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-xor and 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 int bond_net_id __read_mostly;
 203
 204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 205 static int arp_ip_count;
 206 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 208 static int lacp_fast;
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static int bond_init(struct net_device *bond_dev);
 213 static void bond_uninit(struct net_device *bond_dev);
 214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 215                                                 struct rtnl_link_stats64 *stats);
 216 static void bond_slave_arr_handler(struct work_struct *work);
 217 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 218                                   int mod);
 219
 220 /*---------------------------- General routines -----------------------------*/
 221
 222 const char *bond_mode_name(int mode)
 223 {
 224         static const char *names[] = {
 225                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 226                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 227                 [BOND_MODE_XOR] = "load balancing (xor)",
 228                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 229                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 230                 [BOND_MODE_TLB] = "transmit load balancing",
 231                 [BOND_MODE_ALB] = "adaptive load balancing",
 232         };
 233
 234         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 235                 return "unknown";
 236
 237         return names[mode];
 238 }
 239
 240 /*---------------------------------- VLAN -----------------------------------*/
 241
 242 /**
 243  * bond_dev_queue_xmit - Prepare skb for xmit.
 244  *
 245  * @bond: bond device that got this skb for tx.
 246  * @skb: hw accel VLAN tagged skb to transmit
 247  * @slave_dev: slave that is supposed to xmit this skbuff
 248  */
 249 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 250                         struct net_device *slave_dev)
 251 {
 252         skb->dev = slave_dev;
 253
 254         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 255                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 256         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 257
 258         if (unlikely(netpoll_tx_running(bond->dev)))
 259                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 260         else
 261                 dev_queue_xmit(skb);
 262 }
 263
 264 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 265  * We don't protect the slave list iteration with a lock because:
 266  * a. This operation is performed in IOCTL context,
 267  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 268  * c. Holding a lock with BH disabled while directly calling a base driver
 269  *    entry point is generally a BAD idea.
 270  *
 271  * The design of synchronization/protection for this operation in the 8021q
 272  * module is good for one or more VLAN devices over a single physical device
 273  * and cannot be extended for a teaming solution like bonding, so there is a
 274  * potential race condition here where a net device from the vlan group might
 275  * be referenced (either by a base driver or the 8021q code) while it is being
 276  * removed from the system. However, it turns out we're not making matters
 277  * worse, and if it works for regular VLAN usage it will work here too.
 278 */
 279
 280 /**
 281  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 282  * @bond_dev: bonding net device that got called
 283  * @vid: vlan id being added
 284  */
 285 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 286                                 __be16 proto, u16 vid)
 287 {
 288         struct bonding *bond = netdev_priv(bond_dev);
 289         struct slave *slave, *rollback_slave;
 290         struct list_head *iter;
 291         int res;
 292
 293         bond_for_each_slave(bond, slave, iter) {
 294                 res = vlan_vid_add(slave->dev, proto, vid);
 295                 if (res)
 296                         goto unwind;
 297         }
 298
 299         return 0;
 300
 301 unwind:
 302         /* unwind to the slave that failed */
 303         bond_for_each_slave(bond, rollback_slave, iter) {
 304                 if (rollback_slave == slave)
 305                         break;
 306
 307                 vlan_vid_del(rollback_slave->dev, proto, vid);
 308         }
 309
 310         return res;
 311 }
 312
 313 /**
 314  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 315  * @bond_dev: bonding net device that got called
 316  * @vid: vlan id being removed
 317  */
 318 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 319                                  __be16 proto, u16 vid)
 320 {
 321         struct bonding *bond = netdev_priv(bond_dev);
 322         struct list_head *iter;
 323         struct slave *slave;
 324
 325         bond_for_each_slave(bond, slave, iter)
 326                 vlan_vid_del(slave->dev, proto, vid);
 327
 328         if (bond_is_lb(bond))
 329                 bond_alb_clear_vlan(bond, vid);
 330
 331         return 0;
 332 }
 333
 334 /*------------------------------- Link status -------------------------------*/
 335
 336 /* Set the carrier state for the master according to the state of its
 337  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 338  * do special 802.3ad magic.
 339  *
 340  * Returns zero if carrier state does not change, nonzero if it does.
 341  */
 342 int bond_set_carrier(struct bonding *bond)
 343 {
 344         struct list_head *iter;
 345         struct slave *slave;
 346
 347         if (!bond_has_slaves(bond))
 348                 goto down;
 349
 350         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 351                 return bond_3ad_set_carrier(bond);
 352
 353         bond_for_each_slave(bond, slave, iter) {
 354                 if (slave->link == BOND_LINK_UP) {
 355                         if (!netif_carrier_ok(bond->dev)) {
 356                                 netif_carrier_on(bond->dev);
 357                                 return 1;
 358                         }
 359                         return 0;
 360                 }
 361         }
 362
 363 down:
 364         if (netif_carrier_ok(bond->dev)) {
 365                 netif_carrier_off(bond->dev);
 366                 return 1;
 367         }
 368         return 0;
 369 }
 370
 371 /* Get link speed and duplex from the slave's base driver
 372  * using ethtool. If for some reason the call fails or the
 373  * values are invalid, set speed and duplex to -1,
 374  * and return. Return 1 if speed or duplex settings are
 375  * UNKNOWN; 0 otherwise.
 376  */
 377 static int bond_update_speed_duplex(struct slave *slave)
 378 {
 379         struct net_device *slave_dev = slave->dev;
 380         struct ethtool_link_ksettings ecmd;
 381         int res;
 382
 383         slave->speed = SPEED_UNKNOWN;
 384         slave->duplex = DUPLEX_UNKNOWN;
 385
 386         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 387         if (res < 0) {
 388                 slave->link = BOND_LINK_DOWN;
 389                 return 1;
 390         }
 391         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1)) {
 392                 slave->link = BOND_LINK_DOWN;
 393                 return 1;
 394         }
 395         switch (ecmd.base.duplex) {
 396         case DUPLEX_FULL:
 397         case DUPLEX_HALF:
 398                 break;
 399         default:
 400                 slave->link = BOND_LINK_DOWN;
 401                 return 1;
 402         }
 403
 404         slave->speed = ecmd.base.speed;
 405         slave->duplex = ecmd.base.duplex;
 406
 407         return 0;
 408 }
 409
 410 const char *bond_slave_link_status(s8 link)
 411 {
 412         switch (link) {
 413         case BOND_LINK_UP:
 414                 return "up";
 415         case BOND_LINK_FAIL:
 416                 return "going down";
 417         case BOND_LINK_DOWN:
 418                 return "down";
 419         case BOND_LINK_BACK:
 420                 return "going back";
 421         default:
 422                 return "unknown";
 423         }
 424 }
 425
 426 /* if <dev> supports MII link status reporting, check its link status.
 427  *
 428  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 429  * depending upon the setting of the use_carrier parameter.
 430  *
 431  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 432  * can't tell and just pretend it is), or 0, meaning that the link is
 433  * down.
 434  *
 435  * If reporting is non-zero, instead of faking link up, return -1 if
 436  * both ETHTOOL and MII ioctls fail (meaning the device does not
 437  * support them).  If use_carrier is set, return whatever it says.
 438  * It'd be nice if there was a good way to tell if a driver supports
 439  * netif_carrier, but there really isn't.
 440  */
 441 static int bond_check_dev_link(struct bonding *bond,
 442                                struct net_device *slave_dev, int reporting)
 443 {
 444         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 445         int (*ioctl)(struct net_device *, struct ifreq *, int);
 446         struct ifreq ifr;
 447         struct mii_ioctl_data *mii;
 448
 449         if (!reporting && !netif_running(slave_dev))
 450                 return 0;
 451
 452         if (bond->params.use_carrier)
 453                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 454
 455         /* Try to get link status using Ethtool first. */
 456         if (slave_dev->ethtool_ops->get_link)
 457                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 458                         BMSR_LSTATUS : 0;
 459
 460         /* Ethtool can't be used, fallback to MII ioctls. */
 461         ioctl = slave_ops->ndo_do_ioctl;
 462         if (ioctl) {
 463                 /* TODO: set pointer to correct ioctl on a per team member
 464                  *       bases to make this more efficient. that is, once
 465                  *       we determine the correct ioctl, we will always
 466                  *       call it and not the others for that team
 467                  *       member.
 468                  */
 469
 470                 /* We cannot assume that SIOCGMIIPHY will also read a
 471                  * register; not all network drivers (e.g., e100)
 472                  * support that.
 473                  */
 474
 475                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 476                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 477                 mii = if_mii(&ifr);
 478                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 479                         mii->reg_num = MII_BMSR;
 480                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 481                                 return mii->val_out & BMSR_LSTATUS;
 482                 }
 483         }
 484
 485         /* If reporting, report that either there's no dev->do_ioctl,
 486          * or both SIOCGMIIREG and get_link failed (meaning that we
 487          * cannot report link status).  If not reporting, pretend
 488          * we're ok.
 489          */
 490         return reporting ? -1 : BMSR_LSTATUS;
 491 }
 492
 493 /*----------------------------- Multicast list ------------------------------*/
 494
 495 /* Push the promiscuity flag down to appropriate slaves */
 496 static int bond_set_promiscuity(struct bonding *bond, int inc)
 497 {
 498         struct list_head *iter;
 499         int err = 0;
 500
 501         if (bond_uses_primary(bond)) {
 502                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 503
 504                 if (curr_active)
 505                         err = dev_set_promiscuity(curr_active->dev, inc);
 506         } else {
 507                 struct slave *slave;
 508
 509                 bond_for_each_slave(bond, slave, iter) {
 510                         err = dev_set_promiscuity(slave->dev, inc);
 511                         if (err)
 512                                 return err;
 513                 }
 514         }
 515         return err;
 516 }
 517
 518 /* Push the allmulti flag down to all slaves */
 519 static int bond_set_allmulti(struct bonding *bond, int inc)
 520 {
 521         struct list_head *iter;
 522         int err = 0;
 523
 524         if (bond_uses_primary(bond)) {
 525                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 526
 527                 if (curr_active)
 528                         err = dev_set_allmulti(curr_active->dev, inc);
 529         } else {
 530                 struct slave *slave;
 531
 532                 bond_for_each_slave(bond, slave, iter) {
 533                         err = dev_set_allmulti(slave->dev, inc);
 534                         if (err)
 535                                 return err;
 536                 }
 537         }
 538         return err;
 539 }
 540
 541 /* Retrieve the list of registered multicast addresses for the bonding
 542  * device and retransmit an IGMP JOIN request to the current active
 543  * slave.
 544  */
 545 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 546 {
 547         struct bonding *bond = container_of(work, struct bonding,
 548                                             mcast_work.work);
 549
 550         if (!rtnl_trylock()) {
 551                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 552                 return;
 553         }
 554         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 555
 556         if (bond->igmp_retrans > 1) {
 557                 bond->igmp_retrans--;
 558                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 559         }
 560         rtnl_unlock();
 561 }
 562
 563 /* Flush bond's hardware addresses from slave */
 564 static void bond_hw_addr_flush(struct net_device *bond_dev,
 565                                struct net_device *slave_dev)
 566 {
 567         struct bonding *bond = netdev_priv(bond_dev);
 568
 569         dev_uc_unsync(slave_dev, bond_dev);
 570         dev_mc_unsync(slave_dev, bond_dev);
 571
 572         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 573                 /* del lacpdu mc addr from mc list */
 574                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 575
 576                 dev_mc_del(slave_dev, lacpdu_multicast);
 577         }
 578 }
 579
 580 /*--------------------------- Active slave change ---------------------------*/
 581
 582 /* Update the hardware address list and promisc/allmulti for the new and
 583  * old active slaves (if any).  Modes that are not using primary keep all
 584  * slaves up date at all times; only the modes that use primary need to call
 585  * this function to swap these settings during a failover.
 586  */
 587 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 588                               struct slave *old_active)
 589 {
 590         if (old_active) {
 591                 if (bond->dev->flags & IFF_PROMISC)
 592                         dev_set_promiscuity(old_active->dev, -1);
 593
 594                 if (bond->dev->flags & IFF_ALLMULTI)
 595                         dev_set_allmulti(old_active->dev, -1);
 596
 597                 bond_hw_addr_flush(bond->dev, old_active->dev);
 598         }
 599
 600         if (new_active) {
 601                 /* FIXME: Signal errors upstream. */
 602                 if (bond->dev->flags & IFF_PROMISC)
 603                         dev_set_promiscuity(new_active->dev, 1);
 604
 605                 if (bond->dev->flags & IFF_ALLMULTI)
 606                         dev_set_allmulti(new_active->dev, 1);
 607
 608                 netif_addr_lock_bh(bond->dev);
 609                 dev_uc_sync(new_active->dev, bond->dev);
 610                 dev_mc_sync(new_active->dev, bond->dev);
 611                 netif_addr_unlock_bh(bond->dev);
 612         }
 613 }
 614
 615 /**
 616  * bond_set_dev_addr - clone slave's address to bond
 617  * @bond_dev: bond net device
 618  * @slave_dev: slave net device
 619  *
 620  * Should be called with RTNL held.
 621  */
 622 static void bond_set_dev_addr(struct net_device *bond_dev,
 623                               struct net_device *slave_dev)
 624 {
 625         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
 626                    bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
 627         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 628         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 629         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 630 }
 631
 632 static struct slave *bond_get_old_active(struct bonding *bond,
 633                                          struct slave *new_active)
 634 {
 635         struct slave *slave;
 636         struct list_head *iter;
 637
 638         bond_for_each_slave(bond, slave, iter) {
 639                 if (slave == new_active)
 640                         continue;
 641
 642                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 643                         return slave;
 644         }
 645
 646         return NULL;
 647 }
 648
 649 /* bond_do_fail_over_mac
 650  *
 651  * Perform special MAC address swapping for fail_over_mac settings
 652  *
 653  * Called with RTNL
 654  */
 655 static void bond_do_fail_over_mac(struct bonding *bond,
 656                                   struct slave *new_active,
 657                                   struct slave *old_active)
 658 {
 659         u8 tmp_mac[ETH_ALEN];
 660         struct sockaddr saddr;
 661         int rv;
 662
 663         switch (bond->params.fail_over_mac) {
 664         case BOND_FOM_ACTIVE:
 665                 if (new_active)
 666                         bond_set_dev_addr(bond->dev, new_active->dev);
 667                 break;
 668         case BOND_FOM_FOLLOW:
 669                 /* if new_active && old_active, swap them
 670                  * if just old_active, do nothing (going to no active slave)
 671                  * if just new_active, set new_active to bond's MAC
 672                  */
 673                 if (!new_active)
 674                         return;
 675
 676                 if (!old_active)
 677                         old_active = bond_get_old_active(bond, new_active);
 678
 679                 if (old_active) {
 680                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 681                         ether_addr_copy(saddr.sa_data,
 682                                         old_active->dev->dev_addr);
 683                         saddr.sa_family = new_active->dev->type;
 684                 } else {
 685                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 686                         saddr.sa_family = bond->dev->type;
 687                 }
 688
 689                 rv = dev_set_mac_address(new_active->dev, &saddr);
 690                 if (rv) {
 691                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 692                                    -rv, new_active->dev->name);
 693                         goto out;
 694                 }
 695
 696                 if (!old_active)
 697                         goto out;
 698
 699                 ether_addr_copy(saddr.sa_data, tmp_mac);
 700                 saddr.sa_family = old_active->dev->type;
 701
 702                 rv = dev_set_mac_address(old_active->dev, &saddr);
 703                 if (rv)
 704                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 705                                    -rv, new_active->dev->name);
 706 out:
 707                 break;
 708         default:
 709                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 710                            bond->params.fail_over_mac);
 711                 break;
 712         }
 713
 714 }
 715
 716 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 717 {
 718         struct slave *prim = rtnl_dereference(bond->primary_slave);
 719         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 720
 721         if (!prim || prim->link != BOND_LINK_UP) {
 722                 if (!curr || curr->link != BOND_LINK_UP)
 723                         return NULL;
 724                 return curr;
 725         }
 726
 727         if (bond->force_primary) {
 728                 bond->force_primary = false;
 729                 return prim;
 730         }
 731
 732         if (!curr || curr->link != BOND_LINK_UP)
 733                 return prim;
 734
 735         /* At this point, prim and curr are both up */
 736         switch (bond->params.primary_reselect) {
 737         case BOND_PRI_RESELECT_ALWAYS:
 738                 return prim;
 739         case BOND_PRI_RESELECT_BETTER:
 740                 if (prim->speed < curr->speed)
 741                         return curr;
 742                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 743                         return curr;
 744                 return prim;
 745         case BOND_PRI_RESELECT_FAILURE:
 746                 return curr;
 747         default:
 748                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 749                            bond->params.primary_reselect);
 750                 return curr;
 751         }
 752 }
 753
 754 /**
 755  * bond_find_best_slave - select the best available slave to be the active one
 756  * @bond: our bonding struct
 757  */
 758 static struct slave *bond_find_best_slave(struct bonding *bond)
 759 {
 760         struct slave *slave, *bestslave = NULL;
 761         struct list_head *iter;
 762         int mintime = bond->params.updelay;
 763
 764         slave = bond_choose_primary_or_current(bond);
 765         if (slave)
 766                 return slave;
 767
 768         bond_for_each_slave(bond, slave, iter) {
 769                 if (slave->link == BOND_LINK_UP)
 770                         return slave;
 771                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 772                     slave->delay < mintime) {
 773                         mintime = slave->delay;
 774                         bestslave = slave;
 775                 }
 776         }
 777
 778         return bestslave;
 779 }
 780
 781 static bool bond_should_notify_peers(struct bonding *bond)
 782 {
 783         struct slave *slave;
 784
 785         rcu_read_lock();
 786         slave = rcu_dereference(bond->curr_active_slave);
 787         rcu_read_unlock();
 788
 789         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 790                    slave ? slave->dev->name : "NULL");
 791
 792         if (!slave || !bond->send_peer_notif ||
 793             !netif_carrier_ok(bond->dev) ||
 794             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 795                 return false;
 796
 797         return true;
 798 }
 799
 800 /**
 801  * change_active_interface - change the active slave into the specified one
 802  * @bond: our bonding struct
 803  * @new: the new slave to make the active one
 804  *
 805  * Set the new slave to the bond's settings and unset them on the old
 806  * curr_active_slave.
 807  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 808  *
 809  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 810  * because it is apparently the best available slave we have, even though its
 811  * updelay hasn't timed out yet.
 812  *
 813  * Caller must hold RTNL.
 814  */
 815 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 816 {
 817         struct slave *old_active;
 818
 819         ASSERT_RTNL();
 820
 821         old_active = rtnl_dereference(bond->curr_active_slave);
 822
 823         if (old_active == new_active)
 824                 return;
 825
 826         if (new_active) {
 827                 new_active->last_link_up = jiffies;
 828
 829                 if (new_active->link == BOND_LINK_BACK) {
 830                         if (bond_uses_primary(bond)) {
 831                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 832                                             new_active->dev->name,
 833                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 834                         }
 835
 836                         new_active->delay = 0;
 837                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 838                                                   BOND_SLAVE_NOTIFY_NOW);
 839
 840                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 841                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 842
 843                         if (bond_is_lb(bond))
 844                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 845                 } else {
 846                         if (bond_uses_primary(bond)) {
 847                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 848                                             new_active->dev->name);
 849                         }
 850                 }
 851         }
 852
 853         if (bond_uses_primary(bond))
 854                 bond_hw_addr_swap(bond, new_active, old_active);
 855
 856         if (bond_is_lb(bond)) {
 857                 bond_alb_handle_active_change(bond, new_active);
 858                 if (old_active)
 859                         bond_set_slave_inactive_flags(old_active,
 860                                                       BOND_SLAVE_NOTIFY_NOW);
 861                 if (new_active)
 862                         bond_set_slave_active_flags(new_active,
 863                                                     BOND_SLAVE_NOTIFY_NOW);
 864         } else {
 865                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 866         }
 867
 868         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 869                 if (old_active)
 870                         bond_set_slave_inactive_flags(old_active,
 871                                                       BOND_SLAVE_NOTIFY_NOW);
 872
 873                 if (new_active) {
 874                         bool should_notify_peers = false;
 875
 876                         bond_set_slave_active_flags(new_active,
 877                                                     BOND_SLAVE_NOTIFY_NOW);
 878
 879                         if (bond->params.fail_over_mac)
 880                                 bond_do_fail_over_mac(bond, new_active,
 881                                                       old_active);
 882
 883                         if (netif_running(bond->dev)) {
 884                                 bond->send_peer_notif =
 885                                         bond->params.num_peer_notif;
 886                                 should_notify_peers =
 887                                         bond_should_notify_peers(bond);
 888                         }
 889
 890                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 891                         if (should_notify_peers)
 892                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 893                                                          bond->dev);
 894                 }
 895         }
 896
 897         /* resend IGMP joins since active slave has changed or
 898          * all were sent on curr_active_slave.
 899          * resend only if bond is brought up with the affected
 900          * bonding modes and the retransmission is enabled
 901          */
 902         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 903             ((bond_uses_primary(bond) && new_active) ||
 904              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 905                 bond->igmp_retrans = bond->params.resend_igmp;
 906                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 907         }
 908 }
 909
 910 /**
 911  * bond_select_active_slave - select a new active slave, if needed
 912  * @bond: our bonding struct
 913  *
 914  * This functions should be called when one of the following occurs:
 915  * - The old curr_active_slave has been released or lost its link.
 916  * - The primary_slave has got its link back.
 917  * - A slave has got its link back and there's no old curr_active_slave.
 918  *
 919  * Caller must hold RTNL.
 920  */
 921 void bond_select_active_slave(struct bonding *bond)
 922 {
 923         struct slave *best_slave;
 924         int rv;
 925
 926         ASSERT_RTNL();
 927
 928         best_slave = bond_find_best_slave(bond);
 929         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 930                 bond_change_active_slave(bond, best_slave);
 931                 rv = bond_set_carrier(bond);
 932                 if (!rv)
 933                         return;
 934
 935                 if (netif_carrier_ok(bond->dev))
 936                         netdev_info(bond->dev, "first active interface up!\n");
 937                 else
 938                         netdev_info(bond->dev, "now running without any active interface!\n");
 939         }
 940 }
 941
 942 #ifdef CONFIG_NET_POLL_CONTROLLER
 943 static inline int slave_enable_netpoll(struct slave *slave)
 944 {
 945         struct netpoll *np;
 946         int err = 0;
 947
 948         np = kzalloc(sizeof(*np), GFP_KERNEL);
 949         err = -ENOMEM;
 950         if (!np)
 951                 goto out;
 952
 953         err = __netpoll_setup(np, slave->dev);
 954         if (err) {
 955                 kfree(np);
 956                 goto out;
 957         }
 958         slave->np = np;
 959 out:
 960         return err;
 961 }
 962 static inline void slave_disable_netpoll(struct slave *slave)
 963 {
 964         struct netpoll *np = slave->np;
 965
 966         if (!np)
 967                 return;
 968
 969         slave->np = NULL;
 970         __netpoll_free_async(np);
 971 }
 972
 973 static void bond_poll_controller(struct net_device *bond_dev)
 974 {
 975         struct bonding *bond = netdev_priv(bond_dev);
 976         struct slave *slave = NULL;
 977         struct list_head *iter;
 978         struct ad_info ad_info;
 979         struct netpoll_info *ni;
 980         const struct net_device_ops *ops;
 981
 982         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 983                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 984                         return;
 985
 986         bond_for_each_slave_rcu(bond, slave, iter) {
 987                 ops = slave->dev->netdev_ops;
 988                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 989                         continue;
 990
 991                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 992                         struct aggregator *agg =
 993                             SLAVE_AD_INFO(slave)->port.aggregator;
 994
 995                         if (agg &&
 996                             agg->aggregator_identifier != ad_info.aggregator_id)
 997                                 continue;
 998                 }
 999
1000                 ni = rcu_dereference_bh(slave->dev->npinfo);
1001                 if (down_trylock(&ni->dev_lock))
1002                         continue;
1003                 ops->ndo_poll_controller(slave->dev);
1004                 up(&ni->dev_lock);
1005         }
1006 }
1007
1008 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1009 {
1010         struct bonding *bond = netdev_priv(bond_dev);
1011         struct list_head *iter;
1012         struct slave *slave;
1013
1014         bond_for_each_slave(bond, slave, iter)
1015                 if (bond_slave_is_up(slave))
1016                         slave_disable_netpoll(slave);
1017 }
1018
1019 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1020 {
1021         struct bonding *bond = netdev_priv(dev);
1022         struct list_head *iter;
1023         struct slave *slave;
1024         int err = 0;
1025
1026         bond_for_each_slave(bond, slave, iter) {
1027                 err = slave_enable_netpoll(slave);
1028                 if (err) {
1029                         bond_netpoll_cleanup(dev);
1030                         break;
1031                 }
1032         }
1033         return err;
1034 }
1035 #else
1036 static inline int slave_enable_netpoll(struct slave *slave)
1037 {
1038         return 0;
1039 }
1040 static inline void slave_disable_netpoll(struct slave *slave)
1041 {
1042 }
1043 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1044 {
1045 }
1046 #endif
1047
1048 /*---------------------------------- IOCTL ----------------------------------*/
1049
1050 static netdev_features_t bond_fix_features(struct net_device *dev,
1051                                            netdev_features_t features)
1052 {
1053         struct bonding *bond = netdev_priv(dev);
1054         struct list_head *iter;
1055         netdev_features_t mask;
1056         struct slave *slave;
1057
1058         mask = features;
1059
1060         features &= ~NETIF_F_ONE_FOR_ALL;
1061         features |= NETIF_F_ALL_FOR_ALL;
1062
1063         bond_for_each_slave(bond, slave, iter) {
1064                 features = netdev_increment_features(features,
1065                                                      slave->dev->features,
1066                                                      mask);
1067         }
1068         features = netdev_add_tso_features(features, mask);
1069
1070         return features;
1071 }
1072
1073 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1074                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1075                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1076
1077 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1078                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1079
1080 static void bond_compute_features(struct bonding *bond)
1081 {
1082         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1083                                         IFF_XMIT_DST_RELEASE_PERM;
1084         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1085         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1086         struct net_device *bond_dev = bond->dev;
1087         struct list_head *iter;
1088         struct slave *slave;
1089         unsigned short max_hard_header_len = ETH_HLEN;
1090         unsigned int gso_max_size = GSO_MAX_SIZE;
1091         u16 gso_max_segs = GSO_MAX_SEGS;
1092
1093         if (!bond_has_slaves(bond))
1094                 goto done;
1095         vlan_features &= NETIF_F_ALL_FOR_ALL;
1096
1097         bond_for_each_slave(bond, slave, iter) {
1098                 vlan_features = netdev_increment_features(vlan_features,
1099                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1100
1101                 enc_features = netdev_increment_features(enc_features,
1102                                                          slave->dev->hw_enc_features,
1103                                                          BOND_ENC_FEATURES);
1104                 dst_release_flag &= slave->dev->priv_flags;
1105                 if (slave->dev->hard_header_len > max_hard_header_len)
1106                         max_hard_header_len = slave->dev->hard_header_len;
1107
1108                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1109                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1110         }
1111
1112 done:
1113         bond_dev->vlan_features = vlan_features;
1114         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1115         bond_dev->hard_header_len = max_hard_header_len;
1116         bond_dev->gso_max_segs = gso_max_segs;
1117         netif_set_gso_max_size(bond_dev, gso_max_size);
1118
1119         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1120         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1121             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1122                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1123
1124         netdev_change_features(bond_dev);
1125 }
1126
1127 static void bond_setup_by_slave(struct net_device *bond_dev,
1128                                 struct net_device *slave_dev)
1129 {
1130         bond_dev->header_ops        = slave_dev->header_ops;
1131
1132         bond_dev->type              = slave_dev->type;
1133         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1134         bond_dev->addr_len          = slave_dev->addr_len;
1135
1136         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1137                 slave_dev->addr_len);
1138 }
1139
1140 /* On bonding slaves other than the currently active slave, suppress
1141  * duplicates except for alb non-mcast/bcast.
1142  */
1143 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1144                                             struct slave *slave,
1145                                             struct bonding *bond)
1146 {
1147         if (bond_is_slave_inactive(slave)) {
1148                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1149                     skb->pkt_type != PACKET_BROADCAST &&
1150                     skb->pkt_type != PACKET_MULTICAST)
1151                         return false;
1152                 return true;
1153         }
1154         return false;
1155 }
1156
1157 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1158 {
1159         struct sk_buff *skb = *pskb;
1160         struct slave *slave;
1161         struct bonding *bond;
1162         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1163                           struct slave *);
1164         int ret = RX_HANDLER_ANOTHER;
1165
1166         skb = skb_share_check(skb, GFP_ATOMIC);
1167         if (unlikely(!skb))
1168                 return RX_HANDLER_CONSUMED;
1169
1170         *pskb = skb;
1171
1172         slave = bond_slave_get_rcu(skb->dev);
1173         bond = slave->bond;
1174
1175         recv_probe = ACCESS_ONCE(bond->recv_probe);
1176         if (recv_probe) {
1177                 ret = recv_probe(skb, bond, slave);
1178                 if (ret == RX_HANDLER_CONSUMED) {
1179                         consume_skb(skb);
1180                         return ret;
1181                 }
1182         }
1183
1184         if (bond_should_deliver_exact_match(skb, slave, bond))
1185                 return RX_HANDLER_EXACT;
1186
1187         skb->dev = bond->dev;
1188
1189         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1190             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1191             skb->pkt_type == PACKET_HOST) {
1192
1193                 if (unlikely(skb_cow_head(skb,
1194                                           skb->data - skb_mac_header(skb)))) {
1195                         kfree_skb(skb);
1196                         return RX_HANDLER_CONSUMED;
1197                 }
1198                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
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 int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave)
1222 {
1223         struct netdev_lag_upper_info lag_upper_info;
1224         int err;
1225
1226         lag_upper_info.tx_type = bond_lag_tx_type(bond);
1227         err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1228                                            &lag_upper_info);
1229         if (err)
1230                 return err;
1231         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1232         return 0;
1233 }
1234
1235 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1236 {
1237         netdev_upper_dev_unlink(slave->dev, bond->dev);
1238         slave->dev->flags &= ~IFF_SLAVE;
1239         rtmsg_ifinfo(RTM_NEWLINK, slave->dev, IFF_SLAVE, GFP_KERNEL);
1240 }
1241
1242 static struct slave *bond_alloc_slave(struct bonding *bond)
1243 {
1244         struct slave *slave = NULL;
1245
1246         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1247         if (!slave)
1248                 return NULL;
1249
1250         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1251                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1252                                                GFP_KERNEL);
1253                 if (!SLAVE_AD_INFO(slave)) {
1254                         kfree(slave);
1255                         return NULL;
1256                 }
1257         }
1258         return slave;
1259 }
1260
1261 static void bond_free_slave(struct slave *slave)
1262 {
1263         struct bonding *bond = bond_get_bond_by_slave(slave);
1264
1265         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1266                 kfree(SLAVE_AD_INFO(slave));
1267
1268         kfree(slave);
1269 }
1270
1271 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1272 {
1273         info->bond_mode = BOND_MODE(bond);
1274         info->miimon = bond->params.miimon;
1275         info->num_slaves = bond->slave_cnt;
1276 }
1277
1278 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1279 {
1280         strcpy(info->slave_name, slave->dev->name);
1281         info->link = slave->link;
1282         info->state = bond_slave_state(slave);
1283         info->link_failure_count = slave->link_failure_count;
1284 }
1285
1286 static void bond_netdev_notify(struct net_device *dev,
1287                                struct netdev_bonding_info *info)
1288 {
1289         rtnl_lock();
1290         netdev_bonding_info_change(dev, info);
1291         rtnl_unlock();
1292 }
1293
1294 static void bond_netdev_notify_work(struct work_struct *_work)
1295 {
1296         struct netdev_notify_work *w =
1297                 container_of(_work, struct netdev_notify_work, work.work);
1298
1299         bond_netdev_notify(w->dev, &w->bonding_info);
1300         dev_put(w->dev);
1301         kfree(w);
1302 }
1303
1304 void bond_queue_slave_event(struct slave *slave)
1305 {
1306         struct bonding *bond = slave->bond;
1307         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1308
1309         if (!nnw)
1310                 return;
1311
1312         dev_hold(slave->dev);
1313         nnw->dev = slave->dev;
1314         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1315         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1316         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1317
1318         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1319 }
1320
1321 void bond_lower_state_changed(struct slave *slave)
1322 {
1323         struct netdev_lag_lower_state_info info;
1324
1325         info.link_up = slave->link == BOND_LINK_UP ||
1326                        slave->link == BOND_LINK_FAIL;
1327         info.tx_enabled = bond_is_active_slave(slave);
1328         netdev_lower_state_changed(slave->dev, &info);
1329 }
1330
1331 /* enslave device <slave> to bond device <master> */
1332 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1333 {
1334         struct bonding *bond = netdev_priv(bond_dev);
1335         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1336         struct slave *new_slave = NULL, *prev_slave;
1337         struct sockaddr addr;
1338         int link_reporting;
1339         int res = 0, i;
1340
1341         if (!bond->params.use_carrier &&
1342             slave_dev->ethtool_ops->get_link == NULL &&
1343             slave_ops->ndo_do_ioctl == NULL) {
1344                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1345                             slave_dev->name);
1346         }
1347
1348         /* already in-use? */
1349         if (netdev_is_rx_handler_busy(slave_dev)) {
1350                 netdev_err(bond_dev,
1351                            "Error: Device is in use and cannot be enslaved\n");
1352                 return -EBUSY;
1353         }
1354
1355         if (bond_dev == slave_dev) {
1356                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1357                 return -EPERM;
1358         }
1359
1360         /* vlan challenged mutual exclusion */
1361         /* no need to lock since we're protected by rtnl_lock */
1362         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1363                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1364                            slave_dev->name);
1365                 if (vlan_uses_dev(bond_dev)) {
1366                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1367                                    slave_dev->name, bond_dev->name);
1368                         return -EPERM;
1369                 } else {
1370                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1371                                     slave_dev->name, slave_dev->name,
1372                                     bond_dev->name);
1373                 }
1374         } else {
1375                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1376                            slave_dev->name);
1377         }
1378
1379         /* Old ifenslave binaries are no longer supported.  These can
1380          * be identified with moderate accuracy by the state of the slave:
1381          * the current ifenslave will set the interface down prior to
1382          * enslaving it; the old ifenslave will not.
1383          */
1384         if (slave_dev->flags & IFF_UP) {
1385                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1386                            slave_dev->name);
1387                 return -EPERM;
1388         }
1389
1390         /* set bonding device ether type by slave - bonding netdevices are
1391          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1392          * there is a need to override some of the type dependent attribs/funcs.
1393          *
1394          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1395          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1396          */
1397         if (!bond_has_slaves(bond)) {
1398                 if (bond_dev->type != slave_dev->type) {
1399                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1400                                    bond_dev->type, slave_dev->type);
1401
1402                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1403                                                        bond_dev);
1404                         res = notifier_to_errno(res);
1405                         if (res) {
1406                                 netdev_err(bond_dev, "refused to change device type\n");
1407                                 return -EBUSY;
1408                         }
1409
1410                         /* Flush unicast and multicast addresses */
1411                         dev_uc_flush(bond_dev);
1412                         dev_mc_flush(bond_dev);
1413
1414                         if (slave_dev->type != ARPHRD_ETHER)
1415                                 bond_setup_by_slave(bond_dev, slave_dev);
1416                         else {
1417                                 ether_setup(bond_dev);
1418                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1419                         }
1420
1421                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1422                                                  bond_dev);
1423                 }
1424         } else if (bond_dev->type != slave_dev->type) {
1425                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1426                            slave_dev->name, slave_dev->type, bond_dev->type);
1427                 return -EINVAL;
1428         }
1429
1430         if (slave_dev->type == ARPHRD_INFINIBAND &&
1431             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1432                 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1433                             slave_dev->type);
1434                 res = -EOPNOTSUPP;
1435                 goto err_undo_flags;
1436         }
1437
1438         if (!slave_ops->ndo_set_mac_address ||
1439             slave_dev->type == ARPHRD_INFINIBAND) {
1440                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1441                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1442                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1443                         if (!bond_has_slaves(bond)) {
1444                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1445                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1446                         } else {
1447                                 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");
1448                                 res = -EOPNOTSUPP;
1449                                 goto err_undo_flags;
1450                         }
1451                 }
1452         }
1453
1454         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1455
1456         /* If this is the first slave, then we need to set the master's hardware
1457          * address to be the same as the slave's.
1458          */
1459         if (!bond_has_slaves(bond) &&
1460             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1461                 bond_set_dev_addr(bond->dev, slave_dev);
1462
1463         new_slave = bond_alloc_slave(bond);
1464         if (!new_slave) {
1465                 res = -ENOMEM;
1466                 goto err_undo_flags;
1467         }
1468
1469         new_slave->bond = bond;
1470         new_slave->dev = slave_dev;
1471         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1472          * is set via sysfs or module option if desired.
1473          */
1474         new_slave->queue_id = 0;
1475
1476         /* Save slave's original mtu and then set it to match the bond */
1477         new_slave->original_mtu = slave_dev->mtu;
1478         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1479         if (res) {
1480                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1481                 goto err_free;
1482         }
1483
1484         /* Save slave's original ("permanent") mac address for modes
1485          * that need it, and for restoring it upon release, and then
1486          * set it to the master's address
1487          */
1488         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1489
1490         if (!bond->params.fail_over_mac ||
1491             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1492                 /* Set slave to master's mac address.  The application already
1493                  * set the master's mac address to that of the first slave
1494                  */
1495                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1496                 addr.sa_family = slave_dev->type;
1497                 res = dev_set_mac_address(slave_dev, &addr);
1498                 if (res) {
1499                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1500                         goto err_restore_mtu;
1501                 }
1502         }
1503
1504         /* set slave flag before open to prevent IPv6 addrconf */
1505         slave_dev->flags |= IFF_SLAVE;
1506
1507         /* open the slave since the application closed it */
1508         res = dev_open(slave_dev);
1509         if (res) {
1510                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1511                 goto err_restore_mac;
1512         }
1513
1514         slave_dev->priv_flags |= IFF_BONDING;
1515         /* initialize slave stats */
1516         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1517
1518         if (bond_is_lb(bond)) {
1519                 /* bond_alb_init_slave() must be called before all other stages since
1520                  * it might fail and we do not want to have to undo everything
1521                  */
1522                 res = bond_alb_init_slave(bond, new_slave);
1523                 if (res)
1524                         goto err_close;
1525         }
1526
1527         /* If the mode uses primary, then the following is handled by
1528          * bond_change_active_slave().
1529          */
1530         if (!bond_uses_primary(bond)) {
1531                 /* set promiscuity level to new slave */
1532                 if (bond_dev->flags & IFF_PROMISC) {
1533                         res = dev_set_promiscuity(slave_dev, 1);
1534                         if (res)
1535                                 goto err_close;
1536                 }
1537
1538                 /* set allmulti level to new slave */
1539                 if (bond_dev->flags & IFF_ALLMULTI) {
1540                         res = dev_set_allmulti(slave_dev, 1);
1541                         if (res)
1542                                 goto err_close;
1543                 }
1544
1545                 netif_addr_lock_bh(bond_dev);
1546
1547                 dev_mc_sync_multiple(slave_dev, bond_dev);
1548                 dev_uc_sync_multiple(slave_dev, bond_dev);
1549
1550                 netif_addr_unlock_bh(bond_dev);
1551         }
1552
1553         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1554                 /* add lacpdu mc addr to mc list */
1555                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1556
1557                 dev_mc_add(slave_dev, lacpdu_multicast);
1558         }
1559
1560         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1561         if (res) {
1562                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1563                            slave_dev->name);
1564                 goto err_close;
1565         }
1566
1567         prev_slave = bond_last_slave(bond);
1568
1569         new_slave->delay = 0;
1570         new_slave->link_failure_count = 0;
1571
1572         bond_update_speed_duplex(new_slave);
1573
1574         new_slave->last_rx = jiffies -
1575                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1576         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1577                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1578
1579         if (bond->params.miimon && !bond->params.use_carrier) {
1580                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1581
1582                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1583                         /* miimon is set but a bonded network driver
1584                          * does not support ETHTOOL/MII and
1585                          * arp_interval is not set.  Note: if
1586                          * use_carrier is enabled, we will never go
1587                          * here (because netif_carrier is always
1588                          * supported); thus, we don't need to change
1589                          * the messages for netif_carrier.
1590                          */
1591                         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",
1592                                     slave_dev->name);
1593                 } else if (link_reporting == -1) {
1594                         /* unable get link status using mii/ethtool */
1595                         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",
1596                                     slave_dev->name);
1597                 }
1598         }
1599
1600         /* check for initial state */
1601         new_slave->link = BOND_LINK_NOCHANGE;
1602         if (bond->params.miimon) {
1603                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1604                         if (bond->params.updelay) {
1605                                 bond_set_slave_link_state(new_slave,
1606                                                           BOND_LINK_BACK,
1607                                                           BOND_SLAVE_NOTIFY_NOW);
1608                                 new_slave->delay = bond->params.updelay;
1609                         } else {
1610                                 bond_set_slave_link_state(new_slave,
1611                                                           BOND_LINK_UP,
1612                                                           BOND_SLAVE_NOTIFY_NOW);
1613                         }
1614                 } else {
1615                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1616                                                   BOND_SLAVE_NOTIFY_NOW);
1617                 }
1618         } else if (bond->params.arp_interval) {
1619                 bond_set_slave_link_state(new_slave,
1620                                           (netif_carrier_ok(slave_dev) ?
1621                                           BOND_LINK_UP : BOND_LINK_DOWN),
1622                                           BOND_SLAVE_NOTIFY_NOW);
1623         } else {
1624                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1625                                           BOND_SLAVE_NOTIFY_NOW);
1626         }
1627
1628         if (new_slave->link != BOND_LINK_DOWN)
1629                 new_slave->last_link_up = jiffies;
1630         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1631                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1632                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1633
1634         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1635                 /* if there is a primary slave, remember it */
1636                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1637                         rcu_assign_pointer(bond->primary_slave, new_slave);
1638                         bond->force_primary = true;
1639                 }
1640         }
1641
1642         switch (BOND_MODE(bond)) {
1643         case BOND_MODE_ACTIVEBACKUP:
1644                 bond_set_slave_inactive_flags(new_slave,
1645                                               BOND_SLAVE_NOTIFY_NOW);
1646                 break;
1647         case BOND_MODE_8023AD:
1648                 /* in 802.3ad mode, the internal mechanism
1649                  * will activate the slaves in the selected
1650                  * aggregator
1651                  */
1652                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1653                 /* if this is the first slave */
1654                 if (!prev_slave) {
1655                         SLAVE_AD_INFO(new_slave)->id = 1;
1656                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1657                          * can be called only after the mac address of the bond is set
1658                          */
1659                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1660                 } else {
1661                         SLAVE_AD_INFO(new_slave)->id =
1662                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1663                 }
1664
1665                 bond_3ad_bind_slave(new_slave);
1666                 break;
1667         case BOND_MODE_TLB:
1668         case BOND_MODE_ALB:
1669                 bond_set_active_slave(new_slave);
1670                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1671                 break;
1672         default:
1673                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1674
1675                 /* always active in trunk mode */
1676                 bond_set_active_slave(new_slave);
1677
1678                 /* In trunking mode there is little meaning to curr_active_slave
1679                  * anyway (it holds no special properties of the bond device),
1680                  * so we can change it without calling change_active_interface()
1681                  */
1682                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1683                     new_slave->link == BOND_LINK_UP)
1684                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1685
1686                 break;
1687         } /* switch(bond_mode) */
1688
1689 #ifdef CONFIG_NET_POLL_CONTROLLER
1690         slave_dev->npinfo = bond->dev->npinfo;
1691         if (slave_dev->npinfo) {
1692                 if (slave_enable_netpoll(new_slave)) {
1693                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1694                         res = -EBUSY;
1695                         goto err_detach;
1696                 }
1697         }
1698 #endif
1699
1700         if (!(bond_dev->features & NETIF_F_LRO))
1701                 dev_disable_lro(slave_dev);
1702
1703         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1704                                          new_slave);
1705         if (res) {
1706                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1707                 goto err_detach;
1708         }
1709
1710         res = bond_master_upper_dev_link(bond, new_slave);
1711         if (res) {
1712                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1713                 goto err_unregister;
1714         }
1715
1716         res = bond_sysfs_slave_add(new_slave);
1717         if (res) {
1718                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1719                 goto err_upper_unlink;
1720         }
1721
1722         bond->slave_cnt++;
1723         bond_compute_features(bond);
1724         bond_set_carrier(bond);
1725
1726         if (bond_uses_primary(bond)) {
1727                 block_netpoll_tx();
1728                 bond_select_active_slave(bond);
1729                 unblock_netpoll_tx();
1730         }
1731
1732         if (bond_mode_uses_xmit_hash(bond))
1733                 bond_update_slave_arr(bond, NULL);
1734
1735         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1736                     slave_dev->name,
1737                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1738                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1739
1740         /* enslave is successful */
1741         bond_queue_slave_event(new_slave);
1742         return 0;
1743
1744 /* Undo stages on error */
1745 err_upper_unlink:
1746         bond_upper_dev_unlink(bond, new_slave);
1747
1748 err_unregister:
1749         netdev_rx_handler_unregister(slave_dev);
1750
1751 err_detach:
1752         if (!bond_uses_primary(bond))
1753                 bond_hw_addr_flush(bond_dev, slave_dev);
1754
1755         vlan_vids_del_by_dev(slave_dev, bond_dev);
1756         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1757                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1758         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1759                 block_netpoll_tx();
1760                 bond_change_active_slave(bond, NULL);
1761                 bond_select_active_slave(bond);
1762                 unblock_netpoll_tx();
1763         }
1764         /* either primary_slave or curr_active_slave might've changed */
1765         synchronize_rcu();
1766         slave_disable_netpoll(new_slave);
1767
1768 err_close:
1769         slave_dev->priv_flags &= ~IFF_BONDING;
1770         dev_close(slave_dev);
1771
1772 err_restore_mac:
1773         slave_dev->flags &= ~IFF_SLAVE;
1774         if (!bond->params.fail_over_mac ||
1775             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1776                 /* XXX TODO - fom follow mode needs to change master's
1777                  * MAC if this slave's MAC is in use by the bond, or at
1778                  * least print a warning.
1779                  */
1780                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1781                 addr.sa_family = slave_dev->type;
1782                 dev_set_mac_address(slave_dev, &addr);
1783         }
1784
1785 err_restore_mtu:
1786         dev_set_mtu(slave_dev, new_slave->original_mtu);
1787
1788 err_free:
1789         bond_free_slave(new_slave);
1790
1791 err_undo_flags:
1792         /* Enslave of first slave has failed and we need to fix master's mac */
1793         if (!bond_has_slaves(bond)) {
1794                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1795                                             slave_dev->dev_addr))
1796                         eth_hw_addr_random(bond_dev);
1797                 if (bond_dev->type != ARPHRD_ETHER) {
1798                         dev_close(bond_dev);
1799                         ether_setup(bond_dev);
1800                         bond_dev->flags |= IFF_MASTER;
1801                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1802                 }
1803         }
1804
1805         return res;
1806 }
1807
1808 /* Try to release the slave device <slave> from the bond device <master>
1809  * It is legal to access curr_active_slave without a lock because all the function
1810  * is RTNL-locked. If "all" is true it means that the function is being called
1811  * while destroying a bond interface and all slaves are being released.
1812  *
1813  * The rules for slave state should be:
1814  *   for Active/Backup:
1815  *     Active stays on all backups go down
1816  *   for Bonded connections:
1817  *     The first up interface should be left on and all others downed.
1818  */
1819 static int __bond_release_one(struct net_device *bond_dev,
1820                               struct net_device *slave_dev,
1821                               bool all)
1822 {
1823         struct bonding *bond = netdev_priv(bond_dev);
1824         struct slave *slave, *oldcurrent;
1825         struct sockaddr addr;
1826         int old_flags = bond_dev->flags;
1827         netdev_features_t old_features = bond_dev->features;
1828
1829         /* slave is not a slave or master is not master of this slave */
1830         if (!(slave_dev->flags & IFF_SLAVE) ||
1831             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1832                 netdev_dbg(bond_dev, "cannot release %s\n",
1833                            slave_dev->name);
1834                 return -EINVAL;
1835         }
1836
1837         block_netpoll_tx();
1838
1839         slave = bond_get_slave_by_dev(bond, slave_dev);
1840         if (!slave) {
1841                 /* not a slave of this bond */
1842                 netdev_info(bond_dev, "%s not enslaved\n",
1843                             slave_dev->name);
1844                 unblock_netpoll_tx();
1845                 return -EINVAL;
1846         }
1847
1848         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1849
1850         bond_sysfs_slave_del(slave);
1851
1852         /* recompute stats just before removing the slave */
1853         bond_get_stats(bond->dev, &bond->bond_stats);
1854
1855         bond_upper_dev_unlink(bond, slave);
1856         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1857          * for this slave anymore.
1858          */
1859         netdev_rx_handler_unregister(slave_dev);
1860
1861         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1862                 bond_3ad_unbind_slave(slave);
1863
1864         if (bond_mode_uses_xmit_hash(bond))
1865                 bond_update_slave_arr(bond, slave);
1866
1867         netdev_info(bond_dev, "Releasing %s interface %s\n",
1868                     bond_is_active_slave(slave) ? "active" : "backup",
1869                     slave_dev->name);
1870
1871         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1872
1873         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1874
1875         if (!all && (!bond->params.fail_over_mac ||
1876                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1877                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1878                     bond_has_slaves(bond))
1879                         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",
1880                                     slave_dev->name, slave->perm_hwaddr,
1881                                     bond_dev->name, slave_dev->name);
1882         }
1883
1884         if (rtnl_dereference(bond->primary_slave) == slave)
1885                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1886
1887         if (oldcurrent == slave)
1888                 bond_change_active_slave(bond, NULL);
1889
1890         if (bond_is_lb(bond)) {
1891                 /* Must be called only after the slave has been
1892                  * detached from the list and the curr_active_slave
1893                  * has been cleared (if our_slave == old_current),
1894                  * but before a new active slave is selected.
1895                  */
1896                 bond_alb_deinit_slave(bond, slave);
1897         }
1898
1899         if (all) {
1900                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1901         } else if (oldcurrent == slave) {
1902                 /* Note that we hold RTNL over this sequence, so there
1903                  * is no concern that another slave add/remove event
1904                  * will interfere.
1905                  */
1906                 bond_select_active_slave(bond);
1907         }
1908
1909         if (!bond_has_slaves(bond)) {
1910                 bond_set_carrier(bond);
1911                 eth_hw_addr_random(bond_dev);
1912         }
1913
1914         unblock_netpoll_tx();
1915         synchronize_rcu();
1916         bond->slave_cnt--;
1917
1918         if (!bond_has_slaves(bond)) {
1919                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1920                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1921         }
1922
1923         bond_compute_features(bond);
1924         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1925             (old_features & NETIF_F_VLAN_CHALLENGED))
1926                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1927                             slave_dev->name, bond_dev->name);
1928
1929         vlan_vids_del_by_dev(slave_dev, bond_dev);
1930
1931         /* If the mode uses primary, then this case was handled above by
1932          * bond_change_active_slave(..., NULL)
1933          */
1934         if (!bond_uses_primary(bond)) {
1935                 /* unset promiscuity level from slave
1936                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1937                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1938                  * value of that flag before that change, as that was the value
1939                  * when this slave was attached, so we cache at the start of the
1940                  * function and use it here. Same goes for ALLMULTI below
1941                  */
1942                 if (old_flags & IFF_PROMISC)
1943                         dev_set_promiscuity(slave_dev, -1);
1944
1945                 /* unset allmulti level from slave */
1946                 if (old_flags & IFF_ALLMULTI)
1947                         dev_set_allmulti(slave_dev, -1);
1948
1949                 bond_hw_addr_flush(bond_dev, slave_dev);
1950         }
1951
1952         slave_disable_netpoll(slave);
1953
1954         /* close slave before restoring its mac address */
1955         dev_close(slave_dev);
1956
1957         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1958             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1959                 /* restore original ("permanent") mac address */
1960                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1961                 addr.sa_family = slave_dev->type;
1962                 dev_set_mac_address(slave_dev, &addr);
1963         }
1964
1965         dev_set_mtu(slave_dev, slave->original_mtu);
1966
1967         slave_dev->priv_flags &= ~IFF_BONDING;
1968
1969         bond_free_slave(slave);
1970
1971         return 0;
1972 }
1973
1974 /* A wrapper used because of ndo_del_link */
1975 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1976 {
1977         return __bond_release_one(bond_dev, slave_dev, false);
1978 }
1979
1980 /* First release a slave and then destroy the bond if no more slaves are left.
1981  * Must be under rtnl_lock when this function is called.
1982  */
1983 static int  bond_release_and_destroy(struct net_device *bond_dev,
1984                                      struct net_device *slave_dev)
1985 {
1986         struct bonding *bond = netdev_priv(bond_dev);
1987         int ret;
1988
1989         ret = bond_release(bond_dev, slave_dev);
1990         if (ret == 0 && !bond_has_slaves(bond)) {
1991                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1992                 netdev_info(bond_dev, "Destroying bond %s\n",
1993                             bond_dev->name);
1994                 bond_remove_proc_entry(bond);
1995                 unregister_netdevice(bond_dev);
1996         }
1997         return ret;
1998 }
1999
2000 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2001 {
2002         struct bonding *bond = netdev_priv(bond_dev);
2003         bond_fill_ifbond(bond, info);
2004         return 0;
2005 }
2006
2007 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2008 {
2009         struct bonding *bond = netdev_priv(bond_dev);
2010         struct list_head *iter;
2011         int i = 0, res = -ENODEV;
2012         struct slave *slave;
2013
2014         bond_for_each_slave(bond, slave, iter) {
2015                 if (i++ == (int)info->slave_id) {
2016                         res = 0;
2017                         bond_fill_ifslave(slave, info);
2018                         break;
2019                 }
2020         }
2021
2022         return res;
2023 }
2024
2025 /*-------------------------------- Monitoring -------------------------------*/
2026
2027 /* called with rcu_read_lock() */
2028 static int bond_miimon_inspect(struct bonding *bond)
2029 {
2030         int link_state, commit = 0;
2031         struct list_head *iter;
2032         struct slave *slave;
2033         bool ignore_updelay;
2034
2035         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2036
2037         bond_for_each_slave_rcu(bond, slave, iter) {
2038                 slave->new_link = BOND_LINK_NOCHANGE;
2039
2040                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2041
2042                 switch (slave->link) {
2043                 case BOND_LINK_UP:
2044                         if (link_state)
2045                                 continue;
2046
2047                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
2048                                                   BOND_SLAVE_NOTIFY_LATER);
2049                         slave->delay = bond->params.downdelay;
2050                         if (slave->delay) {
2051                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2052                                             (BOND_MODE(bond) ==
2053                                              BOND_MODE_ACTIVEBACKUP) ?
2054                                              (bond_is_active_slave(slave) ?
2055                                               "active " : "backup ") : "",
2056                                             slave->dev->name,
2057                                             bond->params.downdelay * bond->params.miimon);
2058                         }
2059                         /*FALLTHRU*/
2060                 case BOND_LINK_FAIL:
2061                         if (link_state) {
2062                                 /* recovered before downdelay expired */
2063                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2064                                                           BOND_SLAVE_NOTIFY_LATER);
2065                                 slave->last_link_up = jiffies;
2066                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2067                                             (bond->params.downdelay - slave->delay) *
2068                                             bond->params.miimon,
2069                                             slave->dev->name);
2070                                 commit++;
2071                                 continue;
2072                         }
2073
2074                         if (slave->delay <= 0) {
2075                                 slave->new_link = BOND_LINK_DOWN;
2076                                 commit++;
2077                                 continue;
2078                         }
2079
2080                         slave->delay--;
2081                         break;
2082
2083                 case BOND_LINK_DOWN:
2084                         if (!link_state)
2085                                 continue;
2086
2087                         bond_set_slave_link_state(slave, BOND_LINK_BACK,
2088                                                   BOND_SLAVE_NOTIFY_LATER);
2089                         slave->delay = bond->params.updelay;
2090
2091                         if (slave->delay) {
2092                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2093                                             slave->dev->name,
2094                                             ignore_updelay ? 0 :
2095                                             bond->params.updelay *
2096                                             bond->params.miimon);
2097                         }
2098                         /*FALLTHRU*/
2099                 case BOND_LINK_BACK:
2100                         if (!link_state) {
2101                                 bond_set_slave_link_state(slave,
2102                                                           BOND_LINK_DOWN,
2103                                                           BOND_SLAVE_NOTIFY_LATER);
2104                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2105                                             (bond->params.updelay - slave->delay) *
2106                                             bond->params.miimon,
2107                                             slave->dev->name);
2108                                 commit++;
2109                                 continue;
2110                         }
2111
2112                         if (ignore_updelay)
2113                                 slave->delay = 0;
2114
2115                         if (slave->delay <= 0) {
2116                                 slave->new_link = BOND_LINK_UP;
2117                                 commit++;
2118                                 ignore_updelay = false;
2119                                 continue;
2120                         }
2121
2122                         slave->delay--;
2123                         break;
2124                 }
2125         }
2126
2127         return commit;
2128 }
2129
2130 static void bond_miimon_commit(struct bonding *bond)
2131 {
2132         struct list_head *iter;
2133         struct slave *slave, *primary;
2134
2135         bond_for_each_slave(bond, slave, iter) {
2136                 switch (slave->new_link) {
2137                 case BOND_LINK_NOCHANGE:
2138                         continue;
2139
2140                 case BOND_LINK_UP:
2141                         bond_update_speed_duplex(slave);
2142                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2143                                                   BOND_SLAVE_NOTIFY_NOW);
2144                         slave->last_link_up = jiffies;
2145
2146                         primary = rtnl_dereference(bond->primary_slave);
2147                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2148                                 /* prevent it from being the active one */
2149                                 bond_set_backup_slave(slave);
2150                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2151                                 /* make it immediately active */
2152                                 bond_set_active_slave(slave);
2153                         } else if (slave != primary) {
2154                                 /* prevent it from being the active one */
2155                                 bond_set_backup_slave(slave);
2156                         }
2157
2158                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2159                                     slave->dev->name,
2160                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2161                                     slave->duplex ? "full" : "half");
2162
2163                         /* notify ad that the link status has changed */
2164                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2165                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2166
2167                         if (bond_is_lb(bond))
2168                                 bond_alb_handle_link_change(bond, slave,
2169                                                             BOND_LINK_UP);
2170
2171                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2172                                 bond_update_slave_arr(bond, NULL);
2173
2174                         if (!bond->curr_active_slave || slave == primary)
2175                                 goto do_failover;
2176
2177                         continue;
2178
2179                 case BOND_LINK_DOWN:
2180                         if (slave->link_failure_count < UINT_MAX)
2181                                 slave->link_failure_count++;
2182
2183                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2184                                                   BOND_SLAVE_NOTIFY_NOW);
2185
2186                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2187                             BOND_MODE(bond) == BOND_MODE_8023AD)
2188                                 bond_set_slave_inactive_flags(slave,
2189                                                               BOND_SLAVE_NOTIFY_NOW);
2190
2191                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2192                                     slave->dev->name);
2193
2194                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2195                                 bond_3ad_handle_link_change(slave,
2196                                                             BOND_LINK_DOWN);
2197
2198                         if (bond_is_lb(bond))
2199                                 bond_alb_handle_link_change(bond, slave,
2200                                                             BOND_LINK_DOWN);
2201
2202                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2203                                 bond_update_slave_arr(bond, NULL);
2204
2205                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2206                                 goto do_failover;
2207
2208                         continue;
2209
2210                 default:
2211                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2212                                    slave->new_link, slave->dev->name);
2213                         slave->new_link = BOND_LINK_NOCHANGE;
2214
2215                         continue;
2216                 }
2217
2218 do_failover:
2219                 block_netpoll_tx();
2220                 bond_select_active_slave(bond);
2221                 unblock_netpoll_tx();
2222         }
2223
2224         bond_set_carrier(bond);
2225 }
2226
2227 /* bond_mii_monitor
2228  *
2229  * Really a wrapper that splits the mii monitor into two phases: an
2230  * inspection, then (if inspection indicates something needs to be done)
2231  * an acquisition of appropriate locks followed by a commit phase to
2232  * implement whatever link state changes are indicated.
2233  */
2234 static void bond_mii_monitor(struct work_struct *work)
2235 {
2236         struct bonding *bond = container_of(work, struct bonding,
2237                                             mii_work.work);
2238         bool should_notify_peers = false;
2239         unsigned long delay;
2240
2241         delay = msecs_to_jiffies(bond->params.miimon);
2242
2243         if (!bond_has_slaves(bond))
2244                 goto re_arm;
2245
2246         rcu_read_lock();
2247
2248         should_notify_peers = bond_should_notify_peers(bond);
2249
2250         if (bond_miimon_inspect(bond)) {
2251                 rcu_read_unlock();
2252
2253                 /* Race avoidance with bond_close cancel of workqueue */
2254                 if (!rtnl_trylock()) {
2255                         delay = 1;
2256                         should_notify_peers = false;
2257                         goto re_arm;
2258                 }
2259
2260                 bond_miimon_commit(bond);
2261
2262                 rtnl_unlock();  /* might sleep, hold no other locks */
2263         } else
2264                 rcu_read_unlock();
2265
2266 re_arm:
2267         if (bond->params.miimon)
2268                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2269
2270         if (should_notify_peers) {
2271                 if (!rtnl_trylock())
2272                         return;
2273                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2274                 rtnl_unlock();
2275         }
2276 }
2277
2278 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2279 {
2280         struct net_device *upper;
2281         struct list_head *iter;
2282         bool ret = false;
2283
2284         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2285                 return true;
2286
2287         rcu_read_lock();
2288         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2289                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2290                         ret = true;
2291                         break;
2292                 }
2293         }
2294         rcu_read_unlock();
2295
2296         return ret;
2297 }
2298
2299 /* We go to the (large) trouble of VLAN tagging ARP frames because
2300  * switches in VLAN mode (especially if ports are configured as
2301  * "native" to a VLAN) might not pass non-tagged frames.
2302  */
2303 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2304                           __be32 dest_ip, __be32 src_ip,
2305                           struct bond_vlan_tag *tags)
2306 {
2307         struct sk_buff *skb;
2308         struct bond_vlan_tag *outer_tag = tags;
2309
2310         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2311                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2312
2313         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2314                          NULL, slave_dev->dev_addr, NULL);
2315
2316         if (!skb) {
2317                 net_err_ratelimited("ARP packet allocation failed\n");
2318                 return;
2319         }
2320
2321         if (!tags || tags->vlan_proto == VLAN_N_VID)
2322                 goto xmit;
2323
2324         tags++;
2325
2326         /* Go through all the tags backwards and add them to the packet */
2327         while (tags->vlan_proto != VLAN_N_VID) {
2328                 if (!tags->vlan_id) {
2329                         tags++;
2330                         continue;
2331                 }
2332
2333                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2334                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2335                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2336                                                 tags->vlan_id);
2337                 if (!skb) {
2338                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2339                         return;
2340                 }
2341
2342                 tags++;
2343         }
2344         /* Set the outer tag */
2345         if (outer_tag->vlan_id) {
2346                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2347                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2348                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2349                                        outer_tag->vlan_id);
2350         }
2351
2352 xmit:
2353         arp_xmit(skb);
2354 }
2355
2356 /* Validate the device path between the @start_dev and the @end_dev.
2357  * The path is valid if the @end_dev is reachable through device
2358  * stacking.
2359  * When the path is validated, collect any vlan information in the
2360  * path.
2361  */
2362 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2363                                               struct net_device *end_dev,
2364                                               int level)
2365 {
2366         struct bond_vlan_tag *tags;
2367         struct net_device *upper;
2368         struct list_head  *iter;
2369
2370         if (start_dev == end_dev) {
2371                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2372                 if (!tags)
2373                         return ERR_PTR(-ENOMEM);
2374                 tags[level].vlan_proto = VLAN_N_VID;
2375                 return tags;
2376         }
2377
2378         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2379                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2380                 if (IS_ERR_OR_NULL(tags)) {
2381                         if (IS_ERR(tags))
2382                                 return tags;
2383                         continue;
2384                 }
2385                 if (is_vlan_dev(upper)) {
2386                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2387                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2388                 }
2389
2390                 return tags;
2391         }
2392
2393         return NULL;
2394 }
2395
2396 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2397 {
2398         struct rtable *rt;
2399         struct bond_vlan_tag *tags;
2400         __be32 *targets = bond->params.arp_targets, addr;
2401         int i;
2402
2403         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2404                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2405                 tags = NULL;
2406
2407                 /* Find out through which dev should the packet go */
2408                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2409                                      RTO_ONLINK, 0);
2410                 if (IS_ERR(rt)) {
2411                         /* there's no route to target - try to send arp
2412                          * probe to generate any traffic (arp_validate=0)
2413                          */
2414                         if (bond->params.arp_validate)
2415                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2416                                                      bond->dev->name,
2417                                                      &targets[i]);
2418                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2419                                       0, tags);
2420                         continue;
2421                 }
2422
2423                 /* bond device itself */
2424                 if (rt->dst.dev == bond->dev)
2425                         goto found;
2426
2427                 rcu_read_lock();
2428                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2429                 rcu_read_unlock();
2430
2431                 if (!IS_ERR_OR_NULL(tags))
2432                         goto found;
2433
2434                 /* Not our device - skip */
2435                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2436                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2437
2438                 ip_rt_put(rt);
2439                 continue;
2440
2441 found:
2442                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2443                 ip_rt_put(rt);
2444                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2445                               addr, tags);
2446                 kfree(tags);
2447         }
2448 }
2449
2450 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2451 {
2452         int i;
2453
2454         if (!sip || !bond_has_this_ip(bond, tip)) {
2455                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2456                            &sip, &tip);
2457                 return;
2458         }
2459
2460         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2461         if (i == -1) {
2462                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2463                            &sip);
2464                 return;
2465         }
2466         slave->last_rx = jiffies;
2467         slave->target_last_arp_rx[i] = jiffies;
2468 }
2469
2470 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2471                  struct slave *slave)
2472 {
2473         struct arphdr *arp = (struct arphdr *)skb->data;
2474         struct slave *curr_active_slave, *curr_arp_slave;
2475         unsigned char *arp_ptr;
2476         __be32 sip, tip;
2477         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2478
2479         if (!slave_do_arp_validate(bond, slave)) {
2480                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2481                     !slave_do_arp_validate_only(bond))
2482                         slave->last_rx = jiffies;
2483                 return RX_HANDLER_ANOTHER;
2484         } else if (!is_arp) {
2485                 return RX_HANDLER_ANOTHER;
2486         }
2487
2488         alen = arp_hdr_len(bond->dev);
2489
2490         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2491                    skb->dev->name);
2492
2493         if (alen > skb_headlen(skb)) {
2494                 arp = kmalloc(alen, GFP_ATOMIC);
2495                 if (!arp)
2496                         goto out_unlock;
2497                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2498                         goto out_unlock;
2499         }
2500
2501         if (arp->ar_hln != bond->dev->addr_len ||
2502             skb->pkt_type == PACKET_OTHERHOST ||
2503             skb->pkt_type == PACKET_LOOPBACK ||
2504             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2505             arp->ar_pro != htons(ETH_P_IP) ||
2506             arp->ar_pln != 4)
2507                 goto out_unlock;
2508
2509         arp_ptr = (unsigned char *)(arp + 1);
2510         arp_ptr += bond->dev->addr_len;
2511         memcpy(&sip, arp_ptr, 4);
2512         arp_ptr += 4 + bond->dev->addr_len;
2513         memcpy(&tip, arp_ptr, 4);
2514
2515         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2516                    slave->dev->name, bond_slave_state(slave),
2517                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2518                      &sip, &tip);
2519
2520         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2521         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2522
2523         /* We 'trust' the received ARP enough to validate it if:
2524          *
2525          * (a) the slave receiving the ARP is active (which includes the
2526          * current ARP slave, if any), or
2527          *
2528          * (b) the receiving slave isn't active, but there is a currently
2529          * active slave and it received valid arp reply(s) after it became
2530          * the currently active slave, or
2531          *
2532          * (c) there is an ARP slave that sent an ARP during the prior ARP
2533          * interval, and we receive an ARP reply on any slave.  We accept
2534          * these because switch FDB update delays may deliver the ARP
2535          * reply to a slave other than the sender of the ARP request.
2536          *
2537          * Note: for (b), backup slaves are receiving the broadcast ARP
2538          * request, not a reply.  This request passes from the sending
2539          * slave through the L2 switch(es) to the receiving slave.  Since
2540          * this is checking the request, sip/tip are swapped for
2541          * validation.
2542          *
2543          * This is done to avoid endless looping when we can't reach the
2544          * arp_ip_target and fool ourselves with our own arp requests.
2545          */
2546         if (bond_is_active_slave(slave))
2547                 bond_validate_arp(bond, slave, sip, tip);
2548         else if (curr_active_slave &&
2549                  time_after(slave_last_rx(bond, curr_active_slave),
2550                             curr_active_slave->last_link_up))
2551                 bond_validate_arp(bond, slave, tip, sip);
2552         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2553                  bond_time_in_interval(bond,
2554                                        dev_trans_start(curr_arp_slave->dev), 1))
2555                 bond_validate_arp(bond, slave, sip, tip);
2556
2557 out_unlock:
2558         if (arp != (struct arphdr *)skb->data)
2559                 kfree(arp);
2560         return RX_HANDLER_ANOTHER;
2561 }
2562
2563 /* function to verify if we're in the arp_interval timeslice, returns true if
2564  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2565  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2566  */
2567 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2568                                   int mod)
2569 {
2570         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2571
2572         return time_in_range(jiffies,
2573                              last_act - delta_in_ticks,
2574                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2575 }
2576
2577 /* This function is called regularly to monitor each slave's link
2578  * ensuring that traffic is being sent and received when arp monitoring
2579  * is used in load-balancing mode. if the adapter has been dormant, then an
2580  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2581  * arp monitoring in active backup mode.
2582  */
2583 static void bond_loadbalance_arp_mon(struct work_struct *work)
2584 {
2585         struct bonding *bond = container_of(work, struct bonding,
2586                                             arp_work.work);
2587         struct slave *slave, *oldcurrent;
2588         struct list_head *iter;
2589         int do_failover = 0, slave_state_changed = 0;
2590
2591         if (!bond_has_slaves(bond))
2592                 goto re_arm;
2593
2594         rcu_read_lock();
2595
2596         oldcurrent = rcu_dereference(bond->curr_active_slave);
2597         /* see if any of the previous devices are up now (i.e. they have
2598          * xmt and rcv traffic). the curr_active_slave does not come into
2599          * the picture unless it is null. also, slave->last_link_up is not
2600          * needed here because we send an arp on each slave and give a slave
2601          * as long as it needs to get the tx/rx within the delta.
2602          * TODO: what about up/down delay in arp mode? it wasn't here before
2603          *       so it can wait
2604          */
2605         bond_for_each_slave_rcu(bond, slave, iter) {
2606                 unsigned long trans_start = dev_trans_start(slave->dev);
2607
2608                 slave->new_link = BOND_LINK_NOCHANGE;
2609
2610                 if (slave->link != BOND_LINK_UP) {
2611                         if (bond_time_in_interval(bond, trans_start, 1) &&
2612                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2613
2614                                 slave->new_link = BOND_LINK_UP;
2615                                 slave_state_changed = 1;
2616
2617                                 /* primary_slave has no meaning in round-robin
2618                                  * mode. the window of a slave being up and
2619                                  * curr_active_slave being null after enslaving
2620                                  * is closed.
2621                                  */
2622                                 if (!oldcurrent) {
2623                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2624                                                     slave->dev->name);
2625                                         do_failover = 1;
2626                                 } else {
2627                                         netdev_info(bond->dev, "interface %s is now up\n",
2628                                                     slave->dev->name);
2629                                 }
2630                         }
2631                 } else {
2632                         /* slave->link == BOND_LINK_UP */
2633
2634                         /* not all switches will respond to an arp request
2635                          * when the source ip is 0, so don't take the link down
2636                          * if we don't know our ip yet
2637                          */
2638                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2639                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2640
2641                                 slave->new_link = BOND_LINK_DOWN;
2642                                 slave_state_changed = 1;
2643
2644                                 if (slave->link_failure_count < UINT_MAX)
2645                                         slave->link_failure_count++;
2646
2647                                 netdev_info(bond->dev, "interface %s is now down\n",
2648                                             slave->dev->name);
2649
2650                                 if (slave == oldcurrent)
2651                                         do_failover = 1;
2652                         }
2653                 }
2654
2655                 /* note: if switch is in round-robin mode, all links
2656                  * must tx arp to ensure all links rx an arp - otherwise
2657                  * links may oscillate or not come up at all; if switch is
2658                  * in something like xor mode, there is nothing we can
2659                  * do - all replies will be rx'ed on same link causing slaves
2660                  * to be unstable during low/no traffic periods
2661                  */
2662                 if (bond_slave_is_up(slave))
2663                         bond_arp_send_all(bond, slave);
2664         }
2665
2666         rcu_read_unlock();
2667
2668         if (do_failover || slave_state_changed) {
2669                 if (!rtnl_trylock())
2670                         goto re_arm;
2671
2672                 bond_for_each_slave(bond, slave, iter) {
2673                         if (slave->new_link != BOND_LINK_NOCHANGE)
2674                                 slave->link = slave->new_link;
2675                 }
2676
2677                 if (slave_state_changed) {
2678                         bond_slave_state_change(bond);
2679                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2680                                 bond_update_slave_arr(bond, NULL);
2681                 }
2682                 if (do_failover) {
2683                         block_netpoll_tx();
2684                         bond_select_active_slave(bond);
2685                         unblock_netpoll_tx();
2686                 }
2687                 rtnl_unlock();
2688         }
2689
2690 re_arm:
2691         if (bond->params.arp_interval)
2692                 queue_delayed_work(bond->wq, &bond->arp_work,
2693                                    msecs_to_jiffies(bond->params.arp_interval));
2694 }
2695
2696 /* Called to inspect slaves for active-backup mode ARP monitor link state
2697  * changes.  Sets new_link in slaves to specify what action should take
2698  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2699  * to link states must be committed.
2700  *
2701  * Called with rcu_read_lock held.
2702  */
2703 static int bond_ab_arp_inspect(struct bonding *bond)
2704 {
2705         unsigned long trans_start, last_rx;
2706         struct list_head *iter;
2707         struct slave *slave;
2708         int commit = 0;
2709
2710         bond_for_each_slave_rcu(bond, slave, iter) {
2711                 slave->new_link = BOND_LINK_NOCHANGE;
2712                 last_rx = slave_last_rx(bond, slave);
2713
2714                 if (slave->link != BOND_LINK_UP) {
2715                         if (bond_time_in_interval(bond, last_rx, 1)) {
2716                                 slave->new_link = BOND_LINK_UP;
2717                                 commit++;
2718                         }
2719                         continue;
2720                 }
2721
2722                 /* Give slaves 2*delta after being enslaved or made
2723                  * active.  This avoids bouncing, as the last receive
2724                  * times need a full ARP monitor cycle to be updated.
2725                  */
2726                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2727                         continue;
2728
2729                 /* Backup slave is down if:
2730                  * - No current_arp_slave AND
2731                  * - more than 3*delta since last receive AND
2732                  * - the bond has an IP address
2733                  *
2734                  * Note: a non-null current_arp_slave indicates
2735                  * the curr_active_slave went down and we are
2736                  * searching for a new one; under this condition
2737                  * we only take the curr_active_slave down - this
2738                  * gives each slave a chance to tx/rx traffic
2739                  * before being taken out
2740                  */
2741                 if (!bond_is_active_slave(slave) &&
2742                     !rcu_access_pointer(bond->current_arp_slave) &&
2743                     !bond_time_in_interval(bond, last_rx, 3)) {
2744                         slave->new_link = BOND_LINK_DOWN;
2745                         commit++;
2746                 }
2747
2748                 /* Active slave is down if:
2749                  * - more than 2*delta since transmitting OR
2750                  * - (more than 2*delta since receive AND
2751                  *    the bond has an IP address)
2752                  */
2753                 trans_start = dev_trans_start(slave->dev);
2754                 if (bond_is_active_slave(slave) &&
2755                     (!bond_time_in_interval(bond, trans_start, 2) ||
2756                      !bond_time_in_interval(bond, last_rx, 2))) {
2757                         slave->new_link = BOND_LINK_DOWN;
2758                         commit++;
2759                 }
2760         }
2761
2762         return commit;
2763 }
2764
2765 /* Called to commit link state changes noted by inspection step of
2766  * active-backup mode ARP monitor.
2767  *
2768  * Called with RTNL hold.
2769  */
2770 static void bond_ab_arp_commit(struct bonding *bond)
2771 {
2772         unsigned long trans_start;
2773         struct list_head *iter;
2774         struct slave *slave;
2775
2776         bond_for_each_slave(bond, slave, iter) {
2777                 switch (slave->new_link) {
2778                 case BOND_LINK_NOCHANGE:
2779                         continue;
2780
2781                 case BOND_LINK_UP:
2782                         trans_start = dev_trans_start(slave->dev);
2783                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2784                             (!rtnl_dereference(bond->curr_active_slave) &&
2785                              bond_time_in_interval(bond, trans_start, 1))) {
2786                                 struct slave *current_arp_slave;
2787
2788                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2789                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2790                                                           BOND_SLAVE_NOTIFY_NOW);
2791                                 if (current_arp_slave) {
2792                                         bond_set_slave_inactive_flags(
2793                                                 current_arp_slave,
2794                                                 BOND_SLAVE_NOTIFY_NOW);
2795                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2796                                 }
2797
2798                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2799                                             slave->dev->name);
2800
2801                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2802                                     slave == rtnl_dereference(bond->primary_slave))
2803                                         goto do_failover;
2804
2805                         }
2806
2807                         continue;
2808
2809                 case BOND_LINK_DOWN:
2810                         if (slave->link_failure_count < UINT_MAX)
2811                                 slave->link_failure_count++;
2812
2813                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2814                                                   BOND_SLAVE_NOTIFY_NOW);
2815                         bond_set_slave_inactive_flags(slave,
2816                                                       BOND_SLAVE_NOTIFY_NOW);
2817
2818                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2819                                     slave->dev->name);
2820
2821                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2822                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2823                                 goto do_failover;
2824                         }
2825
2826                         continue;
2827
2828                 default:
2829                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2830                                    slave->new_link, slave->dev->name);
2831                         continue;
2832                 }
2833
2834 do_failover:
2835                 block_netpoll_tx();
2836                 bond_select_active_slave(bond);
2837                 unblock_netpoll_tx();
2838         }
2839
2840         bond_set_carrier(bond);
2841 }
2842
2843 /* Send ARP probes for active-backup mode ARP monitor.
2844  *
2845  * Called with rcu_read_lock held.
2846  */
2847 static bool bond_ab_arp_probe(struct bonding *bond)
2848 {
2849         struct slave *slave, *before = NULL, *new_slave = NULL,
2850                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2851                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2852         struct list_head *iter;
2853         bool found = false;
2854         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2855
2856         if (curr_arp_slave && curr_active_slave)
2857                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2858                             curr_arp_slave->dev->name,
2859                             curr_active_slave->dev->name);
2860
2861         if (curr_active_slave) {
2862                 bond_arp_send_all(bond, curr_active_slave);
2863                 return should_notify_rtnl;
2864         }
2865
2866         /* if we don't have a curr_active_slave, search for the next available
2867          * backup slave from the current_arp_slave and make it the candidate
2868          * for becoming the curr_active_slave
2869          */
2870
2871         if (!curr_arp_slave) {
2872                 curr_arp_slave = bond_first_slave_rcu(bond);
2873                 if (!curr_arp_slave)
2874                         return should_notify_rtnl;
2875         }
2876
2877         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2878
2879         bond_for_each_slave_rcu(bond, slave, iter) {
2880                 if (!found && !before && bond_slave_is_up(slave))
2881                         before = slave;
2882
2883                 if (found && !new_slave && bond_slave_is_up(slave))
2884                         new_slave = slave;
2885                 /* if the link state is up at this point, we
2886                  * mark it down - this can happen if we have
2887                  * simultaneous link failures and
2888                  * reselect_active_interface doesn't make this
2889                  * one the current slave so it is still marked
2890                  * up when it is actually down
2891                  */
2892                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2893                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2894                                                   BOND_SLAVE_NOTIFY_LATER);
2895                         if (slave->link_failure_count < UINT_MAX)
2896                                 slave->link_failure_count++;
2897
2898                         bond_set_slave_inactive_flags(slave,
2899                                                       BOND_SLAVE_NOTIFY_LATER);
2900
2901                         netdev_info(bond->dev, "backup interface %s is now down\n",
2902                                     slave->dev->name);
2903                 }
2904                 if (slave == curr_arp_slave)
2905                         found = true;
2906         }
2907
2908         if (!new_slave && before)
2909                 new_slave = before;
2910
2911         if (!new_slave)
2912                 goto check_state;
2913
2914         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2915                                   BOND_SLAVE_NOTIFY_LATER);
2916         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2917         bond_arp_send_all(bond, new_slave);
2918         new_slave->last_link_up = jiffies;
2919         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2920
2921 check_state:
2922         bond_for_each_slave_rcu(bond, slave, iter) {
2923                 if (slave->should_notify || slave->should_notify_link) {
2924                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2925                         break;
2926                 }
2927         }
2928         return should_notify_rtnl;
2929 }
2930
2931 static void bond_activebackup_arp_mon(struct work_struct *work)
2932 {
2933         struct bonding *bond = container_of(work, struct bonding,
2934                                             arp_work.work);
2935         bool should_notify_peers = false;
2936         bool should_notify_rtnl = false;
2937         int delta_in_ticks;
2938
2939         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2940
2941         if (!bond_has_slaves(bond))
2942                 goto re_arm;
2943
2944         rcu_read_lock();
2945
2946         should_notify_peers = bond_should_notify_peers(bond);
2947
2948         if (bond_ab_arp_inspect(bond)) {
2949                 rcu_read_unlock();
2950
2951                 /* Race avoidance with bond_close flush of workqueue */
2952                 if (!rtnl_trylock()) {
2953                         delta_in_ticks = 1;
2954                         should_notify_peers = false;
2955                         goto re_arm;
2956                 }
2957
2958                 bond_ab_arp_commit(bond);
2959
2960                 rtnl_unlock();
2961                 rcu_read_lock();
2962         }
2963
2964         should_notify_rtnl = bond_ab_arp_probe(bond);
2965         rcu_read_unlock();
2966
2967 re_arm:
2968         if (bond->params.arp_interval)
2969                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2970
2971         if (should_notify_peers || should_notify_rtnl) {
2972                 if (!rtnl_trylock())
2973                         return;
2974
2975                 if (should_notify_peers)
2976                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2977                                                  bond->dev);
2978                 if (should_notify_rtnl) {
2979                         bond_slave_state_notify(bond);
2980                         bond_slave_link_notify(bond);
2981                 }
2982
2983                 rtnl_unlock();
2984         }
2985 }
2986
2987 /*-------------------------- netdev event handling --------------------------*/
2988
2989 /* Change device name */
2990 static int bond_event_changename(struct bonding *bond)
2991 {
2992         bond_remove_proc_entry(bond);
2993         bond_create_proc_entry(bond);
2994
2995         bond_debug_reregister(bond);
2996
2997         return NOTIFY_DONE;
2998 }
2999
3000 static int bond_master_netdev_event(unsigned long event,
3001                                     struct net_device *bond_dev)
3002 {
3003         struct bonding *event_bond = netdev_priv(bond_dev);
3004
3005         switch (event) {
3006         case NETDEV_CHANGENAME:
3007                 return bond_event_changename(event_bond);
3008         case NETDEV_UNREGISTER:
3009                 bond_remove_proc_entry(event_bond);
3010                 break;
3011         case NETDEV_REGISTER:
3012                 bond_create_proc_entry(event_bond);
3013                 break;
3014         case NETDEV_NOTIFY_PEERS:
3015                 if (event_bond->send_peer_notif)
3016                         event_bond->send_peer_notif--;
3017                 break;
3018         default:
3019                 break;
3020         }
3021
3022         return NOTIFY_DONE;
3023 }
3024
3025 static int bond_slave_netdev_event(unsigned long event,
3026                                    struct net_device *slave_dev)
3027 {
3028         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3029         struct bonding *bond;
3030         struct net_device *bond_dev;
3031
3032         /* A netdev event can be generated while enslaving a device
3033          * before netdev_rx_handler_register is called in which case
3034          * slave will be NULL
3035          */
3036         if (!slave)
3037                 return NOTIFY_DONE;
3038         bond_dev = slave->bond->dev;
3039         bond = slave->bond;
3040         primary = rtnl_dereference(bond->primary_slave);
3041
3042         switch (event) {
3043         case NETDEV_UNREGISTER:
3044                 if (bond_dev->type != ARPHRD_ETHER)
3045                         bond_release_and_destroy(bond_dev, slave_dev);
3046                 else
3047                         bond_release(bond_dev, slave_dev);
3048                 break;
3049         case NETDEV_UP:
3050         case NETDEV_CHANGE:
3051                 bond_update_speed_duplex(slave);
3052                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3053                         bond_3ad_adapter_speed_duplex_changed(slave);
3054                 /* Fallthrough */
3055         case NETDEV_DOWN:
3056                 /* Refresh slave-array if applicable!
3057                  * If the setup does not use miimon or arpmon (mode-specific!),
3058                  * then these events will not cause the slave-array to be
3059                  * refreshed. This will cause xmit to use a slave that is not
3060                  * usable. Avoid such situation by refeshing the array at these
3061                  * events. If these (miimon/arpmon) parameters are configured
3062                  * then array gets refreshed twice and that should be fine!
3063                  */
3064                 if (bond_mode_uses_xmit_hash(bond))
3065                         bond_update_slave_arr(bond, NULL);
3066                 break;
3067         case NETDEV_CHANGEMTU:
3068                 /* TODO: Should slaves be allowed to
3069                  * independently alter their MTU?  For
3070                  * an active-backup bond, slaves need
3071                  * not be the same type of device, so
3072                  * MTUs may vary.  For other modes,
3073                  * slaves arguably should have the
3074                  * same MTUs. To do this, we'd need to
3075                  * take over the slave's change_mtu
3076                  * function for the duration of their
3077                  * servitude.
3078                  */
3079                 break;
3080         case NETDEV_CHANGENAME:
3081                 /* we don't care if we don't have primary set */
3082                 if (!bond_uses_primary(bond) ||
3083                     !bond->params.primary[0])
3084                         break;
3085
3086                 if (slave == primary) {
3087                         /* slave's name changed - he's no longer primary */
3088                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3089                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3090                         /* we have a new primary slave */
3091                         rcu_assign_pointer(bond->primary_slave, slave);
3092                 } else { /* we didn't change primary - exit */
3093                         break;
3094                 }
3095
3096                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3097                             primary ? slave_dev->name : "none");
3098
3099                 block_netpoll_tx();
3100                 bond_select_active_slave(bond);
3101                 unblock_netpoll_tx();
3102                 break;
3103         case NETDEV_FEAT_CHANGE:
3104                 bond_compute_features(bond);
3105                 break;
3106         case NETDEV_RESEND_IGMP:
3107                 /* Propagate to master device */
3108                 call_netdevice_notifiers(event, slave->bond->dev);
3109                 break;
3110         default:
3111                 break;
3112         }
3113
3114         return NOTIFY_DONE;
3115 }
3116
3117 /* bond_netdev_event: handle netdev notifier chain events.
3118  *
3119  * This function receives events for the netdev chain.  The caller (an
3120  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3121  * locks for us to safely manipulate the slave devices (RTNL lock,
3122  * dev_probe_lock).
3123  */
3124 static int bond_netdev_event(struct notifier_block *this,
3125                              unsigned long event, void *ptr)
3126 {
3127         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3128
3129         netdev_dbg(event_dev, "event: %lx\n", event);
3130
3131         if (!(event_dev->priv_flags & IFF_BONDING))
3132                 return NOTIFY_DONE;
3133
3134         if (event_dev->flags & IFF_MASTER) {
3135                 netdev_dbg(event_dev, "IFF_MASTER\n");
3136                 return bond_master_netdev_event(event, event_dev);
3137         }
3138
3139         if (event_dev->flags & IFF_SLAVE) {
3140                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3141                 return bond_slave_netdev_event(event, event_dev);
3142         }
3143
3144         return NOTIFY_DONE;
3145 }
3146
3147 static struct notifier_block bond_netdev_notifier = {
3148         .notifier_call = bond_netdev_event,
3149 };
3150
3151 /*---------------------------- Hashing Policies -----------------------------*/
3152
3153 /* L2 hash helper */
3154 static inline u32 bond_eth_hash(struct sk_buff *skb)
3155 {
3156         struct ethhdr *ep, hdr_tmp;
3157
3158         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3159         if (ep)
3160                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3161         return 0;
3162 }
3163
3164 /* Extract the appropriate headers based on bond's xmit policy */
3165 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3166                               struct flow_keys *fk)
3167 {
3168         const struct ipv6hdr *iph6;
3169         const struct iphdr *iph;
3170         int noff, proto = -1;
3171
3172         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3173                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3174
3175         fk->ports.ports = 0;
3176         noff = skb_network_offset(skb);
3177         if (skb->protocol == htons(ETH_P_IP)) {
3178                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3179                         return false;
3180                 iph = ip_hdr(skb);
3181                 iph_to_flow_copy_v4addrs(fk, iph);
3182                 noff += iph->ihl << 2;
3183                 if (!ip_is_fragment(iph))
3184                         proto = iph->protocol;
3185         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3186                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3187                         return false;
3188                 iph6 = ipv6_hdr(skb);
3189                 iph_to_flow_copy_v6addrs(fk, iph6);
3190                 noff += sizeof(*iph6);
3191                 proto = iph6->nexthdr;
3192         } else {
3193                 return false;
3194         }
3195         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3196                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3197
3198         return true;
3199 }
3200
3201 /**
3202  * bond_xmit_hash - generate a hash value based on the xmit policy
3203  * @bond: bonding device
3204  * @skb: buffer to use for headers
3205  *
3206  * This function will extract the necessary headers from the skb buffer and use
3207  * them to generate a hash based on the xmit_policy set in the bonding device
3208  */
3209 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3210 {
3211         struct flow_keys flow;
3212         u32 hash;
3213
3214         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3215             skb->l4_hash)
3216                 return skb->hash;
3217
3218         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3219             !bond_flow_dissect(bond, skb, &flow))
3220                 return bond_eth_hash(skb);
3221
3222         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3223             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3224                 hash = bond_eth_hash(skb);
3225         else
3226                 hash = (__force u32)flow.ports.ports;
3227         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3228                 (__force u32)flow_get_u32_src(&flow);
3229         hash ^= (hash >> 16);
3230         hash ^= (hash >> 8);
3231
3232         return hash >> 1;
3233 }
3234
3235 /*-------------------------- Device entry points ----------------------------*/
3236
3237 static void bond_work_init_all(struct bonding *bond)
3238 {
3239         INIT_DELAYED_WORK(&bond->mcast_work,
3240                           bond_resend_igmp_join_requests_delayed);
3241         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3242         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3243         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3244                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3245         else
3246                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3247         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3248         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3249 }
3250
3251 static void bond_work_cancel_all(struct bonding *bond)
3252 {
3253         cancel_delayed_work_sync(&bond->mii_work);
3254         cancel_delayed_work_sync(&bond->arp_work);
3255         cancel_delayed_work_sync(&bond->alb_work);
3256         cancel_delayed_work_sync(&bond->ad_work);
3257         cancel_delayed_work_sync(&bond->mcast_work);
3258         cancel_delayed_work_sync(&bond->slave_arr_work);
3259 }
3260
3261 static int bond_open(struct net_device *bond_dev)
3262 {
3263         struct bonding *bond = netdev_priv(bond_dev);
3264         struct list_head *iter;
3265         struct slave *slave;
3266
3267         /* reset slave->backup and slave->inactive */
3268         if (bond_has_slaves(bond)) {
3269                 bond_for_each_slave(bond, slave, iter) {
3270                         if (bond_uses_primary(bond) &&
3271                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3272                                 bond_set_slave_inactive_flags(slave,
3273                                                               BOND_SLAVE_NOTIFY_NOW);
3274                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3275                                 bond_set_slave_active_flags(slave,
3276                                                             BOND_SLAVE_NOTIFY_NOW);
3277                         }
3278                 }
3279         }
3280
3281         bond_work_init_all(bond);
3282
3283         if (bond_is_lb(bond)) {
3284                 /* bond_alb_initialize must be called before the timer
3285                  * is started.
3286                  */
3287                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3288                         return -ENOMEM;
3289                 if (bond->params.tlb_dynamic_lb)
3290                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3291         }
3292
3293         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3294                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3295
3296         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3297                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3298                 bond->recv_probe = bond_arp_rcv;
3299         }
3300
3301         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3302                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3303                 /* register to receive LACPDUs */
3304                 bond->recv_probe = bond_3ad_lacpdu_recv;
3305                 bond_3ad_initiate_agg_selection(bond, 1);
3306         }
3307
3308         if (bond_mode_uses_xmit_hash(bond))
3309                 bond_update_slave_arr(bond, NULL);
3310
3311         return 0;
3312 }
3313
3314 static int bond_close(struct net_device *bond_dev)
3315 {
3316         struct bonding *bond = netdev_priv(bond_dev);
3317
3318         bond_work_cancel_all(bond);
3319         bond->send_peer_notif = 0;
3320         if (bond_is_lb(bond))
3321                 bond_alb_deinitialize(bond);
3322         bond->recv_probe = NULL;
3323
3324         return 0;
3325 }
3326
3327 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3328  * that some drivers can provide 32bit values only.
3329  */
3330 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3331                             const struct rtnl_link_stats64 *_new,
3332                             const struct rtnl_link_stats64 *_old)
3333 {
3334         const u64 *new = (const u64 *)_new;
3335         const u64 *old = (const u64 *)_old;
3336         u64 *res = (u64 *)_res;
3337         int i;
3338
3339         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3340                 u64 nv = new[i];
3341                 u64 ov = old[i];
3342                 s64 delta = nv - ov;
3343
3344                 /* detects if this particular field is 32bit only */
3345                 if (((nv | ov) >> 32) == 0)
3346                         delta = (s64)(s32)((u32)nv - (u32)ov);
3347
3348                 /* filter anomalies, some drivers reset their stats
3349                  * at down/up events.
3350                  */
3351                 if (delta > 0)
3352                         res[i] += delta;
3353         }
3354 }
3355
3356 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3357                                                 struct rtnl_link_stats64 *stats)
3358 {
3359         struct bonding *bond = netdev_priv(bond_dev);
3360         struct rtnl_link_stats64 temp;
3361         struct list_head *iter;
3362         struct slave *slave;
3363
3364         spin_lock(&bond->stats_lock);
3365         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3366
3367         rcu_read_lock();
3368         bond_for_each_slave_rcu(bond, slave, iter) {
3369                 const struct rtnl_link_stats64 *new =
3370                         dev_get_stats(slave->dev, &temp);
3371
3372                 bond_fold_stats(stats, new, &slave->slave_stats);
3373
3374                 /* save off the slave stats for the next run */
3375                 memcpy(&slave->slave_stats, new, sizeof(*new));
3376         }
3377         rcu_read_unlock();
3378
3379         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3380         spin_unlock(&bond->stats_lock);
3381
3382         return stats;
3383 }
3384
3385 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3386 {
3387         struct bonding *bond = netdev_priv(bond_dev);
3388         struct net_device *slave_dev = NULL;
3389         struct ifbond k_binfo;
3390         struct ifbond __user *u_binfo = NULL;
3391         struct ifslave k_sinfo;
3392         struct ifslave __user *u_sinfo = NULL;
3393         struct mii_ioctl_data *mii = NULL;
3394         struct bond_opt_value newval;
3395         struct net *net;
3396         int res = 0;
3397
3398         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3399
3400         switch (cmd) {
3401         case SIOCGMIIPHY:
3402                 mii = if_mii(ifr);
3403                 if (!mii)
3404                         return -EINVAL;
3405
3406                 mii->phy_id = 0;
3407                 /* Fall Through */
3408         case SIOCGMIIREG:
3409                 /* We do this again just in case we were called by SIOCGMIIREG
3410                  * instead of SIOCGMIIPHY.
3411                  */
3412                 mii = if_mii(ifr);
3413                 if (!mii)
3414                         return -EINVAL;
3415
3416                 if (mii->reg_num == 1) {
3417                         mii->val_out = 0;
3418                         if (netif_carrier_ok(bond->dev))
3419                                 mii->val_out = BMSR_LSTATUS;
3420                 }
3421
3422                 return 0;
3423         case BOND_INFO_QUERY_OLD:
3424         case SIOCBONDINFOQUERY:
3425                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3426
3427                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3428                         return -EFAULT;
3429
3430                 res = bond_info_query(bond_dev, &k_binfo);
3431                 if (res == 0 &&
3432                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3433                         return -EFAULT;
3434
3435                 return res;
3436         case BOND_SLAVE_INFO_QUERY_OLD:
3437         case SIOCBONDSLAVEINFOQUERY:
3438                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3439
3440                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3441                         return -EFAULT;
3442
3443                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3444                 if (res == 0 &&
3445                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3446                         return -EFAULT;
3447
3448                 return res;
3449         default:
3450                 break;
3451         }
3452
3453         net = dev_net(bond_dev);
3454
3455         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3456                 return -EPERM;
3457
3458         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3459
3460         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3461
3462         if (!slave_dev)
3463                 return -ENODEV;
3464
3465         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3466         switch (cmd) {
3467         case BOND_ENSLAVE_OLD:
3468         case SIOCBONDENSLAVE:
3469                 res = bond_enslave(bond_dev, slave_dev);
3470                 break;
3471         case BOND_RELEASE_OLD:
3472         case SIOCBONDRELEASE:
3473                 res = bond_release(bond_dev, slave_dev);
3474                 break;
3475         case BOND_SETHWADDR_OLD:
3476         case SIOCBONDSETHWADDR:
3477                 bond_set_dev_addr(bond_dev, slave_dev);
3478                 res = 0;
3479                 break;
3480         case BOND_CHANGE_ACTIVE_OLD:
3481         case SIOCBONDCHANGEACTIVE:
3482                 bond_opt_initstr(&newval, slave_dev->name);
3483                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3484                 break;
3485         default:
3486                 res = -EOPNOTSUPP;
3487         }
3488
3489         return res;
3490 }
3491
3492 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3493 {
3494         struct bonding *bond = netdev_priv(bond_dev);
3495
3496         if (change & IFF_PROMISC)
3497                 bond_set_promiscuity(bond,
3498                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3499
3500         if (change & IFF_ALLMULTI)
3501                 bond_set_allmulti(bond,
3502                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3503 }
3504
3505 static void bond_set_rx_mode(struct net_device *bond_dev)
3506 {
3507         struct bonding *bond = netdev_priv(bond_dev);
3508         struct list_head *iter;
3509         struct slave *slave;
3510
3511         rcu_read_lock();
3512         if (bond_uses_primary(bond)) {
3513                 slave = rcu_dereference(bond->curr_active_slave);
3514                 if (slave) {
3515                         dev_uc_sync(slave->dev, bond_dev);
3516                         dev_mc_sync(slave->dev, bond_dev);
3517                 }
3518         } else {
3519                 bond_for_each_slave_rcu(bond, slave, iter) {
3520                         dev_uc_sync_multiple(slave->dev, bond_dev);
3521                         dev_mc_sync_multiple(slave->dev, bond_dev);
3522                 }
3523         }
3524         rcu_read_unlock();
3525 }
3526
3527 static int bond_neigh_init(struct neighbour *n)
3528 {
3529         struct bonding *bond = netdev_priv(n->dev);
3530         const struct net_device_ops *slave_ops;
3531         struct neigh_parms parms;
3532         struct slave *slave;
3533         int ret;
3534
3535         slave = bond_first_slave(bond);
3536         if (!slave)
3537                 return 0;
3538         slave_ops = slave->dev->netdev_ops;
3539         if (!slave_ops->ndo_neigh_setup)
3540                 return 0;
3541
3542         parms.neigh_setup = NULL;
3543         parms.neigh_cleanup = NULL;
3544         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3545         if (ret)
3546                 return ret;
3547
3548         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3549          * after the last slave has been detached.  Assumes that all slaves
3550          * utilize the same neigh_cleanup (true at this writing as only user
3551          * is ipoib).
3552          */
3553         n->parms->neigh_cleanup = parms.neigh_cleanup;
3554
3555         if (!parms.neigh_setup)
3556                 return 0;
3557
3558         return parms.neigh_setup(n);
3559 }
3560
3561 /* The bonding ndo_neigh_setup is called at init time beofre any
3562  * slave exists. So we must declare proxy setup function which will
3563  * be used at run time to resolve the actual slave neigh param setup.
3564  *
3565  * It's also called by master devices (such as vlans) to setup their
3566  * underlying devices. In that case - do nothing, we're already set up from
3567  * our init.
3568  */
3569 static int bond_neigh_setup(struct net_device *dev,
3570                             struct neigh_parms *parms)
3571 {
3572         /* modify only our neigh_parms */
3573         if (parms->dev == dev)
3574                 parms->neigh_setup = bond_neigh_init;
3575
3576         return 0;
3577 }
3578
3579 /* Change the MTU of all of a master's slaves to match the master */
3580 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3581 {
3582         struct bonding *bond = netdev_priv(bond_dev);
3583         struct slave *slave, *rollback_slave;
3584         struct list_head *iter;
3585         int res = 0;
3586
3587         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3588
3589         bond_for_each_slave(bond, slave, iter) {
3590                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3591                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3592
3593                 res = dev_set_mtu(slave->dev, new_mtu);
3594
3595                 if (res) {
3596                         /* If we failed to set the slave's mtu to the new value
3597                          * we must abort the operation even in ACTIVE_BACKUP
3598                          * mode, because if we allow the backup slaves to have
3599                          * different mtu values than the active slave we'll
3600                          * need to change their mtu when doing a failover. That
3601                          * means changing their mtu from timer context, which
3602                          * is probably not a good idea.
3603                          */
3604                         netdev_dbg(bond_dev, "err %d %s\n", res,
3605                                    slave->dev->name);
3606                         goto unwind;
3607                 }
3608         }
3609
3610         bond_dev->mtu = new_mtu;
3611
3612         return 0;
3613
3614 unwind:
3615         /* unwind from head to the slave that failed */
3616         bond_for_each_slave(bond, rollback_slave, iter) {
3617                 int tmp_res;
3618
3619                 if (rollback_slave == slave)
3620                         break;
3621
3622                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3623                 if (tmp_res) {
3624                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3625                                    tmp_res, rollback_slave->dev->name);
3626                 }
3627         }
3628
3629         return res;
3630 }
3631
3632 /* Change HW address
3633  *
3634  * Note that many devices must be down to change the HW address, and
3635  * downing the master releases all slaves.  We can make bonds full of
3636  * bonding devices to test this, however.
3637  */
3638 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3639 {
3640         struct bonding *bond = netdev_priv(bond_dev);
3641         struct slave *slave, *rollback_slave;
3642         struct sockaddr *sa = addr, tmp_sa;
3643         struct list_head *iter;
3644         int res = 0;
3645
3646         if (BOND_MODE(bond) == BOND_MODE_ALB)
3647                 return bond_alb_set_mac_address(bond_dev, addr);
3648
3649
3650         netdev_dbg(bond_dev, "bond=%p\n", bond);
3651
3652         /* If fail_over_mac is enabled, do nothing and return success.
3653          * Returning an error causes ifenslave to fail.
3654          */
3655         if (bond->params.fail_over_mac &&
3656             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3657                 return 0;
3658
3659         if (!is_valid_ether_addr(sa->sa_data))
3660                 return -EADDRNOTAVAIL;
3661
3662         bond_for_each_slave(bond, slave, iter) {
3663                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3664                 res = dev_set_mac_address(slave->dev, addr);
3665                 if (res) {
3666                         /* TODO: consider downing the slave
3667                          * and retry ?
3668                          * User should expect communications
3669                          * breakage anyway until ARP finish
3670                          * updating, so...
3671                          */
3672                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3673                         goto unwind;
3674                 }
3675         }
3676
3677         /* success */
3678         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3679         return 0;
3680
3681 unwind:
3682         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3683         tmp_sa.sa_family = bond_dev->type;
3684
3685         /* unwind from head to the slave that failed */
3686         bond_for_each_slave(bond, rollback_slave, iter) {
3687                 int tmp_res;
3688
3689                 if (rollback_slave == slave)
3690                         break;
3691
3692                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3693                 if (tmp_res) {
3694                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3695                                    tmp_res, rollback_slave->dev->name);
3696                 }
3697         }
3698
3699         return res;
3700 }
3701
3702 /**
3703  * bond_xmit_slave_id - transmit skb through slave with slave_id
3704  * @bond: bonding device that is transmitting
3705  * @skb: buffer to transmit
3706  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3707  *
3708  * This function tries to transmit through slave with slave_id but in case
3709  * it fails, it tries to find the first available slave for transmission.
3710  * The skb is consumed in all cases, thus the function is void.
3711  */
3712 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3713 {
3714         struct list_head *iter;
3715         struct slave *slave;
3716         int i = slave_id;
3717
3718         /* Here we start from the slave with slave_id */
3719         bond_for_each_slave_rcu(bond, slave, iter) {
3720                 if (--i < 0) {
3721                         if (bond_slave_can_tx(slave)) {
3722                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3723                                 return;
3724                         }
3725                 }
3726         }
3727
3728         /* Here we start from the first slave up to slave_id */
3729         i = slave_id;
3730         bond_for_each_slave_rcu(bond, slave, iter) {
3731                 if (--i < 0)
3732                         break;
3733                 if (bond_slave_can_tx(slave)) {
3734                         bond_dev_queue_xmit(bond, skb, slave->dev);
3735                         return;
3736                 }
3737         }
3738         /* no slave that can tx has been found */
3739         bond_tx_drop(bond->dev, skb);
3740 }
3741
3742 /**
3743  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3744  * @bond: bonding device to use
3745  *
3746  * Based on the value of the bonding device's packets_per_slave parameter
3747  * this function generates a slave id, which is usually used as the next
3748  * slave to transmit through.
3749  */
3750 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3751 {
3752         u32 slave_id;
3753         struct reciprocal_value reciprocal_packets_per_slave;
3754         int packets_per_slave = bond->params.packets_per_slave;
3755
3756         switch (packets_per_slave) {
3757         case 0:
3758                 slave_id = prandom_u32();
3759                 break;
3760         case 1:
3761                 slave_id = bond->rr_tx_counter;
3762                 break;
3763         default:
3764                 reciprocal_packets_per_slave =
3765                         bond->params.reciprocal_packets_per_slave;
3766                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3767                                              reciprocal_packets_per_slave);
3768                 break;
3769         }
3770         bond->rr_tx_counter++;
3771
3772         return slave_id;
3773 }
3774
3775 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3776 {
3777         struct bonding *bond = netdev_priv(bond_dev);
3778         struct iphdr *iph = ip_hdr(skb);
3779         struct slave *slave;
3780         u32 slave_id;
3781
3782         /* Start with the curr_active_slave that joined the bond as the
3783          * default for sending IGMP traffic.  For failover purposes one
3784          * needs to maintain some consistency for the interface that will
3785          * send the join/membership reports.  The curr_active_slave found
3786          * will send all of this type of traffic.
3787          */
3788         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3789                 slave = rcu_dereference(bond->curr_active_slave);
3790                 if (slave)
3791                         bond_dev_queue_xmit(bond, skb, slave->dev);
3792                 else
3793                         bond_xmit_slave_id(bond, skb, 0);
3794         } else {
3795                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3796
3797                 if (likely(slave_cnt)) {
3798                         slave_id = bond_rr_gen_slave_id(bond);
3799                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3800                 } else {
3801                         bond_tx_drop(bond_dev, skb);
3802                 }
3803         }
3804
3805         return NETDEV_TX_OK;
3806 }
3807
3808 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3809  * the bond has a usable interface.
3810  */
3811 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3812 {
3813         struct bonding *bond = netdev_priv(bond_dev);
3814         struct slave *slave;
3815
3816         slave = rcu_dereference(bond->curr_active_slave);
3817         if (slave)
3818                 bond_dev_queue_xmit(bond, skb, slave->dev);
3819         else
3820                 bond_tx_drop(bond_dev, skb);
3821
3822         return NETDEV_TX_OK;
3823 }
3824
3825 /* Use this to update slave_array when (a) it's not appropriate to update
3826  * slave_array right away (note that update_slave_array() may sleep)
3827  * and / or (b) RTNL is not held.
3828  */
3829 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3830 {
3831         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3832 }
3833
3834 /* Slave array work handler. Holds only RTNL */
3835 static void bond_slave_arr_handler(struct work_struct *work)
3836 {
3837         struct bonding *bond = container_of(work, struct bonding,
3838                                             slave_arr_work.work);
3839         int ret;
3840
3841         if (!rtnl_trylock())
3842                 goto err;
3843
3844         ret = bond_update_slave_arr(bond, NULL);
3845         rtnl_unlock();
3846         if (ret) {
3847                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3848                 goto err;
3849         }
3850         return;
3851
3852 err:
3853         bond_slave_arr_work_rearm(bond, 1);
3854 }
3855
3856 /* Build the usable slaves array in control path for modes that use xmit-hash
3857  * to determine the slave interface -
3858  * (a) BOND_MODE_8023AD
3859  * (b) BOND_MODE_XOR
3860  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3861  *
3862  * The caller is expected to hold RTNL only and NO other lock!
3863  */
3864 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3865 {
3866         struct slave *slave;
3867         struct list_head *iter;
3868         struct bond_up_slave *new_arr, *old_arr;
3869         int agg_id = 0;
3870         int ret = 0;
3871
3872 #ifdef CONFIG_LOCKDEP
3873         WARN_ON(lockdep_is_held(&bond->mode_lock));
3874 #endif
3875
3876         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3877                           GFP_KERNEL);
3878         if (!new_arr) {
3879                 ret = -ENOMEM;
3880                 pr_err("Failed to build slave-array.\n");
3881                 goto out;
3882         }
3883         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3884                 struct ad_info ad_info;
3885
3886                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3887                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3888                         kfree_rcu(new_arr, rcu);
3889                         /* No active aggragator means it's not safe to use
3890                          * the previous array.
3891                          */
3892                         old_arr = rtnl_dereference(bond->slave_arr);
3893                         if (old_arr) {
3894                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3895                                 kfree_rcu(old_arr, rcu);
3896                         }
3897                         goto out;
3898                 }
3899                 agg_id = ad_info.aggregator_id;
3900         }
3901         bond_for_each_slave(bond, slave, iter) {
3902                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3903                         struct aggregator *agg;
3904
3905                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3906                         if (!agg || agg->aggregator_identifier != agg_id)
3907                                 continue;
3908                 }
3909                 if (!bond_slave_can_tx(slave))
3910                         continue;
3911                 if (skipslave == slave)
3912                         continue;
3913                 new_arr->arr[new_arr->count++] = slave;
3914         }
3915
3916         old_arr = rtnl_dereference(bond->slave_arr);
3917         rcu_assign_pointer(bond->slave_arr, new_arr);
3918         if (old_arr)
3919                 kfree_rcu(old_arr, rcu);
3920 out:
3921         if (ret != 0 && skipslave) {
3922                 int idx;
3923
3924                 /* Rare situation where caller has asked to skip a specific
3925                  * slave but allocation failed (most likely!). BTW this is
3926                  * only possible when the call is initiated from
3927                  * __bond_release_one(). In this situation; overwrite the
3928                  * skipslave entry in the array with the last entry from the
3929                  * array to avoid a situation where the xmit path may choose
3930                  * this to-be-skipped slave to send a packet out.
3931                  */
3932                 old_arr = rtnl_dereference(bond->slave_arr);
3933                 for (idx = 0; idx < old_arr->count; idx++) {
3934                         if (skipslave == old_arr->arr[idx]) {
3935                                 old_arr->arr[idx] =
3936                                     old_arr->arr[old_arr->count-1];
3937                                 old_arr->count--;
3938                                 break;
3939                         }
3940                 }
3941         }
3942         return ret;
3943 }
3944
3945 /* Use this Xmit function for 3AD as well as XOR modes. The current
3946  * usable slave array is formed in the control path. The xmit function
3947  * just calculates hash and sends the packet out.
3948  */
3949 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3950 {
3951         struct bonding *bond = netdev_priv(dev);
3952         struct slave *slave;
3953         struct bond_up_slave *slaves;
3954         unsigned int count;
3955
3956         slaves = rcu_dereference(bond->slave_arr);
3957         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3958         if (likely(count)) {
3959                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3960                 bond_dev_queue_xmit(bond, skb, slave->dev);
3961         } else {
3962                 bond_tx_drop(dev, skb);
3963         }
3964
3965         return NETDEV_TX_OK;
3966 }
3967
3968 /* in broadcast mode, we send everything to all usable interfaces. */
3969 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3970 {
3971         struct bonding *bond = netdev_priv(bond_dev);
3972         struct slave *slave = NULL;
3973         struct list_head *iter;
3974
3975         bond_for_each_slave_rcu(bond, slave, iter) {
3976                 if (bond_is_last_slave(bond, slave))
3977                         break;
3978                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3979                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3980
3981                         if (!skb2) {
3982                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3983                                                     bond_dev->name, __func__);
3984                                 continue;
3985                         }
3986                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3987                 }
3988         }
3989         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3990                 bond_dev_queue_xmit(bond, skb, slave->dev);
3991         else
3992                 bond_tx_drop(bond_dev, skb);
3993
3994         return NETDEV_TX_OK;
3995 }
3996
3997 /*------------------------- Device initialization ---------------------------*/
3998
3999 /* Lookup the slave that corresponds to a qid */
4000 static inline int bond_slave_override(struct bonding *bond,
4001                                       struct sk_buff *skb)
4002 {
4003         struct slave *slave = NULL;
4004         struct list_head *iter;
4005
4006         if (!skb->queue_mapping)
4007                 return 1;
4008
4009         /* Find out if any slaves have the same mapping as this skb. */
4010         bond_for_each_slave_rcu(bond, slave, iter) {
4011                 if (slave->queue_id == skb->queue_mapping) {
4012                         if (bond_slave_is_up(slave) &&
4013                             slave->link == BOND_LINK_UP) {
4014                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4015                                 return 0;
4016                         }
4017                         /* If the slave isn't UP, use default transmit policy. */
4018                         break;
4019                 }
4020         }
4021
4022         return 1;
4023 }
4024
4025
4026 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4027                              void *accel_priv, select_queue_fallback_t fallback)
4028 {
4029         /* This helper function exists to help dev_pick_tx get the correct
4030          * destination queue.  Using a helper function skips a call to
4031          * skb_tx_hash and will put the skbs in the queue we expect on their
4032          * way down to the bonding driver.
4033          */
4034         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4035
4036         /* Save the original txq to restore before passing to the driver */
4037         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4038
4039         if (unlikely(txq >= dev->real_num_tx_queues)) {
4040                 do {
4041                         txq -= dev->real_num_tx_queues;
4042                 } while (txq >= dev->real_num_tx_queues);
4043         }
4044         return txq;
4045 }
4046
4047 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4048 {
4049         struct bonding *bond = netdev_priv(dev);
4050
4051         if (bond_should_override_tx_queue(bond) &&
4052             !bond_slave_override(bond, skb))
4053                 return NETDEV_TX_OK;
4054
4055         switch (BOND_MODE(bond)) {
4056         case BOND_MODE_ROUNDROBIN:
4057                 return bond_xmit_roundrobin(skb, dev);
4058         case BOND_MODE_ACTIVEBACKUP:
4059                 return bond_xmit_activebackup(skb, dev);
4060         case BOND_MODE_8023AD:
4061         case BOND_MODE_XOR:
4062                 return bond_3ad_xor_xmit(skb, dev);
4063         case BOND_MODE_BROADCAST:
4064                 return bond_xmit_broadcast(skb, dev);
4065         case BOND_MODE_ALB:
4066                 return bond_alb_xmit(skb, dev);
4067         case BOND_MODE_TLB:
4068                 return bond_tlb_xmit(skb, dev);
4069         default:
4070                 /* Should never happen, mode already checked */
4071                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4072                 WARN_ON_ONCE(1);
4073                 bond_tx_drop(dev, skb);
4074                 return NETDEV_TX_OK;
4075         }
4076 }
4077
4078 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4079 {
4080         struct bonding *bond = netdev_priv(dev);
4081         netdev_tx_t ret = NETDEV_TX_OK;
4082
4083         /* If we risk deadlock from transmitting this in the
4084          * netpoll path, tell netpoll to queue the frame for later tx
4085          */
4086         if (unlikely(is_netpoll_tx_blocked(dev)))
4087                 return NETDEV_TX_BUSY;
4088
4089         rcu_read_lock();
4090         if (bond_has_slaves(bond))
4091                 ret = __bond_start_xmit(skb, dev);
4092         else
4093                 bond_tx_drop(dev, skb);
4094         rcu_read_unlock();
4095
4096         return ret;
4097 }
4098
4099 static int bond_ethtool_get_settings(struct net_device *bond_dev,
4100                                      struct ethtool_cmd *ecmd)
4101 {
4102         struct bonding *bond = netdev_priv(bond_dev);
4103         unsigned long speed = 0;
4104         struct list_head *iter;
4105         struct slave *slave;
4106
4107         ecmd->duplex = DUPLEX_UNKNOWN;
4108         ecmd->port = PORT_OTHER;
4109
4110         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4111          * do not need to check mode.  Though link speed might not represent
4112          * the true receive or transmit bandwidth (not all modes are symmetric)
4113          * this is an accurate maximum.
4114          */
4115         bond_for_each_slave(bond, slave, iter) {
4116                 if (bond_slave_can_tx(slave)) {
4117                         if (slave->speed != SPEED_UNKNOWN)
4118                                 speed += slave->speed;
4119                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4120                             slave->duplex != DUPLEX_UNKNOWN)
4121                                 ecmd->duplex = slave->duplex;
4122                 }
4123         }
4124         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4125
4126         return 0;
4127 }
4128
4129 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4130                                      struct ethtool_drvinfo *drvinfo)
4131 {
4132         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4133         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4134         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4135                  BOND_ABI_VERSION);
4136 }
4137
4138 static const struct ethtool_ops bond_ethtool_ops = {
4139         .get_drvinfo            = bond_ethtool_get_drvinfo,
4140         .get_settings           = bond_ethtool_get_settings,
4141         .get_link               = ethtool_op_get_link,
4142 };
4143
4144 static const struct net_device_ops bond_netdev_ops = {
4145         .ndo_init               = bond_init,
4146         .ndo_uninit             = bond_uninit,
4147         .ndo_open               = bond_open,
4148         .ndo_stop               = bond_close,
4149         .ndo_start_xmit         = bond_start_xmit,
4150         .ndo_select_queue       = bond_select_queue,
4151         .ndo_get_stats64        = bond_get_stats,
4152         .ndo_do_ioctl           = bond_do_ioctl,
4153         .ndo_change_rx_flags    = bond_change_rx_flags,
4154         .ndo_set_rx_mode        = bond_set_rx_mode,
4155         .ndo_change_mtu         = bond_change_mtu,
4156         .ndo_set_mac_address    = bond_set_mac_address,
4157         .ndo_neigh_setup        = bond_neigh_setup,
4158         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4159         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4160 #ifdef CONFIG_NET_POLL_CONTROLLER
4161         .ndo_netpoll_setup      = bond_netpoll_setup,
4162         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4163         .ndo_poll_controller    = bond_poll_controller,
4164 #endif
4165         .ndo_add_slave          = bond_enslave,
4166         .ndo_del_slave          = bond_release,
4167         .ndo_fix_features       = bond_fix_features,
4168         .ndo_neigh_construct    = netdev_default_l2upper_neigh_construct,
4169         .ndo_neigh_destroy      = netdev_default_l2upper_neigh_destroy,
4170         .ndo_bridge_setlink     = switchdev_port_bridge_setlink,
4171         .ndo_bridge_getlink     = switchdev_port_bridge_getlink,
4172         .ndo_bridge_dellink     = switchdev_port_bridge_dellink,
4173         .ndo_fdb_add            = switchdev_port_fdb_add,
4174         .ndo_fdb_del            = switchdev_port_fdb_del,
4175         .ndo_fdb_dump           = switchdev_port_fdb_dump,
4176         .ndo_features_check     = passthru_features_check,
4177 };
4178
4179 static const struct device_type bond_type = {
4180         .name = "bond",
4181 };
4182
4183 static void bond_destructor(struct net_device *bond_dev)
4184 {
4185         struct bonding *bond = netdev_priv(bond_dev);
4186         if (bond->wq)
4187                 destroy_workqueue(bond->wq);
4188         free_netdev(bond_dev);
4189 }
4190
4191 void bond_setup(struct net_device *bond_dev)
4192 {
4193         struct bonding *bond = netdev_priv(bond_dev);
4194
4195         spin_lock_init(&bond->mode_lock);
4196         spin_lock_init(&bond->stats_lock);
4197         bond->params = bonding_defaults;
4198
4199         /* Initialize pointers */
4200         bond->dev = bond_dev;
4201
4202         /* Initialize the device entry points */
4203         ether_setup(bond_dev);
4204         bond_dev->netdev_ops = &bond_netdev_ops;
4205         bond_dev->ethtool_ops = &bond_ethtool_ops;
4206
4207         bond_dev->destructor = bond_destructor;
4208
4209         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4210
4211         /* Initialize the device options */
4212         bond_dev->flags |= IFF_MASTER;
4213         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4214         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4215
4216         /* don't acquire bond device's netif_tx_lock when transmitting */
4217         bond_dev->features |= NETIF_F_LLTX;
4218
4219         /* By default, we declare the bond to be fully
4220          * VLAN hardware accelerated capable. Special
4221          * care is taken in the various xmit functions
4222          * when there are slaves that are not hw accel
4223          * capable
4224          */
4225
4226         /* Don't allow bond devices to change network namespaces. */
4227         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4228
4229         bond_dev->hw_features = BOND_VLAN_FEATURES |
4230                                 NETIF_F_HW_VLAN_CTAG_TX |
4231                                 NETIF_F_HW_VLAN_CTAG_RX |
4232                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4233
4234         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4235         bond_dev->features |= bond_dev->hw_features;
4236 }
4237
4238 /* Destroy a bonding device.
4239  * Must be under rtnl_lock when this function is called.
4240  */
4241 static void bond_uninit(struct net_device *bond_dev)
4242 {
4243         struct bonding *bond = netdev_priv(bond_dev);
4244         struct list_head *iter;
4245         struct slave *slave;
4246         struct bond_up_slave *arr;
4247
4248         bond_netpoll_cleanup(bond_dev);
4249
4250         /* Release the bonded slaves */
4251         bond_for_each_slave(bond, slave, iter)
4252                 __bond_release_one(bond_dev, slave->dev, true);
4253         netdev_info(bond_dev, "Released all slaves\n");
4254
4255         arr = rtnl_dereference(bond->slave_arr);
4256         if (arr) {
4257                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4258                 kfree_rcu(arr, rcu);
4259         }
4260
4261         list_del(&bond->bond_list);
4262
4263         bond_debug_unregister(bond);
4264 }
4265
4266 /*------------------------- Module initialization ---------------------------*/
4267
4268 static int bond_check_params(struct bond_params *params)
4269 {
4270         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4271         struct bond_opt_value newval;
4272         const struct bond_opt_value *valptr;
4273         int arp_all_targets_value;
4274         u16 ad_actor_sys_prio = 0;
4275         u16 ad_user_port_key = 0;
4276
4277         /* Convert string parameters. */
4278         if (mode) {
4279                 bond_opt_initstr(&newval, mode);
4280                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4281                 if (!valptr) {
4282                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4283                         return -EINVAL;
4284                 }
4285                 bond_mode = valptr->value;
4286         }
4287
4288         if (xmit_hash_policy) {
4289                 if ((bond_mode != BOND_MODE_XOR) &&
4290                     (bond_mode != BOND_MODE_8023AD) &&
4291                     (bond_mode != BOND_MODE_TLB)) {
4292                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4293                                 bond_mode_name(bond_mode));
4294                 } else {
4295                         bond_opt_initstr(&newval, xmit_hash_policy);
4296                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4297                                                 &newval);
4298                         if (!valptr) {
4299                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4300                                        xmit_hash_policy);
4301                                 return -EINVAL;
4302                         }
4303                         xmit_hashtype = valptr->value;
4304                 }
4305         }
4306
4307         if (lacp_rate) {
4308                 if (bond_mode != BOND_MODE_8023AD) {
4309                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4310                                 bond_mode_name(bond_mode));
4311                 } else {
4312                         bond_opt_initstr(&newval, lacp_rate);
4313                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4314                                                 &newval);
4315                         if (!valptr) {
4316                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4317                                        lacp_rate);
4318                                 return -EINVAL;
4319                         }
4320                         lacp_fast = valptr->value;
4321                 }
4322         }
4323
4324         if (ad_select) {
4325                 bond_opt_initstr(&newval, ad_select);
4326                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4327                                         &newval);
4328                 if (!valptr) {
4329                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4330                         return -EINVAL;
4331                 }
4332                 params->ad_select = valptr->value;
4333                 if (bond_mode != BOND_MODE_8023AD)
4334                         pr_warn("ad_select param only affects 802.3ad mode\n");
4335         } else {
4336                 params->ad_select = BOND_AD_STABLE;
4337         }
4338
4339         if (max_bonds < 0) {
4340                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4341                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4342                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4343         }
4344
4345         if (miimon < 0) {
4346                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4347                         miimon, INT_MAX);
4348                 miimon = 0;
4349         }
4350
4351         if (updelay < 0) {
4352                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4353                         updelay, INT_MAX);
4354                 updelay = 0;
4355         }
4356
4357         if (downdelay < 0) {
4358                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4359                         downdelay, INT_MAX);
4360                 downdelay = 0;
4361         }
4362
4363         if ((use_carrier != 0) && (use_carrier != 1)) {
4364                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4365                         use_carrier);
4366                 use_carrier = 1;
4367         }
4368
4369         if (num_peer_notif < 0 || num_peer_notif > 255) {
4370                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4371                         num_peer_notif);
4372                 num_peer_notif = 1;
4373         }
4374
4375         /* reset values for 802.3ad/TLB/ALB */
4376         if (!bond_mode_uses_arp(bond_mode)) {
4377                 if (!miimon) {
4378                         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");
4379                         pr_warn("Forcing miimon to 100msec\n");
4380                         miimon = BOND_DEFAULT_MIIMON;
4381                 }
4382         }
4383
4384         if (tx_queues < 1 || tx_queues > 255) {
4385                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4386                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4387                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4388         }
4389
4390         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4391                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4392                         all_slaves_active);
4393                 all_slaves_active = 0;
4394         }
4395
4396         if (resend_igmp < 0 || resend_igmp > 255) {
4397                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4398                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4399                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4400         }
4401
4402         bond_opt_initval(&newval, packets_per_slave);
4403         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4404                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4405                         packets_per_slave, USHRT_MAX);
4406                 packets_per_slave = 1;
4407         }
4408
4409         if (bond_mode == BOND_MODE_ALB) {
4410                 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",
4411                           updelay);
4412         }
4413
4414         if (!miimon) {
4415                 if (updelay || downdelay) {
4416                         /* just warn the user the up/down delay will have
4417                          * no effect since miimon is zero...
4418                          */
4419                         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",
4420                                 updelay, downdelay);
4421                 }
4422         } else {
4423                 /* don't allow arp monitoring */
4424                 if (arp_interval) {
4425                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4426                                 miimon, arp_interval);
4427                         arp_interval = 0;
4428                 }
4429
4430                 if ((updelay % miimon) != 0) {
4431                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4432                                 updelay, miimon, (updelay / miimon) * miimon);
4433                 }
4434
4435                 updelay /= miimon;
4436
4437                 if ((downdelay % miimon) != 0) {
4438                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4439                                 downdelay, miimon,
4440                                 (downdelay / miimon) * miimon);
4441                 }
4442
4443                 downdelay /= miimon;
4444         }
4445
4446         if (arp_interval < 0) {
4447                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4448                         arp_interval, INT_MAX);
4449                 arp_interval = 0;
4450         }
4451
4452         for (arp_ip_count = 0, i = 0;
4453              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4454                 __be32 ip;
4455
4456                 /* not a complete check, but good enough to catch mistakes */
4457                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4458                     !bond_is_ip_target_ok(ip)) {
4459                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4460                                 arp_ip_target[i]);
4461                         arp_interval = 0;
4462                 } else {
4463                         if (bond_get_targets_ip(arp_target, ip) == -1)
4464                                 arp_target[arp_ip_count++] = ip;
4465                         else
4466                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4467                                         &ip);
4468                 }
4469         }
4470
4471         if (arp_interval && !arp_ip_count) {
4472                 /* don't allow arping if no arp_ip_target given... */
4473                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4474                         arp_interval);
4475                 arp_interval = 0;
4476         }
4477
4478         if (arp_validate) {
4479                 if (!arp_interval) {
4480                         pr_err("arp_validate requires arp_interval\n");
4481                         return -EINVAL;
4482                 }
4483
4484                 bond_opt_initstr(&newval, arp_validate);
4485                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4486                                         &newval);
4487                 if (!valptr) {
4488                         pr_err("Error: invalid arp_validate \"%s\"\n",
4489                                arp_validate);
4490                         return -EINVAL;
4491                 }
4492                 arp_validate_value = valptr->value;
4493         } else {
4494                 arp_validate_value = 0;
4495         }
4496
4497         arp_all_targets_value = 0;
4498         if (arp_all_targets) {
4499                 bond_opt_initstr(&newval, arp_all_targets);
4500                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4501                                         &newval);
4502                 if (!valptr) {
4503                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4504                                arp_all_targets);
4505                         arp_all_targets_value = 0;
4506                 } else {
4507                         arp_all_targets_value = valptr->value;
4508                 }
4509         }
4510
4511         if (miimon) {
4512                 pr_info("MII link monitoring set to %d ms\n", miimon);
4513         } else if (arp_interval) {
4514                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4515                                           arp_validate_value);
4516                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4517                         arp_interval, valptr->string, arp_ip_count);
4518
4519                 for (i = 0; i < arp_ip_count; i++)
4520                         pr_cont(" %s", arp_ip_target[i]);
4521
4522                 pr_cont("\n");
4523
4524         } else if (max_bonds) {
4525                 /* miimon and arp_interval not set, we need one so things
4526                  * work as expected, see bonding.txt for details
4527                  */
4528                 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");
4529         }
4530
4531         if (primary && !bond_mode_uses_primary(bond_mode)) {
4532                 /* currently, using a primary only makes sense
4533                  * in active backup, TLB or ALB modes
4534                  */
4535                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4536                         primary, bond_mode_name(bond_mode));
4537                 primary = NULL;
4538         }
4539
4540         if (primary && primary_reselect) {
4541                 bond_opt_initstr(&newval, primary_reselect);
4542                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4543                                         &newval);
4544                 if (!valptr) {
4545                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4546                                primary_reselect);
4547                         return -EINVAL;
4548                 }
4549                 primary_reselect_value = valptr->value;
4550         } else {
4551                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4552         }
4553
4554         if (fail_over_mac) {
4555                 bond_opt_initstr(&newval, fail_over_mac);
4556                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4557                                         &newval);
4558                 if (!valptr) {
4559                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4560                                fail_over_mac);
4561                         return -EINVAL;
4562                 }
4563                 fail_over_mac_value = valptr->value;
4564                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4565                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4566         } else {
4567                 fail_over_mac_value = BOND_FOM_NONE;
4568         }
4569
4570         bond_opt_initstr(&newval, "default");
4571         valptr = bond_opt_parse(
4572                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4573                                      &newval);
4574         if (!valptr) {
4575                 pr_err("Error: No ad_actor_sys_prio default value");
4576                 return -EINVAL;
4577         }
4578         ad_actor_sys_prio = valptr->value;
4579
4580         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4581                                 &newval);
4582         if (!valptr) {
4583                 pr_err("Error: No ad_user_port_key default value");
4584                 return -EINVAL;
4585         }
4586         ad_user_port_key = valptr->value;
4587
4588         if (lp_interval == 0) {
4589                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4590                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4591                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4592         }
4593
4594         /* fill params struct with the proper values */
4595         params->mode = bond_mode;
4596         params->xmit_policy = xmit_hashtype;
4597         params->miimon = miimon;
4598         params->num_peer_notif = num_peer_notif;
4599         params->arp_interval = arp_interval;
4600         params->arp_validate = arp_validate_value;
4601         params->arp_all_targets = arp_all_targets_value;
4602         params->updelay = updelay;
4603         params->downdelay = downdelay;
4604         params->use_carrier = use_carrier;
4605         params->lacp_fast = lacp_fast;
4606         params->primary[0] = 0;
4607         params->primary_reselect = primary_reselect_value;
4608         params->fail_over_mac = fail_over_mac_value;
4609         params->tx_queues = tx_queues;
4610         params->all_slaves_active = all_slaves_active;
4611         params->resend_igmp = resend_igmp;
4612         params->min_links = min_links;
4613         params->lp_interval = lp_interval;
4614         params->packets_per_slave = packets_per_slave;
4615         params->tlb_dynamic_lb = 1; /* Default value */
4616         params->ad_actor_sys_prio = ad_actor_sys_prio;
4617         eth_zero_addr(params->ad_actor_system);
4618         params->ad_user_port_key = ad_user_port_key;
4619         if (packets_per_slave > 0) {
4620                 params->reciprocal_packets_per_slave =
4621                         reciprocal_value(packets_per_slave);
4622         } else {
4623                 /* reciprocal_packets_per_slave is unused if
4624                  * packets_per_slave is 0 or 1, just initialize it
4625                  */
4626                 params->reciprocal_packets_per_slave =
4627                         (struct reciprocal_value) { 0 };
4628         }
4629
4630         if (primary) {
4631                 strncpy(params->primary, primary, IFNAMSIZ);
4632                 params->primary[IFNAMSIZ - 1] = 0;
4633         }
4634
4635         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4636
4637         return 0;
4638 }
4639
4640 /* Called from registration process */
4641 static int bond_init(struct net_device *bond_dev)
4642 {
4643         struct bonding *bond = netdev_priv(bond_dev);
4644         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4645
4646         netdev_dbg(bond_dev, "Begin bond_init\n");
4647
4648         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4649         if (!bond->wq)
4650                 return -ENOMEM;
4651
4652         netdev_lockdep_set_classes(bond_dev);
4653
4654         list_add_tail(&bond->bond_list, &bn->dev_list);
4655
4656         bond_prepare_sysfs_group(bond);
4657
4658         bond_debug_register(bond);
4659
4660         /* Ensure valid dev_addr */
4661         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4662             bond_dev->addr_assign_type == NET_ADDR_PERM)
4663                 eth_hw_addr_random(bond_dev);
4664
4665         return 0;
4666 }
4667
4668 unsigned int bond_get_num_tx_queues(void)
4669 {
4670         return tx_queues;
4671 }
4672
4673 /* Create a new bond based on the specified name and bonding parameters.
4674  * If name is NULL, obtain a suitable "bond%d" name for us.
4675  * Caller must NOT hold rtnl_lock; we need to release it here before we
4676  * set up our sysfs entries.
4677  */
4678 int bond_create(struct net *net, const char *name)
4679 {
4680         struct net_device *bond_dev;
4681         struct bonding *bond;
4682         struct alb_bond_info *bond_info;
4683         int res;
4684
4685         rtnl_lock();
4686
4687         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4688                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4689                                    bond_setup, tx_queues);
4690         if (!bond_dev) {
4691                 pr_err("%s: eek! can't alloc netdev!\n", name);
4692                 rtnl_unlock();
4693                 return -ENOMEM;
4694         }
4695
4696         /*
4697          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4698          * It is set to 0 by default which is wrong.
4699          */
4700         bond = netdev_priv(bond_dev);
4701         bond_info = &(BOND_ALB_INFO(bond));
4702         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4703
4704         dev_net_set(bond_dev, net);
4705         bond_dev->rtnl_link_ops = &bond_link_ops;
4706
4707         res = register_netdevice(bond_dev);
4708
4709         netif_carrier_off(bond_dev);
4710
4711         rtnl_unlock();
4712         if (res < 0)
4713                 bond_destructor(bond_dev);
4714         return res;
4715 }
4716
4717 static int __net_init bond_net_init(struct net *net)
4718 {
4719         struct bond_net *bn = net_generic(net, bond_net_id);
4720
4721         bn->net = net;
4722         INIT_LIST_HEAD(&bn->dev_list);
4723
4724         bond_create_proc_dir(bn);
4725         bond_create_sysfs(bn);
4726
4727         return 0;
4728 }
4729
4730 static void __net_exit bond_net_exit(struct net *net)
4731 {
4732         struct bond_net *bn = net_generic(net, bond_net_id);
4733         struct bonding *bond, *tmp_bond;
4734         LIST_HEAD(list);
4735
4736         bond_destroy_sysfs(bn);
4737
4738         /* Kill off any bonds created after unregistering bond rtnl ops */
4739         rtnl_lock();
4740         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4741                 unregister_netdevice_queue(bond->dev, &list);
4742         unregister_netdevice_many(&list);
4743         rtnl_unlock();
4744
4745         bond_destroy_proc_dir(bn);
4746 }
4747
4748 static struct pernet_operations bond_net_ops = {
4749         .init = bond_net_init,
4750         .exit = bond_net_exit,
4751         .id   = &bond_net_id,
4752         .size = sizeof(struct bond_net),
4753 };
4754
4755 static int __init bonding_init(void)
4756 {
4757         int i;
4758         int res;
4759
4760         pr_info("%s", bond_version);
4761
4762         res = bond_check_params(&bonding_defaults);
4763         if (res)
4764                 goto out;
4765
4766         res = register_pernet_subsys(&bond_net_ops);
4767         if (res)
4768                 goto out;
4769
4770         res = bond_netlink_init();
4771         if (res)
4772                 goto err_link;
4773
4774         bond_create_debugfs();
4775
4776         for (i = 0; i < max_bonds; i++) {
4777                 res = bond_create(&init_net, NULL);
4778                 if (res)
4779                         goto err;
4780         }
4781
4782         register_netdevice_notifier(&bond_netdev_notifier);
4783 out:
4784         return res;
4785 err:
4786         bond_destroy_debugfs();
4787         bond_netlink_fini();
4788 err_link:
4789         unregister_pernet_subsys(&bond_net_ops);
4790         goto out;
4791
4792 }
4793
4794 static void __exit bonding_exit(void)
4795 {
4796         unregister_netdevice_notifier(&bond_netdev_notifier);
4797
4798         bond_destroy_debugfs();
4799
4800         bond_netlink_fini();
4801         unregister_pernet_subsys(&bond_net_ops);
4802
4803 #ifdef CONFIG_NET_POLL_CONTROLLER
4804         /* Make sure we don't have an imbalance on our netpoll blocking */
4805         WARN_ON(atomic_read(&netpoll_block_tx));
4806 #endif
4807 }
4808
4809 module_init(bonding_init);
4810 module_exit(bonding_exit);
4811 MODULE_LICENSE("GPL");
4812 MODULE_VERSION(DRV_VERSION);
4813 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4814 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");