_arp_probe(bond);
3013         rcu_read_unlock();
3014
3015 re_arm:
3016         if (bond->params.arp_interval)
3017                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3018
3019         if (should_notify_peers || should_notify_rtnl) {
3020                 if (!rtnl_trylock())
3021                         return;
3022
3023                 if (should_notify_peers)
3024                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3025                                                  bond->dev);
3026                 if (should_notify_rtnl) {
3027                         bond_slave_state_notify(bond);
3028                         bond_slave_link_notify(bond);
3029                 }
3030
3031                 rtnl_unlock();
3032         }
3033 }
3034
3035 static void bond_arp_monitor(struct work_struct *work)
3036 {
3037         struct bonding *bond = container_of(work, struct bonding,
3038                                             arp_work.work);
3039
3040         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3041                 bond_activebackup_arp_mon(bond);
3042         else
3043                 bond_loadbalance_arp_mon(bond);
3044 }
3045
3046 /*-------------------------- netdev event handling --------------------------*/
3047
3048 /* Change device name */
3049 static int bond_event_changename(struct bonding *bond)
3050 {
3051         bond_remove_proc_entry(bond);
3052         bond_create_proc_entry(bond);
3053
3054         bond_debug_reregister(bond);
3055
3056         return NOTIFY_DONE;
3057 }
3058
3059 static int bond_master_netdev_event(unsigned long event,
3060                                     struct net_device *bond_dev)
3061 {
3062         struct bonding *event_bond = netdev_priv(bond_dev);
3063
3064         switch (event) {
3065         case NETDEV_CHANGENAME:
3066                 return bond_event_changename(event_bond);
3067         case NETDEV_UNREGISTER:
3068                 bond_remove_proc_entry(event_bond);
3069                 break;
3070         case NETDEV_REGISTER:
3071                 bond_create_proc_entry(event_bond);
3072                 break;
3073         case NETDEV_NOTIFY_PEERS:
3074                 if (event_bond->send_peer_notif)
3075                         event_bond->send_peer_notif--;
3076                 break;
3077         default:
3078                 break;
3079         }
3080
3081         return NOTIFY_DONE;
3082 }
3083
3084 static int bond_slave_netdev_event(unsigned long event,
3085                                    struct net_device *slave_dev)
3086 {
3087         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3088         struct bonding *bond;
3089         struct net_device *bond_dev;
3090
3091         /* A netdev event can be generated while enslaving a device
3092          * before netdev_rx_handler_register is called in which case
3093          * slave will be NULL
3094          */
3095         if (!slave)
3096                 return NOTIFY_DONE;
3097         bond_dev = slave->bond->dev;
3098         bond = slave->bond;
3099         primary = rtnl_dereference(bond->primary_slave);
3100
3101         switch (event) {
3102         case NETDEV_UNREGISTER:
3103                 if (bond_dev->type != ARPHRD_ETHER)
3104                         bond_release_and_destroy(bond_dev, slave_dev);
3105                 else
3106                         __bond_release_one(bond_dev, slave_dev, false, true);
3107                 break;
3108         case NETDEV_UP:
3109         case NETDEV_CHANGE:
3110                 /* For 802.3ad mode only:
3111                  * Getting invalid Speed/Duplex values here will put slave
3112                  * in weird state. So mark it as link-down for the time
3113                  * being and let link-monitoring (miimon) set it right when
3114                  * correct speeds/duplex are available.
3115                  */
3116                 if (bond_update_speed_duplex(slave) &&
3117                     BOND_MODE(bond) == BOND_MODE_8023AD)
3118                         slave->link = BOND_LINK_DOWN;
3119
3120                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3121                         bond_3ad_adapter_speed_duplex_changed(slave);
3122                 /* Fallthrough */
3123         case NETDEV_DOWN:
3124                 /* Refresh slave-array if applicable!
3125                  * If the setup does not use miimon or arpmon (mode-specific!),
3126                  * then these events will not cause the slave-array to be
3127                  * refreshed. This will cause xmit to use a slave that is not
3128                  * usable. Avoid such situation by refeshing the array at these
3129                  * events. If these (miimon/arpmon) parameters are configured
3130                  * then array gets refreshed twice and that should be fine!
3131                  */
3132                 if (bond_mode_can_use_xmit_hash(bond))
3133                         bond_update_slave_arr(bond, NULL);
3134                 break;
3135         case NETDEV_CHANGEMTU:
3136                 /* TODO: Should slaves be allowed to
3137                  * independently alter their MTU?  For
3138                  * an active-backup bond, slaves need
3139                  * not be the same type of device, so
3140                  * MTUs may vary.  For other modes,
3141                  * slaves arguably should have the
3142                  * same MTUs. To do this, we'd need to
3143                  * take over the slave's change_mtu
3144                  * function for the duration of their
3145                  * servitude.
3146                  */
3147                 break;
3148         case NETDEV_CHANGENAME:
3149                 /* we don't care if we don't have primary set */
3150                 if (!bond_uses_primary(bond) ||
3151                     !bond->params.primary[0])
3152                         break;
3153
3154                 if (slave == primary) {
3155                         /* slave's name changed - he's no longer primary */
3156                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3157                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3158                         /* we have a new primary slave */
3159                         rcu_assign_pointer(bond->primary_slave, slave);
3160                 } else { /* we didn't change primary - exit */
3161                         break;
3162                 }
3163
3164                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3165                             primary ? slave_dev->name : "none");
3166
3167                 block_netpoll_tx();
3168                 bond_select_active_slave(bond);
3169                 unblock_netpoll_tx();
3170                 break;
3171         case NETDEV_FEAT_CHANGE:
3172                 bond_compute_features(bond);
3173                 break;
3174         case NETDEV_RESEND_IGMP:
3175                 /* Propagate to master device */
3176                 call_netdevice_notifiers(event, slave->bond->dev);
3177                 break;
3178         default:
3179                 break;
3180         }
3181
3182         return NOTIFY_DONE;
3183 }
3184
3185 /* bond_netdev_event: handle netdev notifier chain events.
3186  *
3187  * This function receives events for the netdev chain.  The caller (an
3188  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3189  * locks for us to safely manipulate the slave devices (RTNL lock,
3190  * dev_probe_lock).
3191  */
3192 static int bond_netdev_event(struct notifier_block *this,
3193                              unsigned long event, void *ptr)
3194 {
3195         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3196
3197         netdev_dbg(event_dev, "event: %lx\n", event);
3198
3199         if (!(event_dev->priv_flags & IFF_BONDING))
3200                 return NOTIFY_DONE;
3201
3202         if (event_dev->flags & IFF_MASTER) {
3203                 netdev_dbg(event_dev, "IFF_MASTER\n");
3204                 return bond_master_netdev_event(event, event_dev);
3205         }
3206
3207         if (event_dev->flags & IFF_SLAVE) {
3208                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3209                 return bond_slave_netdev_event(event, event_dev);
3210         }
3211
3212         return NOTIFY_DONE;
3213 }
3214
3215 static struct notifier_block bond_netdev_notifier = {
3216         .notifier_call = bond_netdev_event,
3217 };
3218
3219 /*---------------------------- Hashing Policies -----------------------------*/
3220
3221 /* L2 hash helper */
3222 static inline u32 bond_eth_hash(struct sk_buff *skb)
3223 {
3224         struct ethhdr *ep, hdr_tmp;
3225
3226         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3227         if (ep)
3228                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3229         return 0;
3230 }
3231
3232 /* Extract the appropriate headers based on bond's xmit policy */
3233 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3234                               struct flow_keys *fk)
3235 {
3236         const struct ipv6hdr *iph6;
3237         const struct iphdr *iph;
3238         int noff, proto = -1;
3239
3240         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3241                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3242
3243         fk->ports.ports = 0;
3244         noff = skb_network_offset(skb);
3245         if (skb->protocol == htons(ETH_P_IP)) {
3246                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3247                         return false;
3248                 iph = ip_hdr(skb);
3249                 iph_to_flow_copy_v4addrs(fk, iph);
3250                 noff += iph->ihl << 2;
3251                 if (!ip_is_fragment(iph))
3252                         proto = iph->protocol;
3253         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3254                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3255                         return false;
3256                 iph6 = ipv6_hdr(skb);
3257                 iph_to_flow_copy_v6addrs(fk, iph6);
3258                 noff += sizeof(*iph6);
3259                 proto = iph6->nexthdr;
3260         } else {
3261                 return false;
3262         }
3263         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3264                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3265
3266         return true;
3267 }
3268
3269 /**
3270  * bond_xmit_hash - generate a hash value based on the xmit policy
3271  * @bond: bonding device
3272  * @skb: buffer to use for headers
3273  *
3274  * This function will extract the necessary headers from the skb buffer and use
3275  * them to generate a hash based on the xmit_policy set in the bonding device
3276  */
3277 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3278 {
3279         struct flow_keys flow;
3280         u32 hash;
3281
3282         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3283             skb->l4_hash)
3284                 return skb->hash;
3285
3286         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3287             !bond_flow_dissect(bond, skb, &flow))
3288                 return bond_eth_hash(skb);
3289
3290         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3291             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3292                 hash = bond_eth_hash(skb);
3293         else
3294                 hash = (__force u32)flow.ports.ports;
3295         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3296                 (__force u32)flow_get_u32_src(&flow);
3297         hash ^= (hash >> 16);
3298         hash ^= (hash >> 8);
3299
3300         return hash >> 1;
3301 }
3302
3303 /*-------------------------- Device entry points ----------------------------*/
3304
3305 void bond_work_init_all(struct bonding *bond)
3306 {
3307         INIT_DELAYED_WORK(&bond->mcast_work,
3308                           bond_resend_igmp_join_requests_delayed);
3309         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3310         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3311         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3312         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3313         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3314 }
3315
3316 static void bond_work_cancel_all(struct bonding *bond)
3317 {
3318         cancel_delayed_work_sync(&bond->mii_work);
3319         cancel_delayed_work_sync(&bond->arp_work);
3320         cancel_delayed_work_sync(&bond->alb_work);
3321         cancel_delayed_work_sync(&bond->ad_work);
3322         cancel_delayed_work_sync(&bond->mcast_work);
3323         cancel_delayed_work_sync(&bond->slave_arr_work);
3324 }
3325
3326 static int bond_open(struct net_device *bond_dev)
3327 {
3328         struct bonding *bond = netdev_priv(bond_dev);
3329         struct list_head *iter;
3330         struct slave *slave;
3331
3332         /* reset slave->backup and slave->inactive */
3333         if (bond_has_slaves(bond)) {
3334                 bond_for_each_slave(bond, slave, iter) {
3335                         if (bond_uses_primary(bond) &&
3336                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3337                                 bond_set_slave_inactive_flags(slave,
3338                                                               BOND_SLAVE_NOTIFY_NOW);
3339                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3340                                 bond_set_slave_active_flags(slave,
3341                                                             BOND_SLAVE_NOTIFY_NOW);
3342                         }
3343                 }
3344         }
3345
3346         if (bond_is_lb(bond)) {
3347                 /* bond_alb_initialize must be called before the timer
3348                  * is started.
3349                  */
3350                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3351                         return -ENOMEM;
3352                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3353                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3354         }
3355
3356         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3357                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3358
3359         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3360                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3361                 bond->recv_probe = bond_arp_rcv;
3362         }
3363
3364         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3365                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3366                 /* register to receive LACPDUs */
3367                 bond->recv_probe = bond_3ad_lacpdu_recv;
3368                 bond_3ad_initiate_agg_selection(bond, 1);
3369         }
3370
3371         if (bond_mode_can_use_xmit_hash(bond))
3372                 bond_update_slave_arr(bond, NULL);
3373
3374         return 0;
3375 }
3376
3377 static int bond_close(struct net_device *bond_dev)
3378 {
3379         struct bonding *bond = netdev_priv(bond_dev);
3380
3381         bond_work_cancel_all(bond);
3382         bond->send_peer_notif = 0;
3383         if (bond_is_lb(bond))
3384                 bond_alb_deinitialize(bond);
3385         bond->recv_probe = NULL;
3386
3387         return 0;
3388 }
3389
3390 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3391  * that some drivers can provide 32bit values only.
3392  */
3393 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3394                             const struct rtnl_link_stats64 *_new,
3395                             const struct rtnl_link_stats64 *_old)
3396 {
3397         const u64 *new = (const u64 *)_new;
3398         const u64 *old = (const u64 *)_old;
3399         u64 *res = (u64 *)_res;
3400         int i;
3401
3402         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3403                 u64 nv = new[i];
3404                 u64 ov = old[i];
3405                 s64 delta = nv - ov;
3406
3407                 /* detects if this particular field is 32bit only */
3408                 if (((nv | ov) >> 32) == 0)
3409                         delta = (s64)(s32)((u32)nv - (u32)ov);
3410
3411                 /* filter anomalies, some drivers reset their stats
3412                  * at down/up events.
3413                  */
3414                 if (delta > 0)
3415                         res[i] += delta;
3416         }
3417 }
3418
3419 static int bond_get_nest_level(struct net_device *bond_dev)
3420 {
3421         struct bonding *bond = netdev_priv(bond_dev);
3422
3423         return bond->nest_level;
3424 }
3425
3426 static void bond_get_stats(struct net_device *bond_dev,
3427                            struct rtnl_link_stats64 *stats)
3428 {
3429         struct bonding *bond = netdev_priv(bond_dev);
3430         struct rtnl_link_stats64 temp;
3431         struct list_head *iter;
3432         struct slave *slave;
3433
3434         spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3435         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3436
3437         rcu_read_lock();
3438         bond_for_each_slave_rcu(bond, slave, iter) {
3439                 const struct rtnl_link_stats64 *new =
3440                         dev_get_stats(slave->dev, &temp);
3441
3442                 bond_fold_stats(stats, new, &slave->slave_stats);
3443
3444                 /* save off the slave stats for the next run */
3445                 memcpy(&slave->slave_stats, new, sizeof(*new));
3446         }
3447         rcu_read_unlock();
3448
3449         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3450         spin_unlock(&bond->stats_lock);
3451 }
3452
3453 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3454 {
3455         struct bonding *bond = netdev_priv(bond_dev);
3456         struct net_device *slave_dev = NULL;
3457         struct ifbond k_binfo;
3458         struct ifbond __user *u_binfo = NULL;
3459         struct ifslave k_sinfo;
3460         struct ifslave __user *u_sinfo = NULL;
3461         struct mii_ioctl_data *mii = NULL;
3462         struct bond_opt_value newval;
3463         struct net *net;
3464         int res = 0;
3465
3466         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3467
3468         switch (cmd) {
3469         case SIOCGMIIPHY:
3470                 mii = if_mii(ifr);
3471                 if (!mii)
3472                         return -EINVAL;
3473
3474                 mii->phy_id = 0;
3475                 /* Fall Through */
3476         case SIOCGMIIREG:
3477                 /* We do this again just in case we were called by SIOCGMIIREG
3478                  * instead of SIOCGMIIPHY.
3479                  */
3480                 mii = if_mii(ifr);
3481                 if (!mii)
3482                         return -EINVAL;
3483
3484                 if (mii->reg_num == 1) {
3485                         mii->val_out = 0;
3486                         if (netif_carrier_ok(bond->dev))
3487                                 mii->val_out = BMSR_LSTATUS;
3488                 }
3489
3490                 return 0;
3491         case BOND_INFO_QUERY_OLD:
3492         case SIOCBONDINFOQUERY:
3493                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3494
3495                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3496                         return -EFAULT;
3497
3498                 bond_info_query(bond_dev, &k_binfo);
3499                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3500                         return -EFAULT;
3501
3502                 return 0;
3503         case BOND_SLAVE_INFO_QUERY_OLD:
3504         case SIOCBONDSLAVEINFOQUERY:
3505                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3506
3507                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3508                         return -EFAULT;
3509
3510                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3511                 if (res == 0 &&
3512                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3513                         return -EFAULT;
3514
3515                 return res;
3516         default:
3517                 break;
3518         }
3519
3520         net = dev_net(bond_dev);
3521
3522         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3523                 return -EPERM;
3524
3525         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3526
3527         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3528
3529         if (!slave_dev)
3530                 return -ENODEV;
3531
3532         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3533         switch (cmd) {
3534         case BOND_ENSLAVE_OLD:
3535         case SIOCBONDENSLAVE:
3536                 res = bond_enslave(bond_dev, slave_dev, NULL);
3537                 break;
3538         case BOND_RELEASE_OLD:
3539         case SIOCBONDRELEASE:
3540                 res = bond_release(bond_dev, slave_dev);
3541                 break;
3542         case BOND_SETHWADDR_OLD:
3543         case SIOCBONDSETHWADDR:
3544                 bond_set_dev_addr(bond_dev, slave_dev);
3545                 res = 0;
3546                 break;
3547         case BOND_CHANGE_ACTIVE_OLD:
3548         case SIOCBONDCHANGEACTIVE:
3549                 bond_opt_initstr(&newval, slave_dev->name);
3550                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3551                                             &newval);
3552                 break;
3553         default:
3554                 res = -EOPNOTSUPP;
3555         }
3556
3557         return res;
3558 }
3559
3560 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3561 {
3562         struct bonding *bond = netdev_priv(bond_dev);
3563
3564         if (change & IFF_PROMISC)
3565                 bond_set_promiscuity(bond,
3566                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3567
3568         if (change & IFF_ALLMULTI)
3569                 bond_set_allmulti(bond,
3570                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3571 }
3572
3573 static void bond_set_rx_mode(struct net_device *bond_dev)
3574 {
3575         struct bonding *bond = netdev_priv(bond_dev);
3576         struct list_head *iter;
3577         struct slave *slave;
3578
3579         rcu_read_lock();
3580         if (bond_uses_primary(bond)) {
3581                 slave = rcu_dereference(bond->curr_active_slave);
3582                 if (slave) {
3583                         dev_uc_sync(slave->dev, bond_dev);
3584                         dev_mc_sync(slave->dev, bond_dev);
3585                 }
3586         } else {
3587                 bond_for_each_slave_rcu(bond, slave, iter) {
3588                         dev_uc_sync_multiple(slave->dev, bond_dev);
3589                         dev_mc_sync_multiple(slave->dev, bond_dev);
3590                 }
3591         }
3592         rcu_read_unlock();
3593 }
3594
3595 static int bond_neigh_init(struct neighbour *n)
3596 {
3597         struct bonding *bond = netdev_priv(n->dev);
3598         const struct net_device_ops *slave_ops;
3599         struct neigh_parms parms;
3600         struct slave *slave;
3601         int ret;
3602
3603         slave = bond_first_slave(bond);
3604         if (!slave)
3605                 return 0;
3606         slave_ops = slave->dev->netdev_ops;
3607         if (!slave_ops->ndo_neigh_setup)
3608                 return 0;
3609
3610         parms.neigh_setup = NULL;
3611         parms.neigh_cleanup = NULL;
3612         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3613         if (ret)
3614                 return ret;
3615
3616         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3617          * after the last slave has been detached.  Assumes that all slaves
3618          * utilize the same neigh_cleanup (true at this writing as only user
3619          * is ipoib).
3620          */
3621         n->parms->neigh_cleanup = parms.neigh_cleanup;
3622
3623         if (!parms.neigh_setup)
3624                 return 0;
3625
3626         return parms.neigh_setup(n);
3627 }
3628
3629 /* The bonding ndo_neigh_setup is called at init time beofre any
3630  * slave exists. So we must declare proxy setup function which will
3631  * be used at run time to resolve the actual slave neigh param setup.
3632  *
3633  * It's also called by master devices (such as vlans) to setup their
3634  * underlying devices. In that case - do nothing, we're already set up from
3635  * our init.
3636  */
3637 static int bond_neigh_setup(struct net_device *dev,
3638                             struct neigh_parms *parms)
3639 {
3640         /* modify only our neigh_parms */
3641         if (parms->dev == dev)
3642                 parms->neigh_setup = bond_neigh_init;
3643
3644         return 0;
3645 }
3646
3647 /* Change the MTU of all of a master's slaves to match the master */
3648 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3649 {
3650         struct bonding *bond = netdev_priv(bond_dev);
3651         struct slave *slave, *rollback_slave;
3652         struct list_head *iter;
3653         int res = 0;
3654
3655         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3656
3657         bond_for_each_slave(bond, slave, iter) {
3658                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3659                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3660
3661                 res = dev_set_mtu(slave->dev, new_mtu);
3662
3663                 if (res) {
3664                         /* If we failed to set the slave's mtu to the new value
3665                          * we must abort the operation even in ACTIVE_BACKUP
3666                          * mode, because if we allow the backup slaves to have
3667                          * different mtu values than the active slave we'll
3668                          * need to change their mtu when doing a failover. That
3669                          * means changing their mtu from timer context, which
3670                          * is probably not a good idea.
3671                          */
3672                         netdev_dbg(bond_dev, "err %d %s\n", res,
3673                                    slave->dev->name);
3674                         goto unwind;
3675                 }
3676         }
3677
3678         bond_dev->mtu = new_mtu;
3679
3680         return 0;
3681
3682 unwind:
3683         /* unwind from head to the slave that failed */
3684         bond_for_each_slave(bond, rollback_slave, iter) {
3685                 int tmp_res;
3686
3687                 if (rollback_slave == slave)
3688                         break;
3689
3690                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3691                 if (tmp_res) {
3692                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3693                                    tmp_res, rollback_slave->dev->name);
3694                 }
3695         }
3696
3697         return res;
3698 }
3699
3700 /* Change HW address
3701  *
3702  * Note that many devices must be down to change the HW address, and
3703  * downing the master releases all slaves.  We can make bonds full of
3704  * bonding devices to test this, however.
3705  */
3706 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3707 {
3708         struct bonding *bond = netdev_priv(bond_dev);
3709         struct slave *slave, *rollback_slave;
3710         struct sockaddr_storage *ss = addr, tmp_ss;
3711         struct list_head *iter;
3712         int res = 0;
3713
3714         if (BOND_MODE(bond) == BOND_MODE_ALB)
3715                 return bond_alb_set_mac_address(bond_dev, addr);
3716
3717
3718         netdev_dbg(bond_dev, "bond=%p\n", bond);
3719
3720         /* If fail_over_mac is enabled, do nothing and return success.
3721          * Returning an error causes ifenslave to fail.
3722          */
3723         if (bond->params.fail_over_mac &&
3724             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3725                 return 0;
3726
3727         if (!is_valid_ether_addr(ss->__data))
3728                 return -EADDRNOTAVAIL;
3729
3730         bond_for_each_slave(bond, slave, iter) {
3731                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3732                 res = dev_set_mac_address(slave->dev, addr);
3733                 if (res) {
3734                         /* TODO: consider downing the slave
3735                          * and retry ?
3736                          * User should expect communications
3737                          * breakage anyway until ARP finish
3738                          * updating, so...
3739                          */
3740                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3741                         goto unwind;
3742                 }
3743         }
3744
3745         /* success */
3746         memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3747         return 0;
3748
3749 unwind:
3750         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3751         tmp_ss.ss_family = bond_dev->type;
3752
3753         /* unwind from head to the slave that failed */
3754         bond_for_each_slave(bond, rollback_slave, iter) {
3755                 int tmp_res;
3756
3757                 if (rollback_slave == slave)
3758                         break;
3759
3760                 tmp_res = dev_set_mac_address(rollback_slave->dev,
3761                                               (struct sockaddr *)&tmp_ss);
3762                 if (tmp_res) {
3763                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3764                                    tmp_res, rollback_slave->dev->name);
3765                 }
3766         }
3767
3768         return res;
3769 }
3770
3771 /**
3772  * bond_xmit_slave_id - transmit skb through slave with slave_id
3773  * @bond: bonding device that is transmitting
3774  * @skb: buffer to transmit
3775  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3776  *
3777  * This function tries to transmit through slave with slave_id but in case
3778  * it fails, it tries to find the first available slave for transmission.
3779  * The skb is consumed in all cases, thus the function is void.
3780  */
3781 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3782 {
3783         struct list_head *iter;
3784         struct slave *slave;
3785         int i = slave_id;
3786
3787         /* Here we start from the slave with slave_id */
3788         bond_for_each_slave_rcu(bond, slave, iter) {
3789                 if (--i < 0) {
3790                         if (bond_slave_can_tx(slave)) {
3791                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3792                                 return;
3793                         }
3794                 }
3795         }
3796
3797         /* Here we start from the first slave up to slave_id */
3798         i = slave_id;
3799         bond_for_each_slave_rcu(bond, slave, iter) {
3800                 if (--i < 0)
3801                         break;
3802                 if (bond_slave_can_tx(slave)) {
3803                         bond_dev_queue_xmit(bond, skb, slave->dev);
3804                         return;
3805                 }
3806         }
3807         /* no slave that can tx has been found */
3808         bond_tx_drop(bond->dev, skb);
3809 }
3810
3811 /**
3812  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3813  * @bond: bonding device to use
3814  *
3815  * Based on the value of the bonding device's packets_per_slave parameter
3816  * this function generates a slave id, which is usually used as the next
3817  * slave to transmit through.
3818  */
3819 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3820 {
3821         u32 slave_id;
3822         struct reciprocal_value reciprocal_packets_per_slave;
3823         int packets_per_slave = bond->params.packets_per_slave;
3824
3825         switch (packets_per_slave) {
3826         case 0:
3827                 slave_id = prandom_u32();
3828                 break;
3829         case 1:
3830                 slave_id = bond->rr_tx_counter;
3831                 break;
3832         default:
3833                 reciprocal_packets_per_slave =
3834                         bond->params.reciprocal_packets_per_slave;
3835                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3836                                              reciprocal_packets_per_slave);
3837                 break;
3838         }
3839         bond->rr_tx_counter++;
3840
3841         return slave_id;
3842 }
3843
3844 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3845                                         struct net_device *bond_dev)
3846 {
3847         struct bonding *bond = netdev_priv(bond_dev);
3848         struct iphdr *iph = ip_hdr(skb);
3849         struct slave *slave;
3850         u32 slave_id;
3851
3852         /* Start with the curr_active_slave that joined the bond as the
3853          * default for sending IGMP traffic.  For failover purposes one
3854          * needs to maintain some consistency for the interface that will
3855          * send the join/membership reports.  The curr_active_slave found
3856          * will send all of this type of traffic.
3857          */
3858         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3859                 slave = rcu_dereference(bond->curr_active_slave);
3860                 if (slave)
3861                         bond_dev_queue_xmit(bond, skb, slave->dev);
3862                 else
3863                         bond_xmit_slave_id(bond, skb, 0);
3864         } else {
3865                 int slave_cnt = READ_ONCE(bond->slave_cnt);
3866
3867                 if (likely(slave_cnt)) {
3868                         slave_id = bond_rr_gen_slave_id(bond);
3869                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3870                 } else {
3871                         bond_tx_drop(bond_dev, skb);
3872                 }
3873         }
3874
3875         return NETDEV_TX_OK;
3876 }
3877
3878 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3879  * the bond has a usable interface.
3880  */
3881 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3882                                           struct net_device *bond_dev)
3883 {
3884         struct bonding *bond = netdev_priv(bond_dev);
3885         struct slave *slave;
3886
3887         slave = rcu_dereference(bond->curr_active_slave);
3888         if (slave)
3889                 bond_dev_queue_xmit(bond, skb, slave->dev);
3890         else
3891                 bond_tx_drop(bond_dev, skb);
3892
3893         return NETDEV_TX_OK;
3894 }
3895
3896 /* Use this to update slave_array when (a) it's not appropriate to update
3897  * slave_array right away (note that update_slave_array() may sleep)
3898  * and / or (b) RTNL is not held.
3899  */
3900 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3901 {
3902         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3903 }
3904
3905 /* Slave array work handler. Holds only RTNL */
3906 static void bond_slave_arr_handler(struct work_struct *work)
3907 {
3908         struct bonding *bond = container_of(work, struct bonding,
3909                                             slave_arr_work.work);
3910         int ret;
3911
3912         if (!rtnl_trylock())
3913                 goto err;
3914
3915         ret = bond_update_slave_arr(bond, NULL);
3916         rtnl_unlock();
3917         if (ret) {
3918                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3919                 goto err;
3920         }
3921         return;
3922
3923 err:
3924         bond_slave_arr_work_rearm(bond, 1);
3925 }
3926
3927 /* Build the usable slaves array in control path for modes that use xmit-hash
3928  * to determine the slave interface -
3929  * (a) BOND_MODE_8023AD
3930  * (b) BOND_MODE_XOR
3931  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3932  *
3933  * The caller is expected to hold RTNL only and NO other lock!
3934  */
3935 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3936 {
3937         struct slave *slave;
3938         struct list_head *iter;
3939         struct bond_up_slave *new_arr, *old_arr;
3940         int agg_id = 0;
3941         int ret = 0;
3942
3943 #ifdef CONFIG_LOCKDEP
3944         WARN_ON(lockdep_is_held(&bond->mode_lock));
3945 #endif
3946
3947         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3948                           GFP_KERNEL);
3949         if (!new_arr) {
3950                 ret = -ENOMEM;
3951                 pr_err("Failed to build slave-array.\n");
3952                 goto out;
3953         }
3954         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3955                 struct ad_info ad_info;
3956
3957                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3958                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3959                         kfree_rcu(new_arr, rcu);
3960                         /* No active aggragator means it's not safe to use
3961                          * the previous array.
3962                          */
3963                         old_arr = rtnl_dereference(bond->slave_arr);
3964                         if (old_arr) {
3965                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3966                                 kfree_rcu(old_arr, rcu);
3967                         }
3968                         goto out;
3969                 }
3970                 agg_id = ad_info.aggregator_id;
3971         }
3972         bond_for_each_slave(bond, slave, iter) {
3973                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3974                         struct aggregator *agg;
3975
3976                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3977                         if (!agg || agg->aggregator_identifier != agg_id)
3978                                 continue;
3979                 }
3980                 if (!bond_slave_can_tx(slave))
3981                         continue;
3982                 if (skipslave == slave)
3983                         continue;
3984
3985                 netdev_dbg(bond->dev,
3986                            "Adding slave dev %s to tx hash array[%d]\n",
3987                            slave->dev->name, new_arr->count);
3988
3989                 new_arr->arr[new_arr->count++] = slave;
3990         }
3991
3992         old_arr = rtnl_dereference(bond->slave_arr);
3993         rcu_assign_pointer(bond->slave_arr, new_arr);
3994         if (old_arr)
3995                 kfree_rcu(old_arr, rcu);
3996 out:
3997         if (ret != 0 && skipslave) {
3998                 int idx;
3999
4000                 /* Rare situation where caller has asked to skip a specific
4001                  * slave but allocation failed (most likely!). BTW this is
4002                  * only possible when the call is initiated from
4003                  * __bond_release_one(). In this situation; overwrite the
4004                  * skipslave entry in the array with the last entry from the
4005                  * array to avoid a situation where the xmit path may choose
4006                  * this to-be-skipped slave to send a packet out.
4007                  */
4008                 old_arr = rtnl_dereference(bond->slave_arr);
4009                 for (idx = 0; idx < old_arr->count; idx++) {
4010                         if (skipslave == old_arr->arr[idx]) {
4011                                 old_arr->arr[idx] =
4012                                     old_arr->arr[old_arr->count-1];
4013                                 old_arr->count--;
4014                                 break;
4015                         }
4016                 }
4017         }
4018         return ret;
4019 }
4020
4021 /* Use this Xmit function for 3AD as well as XOR modes. The current
4022  * usable slave array is formed in the control path. The xmit function
4023  * just calculates hash and sends the packet out.
4024  */
4025 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4026                                      struct net_device *dev)
4027 {
4028         struct bonding *bond = netdev_priv(dev);
4029         struct slave *slave;
4030         struct bond_up_slave *slaves;
4031         unsigned int count;
4032
4033         slaves = rcu_dereference(bond->slave_arr);
4034         count = slaves ? READ_ONCE(slaves->count) : 0;
4035         if (likely(count)) {
4036                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4037                 bond_dev_queue_xmit(bond, skb, slave->dev);
4038         } else {
4039                 bond_tx_drop(dev, skb);
4040         }
4041
4042         return NETDEV_TX_OK;
4043 }
4044
4045 /* in broadcast mode, we send everything to all usable interfaces. */
4046 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4047                                        struct net_device *bond_dev)
4048 {
4049         struct bonding *bond = netdev_priv(bond_dev);
4050         struct slave *slave = NULL;
4051         struct list_head *iter;
4052
4053         bond_for_each_slave_rcu(bond, slave, iter) {
4054                 if (bond_is_last_slave(bond, slave))
4055                         break;
4056                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4057                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4058
4059                         if (!skb2) {
4060                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4061                                                     bond_dev->name, __func__);
4062                                 continue;
4063                         }
4064                         bond_dev_queue_xmit(bond, skb2, slave->dev);
4065                 }
4066         }
4067         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4068                 bond_dev_queue_xmit(bond, skb, slave->dev);
4069         else
4070                 bond_tx_drop(bond_dev, skb);
4071
4072         return NETDEV_TX_OK;
4073 }
4074
4075 /*------------------------- Device initialization ---------------------------*/
4076
4077 /* Lookup the slave that corresponds to a qid */
4078 static inline int bond_slave_override(struct bonding *bond,
4079                                       struct sk_buff *skb)
4080 {
4081         struct slave *slave = NULL;
4082         struct list_head *iter;
4083
4084         if (!skb_rx_queue_recorded(skb))
4085                 return 1;
4086
4087         /* Find out if any slaves have the same mapping as this skb. */
4088         bond_for_each_slave_rcu(bond, slave, iter) {
4089                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4090                         if (bond_slave_is_up(slave) &&
4091                             slave->link == BOND_LINK_UP) {
4092                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4093                                 return 0;
4094                         }
4095                         /* If the slave isn't UP, use default transmit policy. */
4096                         break;
4097                 }
4098         }
4099
4100         return 1;
4101 }
4102
4103
4104 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4105                              void *accel_priv, select_queue_fallback_t fallback)
4106 {
4107         /* This helper function exists to help dev_pick_tx get the correct
4108          * destination queue.  Using a helper function skips a call to
4109          * skb_tx_hash and will put the skbs in the queue we expect on their
4110          * way down to the bonding driver.
4111          */
4112         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4113
4114         /* Save the original txq to restore before passing to the driver */
4115         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4116
4117         if (unlikely(txq >= dev->real_num_tx_queues)) {
4118                 do {
4119                         txq -= dev->real_num_tx_queues;
4120                 } while (txq >= dev->real_num_tx_queues);
4121         }
4122         return txq;
4123 }
4124
4125 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4126 {
4127         struct bonding *bond = netdev_priv(dev);
4128
4129         if (bond_should_override_tx_queue(bond) &&
4130             !bond_slave_override(bond, skb))
4131                 return NETDEV_TX_OK;
4132
4133         switch (BOND_MODE(bond)) {
4134         case BOND_MODE_ROUNDROBIN:
4135                 return bond_xmit_roundrobin(skb, dev);
4136         case BOND_MODE_ACTIVEBACKUP:
4137                 return bond_xmit_activebackup(skb, dev);
4138         case BOND_MODE_8023AD:
4139         case BOND_MODE_XOR:
4140                 return bond_3ad_xor_xmit(skb, dev);
4141         case BOND_MODE_BROADCAST:
4142                 return bond_xmit_broadcast(skb, dev);
4143         case BOND_MODE_ALB:
4144                 return bond_alb_xmit(skb, dev);
4145         case BOND_MODE_TLB:
4146                 return bond_tlb_xmit(skb, dev);
4147         default:
4148                 /* Should never happen, mode already checked */
4149                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4150                 WARN_ON_ONCE(1);
4151                 bond_tx_drop(dev, skb);
4152                 return NETDEV_TX_OK;
4153         }
4154 }
4155
4156 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4157 {
4158         struct bonding *bond = netdev_priv(dev);
4159         netdev_tx_t ret = NETDEV_TX_OK;
4160
4161         /* If we risk deadlock from transmitting this in the
4162          * netpoll path, tell netpoll to queue the frame for later tx
4163          */
4164         if (unlikely(is_netpoll_tx_blocked(dev)))
4165                 return NETDEV_TX_BUSY;
4166
4167         rcu_read_lock();
4168         if (bond_has_slaves(bond))
4169                 ret = __bond_start_xmit(skb, dev);
4170         else
4171                 bond_tx_drop(dev, skb);
4172         rcu_read_unlock();
4173
4174         return ret;
4175 }
4176
4177 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4178                                            struct ethtool_link_ksettings *cmd)
4179 {
4180         struct bonding *bond = netdev_priv(bond_dev);
4181         unsigned long speed = 0;
4182         struct list_head *iter;
4183         struct slave *slave;
4184
4185         cmd->base.duplex = DUPLEX_UNKNOWN;
4186         cmd->base.port = PORT_OTHER;
4187
4188         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4189          * do not need to check mode.  Though link speed might not represent
4190          * the true receive or transmit bandwidth (not all modes are symmetric)
4191          * this is an accurate maximum.
4192          */
4193         bond_for_each_slave(bond, slave, iter) {
4194                 if (bond_slave_can_tx(slave)) {
4195                         if (slave->speed != SPEED_UNKNOWN)
4196                                 speed += slave->speed;
4197                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4198                             slave->duplex != DUPLEX_UNKNOWN)
4199                                 cmd->base.duplex = slave->duplex;
4200                 }
4201         }
4202         cmd->base.speed = speed ? : SPEED_UNKNOWN;
4203
4204         return 0;
4205 }
4206
4207 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4208                                      struct ethtool_drvinfo *drvinfo)
4209 {
4210         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4211         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4212         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4213                  BOND_ABI_VERSION);
4214 }
4215
4216 static const struct ethtool_ops bond_ethtool_ops = {
4217         .get_drvinfo            = bond_ethtool_get_drvinfo,
4218         .get_link               = ethtool_op_get_link,
4219         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
4220 };
4221
4222 static const struct net_device_ops bond_netdev_ops = {
4223         .ndo_init               = bond_init,
4224         .ndo_uninit             = bond_uninit,
4225         .ndo_open               = bond_open,
4226         .ndo_stop               = bond_close,
4227         .ndo_start_xmit         = bond_start_xmit,
4228         .ndo_select_queue       = bond_select_queue,
4229         .ndo_get_stats64        = bond_get_stats,
4230         .ndo_do_ioctl           = bond_do_ioctl,
4231         .ndo_change_rx_flags    = bond_change_rx_flags,
4232         .ndo_set_rx_mode        = bond_set_rx_mode,
4233         .ndo_change_mtu         = bond_change_mtu,
4234         .ndo_set_mac_address    = bond_set_mac_address,
4235         .ndo_neigh_setup        = bond_neigh_setup,
4236         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4237         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4238         .ndo_get_lock_subclass  = bond_get_nest_level,
4239 #ifdef CONFIG_NET_POLL_CONTROLLER
4240         .ndo_netpoll_setup      = bond_netpoll_setup,
4241         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4242         .ndo_poll_controller    = bond_poll_controller,
4243 #endif
4244         .ndo_add_slave          = bond_enslave,
4245         .ndo_del_slave          = bond_release,
4246         .ndo_fix_features       = bond_fix_features,
4247         .ndo_features_check     = passthru_features_check,
4248 };
4249
4250 static const struct device_type bond_type = {
4251         .name = "bond",
4252 };
4253
4254 static void bond_destructor(struct net_device *bond_dev)
4255 {
4256         struct bonding *bond = netdev_priv(bond_dev);
4257         if (bond->wq)
4258                 destroy_workqueue(bond->wq);
4259 }
4260
4261 void bond_setup(struct net_device *bond_dev)
4262 {
4263         struct bonding *bond = netdev_priv(bond_dev);
4264
4265         spin_lock_init(&bond->mode_lock);
4266         spin_lock_init(&bond->stats_lock);
4267         bond->params = bonding_defaults;
4268
4269         /* Initialize pointers */
4270         bond->dev = bond_dev;
4271
4272         /* Initialize the device entry points */
4273         ether_setup(bond_dev);
4274         bond_dev->max_mtu = ETH_MAX_MTU;
4275         bond_dev->netdev_ops = &bond_netdev_ops;
4276         bond_dev->ethtool_ops = &bond_ethtool_ops;
4277
4278         bond_dev->needs_free_netdev = true;
4279         bond_dev->priv_destructor = bond_destructor;
4280
4281         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4282
4283         /* Initialize the device options */
4284         bond_dev->flags |= IFF_MASTER;
4285         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4286         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4287
4288         /* don't acquire bond device's netif_tx_lock when transmitting */
4289         bond_dev->features |= NETIF_F_LLTX;
4290
4291         /* By default, we declare the bond to be fully
4292          * VLAN hardware accelerated capable. Special
4293          * care is taken in the various xmit functions
4294          * when there are slaves that are not hw accel
4295          * capable
4296          */
4297
4298         /* Don't allow bond devices to change network namespaces. */
4299         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4300
4301         bond_dev->hw_features = BOND_VLAN_FEATURES |
4302                                 NETIF_F_HW_VLAN_CTAG_TX |
4303                                 NETIF_F_HW_VLAN_CTAG_RX |
4304                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4305
4306         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4307         bond_dev->features |= bond_dev->hw_features;
4308 }
4309
4310 /* Destroy a bonding device.
4311  * Must be under rtnl_lock when this function is called.
4312  */
4313 static void bond_uninit(struct net_device *bond_dev)
4314 {
4315         struct bonding *bond = netdev_priv(bond_dev);
4316         struct list_head *iter;
4317         struct slave *slave;
4318         struct bond_up_slave *arr;
4319
4320         bond_netpoll_cleanup(bond_dev);
4321
4322         /* Release the bonded slaves */
4323         bond_for_each_slave(bond, slave, iter)
4324                 __bond_release_one(bond_dev, slave->dev, true, true);
4325         netdev_info(bond_dev, "Released all slaves\n");
4326
4327         arr = rtnl_dereference(bond->slave_arr);
4328         if (arr) {
4329                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4330                 kfree_rcu(arr, rcu);
4331         }
4332
4333         list_del(&bond->bond_list);
4334
4335         bond_debug_unregister(bond);
4336 }
4337
4338 /*------------------------- Module initialization ---------------------------*/
4339
4340 static int bond_check_params(struct bond_params *params)
4341 {
4342         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4343         struct bond_opt_value newval;
4344         const struct bond_opt_value *valptr;
4345         int arp_all_targets_value = 0;
4346         u16 ad_actor_sys_prio = 0;
4347         u16 ad_user_port_key = 0;
4348         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4349         int arp_ip_count;
4350         int bond_mode   = BOND_MODE_ROUNDROBIN;
4351         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4352         int lacp_fast = 0;
4353         int tlb_dynamic_lb;
4354
4355         /* Convert string parameters. */
4356         if (mode) {
4357                 bond_opt_initstr(&newval, mode);
4358                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4359                 if (!valptr) {
4360                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4361                         return -EINVAL;
4362                 }
4363                 bond_mode = valptr->value;
4364         }
4365
4366         if (xmit_hash_policy) {
4367                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4368                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
4369                     bond_mode == BOND_MODE_BROADCAST) {
4370                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4371                                 bond_mode_name(bond_mode));
4372                 } else {
4373                         bond_opt_initstr(&newval, xmit_hash_policy);
4374                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4375                                                 &newval);
4376                         if (!valptr) {
4377                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4378                                        xmit_hash_policy);
4379                                 return -EINVAL;
4380                         }
4381                         xmit_hashtype = valptr->value;
4382                 }
4383         }
4384
4385         if (lacp_rate) {
4386                 if (bond_mode != BOND_MODE_8023AD) {
4387                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4388                                 bond_mode_name(bond_mode));
4389                 } else {
4390                         bond_opt_initstr(&newval, lacp_rate);
4391                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4392                                                 &newval);
4393                         if (!valptr) {
4394                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4395                                        lacp_rate);
4396                                 return -EINVAL;
4397                         }
4398                         lacp_fast = valptr->value;
4399                 }
4400         }
4401
4402         if (ad_select) {
4403                 bond_opt_initstr(&newval, ad_select);
4404                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4405                                         &newval);
4406                 if (!valptr) {
4407                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4408                         return -EINVAL;
4409                 }
4410                 params->ad_select = valptr->value;
4411                 if (bond_mode != BOND_MODE_8023AD)
4412                         pr_warn("ad_select param only affects 802.3ad mode\n");
4413         } else {
4414                 params->ad_select = BOND_AD_STABLE;
4415         }
4416
4417         if (max_bonds < 0) {
4418                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4419                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4420                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4421         }
4422
4423         if (miimon < 0) {
4424                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4425                         miimon, INT_MAX);
4426                 miimon = 0;
4427         }
4428
4429         if (updelay < 0) {
4430                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4431                         updelay, INT_MAX);
4432                 updelay = 0;
4433         }
4434
4435         if (downdelay < 0) {
4436                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4437                         downdelay, INT_MAX);
4438                 downdelay = 0;
4439         }
4440
4441         if ((use_carrier != 0) && (use_carrier != 1)) {
4442                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4443                         use_carrier);
4444                 use_carrier = 1;
4445         }
4446
4447         if (num_peer_notif < 0 || num_peer_notif > 255) {
4448                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4449                         num_peer_notif);
4450                 num_peer_notif = 1;
4451         }
4452
4453         /* reset values for 802.3ad/TLB/ALB */
4454         if (!bond_mode_uses_arp(bond_mode)) {
4455                 if (!miimon) {
4456                         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");
4457                         pr_warn("Forcing miimon to 100msec\n");
4458                         miimon = BOND_DEFAULT_MIIMON;
4459                 }
4460         }
4461
4462         if (tx_queues < 1 || tx_queues > 255) {
4463                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4464                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4465                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4466         }
4467
4468         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4469                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4470                         all_slaves_active);
4471                 all_slaves_active = 0;
4472         }
4473
4474         if (resend_igmp < 0 || resend_igmp > 255) {
4475                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4476                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4477                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4478         }
4479
4480         bond_opt_initval(&newval, packets_per_slave);
4481         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4482                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4483                         packets_per_slave, USHRT_MAX);
4484                 packets_per_slave = 1;
4485         }
4486
4487         if (bond_mode == BOND_MODE_ALB) {
4488                 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",
4489                           updelay);
4490         }
4491
4492         if (!miimon) {
4493                 if (updelay || downdelay) {
4494                         /* just warn the user the up/down delay will have
4495                          * no effect since miimon is zero...
4496                          */
4497                         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",
4498                                 updelay, downdelay);
4499                 }
4500         } else {
4501                 /* don't allow arp monitoring */
4502                 if (arp_interval) {
4503                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4504                                 miimon, arp_interval);
4505                         arp_interval = 0;
4506                 }
4507
4508                 if ((updelay % miimon) != 0) {
4509                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4510                                 updelay, miimon, (updelay / miimon) * miimon);
4511                 }
4512
4513                 updelay /= miimon;
4514
4515                 if ((downdelay % miimon) != 0) {
4516                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4517                                 downdelay, miimon,
4518                                 (downdelay / miimon) * miimon);
4519                 }
4520
4521                 downdelay /= miimon;
4522         }
4523
4524         if (arp_interval < 0) {
4525                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4526                         arp_interval, INT_MAX);
4527                 arp_interval = 0;
4528         }
4529
4530         for (arp_ip_count = 0, i = 0;
4531              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4532                 __be32 ip;
4533
4534                 /* not a complete check, but good enough to catch mistakes */
4535                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4536                     !bond_is_ip_target_ok(ip)) {
4537                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4538                                 arp_ip_target[i]);
4539                         arp_interval = 0;
4540                 } else {
4541                         if (bond_get_targets_ip(arp_target, ip) == -1)
4542                                 arp_target[arp_ip_count++] = ip;
4543                         else
4544                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4545                                         &ip);
4546                 }
4547         }
4548
4549         if (arp_interval && !arp_ip_count) {
4550                 /* don't allow arping if no arp_ip_target given... */
4551                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4552                         arp_interval);
4553                 arp_interval = 0;
4554         }
4555
4556         if (arp_validate) {
4557                 if (!arp_interval) {
4558                         pr_err("arp_validate requires arp_interval\n");
4559                         return -EINVAL;
4560                 }
4561
4562                 bond_opt_initstr(&newval, arp_validate);
4563                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4564                                         &newval);
4565                 if (!valptr) {
4566                         pr_err("Error: invalid arp_validate \"%s\"\n",
4567                                arp_validate);
4568                         return -EINVAL;
4569                 }
4570                 arp_validate_value = valptr->value;
4571         } else {
4572                 arp_validate_value = 0;
4573         }
4574
4575         if (arp_all_targets) {
4576                 bond_opt_initstr(&newval, arp_all_targets);
4577                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4578                                         &newval);
4579                 if (!valptr) {
4580                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4581                                arp_all_targets);
4582                         arp_all_targets_value = 0;
4583                 } else {
4584                         arp_all_targets_value = valptr->value;
4585                 }
4586         }
4587
4588         if (miimon) {
4589                 pr_info("MII link monitoring set to %d ms\n", miimon);
4590         } else if (arp_interval) {
4591                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4592                                           arp_validate_value);
4593                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4594                         arp_interval, valptr->string, arp_ip_count);
4595
4596                 for (i = 0; i < arp_ip_count; i++)
4597                         pr_cont(" %s", arp_ip_target[i]);
4598
4599                 pr_cont("\n");
4600
4601         } else if (max_bonds) {
4602                 /* miimon and arp_interval not set, we need one so things
4603                  * work as expected, see bonding.txt for details
4604                  */
4605                 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");
4606         }
4607
4608         if (primary && !bond_mode_uses_primary(bond_mode)) {
4609                 /* currently, using a primary only makes sense
4610                  * in active backup, TLB or ALB modes
4611                  */
4612                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4613                         primary, bond_mode_name(bond_mode));
4614                 primary = NULL;
4615         }
4616
4617         if (primary && primary_reselect) {
4618                 bond_opt_initstr(&newval, primary_reselect);
4619                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4620                                         &newval);
4621                 if (!valptr) {
4622                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4623                                primary_reselect);
4624                         return -EINVAL;
4625                 }
4626                 primary_reselect_value = valptr->value;
4627         } else {
4628                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4629         }
4630
4631         if (fail_over_mac) {
4632                 bond_opt_initstr(&newval, fail_over_mac);
4633                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4634                                         &newval);
4635                 if (!valptr) {
4636                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4637                                fail_over_mac);
4638                         return -EINVAL;
4639                 }
4640                 fail_over_mac_value = valptr->value;
4641                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4642                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4643         } else {
4644                 fail_over_mac_value = BOND_FOM_NONE;
4645         }
4646
4647         bond_opt_initstr(&newval, "default");
4648         valptr = bond_opt_parse(
4649                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4650                                      &newval);
4651         if (!valptr) {
4652                 pr_err("Error: No ad_actor_sys_prio default value");
4653                 return -EINVAL;
4654         }
4655         ad_actor_sys_prio = valptr->value;
4656
4657         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4658                                 &newval);
4659         if (!valptr) {
4660                 pr_err("Error: No ad_user_port_key default value");
4661                 return -EINVAL;
4662         }
4663         ad_user_port_key = valptr->value;
4664
4665         bond_opt_initstr(&newval, "default");
4666         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4667         if (!valptr) {
4668                 pr_err("Error: No tlb_dynamic_lb default value");
4669                 return -EINVAL;
4670         }
4671         tlb_dynamic_lb = valptr->value;
4672
4673         if (lp_interval == 0) {
4674                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4675                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4676                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4677         }
4678
4679         /* fill params struct with the proper values */
4680         params->mode = bond_mode;
4681         params->xmit_policy = xmit_hashtype;
4682         params->miimon = miimon;
4683         params->num_peer_notif = num_peer_notif;
4684         params->arp_interval = arp_interval;
4685         params->arp_validate = arp_validate_value;
4686         params->arp_all_targets = arp_all_targets_value;
4687         params->updelay = updelay;
4688         params->downdelay = downdelay;
4689         params->use_carrier = use_carrier;
4690         params->lacp_fast = lacp_fast;
4691         params->primary[0] = 0;
4692         params->primary_reselect = primary_reselect_value;
4693         params->fail_over_mac = fail_over_mac_value;
4694         params->tx_queues = tx_queues;
4695         params->all_slaves_active = all_slaves_active;
4696         params->resend_igmp = resend_igmp;
4697         params->min_links = min_links;
4698         params->lp_interval = lp_interval;
4699         params->packets_per_slave = packets_per_slave;
4700         params->tlb_dynamic_lb = tlb_dynamic_lb;
4701         params->ad_actor_sys_prio = ad_actor_sys_prio;
4702         eth_zero_addr(params->ad_actor_system);
4703         params->ad_user_port_key = ad_user_port_key;
4704         if (packets_per_slave > 0) {
4705                 params->reciprocal_packets_per_slave =
4706                         reciprocal_value(packets_per_slave);
4707         } else {
4708                 /* reciprocal_packets_per_slave is unused if
4709                  * packets_per_slave is 0 or 1, just initialize it
4710                  */
4711                 params->reciprocal_packets_per_slave =
4712                         (struct reciprocal_value) { 0 };
4713         }
4714
4715         if (primary) {
4716                 strncpy(params->primary, primary, IFNAMSIZ);
4717                 params->primary[IFNAMSIZ - 1] = 0;
4718         }
4719
4720         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4721
4722         return 0;
4723 }
4724
4725 /* Called from registration process */
4726 static int bond_init(struct net_device *bond_dev)
4727 {
4728         struct bonding *bond = netdev_priv(bond_dev);
4729         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4730
4731         netdev_dbg(bond_dev, "Begin bond_init\n");
4732
4733         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4734         if (!bond->wq)
4735                 return -ENOMEM;
4736
4737         bond->nest_level = SINGLE_DEPTH_NESTING;
4738         netdev_lockdep_set_classes(bond_dev);
4739
4740         list_add_tail(&bond->bond_list, &bn->dev_list);
4741
4742         bond_prepare_sysfs_group(bond);
4743
4744         bond_debug_register(bond);
4745
4746         /* Ensure valid dev_addr */
4747         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4748             bond_dev->addr_assign_type == NET_ADDR_PERM)
4749                 eth_hw_addr_random(bond_dev);
4750
4751         return 0;
4752 }
4753
4754 unsigned int bond_get_num_tx_queues(void)
4755 {
4756         return tx_queues;
4757 }
4758
4759 /* Create a new bond based on the specified name and bonding parameters.
4760  * If name is NULL, obtain a suitable "bond%d" name for us.
4761  * Caller must NOT hold rtnl_lock; we need to release it here before we
4762  * set up our sysfs entries.
4763  */
4764 int bond_create(struct net *net, const char *name)
4765 {
4766         struct net_device *bond_dev;
4767         struct bonding *bond;
4768         struct alb_bond_info *bond_info;
4769         int res;
4770
4771         rtnl_lock();
4772
4773         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4774                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4775                                    bond_setup, tx_queues);
4776         if (!bond_dev) {
4777                 pr_err("%s: eek! can't alloc netdev!\n", name);
4778                 rtnl_unlock();
4779                 return -ENOMEM;
4780         }
4781
4782         /*
4783          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4784          * It is set to 0 by default which is wrong.
4785          */
4786         bond = netdev_priv(bond_dev);
4787         bond_info = &(BOND_ALB_INFO(bond));
4788         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4789
4790         dev_net_set(bond_dev, net);
4791         bond_dev->rtnl_link_ops = &bond_link_ops;
4792
4793         res = register_netdevice(bond_dev);
4794
4795         netif_carrier_off(bond_dev);
4796
4797         bond_work_init_all(bond);
4798
4799         rtnl_unlock();
4800         if (res < 0)
4801                 free_netdev(bond_dev);
4802         return res;
4803 }
4804
4805 static int __net_init bond_net_init(struct net *net)
4806 {
4807         struct bond_net *bn = net_generic(net, bond_net_id);
4808
4809         bn->net = net;
4810         INIT_LIST_HEAD(&bn->dev_list);
4811
4812         bond_create_proc_dir(bn);
4813         bond_create_sysfs(bn);
4814
4815         return 0;
4816 }
4817
4818 static void __net_exit bond_net_exit(struct net *net)
4819 {
4820         struct bond_net *bn = net_generic(net, bond_net_id);
4821         struct bonding *bond, *tmp_bond;
4822         LIST_HEAD(list);
4823
4824         bond_destroy_sysfs(bn);
4825
4826         /* Kill off any bonds created after unregistering bond rtnl ops */
4827         rtnl_lock();
4828         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4829                 unregister_netdevice_queue(bond->dev, &list);
4830         unregister_netdevice_many(&list);
4831         rtnl_unlock();
4832
4833         bond_destroy_proc_dir(bn);
4834 }
4835
4836 static struct pernet_operations bond_net_ops = {
4837         .init = bond_net_init,
4838         .exit = bond_net_exit,
4839         .id   = &bond_net_id,
4840         .size = sizeof(struct bond_net),
4841 };
4842
4843 static int __init bonding_init(void)
4844 {
4845         int i;
4846         int res;
4847
4848         pr_info("%s", bond_version);
4849
4850         res = bond_check_params(&bonding_defaults);
4851         if (res)
4852                 goto out;
4853
4854         res = register_pernet_subsys(&bond_net_ops);
4855         if (res)
4856                 goto out;
4857
4858         res = bond_netlink_init();
4859         if (res)
4860                 goto err_link;
4861
4862         bond_create_debugfs();
4863
4864         for (i = 0; i < max_bonds; i++) {
4865                 res = bond_create(&init_net, NULL);
4866                 if (res)
4867                         goto err;
4868         }
4869
4870         register_netdevice_notifier(&bond_netdev_notifier);
4871 out:
4872         return res;
4873 err:
4874         bond_destroy_debugfs();
4875         bond_netlink_fini();
4876 err_link:
4877         unregister_pernet_subsys(&bond_net_ops);
4878         goto out;
4879
4880 }
4881
4882 static void __exit bonding_exit(void)
4883 {
4884         unregister_netdevice_notifier(&bond_netdev_notifier);
4885
4886         bond_destroy_debugfs();
4887
4888         bond_netlink_fini();
4889         unregister_pernet_subsys(&bond_net_ops);
4890
4891 #ifdef CONFIG_NET_POLL_CONTROLLER
4892         /* Make sure we don't have an imbalance on our netpoll blocking */
4893         WARN_ON(atomic_read(&netpoll_block_tx));
4894 #endif
4895 }
4896
4897 module_init(bonding_init);
4898 module_exit(bonding_exit);
4899 MODULE_LICENSE("GPL");
4900 MODULE_VERSION(DRV_VERSION);
4901 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4902 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");