drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/dma.h>
  56 #include <linux/uaccess.h>
  57 #include <linux/errno.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/inetdevice.h>
  60 #include <linux/igmp.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/skbuff.h>
  63 #include <net/sock.h>
  64 #include <linux/rtnetlink.h>
  65 #include <linux/smp.h>
  66 #include <linux/if_ether.h>
  67 #include <net/arp.h>
  68 #include <linux/mii.h>
  69 #include <linux/ethtool.h>
  70 #include <linux/if_vlan.h>
  71 #include <linux/if_bonding.h>
  72 #include <linux/jiffies.h>
  73 #include <linux/preempt.h>
  74 #include <net/route.h>
  75 #include <net/net_namespace.h>
  76 #include <net/netns/generic.h>
  77 #include <net/pkt_sched.h>
  78 #include <linux/rculist.h>
  79 #include <net/flow_dissector.h>
  80 #include <net/switchdev.h>
  81 #include <net/bonding.h>
  82 #include <net/bond_3ad.h>
  83 #include <net/bond_alb.h>
  84
  85 #include "bonding_priv.h"
  86
  87 /*---------------------------- Module parameters ----------------------------*/
  88
  89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  90
  91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  92 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  93 static int num_peer_notif = 1;
  94 static int miimon;
  95 static int updelay;
  96 static int downdelay;
  97 static int use_carrier  = 1;
  98 static char *mode;
  99 static char *primary;
 100 static char *primary_reselect;
 101 static char *lacp_rate;
 102 static int min_links;
 103 static char *ad_select;
 104 static char *xmit_hash_policy;
 105 static int arp_interval;
 106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 107 static char *arp_validate;
 108 static char *arp_all_targets;
 109 static char *fail_over_mac;
 110 static int all_slaves_active;
 111 static struct bond_params bonding_defaults;
 112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 113 static int packets_per_slave = 1;
 114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 115
 116 module_param(max_bonds, int, 0);
 117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 118 module_param(tx_queues, int, 0);
 119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 122                                "failover event (alias of num_unsol_na)");
 123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 125                                "failover event (alias of num_grat_arp)");
 126 module_param(miimon, int, 0);
 127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 128 module_param(updelay, int, 0);
 129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 130 module_param(downdelay, int, 0);
 131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 132                             "in milliseconds");
 133 module_param(use_carrier, int, 0);
 134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 135                               "0 for off, 1 for on (default)");
 136 module_param(mode, charp, 0);
 137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 138                        "1 for active-backup, 2 for balance-xor, "
 139                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 140                        "6 for balance-alb");
 141 module_param(primary, charp, 0);
 142 MODULE_PARM_DESC(primary, "Primary network device to use");
 143 module_param(primary_reselect, charp, 0);
 144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 145                                    "once it comes up; "
 146                                    "0 for always (default), "
 147                                    "1 for only if speed of primary is "
 148                                    "better, "
 149                                    "2 for only on active slave "
 150                                    "failure");
 151 module_param(lacp_rate, charp, 0);
 152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 153                             "0 for slow, 1 for fast");
 154 module_param(ad_select, charp, 0);
 155 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
 156                             "0 for stable (default), 1 for bandwidth, "
 157                             "2 for count");
 158 module_param(min_links, int, 0);
 159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 160
 161 module_param(xmit_hash_policy, charp, 0);
 162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
 163                                    "0 for layer 2 (default), 1 for layer 3+4, "
 164                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 165                                    "4 for encap layer 3+4");
 166 module_param(arp_interval, int, 0);
 167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 168 module_param_array(arp_ip_target, charp, NULL, 0);
 169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 170 module_param(arp_validate, charp, 0);
 171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 172                                "0 for none (default), 1 for active, "
 173                                "2 for backup, 3 for all");
 174 module_param(arp_all_targets, charp, 0);
 175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 176 module_param(fail_over_mac, charp, 0);
 177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 178                                 "the same MAC; 0 for none (default), "
 179                                 "1 for active, 2 for follow");
 180 module_param(all_slaves_active, int, 0);
 181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 182                                      "by setting active flag for all slaves; "
 183                                      "0 for never (default), 1 for always.");
 184 module_param(resend_igmp, int, 0);
 185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 186                               "link failure");
 187 module_param(packets_per_slave, int, 0);
 188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 189                                     "mode; 0 for a random slave, 1 packet per "
 190                                     "slave (default), >1 packets per slave.");
 191 module_param(lp_interval, uint, 0);
 192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 193                               "the bonding driver sends learning packets to "
 194                               "each slaves peer switch. The default is 1.");
 195
 196 /*----------------------------- Global variables ----------------------------*/
 197
 198 #ifdef CONFIG_NET_POLL_CONTROLLER
 199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 200 #endif
 201
 202 int bond_net_id __read_mostly;
 203
 204 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
 205 static int arp_ip_count;
 206 static int bond_mode    = BOND_MODE_ROUNDROBIN;
 207 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
 208 static int lacp_fast;
 209
 210 /*-------------------------- Forward declarations ---------------------------*/
 211
 212 static int bond_init(struct net_device *bond_dev);
 213 static void bond_uninit(struct net_device *bond_dev);
 214 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
 215                                                 struct rtnl_link_stats64 *stats);
 216 static void bond_slave_arr_handler(struct work_struct *work);
 217
 218 /*---------------------------- General routines -----------------------------*/
 219
 220 const char *bond_mode_name(int mode)
 221 {
 222         static const char *names[] = {
 223                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 224                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 225                 [BOND_MODE_XOR] = "load balancing (xor)",
 226                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 227                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 228                 [BOND_MODE_TLB] = "transmit load balancing",
 229                 [BOND_MODE_ALB] = "adaptive load balancing",
 230         };
 231
 232         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 233                 return "unknown";
 234
 235         return names[mode];
 236 }
 237
 238 /*---------------------------------- VLAN -----------------------------------*/
 239
 240 /**
 241  * bond_dev_queue_xmit - Prepare skb for xmit.
 242  *
 243  * @bond: bond device that got this skb for tx.
 244  * @skb: hw accel VLAN tagged skb to transmit
 245  * @slave_dev: slave that is supposed to xmit this skbuff
 246  */
 247 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 248                         struct net_device *slave_dev)
 249 {
 250         skb->dev = slave_dev;
 251
 252         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 253                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 254         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
 255
 256         if (unlikely(netpoll_tx_running(bond->dev)))
 257                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 258         else
 259                 dev_queue_xmit(skb);
 260 }
 261
 262 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 263  * We don't protect the slave list iteration with a lock because:
 264  * a. This operation is performed in IOCTL context,
 265  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 266  * c. Holding a lock with BH disabled while directly calling a base driver
 267  *    entry point is generally a BAD idea.
 268  *
 269  * The design of synchronization/protection for this operation in the 8021q
 270  * module is good for one or more VLAN devices over a single physical device
 271  * and cannot be extended for a teaming solution like bonding, so there is a
 272  * potential race condition here where a net device from the vlan group might
 273  * be referenced (either by a base driver or the 8021q code) while it is being
 274  * removed from the system. However, it turns out we're not making matters
 275  * worse, and if it works for regular VLAN usage it will work here too.
 276 */
 277
 278 /**
 279  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 280  * @bond_dev: bonding net device that got called
 281  * @vid: vlan id being added
 282  */
 283 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 284                                 __be16 proto, u16 vid)
 285 {
 286         struct bonding *bond = netdev_priv(bond_dev);
 287         struct slave *slave, *rollback_slave;
 288         struct list_head *iter;
 289         int res;
 290
 291         bond_for_each_slave(bond, slave, iter) {
 292                 res = vlan_vid_add(slave->dev, proto, vid);
 293                 if (res)
 294                         goto unwind;
 295         }
 296
 297         return 0;
 298
 299 unwind:
 300         /* unwind to the slave that failed */
 301         bond_for_each_slave(bond, rollback_slave, iter) {
 302                 if (rollback_slave == slave)
 303                         break;
 304
 305                 vlan_vid_del(rollback_slave->dev, proto, vid);
 306         }
 307
 308         return res;
 309 }
 310
 311 /**
 312  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 313  * @bond_dev: bonding net device that got called
 314  * @vid: vlan id being removed
 315  */
 316 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 317                                  __be16 proto, u16 vid)
 318 {
 319         struct bonding *bond = netdev_priv(bond_dev);
 320         struct list_head *iter;
 321         struct slave *slave;
 322
 323         bond_for_each_slave(bond, slave, iter)
 324                 vlan_vid_del(slave->dev, proto, vid);
 325
 326         if (bond_is_lb(bond))
 327                 bond_alb_clear_vlan(bond, vid);
 328
 329         return 0;
 330 }
 331
 332 /*------------------------------- Link status -------------------------------*/
 333
 334 /* Set the carrier state for the master according to the state of its
 335  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 336  * do special 802.3ad magic.
 337  *
 338  * Returns zero if carrier state does not change, nonzero if it does.
 339  */
 340 int bond_set_carrier(struct bonding *bond)
 341 {
 342         struct list_head *iter;
 343         struct slave *slave;
 344
 345         if (!bond_has_slaves(bond))
 346                 goto down;
 347
 348         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 349                 return bond_3ad_set_carrier(bond);
 350
 351         bond_for_each_slave(bond, slave, iter) {
 352                 if (slave->link == BOND_LINK_UP) {
 353                         if (!netif_carrier_ok(bond->dev)) {
 354                                 netif_carrier_on(bond->dev);
 355                                 return 1;
 356                         }
 357                         return 0;
 358                 }
 359         }
 360
 361 down:
 362         if (netif_carrier_ok(bond->dev)) {
 363                 netif_carrier_off(bond->dev);
 364                 return 1;
 365         }
 366         return 0;
 367 }
 368
 369 /* Get link speed and duplex from the slave's base driver
 370  * using ethtool. If for some reason the call fails or the
 371  * values are invalid, set speed and duplex to -1,
 372  * and return.
 373  */
 374 static void bond_update_speed_duplex(struct slave *slave)
 375 {
 376         struct net_device *slave_dev = slave->dev;
 377         struct ethtool_cmd ecmd;
 378         u32 slave_speed;
 379         int res;
 380
 381         slave->speed = SPEED_UNKNOWN;
 382         slave->duplex = DUPLEX_UNKNOWN;
 383
 384         res = __ethtool_get_settings(slave_dev, &ecmd);
 385         if (res < 0)
 386                 return;
 387
 388         slave_speed = ethtool_cmd_speed(&ecmd);
 389         if (slave_speed == 0 || slave_speed == ((__u32) -1))
 390                 return;
 391
 392         switch (ecmd.duplex) {
 393         case DUPLEX_FULL:
 394         case DUPLEX_HALF:
 395                 break;
 396         default:
 397                 return;
 398         }
 399
 400         slave->speed = slave_speed;
 401         slave->duplex = ecmd.duplex;
 402
 403         return;
 404 }
 405
 406 const char *bond_slave_link_status(s8 link)
 407 {
 408         switch (link) {
 409         case BOND_LINK_UP:
 410                 return "up";
 411         case BOND_LINK_FAIL:
 412                 return "going down";
 413         case BOND_LINK_DOWN:
 414                 return "down";
 415         case BOND_LINK_BACK:
 416                 return "going back";
 417         default:
 418                 return "unknown";
 419         }
 420 }
 421
 422 /* if <dev> supports MII link status reporting, check its link status.
 423  *
 424  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 425  * depending upon the setting of the use_carrier parameter.
 426  *
 427  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 428  * can't tell and just pretend it is), or 0, meaning that the link is
 429  * down.
 430  *
 431  * If reporting is non-zero, instead of faking link up, return -1 if
 432  * both ETHTOOL and MII ioctls fail (meaning the device does not
 433  * support them).  If use_carrier is set, return whatever it says.
 434  * It'd be nice if there was a good way to tell if a driver supports
 435  * netif_carrier, but there really isn't.
 436  */
 437 static int bond_check_dev_link(struct bonding *bond,
 438                                struct net_device *slave_dev, int reporting)
 439 {
 440         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 441         int (*ioctl)(struct net_device *, struct ifreq *, int);
 442         struct ifreq ifr;
 443         struct mii_ioctl_data *mii;
 444
 445         if (!reporting && !netif_running(slave_dev))
 446                 return 0;
 447
 448         if (bond->params.use_carrier)
 449                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 450
 451         /* Try to get link status using Ethtool first. */
 452         if (slave_dev->ethtool_ops->get_link)
 453                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 454                         BMSR_LSTATUS : 0;
 455
 456         /* Ethtool can't be used, fallback to MII ioctls. */
 457         ioctl = slave_ops->ndo_do_ioctl;
 458         if (ioctl) {
 459                 /* TODO: set pointer to correct ioctl on a per team member
 460                  *       bases to make this more efficient. that is, once
 461                  *       we determine the correct ioctl, we will always
 462                  *       call it and not the others for that team
 463                  *       member.
 464                  */
 465
 466                 /* We cannot assume that SIOCGMIIPHY will also read a
 467                  * register; not all network drivers (e.g., e100)
 468                  * support that.
 469                  */
 470
 471                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 472                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 473                 mii = if_mii(&ifr);
 474                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 475                         mii->reg_num = MII_BMSR;
 476                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
 477                                 return mii->val_out & BMSR_LSTATUS;
 478                 }
 479         }
 480
 481         /* If reporting, report that either there's no dev->do_ioctl,
 482          * or both SIOCGMIIREG and get_link failed (meaning that we
 483          * cannot report link status).  If not reporting, pretend
 484          * we're ok.
 485          */
 486         return reporting ? -1 : BMSR_LSTATUS;
 487 }
 488
 489 /*----------------------------- Multicast list ------------------------------*/
 490
 491 /* Push the promiscuity flag down to appropriate slaves */
 492 static int bond_set_promiscuity(struct bonding *bond, int inc)
 493 {
 494         struct list_head *iter;
 495         int err = 0;
 496
 497         if (bond_uses_primary(bond)) {
 498                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 499
 500                 if (curr_active)
 501                         err = dev_set_promiscuity(curr_active->dev, inc);
 502         } else {
 503                 struct slave *slave;
 504
 505                 bond_for_each_slave(bond, slave, iter) {
 506                         err = dev_set_promiscuity(slave->dev, inc);
 507                         if (err)
 508                                 return err;
 509                 }
 510         }
 511         return err;
 512 }
 513
 514 /* Push the allmulti flag down to all slaves */
 515 static int bond_set_allmulti(struct bonding *bond, int inc)
 516 {
 517         struct list_head *iter;
 518         int err = 0;
 519
 520         if (bond_uses_primary(bond)) {
 521                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 522
 523                 if (curr_active)
 524                         err = dev_set_allmulti(curr_active->dev, inc);
 525         } else {
 526                 struct slave *slave;
 527
 528                 bond_for_each_slave(bond, slave, iter) {
 529                         err = dev_set_allmulti(slave->dev, inc);
 530                         if (err)
 531                                 return err;
 532                 }
 533         }
 534         return err;
 535 }
 536
 537 /* Retrieve the list of registered multicast addresses for the bonding
 538  * device and retransmit an IGMP JOIN request to the current active
 539  * slave.
 540  */
 541 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 542 {
 543         struct bonding *bond = container_of(work, struct bonding,
 544                                             mcast_work.work);
 545
 546         if (!rtnl_trylock()) {
 547                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 548                 return;
 549         }
 550         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 551
 552         if (bond->igmp_retrans > 1) {
 553                 bond->igmp_retrans--;
 554                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 555         }
 556         rtnl_unlock();
 557 }
 558
 559 /* Flush bond's hardware addresses from slave */
 560 static void bond_hw_addr_flush(struct net_device *bond_dev,
 561                                struct net_device *slave_dev)
 562 {
 563         struct bonding *bond = netdev_priv(bond_dev);
 564
 565         dev_uc_unsync(slave_dev, bond_dev);
 566         dev_mc_unsync(slave_dev, bond_dev);
 567
 568         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 569                 /* del lacpdu mc addr from mc list */
 570                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 571
 572                 dev_mc_del(slave_dev, lacpdu_multicast);
 573         }
 574 }
 575
 576 /*--------------------------- Active slave change ---------------------------*/
 577
 578 /* Update the hardware address list and promisc/allmulti for the new and
 579  * old active slaves (if any).  Modes that are not using primary keep all
 580  * slaves up date at all times; only the modes that use primary need to call
 581  * this function to swap these settings during a failover.
 582  */
 583 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 584                               struct slave *old_active)
 585 {
 586         if (old_active) {
 587                 if (bond->dev->flags & IFF_PROMISC)
 588                         dev_set_promiscuity(old_active->dev, -1);
 589
 590                 if (bond->dev->flags & IFF_ALLMULTI)
 591                         dev_set_allmulti(old_active->dev, -1);
 592
 593                 bond_hw_addr_flush(bond->dev, old_active->dev);
 594         }
 595
 596         if (new_active) {
 597                 /* FIXME: Signal errors upstream. */
 598                 if (bond->dev->flags & IFF_PROMISC)
 599                         dev_set_promiscuity(new_active->dev, 1);
 600
 601                 if (bond->dev->flags & IFF_ALLMULTI)
 602                         dev_set_allmulti(new_active->dev, 1);
 603
 604                 netif_addr_lock_bh(bond->dev);
 605                 dev_uc_sync(new_active->dev, bond->dev);
 606                 dev_mc_sync(new_active->dev, bond->dev);
 607                 netif_addr_unlock_bh(bond->dev);
 608         }
 609 }
 610
 611 /**
 612  * bond_set_dev_addr - clone slave's address to bond
 613  * @bond_dev: bond net device
 614  * @slave_dev: slave net device
 615  *
 616  * Should be called with RTNL held.
 617  */
 618 static void bond_set_dev_addr(struct net_device *bond_dev,
 619                               struct net_device *slave_dev)
 620 {
 621         netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 622                    bond_dev, slave_dev, slave_dev->addr_len);
 623         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 624         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 625         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 626 }
 627
 628 /* bond_do_fail_over_mac
 629  *
 630  * Perform special MAC address swapping for fail_over_mac settings
 631  *
 632  * Called with RTNL
 633  */
 634 static void bond_do_fail_over_mac(struct bonding *bond,
 635                                   struct slave *new_active,
 636                                   struct slave *old_active)
 637 {
 638         u8 tmp_mac[ETH_ALEN];
 639         struct sockaddr saddr;
 640         int rv;
 641
 642         switch (bond->params.fail_over_mac) {
 643         case BOND_FOM_ACTIVE:
 644                 if (new_active)
 645                         bond_set_dev_addr(bond->dev, new_active->dev);
 646                 break;
 647         case BOND_FOM_FOLLOW:
 648                 /* if new_active && old_active, swap them
 649                  * if just old_active, do nothing (going to no active slave)
 650                  * if just new_active, set new_active to bond's MAC
 651                  */
 652                 if (!new_active)
 653                         return;
 654
 655                 if (old_active) {
 656                         ether_addr_copy(tmp_mac, new_active->dev->dev_addr);
 657                         ether_addr_copy(saddr.sa_data,
 658                                         old_active->dev->dev_addr);
 659                         saddr.sa_family = new_active->dev->type;
 660                 } else {
 661                         ether_addr_copy(saddr.sa_data, bond->dev->dev_addr);
 662                         saddr.sa_family = bond->dev->type;
 663                 }
 664
 665                 rv = dev_set_mac_address(new_active->dev, &saddr);
 666                 if (rv) {
 667                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 668                                    -rv, new_active->dev->name);
 669                         goto out;
 670                 }
 671
 672                 if (!old_active)
 673                         goto out;
 674
 675                 ether_addr_copy(saddr.sa_data, tmp_mac);
 676                 saddr.sa_family = old_active->dev->type;
 677
 678                 rv = dev_set_mac_address(old_active->dev, &saddr);
 679                 if (rv)
 680                         netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
 681                                    -rv, new_active->dev->name);
 682 out:
 683                 break;
 684         default:
 685                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 686                            bond->params.fail_over_mac);
 687                 break;
 688         }
 689
 690 }
 691
 692 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 693 {
 694         struct slave *prim = rtnl_dereference(bond->primary_slave);
 695         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 696
 697         if (!prim || prim->link != BOND_LINK_UP) {
 698                 if (!curr || curr->link != BOND_LINK_UP)
 699                         return NULL;
 700                 return curr;
 701         }
 702
 703         if (bond->force_primary) {
 704                 bond->force_primary = false;
 705                 return prim;
 706         }
 707
 708         if (!curr || curr->link != BOND_LINK_UP)
 709                 return prim;
 710
 711         /* At this point, prim and curr are both up */
 712         switch (bond->params.primary_reselect) {
 713         case BOND_PRI_RESELECT_ALWAYS:
 714                 return prim;
 715         case BOND_PRI_RESELECT_BETTER:
 716                 if (prim->speed < curr->speed)
 717                         return curr;
 718                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 719                         return curr;
 720                 return prim;
 721         case BOND_PRI_RESELECT_FAILURE:
 722                 return curr;
 723         default:
 724                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 725                            bond->params.primary_reselect);
 726                 return curr;
 727         }
 728 }
 729
 730 /**
 731  * bond_find_best_slave - select the best available slave to be the active one
 732  * @bond: our bonding struct
 733  */
 734 static struct slave *bond_find_best_slave(struct bonding *bond)
 735 {
 736         struct slave *slave, *bestslave = NULL;
 737         struct list_head *iter;
 738         int mintime = bond->params.updelay;
 739
 740         slave = bond_choose_primary_or_current(bond);
 741         if (slave)
 742                 return slave;
 743
 744         bond_for_each_slave(bond, slave, iter) {
 745                 if (slave->link == BOND_LINK_UP)
 746                         return slave;
 747                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 748                     slave->delay < mintime) {
 749                         mintime = slave->delay;
 750                         bestslave = slave;
 751                 }
 752         }
 753
 754         return bestslave;
 755 }
 756
 757 static bool bond_should_notify_peers(struct bonding *bond)
 758 {
 759         struct slave *slave;
 760
 761         rcu_read_lock();
 762         slave = rcu_dereference(bond->curr_active_slave);
 763         rcu_read_unlock();
 764
 765         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 766                    slave ? slave->dev->name : "NULL");
 767
 768         if (!slave || !bond->send_peer_notif ||
 769             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 770                 return false;
 771
 772         return true;
 773 }
 774
 775 /**
 776  * change_active_interface - change the active slave into the specified one
 777  * @bond: our bonding struct
 778  * @new: the new slave to make the active one
 779  *
 780  * Set the new slave to the bond's settings and unset them on the old
 781  * curr_active_slave.
 782  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 783  *
 784  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 785  * because it is apparently the best available slave we have, even though its
 786  * updelay hasn't timed out yet.
 787  *
 788  * Caller must hold RTNL.
 789  */
 790 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 791 {
 792         struct slave *old_active;
 793
 794         ASSERT_RTNL();
 795
 796         old_active = rtnl_dereference(bond->curr_active_slave);
 797
 798         if (old_active == new_active)
 799                 return;
 800
 801         if (new_active) {
 802                 new_active->last_link_up = jiffies;
 803
 804                 if (new_active->link == BOND_LINK_BACK) {
 805                         if (bond_uses_primary(bond)) {
 806                                 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
 807                                             new_active->dev->name,
 808                                             (bond->params.updelay - new_active->delay) * bond->params.miimon);
 809                         }
 810
 811                         new_active->delay = 0;
 812                         bond_set_slave_link_state(new_active, BOND_LINK_UP);
 813
 814                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 815                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 816
 817                         if (bond_is_lb(bond))
 818                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 819                 } else {
 820                         if (bond_uses_primary(bond)) {
 821                                 netdev_info(bond->dev, "making interface %s the new active one\n",
 822                                             new_active->dev->name);
 823                         }
 824                 }
 825         }
 826
 827         if (bond_uses_primary(bond))
 828                 bond_hw_addr_swap(bond, new_active, old_active);
 829
 830         if (bond_is_lb(bond)) {
 831                 bond_alb_handle_active_change(bond, new_active);
 832                 if (old_active)
 833                         bond_set_slave_inactive_flags(old_active,
 834                                                       BOND_SLAVE_NOTIFY_NOW);
 835                 if (new_active)
 836                         bond_set_slave_active_flags(new_active,
 837                                                     BOND_SLAVE_NOTIFY_NOW);
 838         } else {
 839                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 840         }
 841
 842         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 843                 if (old_active)
 844                         bond_set_slave_inactive_flags(old_active,
 845                                                       BOND_SLAVE_NOTIFY_NOW);
 846
 847                 if (new_active) {
 848                         bool should_notify_peers = false;
 849
 850                         bond_set_slave_active_flags(new_active,
 851                                                     BOND_SLAVE_NOTIFY_NOW);
 852
 853                         if (bond->params.fail_over_mac)
 854                                 bond_do_fail_over_mac(bond, new_active,
 855                                                       old_active);
 856
 857                         if (netif_running(bond->dev)) {
 858                                 bond->send_peer_notif =
 859                                         bond->params.num_peer_notif;
 860                                 should_notify_peers =
 861                                         bond_should_notify_peers(bond);
 862                         }
 863
 864                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 865                         if (should_notify_peers)
 866                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 867                                                          bond->dev);
 868                 }
 869         }
 870
 871         /* resend IGMP joins since active slave has changed or
 872          * all were sent on curr_active_slave.
 873          * resend only if bond is brought up with the affected
 874          * bonding modes and the retransmission is enabled
 875          */
 876         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 877             ((bond_uses_primary(bond) && new_active) ||
 878              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 879                 bond->igmp_retrans = bond->params.resend_igmp;
 880                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 881         }
 882 }
 883
 884 /**
 885  * bond_select_active_slave - select a new active slave, if needed
 886  * @bond: our bonding struct
 887  *
 888  * This functions should be called when one of the following occurs:
 889  * - The old curr_active_slave has been released or lost its link.
 890  * - The primary_slave has got its link back.
 891  * - A slave has got its link back and there's no old curr_active_slave.
 892  *
 893  * Caller must hold RTNL.
 894  */
 895 void bond_select_active_slave(struct bonding *bond)
 896 {
 897         struct slave *best_slave;
 898         int rv;
 899
 900         ASSERT_RTNL();
 901
 902         best_slave = bond_find_best_slave(bond);
 903         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 904                 bond_change_active_slave(bond, best_slave);
 905                 rv = bond_set_carrier(bond);
 906                 if (!rv)
 907                         return;
 908
 909                 if (netif_carrier_ok(bond->dev)) {
 910                         netdev_info(bond->dev, "first active interface up!\n");
 911                 } else {
 912                         netdev_info(bond->dev, "now running without any active interface!\n");
 913                 }
 914         }
 915 }
 916
 917 #ifdef CONFIG_NET_POLL_CONTROLLER
 918 static inline int slave_enable_netpoll(struct slave *slave)
 919 {
 920         struct netpoll *np;
 921         int err = 0;
 922
 923         np = kzalloc(sizeof(*np), GFP_KERNEL);
 924         err = -ENOMEM;
 925         if (!np)
 926                 goto out;
 927
 928         err = __netpoll_setup(np, slave->dev);
 929         if (err) {
 930                 kfree(np);
 931                 goto out;
 932         }
 933         slave->np = np;
 934 out:
 935         return err;
 936 }
 937 static inline void slave_disable_netpoll(struct slave *slave)
 938 {
 939         struct netpoll *np = slave->np;
 940
 941         if (!np)
 942                 return;
 943
 944         slave->np = NULL;
 945         __netpoll_free_async(np);
 946 }
 947
 948 static void bond_poll_controller(struct net_device *bond_dev)
 949 {
 950         struct bonding *bond = netdev_priv(bond_dev);
 951         struct slave *slave = NULL;
 952         struct list_head *iter;
 953         struct ad_info ad_info;
 954         struct netpoll_info *ni;
 955         const struct net_device_ops *ops;
 956
 957         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 958                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 959                         return;
 960
 961         rcu_read_lock_bh();
 962         bond_for_each_slave_rcu(bond, slave, iter) {
 963                 ops = slave->dev->netdev_ops;
 964                 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
 965                         continue;
 966
 967                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 968                         struct aggregator *agg =
 969                             SLAVE_AD_INFO(slave)->port.aggregator;
 970
 971                         if (agg &&
 972                             agg->aggregator_identifier != ad_info.aggregator_id)
 973                                 continue;
 974                 }
 975
 976                 ni = rcu_dereference_bh(slave->dev->npinfo);
 977                 if (down_trylock(&ni->dev_lock))
 978                         continue;
 979                 ops->ndo_poll_controller(slave->dev);
 980                 up(&ni->dev_lock);
 981         }
 982         rcu_read_unlock_bh();
 983 }
 984
 985 static void bond_netpoll_cleanup(struct net_device *bond_dev)
 986 {
 987         struct bonding *bond = netdev_priv(bond_dev);
 988         struct list_head *iter;
 989         struct slave *slave;
 990
 991         bond_for_each_slave(bond, slave, iter)
 992                 if (bond_slave_is_up(slave))
 993                         slave_disable_netpoll(slave);
 994 }
 995
 996 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
 997 {
 998         struct bonding *bond = netdev_priv(dev);
 999         struct list_head *iter;
1000         struct slave *slave;
1001         int err = 0;
1002
1003         bond_for_each_slave(bond, slave, iter) {
1004                 err = slave_enable_netpoll(slave);
1005                 if (err) {
1006                         bond_netpoll_cleanup(dev);
1007                         break;
1008                 }
1009         }
1010         return err;
1011 }
1012 #else
1013 static inline int slave_enable_netpoll(struct slave *slave)
1014 {
1015         return 0;
1016 }
1017 static inline void slave_disable_netpoll(struct slave *slave)
1018 {
1019 }
1020 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1021 {
1022 }
1023 #endif
1024
1025 /*---------------------------------- IOCTL ----------------------------------*/
1026
1027 static netdev_features_t bond_fix_features(struct net_device *dev,
1028                                            netdev_features_t features)
1029 {
1030         struct bonding *bond = netdev_priv(dev);
1031         struct list_head *iter;
1032         netdev_features_t mask;
1033         struct slave *slave;
1034
1035         mask = features;
1036
1037         features &= ~NETIF_F_ONE_FOR_ALL;
1038         features |= NETIF_F_ALL_FOR_ALL;
1039
1040         bond_for_each_slave(bond, slave, iter) {
1041                 features = netdev_increment_features(features,
1042                                                      slave->dev->features,
1043                                                      mask);
1044         }
1045         features = netdev_add_tso_features(features, mask);
1046
1047         return features;
1048 }
1049
1050 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1051                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1052                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1053
1054 #define BOND_ENC_FEATURES       (NETIF_F_ALL_CSUM | NETIF_F_SG | NETIF_F_RXCSUM |\
1055                                  NETIF_F_TSO)
1056
1057 static void bond_compute_features(struct bonding *bond)
1058 {
1059         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1060                                         IFF_XMIT_DST_RELEASE_PERM;
1061         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1062         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1063         struct net_device *bond_dev = bond->dev;
1064         struct list_head *iter;
1065         struct slave *slave;
1066         unsigned short max_hard_header_len = ETH_HLEN;
1067         unsigned int gso_max_size = GSO_MAX_SIZE;
1068         u16 gso_max_segs = GSO_MAX_SEGS;
1069
1070         if (!bond_has_slaves(bond))
1071                 goto done;
1072         vlan_features &= NETIF_F_ALL_FOR_ALL;
1073
1074         bond_for_each_slave(bond, slave, iter) {
1075                 vlan_features = netdev_increment_features(vlan_features,
1076                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1077
1078                 enc_features = netdev_increment_features(enc_features,
1079                                                          slave->dev->hw_enc_features,
1080                                                          BOND_ENC_FEATURES);
1081                 dst_release_flag &= slave->dev->priv_flags;
1082                 if (slave->dev->hard_header_len > max_hard_header_len)
1083                         max_hard_header_len = slave->dev->hard_header_len;
1084
1085                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1086                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1087         }
1088
1089 done:
1090         bond_dev->vlan_features = vlan_features;
1091         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1092         bond_dev->hard_header_len = max_hard_header_len;
1093         bond_dev->gso_max_segs = gso_max_segs;
1094         netif_set_gso_max_size(bond_dev, gso_max_size);
1095
1096         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1097         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1098             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1099                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1100
1101         netdev_change_features(bond_dev);
1102 }
1103
1104 static void bond_setup_by_slave(struct net_device *bond_dev,
1105                                 struct net_device *slave_dev)
1106 {
1107         bond_dev->header_ops        = slave_dev->header_ops;
1108
1109         bond_dev->type              = slave_dev->type;
1110         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1111         bond_dev->addr_len          = slave_dev->addr_len;
1112
1113         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1114                 slave_dev->addr_len);
1115 }
1116
1117 /* On bonding slaves other than the currently active slave, suppress
1118  * duplicates except for alb non-mcast/bcast.
1119  */
1120 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1121                                             struct slave *slave,
1122                                             struct bonding *bond)
1123 {
1124         if (bond_is_slave_inactive(slave)) {
1125                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1126                     skb->pkt_type != PACKET_BROADCAST &&
1127                     skb->pkt_type != PACKET_MULTICAST)
1128                         return false;
1129                 return true;
1130         }
1131         return false;
1132 }
1133
1134 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1135 {
1136         struct sk_buff *skb = *pskb;
1137         struct slave *slave;
1138         struct bonding *bond;
1139         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1140                           struct slave *);
1141         int ret = RX_HANDLER_ANOTHER;
1142
1143         skb = skb_share_check(skb, GFP_ATOMIC);
1144         if (unlikely(!skb))
1145                 return RX_HANDLER_CONSUMED;
1146
1147         *pskb = skb;
1148
1149         slave = bond_slave_get_rcu(skb->dev);
1150         bond = slave->bond;
1151
1152         recv_probe = ACCESS_ONCE(bond->recv_probe);
1153         if (recv_probe) {
1154                 ret = recv_probe(skb, bond, slave);
1155                 if (ret == RX_HANDLER_CONSUMED) {
1156                         consume_skb(skb);
1157                         return ret;
1158                 }
1159         }
1160
1161         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1162                 return RX_HANDLER_EXACT;
1163         }
1164
1165         skb->dev = bond->dev;
1166
1167         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1168             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1169             skb->pkt_type == PACKET_HOST) {
1170
1171                 if (unlikely(skb_cow_head(skb,
1172                                           skb->data - skb_mac_header(skb)))) {
1173                         kfree_skb(skb);
1174                         return RX_HANDLER_CONSUMED;
1175                 }
1176                 ether_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr);
1177         }
1178
1179         return ret;
1180 }
1181
1182 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1183                                       struct net_device *slave_dev,
1184                                       struct slave *slave)
1185 {
1186         int err;
1187
1188         err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1189         if (err)
1190                 return err;
1191         slave_dev->flags |= IFF_SLAVE;
1192         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1193         return 0;
1194 }
1195
1196 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1197                                   struct net_device *slave_dev)
1198 {
1199         netdev_upper_dev_unlink(slave_dev, bond_dev);
1200         slave_dev->flags &= ~IFF_SLAVE;
1201         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1202 }
1203
1204 static struct slave *bond_alloc_slave(struct bonding *bond)
1205 {
1206         struct slave *slave = NULL;
1207
1208         slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1209         if (!slave)
1210                 return NULL;
1211
1212         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1213                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1214                                                GFP_KERNEL);
1215                 if (!SLAVE_AD_INFO(slave)) {
1216                         kfree(slave);
1217                         return NULL;
1218                 }
1219         }
1220         return slave;
1221 }
1222
1223 static void bond_free_slave(struct slave *slave)
1224 {
1225         struct bonding *bond = bond_get_bond_by_slave(slave);
1226
1227         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1228                 kfree(SLAVE_AD_INFO(slave));
1229
1230         kfree(slave);
1231 }
1232
1233 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1234 {
1235         info->bond_mode = BOND_MODE(bond);
1236         info->miimon = bond->params.miimon;
1237         info->num_slaves = bond->slave_cnt;
1238 }
1239
1240 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1241 {
1242         strcpy(info->slave_name, slave->dev->name);
1243         info->link = slave->link;
1244         info->state = bond_slave_state(slave);
1245         info->link_failure_count = slave->link_failure_count;
1246 }
1247
1248 static void bond_netdev_notify(struct net_device *dev,
1249                                struct netdev_bonding_info *info)
1250 {
1251         rtnl_lock();
1252         netdev_bonding_info_change(dev, info);
1253         rtnl_unlock();
1254 }
1255
1256 static void bond_netdev_notify_work(struct work_struct *_work)
1257 {
1258         struct netdev_notify_work *w =
1259                 container_of(_work, struct netdev_notify_work, work.work);
1260
1261         bond_netdev_notify(w->dev, &w->bonding_info);
1262         dev_put(w->dev);
1263         kfree(w);
1264 }
1265
1266 void bond_queue_slave_event(struct slave *slave)
1267 {
1268         struct bonding *bond = slave->bond;
1269         struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1270
1271         if (!nnw)
1272                 return;
1273
1274         dev_hold(slave->dev);
1275         nnw->dev = slave->dev;
1276         bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1277         bond_fill_ifbond(bond, &nnw->bonding_info.master);
1278         INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1279
1280         queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1281 }
1282
1283 /* enslave device <slave> to bond device <master> */
1284 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1285 {
1286         struct bonding *bond = netdev_priv(bond_dev);
1287         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1288         struct slave *new_slave = NULL, *prev_slave;
1289         struct sockaddr addr;
1290         int link_reporting;
1291         int res = 0, i;
1292
1293         if (!bond->params.use_carrier &&
1294             slave_dev->ethtool_ops->get_link == NULL &&
1295             slave_ops->ndo_do_ioctl == NULL) {
1296                 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1297                             slave_dev->name);
1298         }
1299
1300         /* already enslaved */
1301         if (slave_dev->flags & IFF_SLAVE) {
1302                 netdev_dbg(bond_dev, "Error: Device was already enslaved\n");
1303                 return -EBUSY;
1304         }
1305
1306         if (bond_dev == slave_dev) {
1307                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1308                 return -EPERM;
1309         }
1310
1311         /* vlan challenged mutual exclusion */
1312         /* no need to lock since we're protected by rtnl_lock */
1313         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1314                 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1315                            slave_dev->name);
1316                 if (vlan_uses_dev(bond_dev)) {
1317                         netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1318                                    slave_dev->name, bond_dev->name);
1319                         return -EPERM;
1320                 } else {
1321                         netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1322                                     slave_dev->name, slave_dev->name,
1323                                     bond_dev->name);
1324                 }
1325         } else {
1326                 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1327                            slave_dev->name);
1328         }
1329
1330         /* Old ifenslave binaries are no longer supported.  These can
1331          * be identified with moderate accuracy by the state of the slave:
1332          * the current ifenslave will set the interface down prior to
1333          * enslaving it; the old ifenslave will not.
1334          */
1335         if ((slave_dev->flags & IFF_UP)) {
1336                 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1337                            slave_dev->name);
1338                 res = -EPERM;
1339                 goto err_undo_flags;
1340         }
1341
1342         /* set bonding device ether type by slave - bonding netdevices are
1343          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1344          * there is a need to override some of the type dependent attribs/funcs.
1345          *
1346          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1347          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1348          */
1349         if (!bond_has_slaves(bond)) {
1350                 if (bond_dev->type != slave_dev->type) {
1351                         netdev_dbg(bond_dev, "change device type from %d to %d\n",
1352                                    bond_dev->type, slave_dev->type);
1353
1354                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1355                                                        bond_dev);
1356                         res = notifier_to_errno(res);
1357                         if (res) {
1358                                 netdev_err(bond_dev, "refused to change device type\n");
1359                                 res = -EBUSY;
1360                                 goto err_undo_flags;
1361                         }
1362
1363                         /* Flush unicast and multicast addresses */
1364                         dev_uc_flush(bond_dev);
1365                         dev_mc_flush(bond_dev);
1366
1367                         if (slave_dev->type != ARPHRD_ETHER)
1368                                 bond_setup_by_slave(bond_dev, slave_dev);
1369                         else {
1370                                 ether_setup(bond_dev);
1371                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1372                         }
1373
1374                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1375                                                  bond_dev);
1376                 }
1377         } else if (bond_dev->type != slave_dev->type) {
1378                 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1379                            slave_dev->name, slave_dev->type, bond_dev->type);
1380                 res = -EINVAL;
1381                 goto err_undo_flags;
1382         }
1383
1384         if (slave_ops->ndo_set_mac_address == NULL) {
1385                 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1386                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1387                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1388                         if (!bond_has_slaves(bond)) {
1389                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1390                                 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1391                         } else {
1392                                 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");
1393                                 res = -EOPNOTSUPP;
1394                                 goto err_undo_flags;
1395                         }
1396                 }
1397         }
1398
1399         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1400
1401         /* If this is the first slave, then we need to set the master's hardware
1402          * address to be the same as the slave's.
1403          */
1404         if (!bond_has_slaves(bond) &&
1405             bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1406                 bond_set_dev_addr(bond->dev, slave_dev);
1407
1408         new_slave = bond_alloc_slave(bond);
1409         if (!new_slave) {
1410                 res = -ENOMEM;
1411                 goto err_undo_flags;
1412         }
1413
1414         new_slave->bond = bond;
1415         new_slave->dev = slave_dev;
1416         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1417          * is set via sysfs or module option if desired.
1418          */
1419         new_slave->queue_id = 0;
1420
1421         /* Save slave's original mtu and then set it to match the bond */
1422         new_slave->original_mtu = slave_dev->mtu;
1423         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1424         if (res) {
1425                 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1426                 goto err_free;
1427         }
1428
1429         /* Save slave's original ("permanent") mac address for modes
1430          * that need it, and for restoring it upon release, and then
1431          * set it to the master's address
1432          */
1433         ether_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr);
1434
1435         if (!bond->params.fail_over_mac ||
1436             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1437                 /* Set slave to master's mac address.  The application already
1438                  * set the master's mac address to that of the first slave
1439                  */
1440                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1441                 addr.sa_family = slave_dev->type;
1442                 res = dev_set_mac_address(slave_dev, &addr);
1443                 if (res) {
1444                         netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1445                         goto err_restore_mtu;
1446                 }
1447         }
1448
1449         /* open the slave since the application closed it */
1450         res = dev_open(slave_dev);
1451         if (res) {
1452                 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1453                 goto err_restore_mac;
1454         }
1455
1456         slave_dev->priv_flags |= IFF_BONDING;
1457         /* initialize slave stats */
1458         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1459
1460         if (bond_is_lb(bond)) {
1461                 /* bond_alb_init_slave() must be called before all other stages since
1462                  * it might fail and we do not want to have to undo everything
1463                  */
1464                 res = bond_alb_init_slave(bond, new_slave);
1465                 if (res)
1466                         goto err_close;
1467         }
1468
1469         /* If the mode uses primary, then the following is handled by
1470          * bond_change_active_slave().
1471          */
1472         if (!bond_uses_primary(bond)) {
1473                 /* set promiscuity level to new slave */
1474                 if (bond_dev->flags & IFF_PROMISC) {
1475                         res = dev_set_promiscuity(slave_dev, 1);
1476                         if (res)
1477                                 goto err_close;
1478                 }
1479
1480                 /* set allmulti level to new slave */
1481                 if (bond_dev->flags & IFF_ALLMULTI) {
1482                         res = dev_set_allmulti(slave_dev, 1);
1483                         if (res)
1484                                 goto err_close;
1485                 }
1486
1487                 netif_addr_lock_bh(bond_dev);
1488
1489                 dev_mc_sync_multiple(slave_dev, bond_dev);
1490                 dev_uc_sync_multiple(slave_dev, bond_dev);
1491
1492                 netif_addr_unlock_bh(bond_dev);
1493         }
1494
1495         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1496                 /* add lacpdu mc addr to mc list */
1497                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1498
1499                 dev_mc_add(slave_dev, lacpdu_multicast);
1500         }
1501
1502         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1503         if (res) {
1504                 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1505                            slave_dev->name);
1506                 goto err_close;
1507         }
1508
1509         prev_slave = bond_last_slave(bond);
1510
1511         new_slave->delay = 0;
1512         new_slave->link_failure_count = 0;
1513
1514         bond_update_speed_duplex(new_slave);
1515
1516         new_slave->last_rx = jiffies -
1517                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1518         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1519                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1520
1521         if (bond->params.miimon && !bond->params.use_carrier) {
1522                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1523
1524                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1525                         /* miimon is set but a bonded network driver
1526                          * does not support ETHTOOL/MII and
1527                          * arp_interval is not set.  Note: if
1528                          * use_carrier is enabled, we will never go
1529                          * here (because netif_carrier is always
1530                          * supported); thus, we don't need to change
1531                          * the messages for netif_carrier.
1532                          */
1533                         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",
1534                                     slave_dev->name);
1535                 } else if (link_reporting == -1) {
1536                         /* unable get link status using mii/ethtool */
1537                         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",
1538                                     slave_dev->name);
1539                 }
1540         }
1541
1542         /* check for initial state */
1543         if (bond->params.miimon) {
1544                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1545                         if (bond->params.updelay) {
1546                                 bond_set_slave_link_state(new_slave,
1547                                                           BOND_LINK_BACK);
1548                                 new_slave->delay = bond->params.updelay;
1549                         } else {
1550                                 bond_set_slave_link_state(new_slave,
1551                                                           BOND_LINK_UP);
1552                         }
1553                 } else {
1554                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN);
1555                 }
1556         } else if (bond->params.arp_interval) {
1557                 bond_set_slave_link_state(new_slave,
1558                                           (netif_carrier_ok(slave_dev) ?
1559                                           BOND_LINK_UP : BOND_LINK_DOWN));
1560         } else {
1561                 bond_set_slave_link_state(new_slave, BOND_LINK_UP);
1562         }
1563
1564         if (new_slave->link != BOND_LINK_DOWN)
1565                 new_slave->last_link_up = jiffies;
1566         netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1567                    new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1568                    (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1569
1570         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1571                 /* if there is a primary slave, remember it */
1572                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1573                         rcu_assign_pointer(bond->primary_slave, new_slave);
1574                         bond->force_primary = true;
1575                 }
1576         }
1577
1578         switch (BOND_MODE(bond)) {
1579         case BOND_MODE_ACTIVEBACKUP:
1580                 bond_set_slave_inactive_flags(new_slave,
1581                                               BOND_SLAVE_NOTIFY_NOW);
1582                 break;
1583         case BOND_MODE_8023AD:
1584                 /* in 802.3ad mode, the internal mechanism
1585                  * will activate the slaves in the selected
1586                  * aggregator
1587                  */
1588                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1589                 /* if this is the first slave */
1590                 if (!prev_slave) {
1591                         SLAVE_AD_INFO(new_slave)->id = 1;
1592                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1593                          * can be called only after the mac address of the bond is set
1594                          */
1595                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1596                 } else {
1597                         SLAVE_AD_INFO(new_slave)->id =
1598                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1599                 }
1600
1601                 bond_3ad_bind_slave(new_slave);
1602                 break;
1603         case BOND_MODE_TLB:
1604         case BOND_MODE_ALB:
1605                 bond_set_active_slave(new_slave);
1606                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1607                 break;
1608         default:
1609                 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1610
1611                 /* always active in trunk mode */
1612                 bond_set_active_slave(new_slave);
1613
1614                 /* In trunking mode there is little meaning to curr_active_slave
1615                  * anyway (it holds no special properties of the bond device),
1616                  * so we can change it without calling change_active_interface()
1617                  */
1618                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1619                     new_slave->link == BOND_LINK_UP)
1620                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1621
1622                 break;
1623         } /* switch(bond_mode) */
1624
1625 #ifdef CONFIG_NET_POLL_CONTROLLER
1626         slave_dev->npinfo = bond->dev->npinfo;
1627         if (slave_dev->npinfo) {
1628                 if (slave_enable_netpoll(new_slave)) {
1629                         netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1630                         res = -EBUSY;
1631                         goto err_detach;
1632                 }
1633         }
1634 #endif
1635
1636         if (!(bond_dev->features & NETIF_F_LRO))
1637                 dev_disable_lro(slave_dev);
1638
1639         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1640                                          new_slave);
1641         if (res) {
1642                 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1643                 goto err_detach;
1644         }
1645
1646         res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1647         if (res) {
1648                 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1649                 goto err_unregister;
1650         }
1651
1652         res = bond_sysfs_slave_add(new_slave);
1653         if (res) {
1654                 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1655                 goto err_upper_unlink;
1656         }
1657
1658         bond->slave_cnt++;
1659         bond_compute_features(bond);
1660         bond_set_carrier(bond);
1661
1662         if (bond_uses_primary(bond)) {
1663                 block_netpoll_tx();
1664                 bond_select_active_slave(bond);
1665                 unblock_netpoll_tx();
1666         }
1667
1668         if (bond_mode_uses_xmit_hash(bond))
1669                 bond_update_slave_arr(bond, NULL);
1670
1671         netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1672                     slave_dev->name,
1673                     bond_is_active_slave(new_slave) ? "an active" : "a backup",
1674                     new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1675
1676         /* enslave is successful */
1677         bond_queue_slave_event(new_slave);
1678         return 0;
1679
1680 /* Undo stages on error */
1681 err_upper_unlink:
1682         bond_upper_dev_unlink(bond_dev, slave_dev);
1683
1684 err_unregister:
1685         netdev_rx_handler_unregister(slave_dev);
1686
1687 err_detach:
1688         if (!bond_uses_primary(bond))
1689                 bond_hw_addr_flush(bond_dev, slave_dev);
1690
1691         vlan_vids_del_by_dev(slave_dev, bond_dev);
1692         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1693                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1694         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1695                 block_netpoll_tx();
1696                 bond_change_active_slave(bond, NULL);
1697                 bond_select_active_slave(bond);
1698                 unblock_netpoll_tx();
1699         }
1700         /* either primary_slave or curr_active_slave might've changed */
1701         synchronize_rcu();
1702         slave_disable_netpoll(new_slave);
1703
1704 err_close:
1705         slave_dev->priv_flags &= ~IFF_BONDING;
1706         dev_close(slave_dev);
1707
1708 err_restore_mac:
1709         if (!bond->params.fail_over_mac ||
1710             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1711                 /* XXX TODO - fom follow mode needs to change master's
1712                  * MAC if this slave's MAC is in use by the bond, or at
1713                  * least print a warning.
1714                  */
1715                 ether_addr_copy(addr.sa_data, new_slave->perm_hwaddr);
1716                 addr.sa_family = slave_dev->type;
1717                 dev_set_mac_address(slave_dev, &addr);
1718         }
1719
1720 err_restore_mtu:
1721         dev_set_mtu(slave_dev, new_slave->original_mtu);
1722
1723 err_free:
1724         bond_free_slave(new_slave);
1725
1726 err_undo_flags:
1727         /* Enslave of first slave has failed and we need to fix master's mac */
1728         if (!bond_has_slaves(bond) &&
1729             ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1730                 eth_hw_addr_random(bond_dev);
1731
1732         return res;
1733 }
1734
1735 /* Try to release the slave device <slave> from the bond device <master>
1736  * It is legal to access curr_active_slave without a lock because all the function
1737  * is RTNL-locked. If "all" is true it means that the function is being called
1738  * while destroying a bond interface and all slaves are being released.
1739  *
1740  * The rules for slave state should be:
1741  *   for Active/Backup:
1742  *     Active stays on all backups go down
1743  *   for Bonded connections:
1744  *     The first up interface should be left on and all others downed.
1745  */
1746 static int __bond_release_one(struct net_device *bond_dev,
1747                               struct net_device *slave_dev,
1748                               bool all)
1749 {
1750         struct bonding *bond = netdev_priv(bond_dev);
1751         struct slave *slave, *oldcurrent;
1752         struct sockaddr addr;
1753         int old_flags = bond_dev->flags;
1754         netdev_features_t old_features = bond_dev->features;
1755
1756         /* slave is not a slave or master is not master of this slave */
1757         if (!(slave_dev->flags & IFF_SLAVE) ||
1758             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1759                 netdev_dbg(bond_dev, "cannot release %s\n",
1760                            slave_dev->name);
1761                 return -EINVAL;
1762         }
1763
1764         block_netpoll_tx();
1765
1766         slave = bond_get_slave_by_dev(bond, slave_dev);
1767         if (!slave) {
1768                 /* not a slave of this bond */
1769                 netdev_info(bond_dev, "%s not enslaved\n",
1770                             slave_dev->name);
1771                 unblock_netpoll_tx();
1772                 return -EINVAL;
1773         }
1774
1775         bond_sysfs_slave_del(slave);
1776
1777         /* recompute stats just before removing the slave */
1778         bond_get_stats(bond->dev, &bond->bond_stats);
1779
1780         bond_upper_dev_unlink(bond_dev, slave_dev);
1781         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1782          * for this slave anymore.
1783          */
1784         netdev_rx_handler_unregister(slave_dev);
1785
1786         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1787                 bond_3ad_unbind_slave(slave);
1788
1789         if (bond_mode_uses_xmit_hash(bond))
1790                 bond_update_slave_arr(bond, slave);
1791
1792         netdev_info(bond_dev, "Releasing %s interface %s\n",
1793                     bond_is_active_slave(slave) ? "active" : "backup",
1794                     slave_dev->name);
1795
1796         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1797
1798         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1799
1800         if (!all && (!bond->params.fail_over_mac ||
1801                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1802                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1803                     bond_has_slaves(bond))
1804                         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",
1805                                     slave_dev->name, slave->perm_hwaddr,
1806                                     bond_dev->name, slave_dev->name);
1807         }
1808
1809         if (rtnl_dereference(bond->primary_slave) == slave)
1810                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1811
1812         if (oldcurrent == slave)
1813                 bond_change_active_slave(bond, NULL);
1814
1815         if (bond_is_lb(bond)) {
1816                 /* Must be called only after the slave has been
1817                  * detached from the list and the curr_active_slave
1818                  * has been cleared (if our_slave == old_current),
1819                  * but before a new active slave is selected.
1820                  */
1821                 bond_alb_deinit_slave(bond, slave);
1822         }
1823
1824         if (all) {
1825                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1826         } else if (oldcurrent == slave) {
1827                 /* Note that we hold RTNL over this sequence, so there
1828                  * is no concern that another slave add/remove event
1829                  * will interfere.
1830                  */
1831                 bond_select_active_slave(bond);
1832         }
1833
1834         if (!bond_has_slaves(bond)) {
1835                 bond_set_carrier(bond);
1836                 eth_hw_addr_random(bond_dev);
1837         }
1838
1839         unblock_netpoll_tx();
1840         synchronize_rcu();
1841         bond->slave_cnt--;
1842
1843         if (!bond_has_slaves(bond)) {
1844                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1845                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1846         }
1847
1848         bond_compute_features(bond);
1849         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1850             (old_features & NETIF_F_VLAN_CHALLENGED))
1851                 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1852                             slave_dev->name, bond_dev->name);
1853
1854         vlan_vids_del_by_dev(slave_dev, bond_dev);
1855
1856         /* If the mode uses primary, then this case was handled above by
1857          * bond_change_active_slave(..., NULL)
1858          */
1859         if (!bond_uses_primary(bond)) {
1860                 /* unset promiscuity level from slave
1861                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1862                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1863                  * value of that flag before that change, as that was the value
1864                  * when this slave was attached, so we cache at the start of the
1865                  * function and use it here. Same goes for ALLMULTI below
1866                  */
1867                 if (old_flags & IFF_PROMISC)
1868                         dev_set_promiscuity(slave_dev, -1);
1869
1870                 /* unset allmulti level from slave */
1871                 if (old_flags & IFF_ALLMULTI)
1872                         dev_set_allmulti(slave_dev, -1);
1873
1874                 bond_hw_addr_flush(bond_dev, slave_dev);
1875         }
1876
1877         slave_disable_netpoll(slave);
1878
1879         /* close slave before restoring its mac address */
1880         dev_close(slave_dev);
1881
1882         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1883             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1884                 /* restore original ("permanent") mac address */
1885                 ether_addr_copy(addr.sa_data, slave->perm_hwaddr);
1886                 addr.sa_family = slave_dev->type;
1887                 dev_set_mac_address(slave_dev, &addr);
1888         }
1889
1890         dev_set_mtu(slave_dev, slave->original_mtu);
1891
1892         slave_dev->priv_flags &= ~IFF_BONDING;
1893
1894         bond_free_slave(slave);
1895
1896         return 0;
1897 }
1898
1899 /* A wrapper used because of ndo_del_link */
1900 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1901 {
1902         return __bond_release_one(bond_dev, slave_dev, false);
1903 }
1904
1905 /* First release a slave and then destroy the bond if no more slaves are left.
1906  * Must be under rtnl_lock when this function is called.
1907  */
1908 static int  bond_release_and_destroy(struct net_device *bond_dev,
1909                                      struct net_device *slave_dev)
1910 {
1911         struct bonding *bond = netdev_priv(bond_dev);
1912         int ret;
1913
1914         ret = bond_release(bond_dev, slave_dev);
1915         if (ret == 0 && !bond_has_slaves(bond)) {
1916                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1917                 netdev_info(bond_dev, "Destroying bond %s\n",
1918                             bond_dev->name);
1919                 unregister_netdevice(bond_dev);
1920         }
1921         return ret;
1922 }
1923
1924 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1925 {
1926         struct bonding *bond = netdev_priv(bond_dev);
1927         bond_fill_ifbond(bond, info);
1928         return 0;
1929 }
1930
1931 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1932 {
1933         struct bonding *bond = netdev_priv(bond_dev);
1934         struct list_head *iter;
1935         int i = 0, res = -ENODEV;
1936         struct slave *slave;
1937
1938         bond_for_each_slave(bond, slave, iter) {
1939                 if (i++ == (int)info->slave_id) {
1940                         res = 0;
1941                         bond_fill_ifslave(slave, info);
1942                         break;
1943                 }
1944         }
1945
1946         return res;
1947 }
1948
1949 /*-------------------------------- Monitoring -------------------------------*/
1950
1951 /* called with rcu_read_lock() */
1952 static int bond_miimon_inspect(struct bonding *bond)
1953 {
1954         int link_state, commit = 0;
1955         struct list_head *iter;
1956         struct slave *slave;
1957         bool ignore_updelay;
1958
1959         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
1960
1961         bond_for_each_slave_rcu(bond, slave, iter) {
1962                 slave->new_link = BOND_LINK_NOCHANGE;
1963
1964                 link_state = bond_check_dev_link(bond, slave->dev, 0);
1965
1966                 switch (slave->link) {
1967                 case BOND_LINK_UP:
1968                         if (link_state)
1969                                 continue;
1970
1971                         bond_set_slave_link_state(slave, BOND_LINK_FAIL);
1972                         slave->delay = bond->params.downdelay;
1973                         if (slave->delay) {
1974                                 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
1975                                             (BOND_MODE(bond) ==
1976                                              BOND_MODE_ACTIVEBACKUP) ?
1977                                              (bond_is_active_slave(slave) ?
1978                                               "active " : "backup ") : "",
1979                                             slave->dev->name,
1980                                             bond->params.downdelay * bond->params.miimon);
1981                         }
1982                         /*FALLTHRU*/
1983                 case BOND_LINK_FAIL:
1984                         if (link_state) {
1985                                 /* recovered before downdelay expired */
1986                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
1987                                 slave->last_link_up = jiffies;
1988                                 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
1989                                             (bond->params.downdelay - slave->delay) *
1990                                             bond->params.miimon,
1991                                             slave->dev->name);
1992                                 continue;
1993                         }
1994
1995                         if (slave->delay <= 0) {
1996                                 slave->new_link = BOND_LINK_DOWN;
1997                                 commit++;
1998                                 continue;
1999                         }
2000
2001                         slave->delay--;
2002                         break;
2003
2004                 case BOND_LINK_DOWN:
2005                         if (!link_state)
2006                                 continue;
2007
2008                         bond_set_slave_link_state(slave, BOND_LINK_BACK);
2009                         slave->delay = bond->params.updelay;
2010
2011                         if (slave->delay) {
2012                                 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2013                                             slave->dev->name,
2014                                             ignore_updelay ? 0 :
2015                                             bond->params.updelay *
2016                                             bond->params.miimon);
2017                         }
2018                         /*FALLTHRU*/
2019                 case BOND_LINK_BACK:
2020                         if (!link_state) {
2021                                 bond_set_slave_link_state(slave,
2022                                                           BOND_LINK_DOWN);
2023                                 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2024                                             (bond->params.updelay - slave->delay) *
2025                                             bond->params.miimon,
2026                                             slave->dev->name);
2027
2028                                 continue;
2029                         }
2030
2031                         if (ignore_updelay)
2032                                 slave->delay = 0;
2033
2034                         if (slave->delay <= 0) {
2035                                 slave->new_link = BOND_LINK_UP;
2036                                 commit++;
2037                                 ignore_updelay = false;
2038                                 continue;
2039                         }
2040
2041                         slave->delay--;
2042                         break;
2043                 }
2044         }
2045
2046         return commit;
2047 }
2048
2049 static void bond_miimon_commit(struct bonding *bond)
2050 {
2051         struct list_head *iter;
2052         struct slave *slave, *primary;
2053
2054         bond_for_each_slave(bond, slave, iter) {
2055                 switch (slave->new_link) {
2056                 case BOND_LINK_NOCHANGE:
2057                         continue;
2058
2059                 case BOND_LINK_UP:
2060                         bond_set_slave_link_state(slave, BOND_LINK_UP);
2061                         slave->last_link_up = jiffies;
2062
2063                         primary = rtnl_dereference(bond->primary_slave);
2064                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2065                                 /* prevent it from being the active one */
2066                                 bond_set_backup_slave(slave);
2067                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2068                                 /* make it immediately active */
2069                                 bond_set_active_slave(slave);
2070                         } else if (slave != primary) {
2071                                 /* prevent it from being the active one */
2072                                 bond_set_backup_slave(slave);
2073                         }
2074
2075                         netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2076                                     slave->dev->name,
2077                                     slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2078                                     slave->duplex ? "full" : "half");
2079
2080                         /* notify ad that the link status has changed */
2081                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2082                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2083
2084                         if (bond_is_lb(bond))
2085                                 bond_alb_handle_link_change(bond, slave,
2086                                                             BOND_LINK_UP);
2087
2088                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2089                                 bond_update_slave_arr(bond, NULL);
2090
2091                         if (!bond->curr_active_slave || slave == primary)
2092                                 goto do_failover;
2093
2094                         continue;
2095
2096                 case BOND_LINK_DOWN:
2097                         if (slave->link_failure_count < UINT_MAX)
2098                                 slave->link_failure_count++;
2099
2100                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2101
2102                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2103                             BOND_MODE(bond) == BOND_MODE_8023AD)
2104                                 bond_set_slave_inactive_flags(slave,
2105                                                               BOND_SLAVE_NOTIFY_NOW);
2106
2107                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2108                                     slave->dev->name);
2109
2110                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2111                                 bond_3ad_handle_link_change(slave,
2112                                                             BOND_LINK_DOWN);
2113
2114                         if (bond_is_lb(bond))
2115                                 bond_alb_handle_link_change(bond, slave,
2116                                                             BOND_LINK_DOWN);
2117
2118                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2119                                 bond_update_slave_arr(bond, NULL);
2120
2121                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2122                                 goto do_failover;
2123
2124                         continue;
2125
2126                 default:
2127                         netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2128                                    slave->new_link, slave->dev->name);
2129                         slave->new_link = BOND_LINK_NOCHANGE;
2130
2131                         continue;
2132                 }
2133
2134 do_failover:
2135                 block_netpoll_tx();
2136                 bond_select_active_slave(bond);
2137                 unblock_netpoll_tx();
2138         }
2139
2140         bond_set_carrier(bond);
2141 }
2142
2143 /* bond_mii_monitor
2144  *
2145  * Really a wrapper that splits the mii monitor into two phases: an
2146  * inspection, then (if inspection indicates something needs to be done)
2147  * an acquisition of appropriate locks followed by a commit phase to
2148  * implement whatever link state changes are indicated.
2149  */
2150 static void bond_mii_monitor(struct work_struct *work)
2151 {
2152         struct bonding *bond = container_of(work, struct bonding,
2153                                             mii_work.work);
2154         bool should_notify_peers = false;
2155         unsigned long delay;
2156
2157         delay = msecs_to_jiffies(bond->params.miimon);
2158
2159         if (!bond_has_slaves(bond))
2160                 goto re_arm;
2161
2162         rcu_read_lock();
2163
2164         should_notify_peers = bond_should_notify_peers(bond);
2165
2166         if (bond_miimon_inspect(bond)) {
2167                 rcu_read_unlock();
2168
2169                 /* Race avoidance with bond_close cancel of workqueue */
2170                 if (!rtnl_trylock()) {
2171                         delay = 1;
2172                         should_notify_peers = false;
2173                         goto re_arm;
2174                 }
2175
2176                 bond_miimon_commit(bond);
2177
2178                 rtnl_unlock();  /* might sleep, hold no other locks */
2179         } else
2180                 rcu_read_unlock();
2181
2182 re_arm:
2183         if (bond->params.miimon)
2184                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2185
2186         if (should_notify_peers) {
2187                 if (!rtnl_trylock())
2188                         return;
2189                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2190                 rtnl_unlock();
2191         }
2192 }
2193
2194 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2195 {
2196         struct net_device *upper;
2197         struct list_head *iter;
2198         bool ret = false;
2199
2200         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2201                 return true;
2202
2203         rcu_read_lock();
2204         netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2205                 if (ip == bond_confirm_addr(upper, 0, ip)) {
2206                         ret = true;
2207                         break;
2208                 }
2209         }
2210         rcu_read_unlock();
2211
2212         return ret;
2213 }
2214
2215 /* We go to the (large) trouble of VLAN tagging ARP frames because
2216  * switches in VLAN mode (especially if ports are configured as
2217  * "native" to a VLAN) might not pass non-tagged frames.
2218  */
2219 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2220                           __be32 dest_ip, __be32 src_ip,
2221                           struct bond_vlan_tag *tags)
2222 {
2223         struct sk_buff *skb;
2224         struct bond_vlan_tag *outer_tag = tags;
2225
2226         netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2227                    arp_op, slave_dev->name, &dest_ip, &src_ip);
2228
2229         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2230                          NULL, slave_dev->dev_addr, NULL);
2231
2232         if (!skb) {
2233                 net_err_ratelimited("ARP packet allocation failed\n");
2234                 return;
2235         }
2236
2237         if (!tags || tags->vlan_proto == VLAN_N_VID)
2238                 goto xmit;
2239
2240         tags++;
2241
2242         /* Go through all the tags backwards and add them to the packet */
2243         while (tags->vlan_proto != VLAN_N_VID) {
2244                 if (!tags->vlan_id) {
2245                         tags++;
2246                         continue;
2247                 }
2248
2249                 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2250                            ntohs(outer_tag->vlan_proto), tags->vlan_id);
2251                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2252                                                 tags->vlan_id);
2253                 if (!skb) {
2254                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2255                         return;
2256                 }
2257
2258                 tags++;
2259         }
2260         /* Set the outer tag */
2261         if (outer_tag->vlan_id) {
2262                 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2263                            ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2264                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2265                                        outer_tag->vlan_id);
2266         }
2267
2268 xmit:
2269         arp_xmit(skb);
2270 }
2271
2272 /* Validate the device path between the @start_dev and the @end_dev.
2273  * The path is valid if the @end_dev is reachable through device
2274  * stacking.
2275  * When the path is validated, collect any vlan information in the
2276  * path.
2277  */
2278 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2279                                               struct net_device *end_dev,
2280                                               int level)
2281 {
2282         struct bond_vlan_tag *tags;
2283         struct net_device *upper;
2284         struct list_head  *iter;
2285
2286         if (start_dev == end_dev) {
2287                 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2288                 if (!tags)
2289                         return ERR_PTR(-ENOMEM);
2290                 tags[level].vlan_proto = VLAN_N_VID;
2291                 return tags;
2292         }
2293
2294         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2295                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2296                 if (IS_ERR_OR_NULL(tags)) {
2297                         if (IS_ERR(tags))
2298                                 return tags;
2299                         continue;
2300                 }
2301                 if (is_vlan_dev(upper)) {
2302                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2303                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2304                 }
2305
2306                 return tags;
2307         }
2308
2309         return NULL;
2310 }
2311
2312 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2313 {
2314         struct rtable *rt;
2315         struct bond_vlan_tag *tags;
2316         __be32 *targets = bond->params.arp_targets, addr;
2317         int i;
2318
2319         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2320                 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2321                 tags = NULL;
2322
2323                 /* Find out through which dev should the packet go */
2324                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2325                                      RTO_ONLINK, 0);
2326                 if (IS_ERR(rt)) {
2327                         /* there's no route to target - try to send arp
2328                          * probe to generate any traffic (arp_validate=0)
2329                          */
2330                         if (bond->params.arp_validate)
2331                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2332                                                      bond->dev->name,
2333                                                      &targets[i]);
2334                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2335                                       0, tags);
2336                         continue;
2337                 }
2338
2339                 /* bond device itself */
2340                 if (rt->dst.dev == bond->dev)
2341                         goto found;
2342
2343                 rcu_read_lock();
2344                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2345                 rcu_read_unlock();
2346
2347                 if (!IS_ERR_OR_NULL(tags))
2348                         goto found;
2349
2350                 /* Not our device - skip */
2351                 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2352                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2353
2354                 ip_rt_put(rt);
2355                 continue;
2356
2357 found:
2358                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2359                 ip_rt_put(rt);
2360                 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2361                               addr, tags);
2362                 kfree(tags);
2363         }
2364 }
2365
2366 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2367 {
2368         int i;
2369
2370         if (!sip || !bond_has_this_ip(bond, tip)) {
2371                 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2372                            &sip, &tip);
2373                 return;
2374         }
2375
2376         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2377         if (i == -1) {
2378                 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2379                            &sip);
2380                 return;
2381         }
2382         slave->last_rx = jiffies;
2383         slave->target_last_arp_rx[i] = jiffies;
2384 }
2385
2386 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2387                  struct slave *slave)
2388 {
2389         struct arphdr *arp = (struct arphdr *)skb->data;
2390         struct slave *curr_active_slave;
2391         unsigned char *arp_ptr;
2392         __be32 sip, tip;
2393         int alen, is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2394
2395         if (!slave_do_arp_validate(bond, slave)) {
2396                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2397                     !slave_do_arp_validate_only(bond))
2398                         slave->last_rx = jiffies;
2399                 return RX_HANDLER_ANOTHER;
2400         } else if (!is_arp) {
2401                 return RX_HANDLER_ANOTHER;
2402         }
2403
2404         alen = arp_hdr_len(bond->dev);
2405
2406         netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2407                    skb->dev->name);
2408
2409         if (alen > skb_headlen(skb)) {
2410                 arp = kmalloc(alen, GFP_ATOMIC);
2411                 if (!arp)
2412                         goto out_unlock;
2413                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2414                         goto out_unlock;
2415         }
2416
2417         if (arp->ar_hln != bond->dev->addr_len ||
2418             skb->pkt_type == PACKET_OTHERHOST ||
2419             skb->pkt_type == PACKET_LOOPBACK ||
2420             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2421             arp->ar_pro != htons(ETH_P_IP) ||
2422             arp->ar_pln != 4)
2423                 goto out_unlock;
2424
2425         arp_ptr = (unsigned char *)(arp + 1);
2426         arp_ptr += bond->dev->addr_len;
2427         memcpy(&sip, arp_ptr, 4);
2428         arp_ptr += 4 + bond->dev->addr_len;
2429         memcpy(&tip, arp_ptr, 4);
2430
2431         netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2432                    slave->dev->name, bond_slave_state(slave),
2433                      bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2434                      &sip, &tip);
2435
2436         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2437
2438         /* Backup slaves won't see the ARP reply, but do come through
2439          * here for each ARP probe (so we swap the sip/tip to validate
2440          * the probe).  In a "redundant switch, common router" type of
2441          * configuration, the ARP probe will (hopefully) travel from
2442          * the active, through one switch, the router, then the other
2443          * switch before reaching the backup.
2444          *
2445          * We 'trust' the arp requests if there is an active slave and
2446          * it received valid arp reply(s) after it became active. This
2447          * is done to avoid endless looping when we can't reach the
2448          * arp_ip_target and fool ourselves with our own arp requests.
2449          */
2450
2451         if (bond_is_active_slave(slave))
2452                 bond_validate_arp(bond, slave, sip, tip);
2453         else if (curr_active_slave &&
2454                  time_after(slave_last_rx(bond, curr_active_slave),
2455                             curr_active_slave->last_link_up))
2456                 bond_validate_arp(bond, slave, tip, sip);
2457
2458 out_unlock:
2459         if (arp != (struct arphdr *)skb->data)
2460                 kfree(arp);
2461         return RX_HANDLER_ANOTHER;
2462 }
2463
2464 /* function to verify if we're in the arp_interval timeslice, returns true if
2465  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2466  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2467  */
2468 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2469                                   int mod)
2470 {
2471         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2472
2473         return time_in_range(jiffies,
2474                              last_act - delta_in_ticks,
2475                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2476 }
2477
2478 /* This function is called regularly to monitor each slave's link
2479  * ensuring that traffic is being sent and received when arp monitoring
2480  * is used in load-balancing mode. if the adapter has been dormant, then an
2481  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2482  * arp monitoring in active backup mode.
2483  */
2484 static void bond_loadbalance_arp_mon(struct work_struct *work)
2485 {
2486         struct bonding *bond = container_of(work, struct bonding,
2487                                             arp_work.work);
2488         struct slave *slave, *oldcurrent;
2489         struct list_head *iter;
2490         int do_failover = 0, slave_state_changed = 0;
2491
2492         if (!bond_has_slaves(bond))
2493                 goto re_arm;
2494
2495         rcu_read_lock();
2496
2497         oldcurrent = rcu_dereference(bond->curr_active_slave);
2498         /* see if any of the previous devices are up now (i.e. they have
2499          * xmt and rcv traffic). the curr_active_slave does not come into
2500          * the picture unless it is null. also, slave->last_link_up is not
2501          * needed here because we send an arp on each slave and give a slave
2502          * as long as it needs to get the tx/rx within the delta.
2503          * TODO: what about up/down delay in arp mode? it wasn't here before
2504          *       so it can wait
2505          */
2506         bond_for_each_slave_rcu(bond, slave, iter) {
2507                 unsigned long trans_start = dev_trans_start(slave->dev);
2508
2509                 if (slave->link != BOND_LINK_UP) {
2510                         if (bond_time_in_interval(bond, trans_start, 1) &&
2511                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2512
2513                                 slave->link  = BOND_LINK_UP;
2514                                 slave_state_changed = 1;
2515
2516                                 /* primary_slave has no meaning in round-robin
2517                                  * mode. the window of a slave being up and
2518                                  * curr_active_slave being null after enslaving
2519                                  * is closed.
2520                                  */
2521                                 if (!oldcurrent) {
2522                                         netdev_info(bond->dev, "link status definitely up for interface %s\n",
2523                                                     slave->dev->name);
2524                                         do_failover = 1;
2525                                 } else {
2526                                         netdev_info(bond->dev, "interface %s is now up\n",
2527                                                     slave->dev->name);
2528                                 }
2529                         }
2530                 } else {
2531                         /* slave->link == BOND_LINK_UP */
2532
2533                         /* not all switches will respond to an arp request
2534                          * when the source ip is 0, so don't take the link down
2535                          * if we don't know our ip yet
2536                          */
2537                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2538                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2539
2540                                 slave->link  = BOND_LINK_DOWN;
2541                                 slave_state_changed = 1;
2542
2543                                 if (slave->link_failure_count < UINT_MAX)
2544                                         slave->link_failure_count++;
2545
2546                                 netdev_info(bond->dev, "interface %s is now down\n",
2547                                             slave->dev->name);
2548
2549                                 if (slave == oldcurrent)
2550                                         do_failover = 1;
2551                         }
2552                 }
2553
2554                 /* note: if switch is in round-robin mode, all links
2555                  * must tx arp to ensure all links rx an arp - otherwise
2556                  * links may oscillate or not come up at all; if switch is
2557                  * in something like xor mode, there is nothing we can
2558                  * do - all replies will be rx'ed on same link causing slaves
2559                  * to be unstable during low/no traffic periods
2560                  */
2561                 if (bond_slave_is_up(slave))
2562                         bond_arp_send_all(bond, slave);
2563         }
2564
2565         rcu_read_unlock();
2566
2567         if (do_failover || slave_state_changed) {
2568                 if (!rtnl_trylock())
2569                         goto re_arm;
2570
2571                 if (slave_state_changed) {
2572                         bond_slave_state_change(bond);
2573                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2574                                 bond_update_slave_arr(bond, NULL);
2575                 }
2576                 if (do_failover) {
2577                         block_netpoll_tx();
2578                         bond_select_active_slave(bond);
2579                         unblock_netpoll_tx();
2580                 }
2581                 rtnl_unlock();
2582         }
2583
2584 re_arm:
2585         if (bond->params.arp_interval)
2586                 queue_delayed_work(bond->wq, &bond->arp_work,
2587                                    msecs_to_jiffies(bond->params.arp_interval));
2588 }
2589
2590 /* Called to inspect slaves for active-backup mode ARP monitor link state
2591  * changes.  Sets new_link in slaves to specify what action should take
2592  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2593  * to link states must be committed.
2594  *
2595  * Called with rcu_read_lock held.
2596  */
2597 static int bond_ab_arp_inspect(struct bonding *bond)
2598 {
2599         unsigned long trans_start, last_rx;
2600         struct list_head *iter;
2601         struct slave *slave;
2602         int commit = 0;
2603
2604         bond_for_each_slave_rcu(bond, slave, iter) {
2605                 slave->new_link = BOND_LINK_NOCHANGE;
2606                 last_rx = slave_last_rx(bond, slave);
2607
2608                 if (slave->link != BOND_LINK_UP) {
2609                         if (bond_time_in_interval(bond, last_rx, 1)) {
2610                                 slave->new_link = BOND_LINK_UP;
2611                                 commit++;
2612                         }
2613                         continue;
2614                 }
2615
2616                 /* Give slaves 2*delta after being enslaved or made
2617                  * active.  This avoids bouncing, as the last receive
2618                  * times need a full ARP monitor cycle to be updated.
2619                  */
2620                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2621                         continue;
2622
2623                 /* Backup slave is down if:
2624                  * - No current_arp_slave AND
2625                  * - more than 3*delta since last receive AND
2626                  * - the bond has an IP address
2627                  *
2628                  * Note: a non-null current_arp_slave indicates
2629                  * the curr_active_slave went down and we are
2630                  * searching for a new one; under this condition
2631                  * we only take the curr_active_slave down - this
2632                  * gives each slave a chance to tx/rx traffic
2633                  * before being taken out
2634                  */
2635                 if (!bond_is_active_slave(slave) &&
2636                     !rcu_access_pointer(bond->current_arp_slave) &&
2637                     !bond_time_in_interval(bond, last_rx, 3)) {
2638                         slave->new_link = BOND_LINK_DOWN;
2639                         commit++;
2640                 }
2641
2642                 /* Active slave is down if:
2643                  * - more than 2*delta since transmitting OR
2644                  * - (more than 2*delta since receive AND
2645                  *    the bond has an IP address)
2646                  */
2647                 trans_start = dev_trans_start(slave->dev);
2648                 if (bond_is_active_slave(slave) &&
2649                     (!bond_time_in_interval(bond, trans_start, 2) ||
2650                      !bond_time_in_interval(bond, last_rx, 2))) {
2651                         slave->new_link = BOND_LINK_DOWN;
2652                         commit++;
2653                 }
2654         }
2655
2656         return commit;
2657 }
2658
2659 /* Called to commit link state changes noted by inspection step of
2660  * active-backup mode ARP monitor.
2661  *
2662  * Called with RTNL hold.
2663  */
2664 static void bond_ab_arp_commit(struct bonding *bond)
2665 {
2666         unsigned long trans_start;
2667         struct list_head *iter;
2668         struct slave *slave;
2669
2670         bond_for_each_slave(bond, slave, iter) {
2671                 switch (slave->new_link) {
2672                 case BOND_LINK_NOCHANGE:
2673                         continue;
2674
2675                 case BOND_LINK_UP:
2676                         trans_start = dev_trans_start(slave->dev);
2677                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2678                             (!rtnl_dereference(bond->curr_active_slave) &&
2679                              bond_time_in_interval(bond, trans_start, 1))) {
2680                                 struct slave *current_arp_slave;
2681
2682                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2683                                 bond_set_slave_link_state(slave, BOND_LINK_UP);
2684                                 if (current_arp_slave) {
2685                                         bond_set_slave_inactive_flags(
2686                                                 current_arp_slave,
2687                                                 BOND_SLAVE_NOTIFY_NOW);
2688                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2689                                 }
2690
2691                                 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2692                                             slave->dev->name);
2693
2694                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2695                                     slave == rtnl_dereference(bond->primary_slave))
2696                                         goto do_failover;
2697
2698                         }
2699
2700                         continue;
2701
2702                 case BOND_LINK_DOWN:
2703                         if (slave->link_failure_count < UINT_MAX)
2704                                 slave->link_failure_count++;
2705
2706                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2707                         bond_set_slave_inactive_flags(slave,
2708                                                       BOND_SLAVE_NOTIFY_NOW);
2709
2710                         netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2711                                     slave->dev->name);
2712
2713                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2714                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2715                                 goto do_failover;
2716                         }
2717
2718                         continue;
2719
2720                 default:
2721                         netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2722                                    slave->new_link, slave->dev->name);
2723                         continue;
2724                 }
2725
2726 do_failover:
2727                 block_netpoll_tx();
2728                 bond_select_active_slave(bond);
2729                 unblock_netpoll_tx();
2730         }
2731
2732         bond_set_carrier(bond);
2733 }
2734
2735 /* Send ARP probes for active-backup mode ARP monitor.
2736  *
2737  * Called with rcu_read_lock held.
2738  */
2739 static bool bond_ab_arp_probe(struct bonding *bond)
2740 {
2741         struct slave *slave, *before = NULL, *new_slave = NULL,
2742                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2743                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2744         struct list_head *iter;
2745         bool found = false;
2746         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2747
2748         if (curr_arp_slave && curr_active_slave)
2749                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2750                             curr_arp_slave->dev->name,
2751                             curr_active_slave->dev->name);
2752
2753         if (curr_active_slave) {
2754                 bond_arp_send_all(bond, curr_active_slave);
2755                 return should_notify_rtnl;
2756         }
2757
2758         /* if we don't have a curr_active_slave, search for the next available
2759          * backup slave from the current_arp_slave and make it the candidate
2760          * for becoming the curr_active_slave
2761          */
2762
2763         if (!curr_arp_slave) {
2764                 curr_arp_slave = bond_first_slave_rcu(bond);
2765                 if (!curr_arp_slave)
2766                         return should_notify_rtnl;
2767         }
2768
2769         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2770
2771         bond_for_each_slave_rcu(bond, slave, iter) {
2772                 if (!found && !before && bond_slave_is_up(slave))
2773                         before = slave;
2774
2775                 if (found && !new_slave && bond_slave_is_up(slave))
2776                         new_slave = slave;
2777                 /* if the link state is up at this point, we
2778                  * mark it down - this can happen if we have
2779                  * simultaneous link failures and
2780                  * reselect_active_interface doesn't make this
2781                  * one the current slave so it is still marked
2782                  * up when it is actually down
2783                  */
2784                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2785                         bond_set_slave_link_state(slave, BOND_LINK_DOWN);
2786                         if (slave->link_failure_count < UINT_MAX)
2787                                 slave->link_failure_count++;
2788
2789                         bond_set_slave_inactive_flags(slave,
2790                                                       BOND_SLAVE_NOTIFY_LATER);
2791
2792                         netdev_info(bond->dev, "backup interface %s is now down\n",
2793                                     slave->dev->name);
2794                 }
2795                 if (slave == curr_arp_slave)
2796                         found = true;
2797         }
2798
2799         if (!new_slave && before)
2800                 new_slave = before;
2801
2802         if (!new_slave)
2803                 goto check_state;
2804
2805         bond_set_slave_link_state(new_slave, BOND_LINK_BACK);
2806         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2807         bond_arp_send_all(bond, new_slave);
2808         new_slave->last_link_up = jiffies;
2809         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2810
2811 check_state:
2812         bond_for_each_slave_rcu(bond, slave, iter) {
2813                 if (slave->should_notify) {
2814                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2815                         break;
2816                 }
2817         }
2818         return should_notify_rtnl;
2819 }
2820
2821 static void bond_activebackup_arp_mon(struct work_struct *work)
2822 {
2823         struct bonding *bond = container_of(work, struct bonding,
2824                                             arp_work.work);
2825         bool should_notify_peers = false;
2826         bool should_notify_rtnl = false;
2827         int delta_in_ticks;
2828
2829         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2830
2831         if (!bond_has_slaves(bond))
2832                 goto re_arm;
2833
2834         rcu_read_lock();
2835
2836         should_notify_peers = bond_should_notify_peers(bond);
2837
2838         if (bond_ab_arp_inspect(bond)) {
2839                 rcu_read_unlock();
2840
2841                 /* Race avoidance with bond_close flush of workqueue */
2842                 if (!rtnl_trylock()) {
2843                         delta_in_ticks = 1;
2844                         should_notify_peers = false;
2845                         goto re_arm;
2846                 }
2847
2848                 bond_ab_arp_commit(bond);
2849
2850                 rtnl_unlock();
2851                 rcu_read_lock();
2852         }
2853
2854         should_notify_rtnl = bond_ab_arp_probe(bond);
2855         rcu_read_unlock();
2856
2857 re_arm:
2858         if (bond->params.arp_interval)
2859                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2860
2861         if (should_notify_peers || should_notify_rtnl) {
2862                 if (!rtnl_trylock())
2863                         return;
2864
2865                 if (should_notify_peers)
2866                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2867                                                  bond->dev);
2868                 if (should_notify_rtnl)
2869                         bond_slave_state_notify(bond);
2870
2871                 rtnl_unlock();
2872         }
2873 }
2874
2875 /*-------------------------- netdev event handling --------------------------*/
2876
2877 /* Change device name */
2878 static int bond_event_changename(struct bonding *bond)
2879 {
2880         bond_remove_proc_entry(bond);
2881         bond_create_proc_entry(bond);
2882
2883         bond_debug_reregister(bond);
2884
2885         return NOTIFY_DONE;
2886 }
2887
2888 static int bond_master_netdev_event(unsigned long event,
2889                                     struct net_device *bond_dev)
2890 {
2891         struct bonding *event_bond = netdev_priv(bond_dev);
2892
2893         switch (event) {
2894         case NETDEV_CHANGENAME:
2895                 return bond_event_changename(event_bond);
2896         case NETDEV_UNREGISTER:
2897                 bond_remove_proc_entry(event_bond);
2898                 break;
2899         case NETDEV_REGISTER:
2900                 bond_create_proc_entry(event_bond);
2901                 break;
2902         case NETDEV_NOTIFY_PEERS:
2903                 if (event_bond->send_peer_notif)
2904                         event_bond->send_peer_notif--;
2905                 break;
2906         default:
2907                 break;
2908         }
2909
2910         return NOTIFY_DONE;
2911 }
2912
2913 static int bond_slave_netdev_event(unsigned long event,
2914                                    struct net_device *slave_dev)
2915 {
2916         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
2917         struct bonding *bond;
2918         struct net_device *bond_dev;
2919         u32 old_speed;
2920         u8 old_duplex;
2921
2922         /* A netdev event can be generated while enslaving a device
2923          * before netdev_rx_handler_register is called in which case
2924          * slave will be NULL
2925          */
2926         if (!slave)
2927                 return NOTIFY_DONE;
2928         bond_dev = slave->bond->dev;
2929         bond = slave->bond;
2930         primary = rtnl_dereference(bond->primary_slave);
2931
2932         switch (event) {
2933         case NETDEV_UNREGISTER:
2934                 if (bond_dev->type != ARPHRD_ETHER)
2935                         bond_release_and_destroy(bond_dev, slave_dev);
2936                 else
2937                         bond_release(bond_dev, slave_dev);
2938                 break;
2939         case NETDEV_UP:
2940         case NETDEV_CHANGE:
2941                 old_speed = slave->speed;
2942                 old_duplex = slave->duplex;
2943
2944                 bond_update_speed_duplex(slave);
2945
2946                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2947                         if (old_speed != slave->speed)
2948                                 bond_3ad_adapter_speed_changed(slave);
2949                         if (old_duplex != slave->duplex)
2950                                 bond_3ad_adapter_duplex_changed(slave);
2951                 }
2952                 /* Fallthrough */
2953         case NETDEV_DOWN:
2954                 /* Refresh slave-array if applicable!
2955                  * If the setup does not use miimon or arpmon (mode-specific!),
2956                  * then these events will not cause the slave-array to be
2957                  * refreshed. This will cause xmit to use a slave that is not
2958                  * usable. Avoid such situation by refeshing the array at these
2959                  * events. If these (miimon/arpmon) parameters are configured
2960                  * then array gets refreshed twice and that should be fine!
2961                  */
2962                 if (bond_mode_uses_xmit_hash(bond))
2963                         bond_update_slave_arr(bond, NULL);
2964                 break;
2965         case NETDEV_CHANGEMTU:
2966                 /* TODO: Should slaves be allowed to
2967                  * independently alter their MTU?  For
2968                  * an active-backup bond, slaves need
2969                  * not be the same type of device, so
2970                  * MTUs may vary.  For other modes,
2971                  * slaves arguably should have the
2972                  * same MTUs. To do this, we'd need to
2973                  * take over the slave's change_mtu
2974                  * function for the duration of their
2975                  * servitude.
2976                  */
2977                 break;
2978         case NETDEV_CHANGENAME:
2979                 /* we don't care if we don't have primary set */
2980                 if (!bond_uses_primary(bond) ||
2981                     !bond->params.primary[0])
2982                         break;
2983
2984                 if (slave == primary) {
2985                         /* slave's name changed - he's no longer primary */
2986                         RCU_INIT_POINTER(bond->primary_slave, NULL);
2987                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2988                         /* we have a new primary slave */
2989                         rcu_assign_pointer(bond->primary_slave, slave);
2990                 } else { /* we didn't change primary - exit */
2991                         break;
2992                 }
2993
2994                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
2995                             primary ? slave_dev->name : "none");
2996
2997                 block_netpoll_tx();
2998                 bond_select_active_slave(bond);
2999                 unblock_netpoll_tx();
3000                 break;
3001         case NETDEV_FEAT_CHANGE:
3002                 bond_compute_features(bond);
3003                 break;
3004         case NETDEV_RESEND_IGMP:
3005                 /* Propagate to master device */
3006                 call_netdevice_notifiers(event, slave->bond->dev);
3007                 break;
3008         default:
3009                 break;
3010         }
3011
3012         return NOTIFY_DONE;
3013 }
3014
3015 /* bond_netdev_event: handle netdev notifier chain events.
3016  *
3017  * This function receives events for the netdev chain.  The caller (an
3018  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3019  * locks for us to safely manipulate the slave devices (RTNL lock,
3020  * dev_probe_lock).
3021  */
3022 static int bond_netdev_event(struct notifier_block *this,
3023                              unsigned long event, void *ptr)
3024 {
3025         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3026
3027         netdev_dbg(event_dev, "event: %lx\n", event);
3028
3029         if (!(event_dev->priv_flags & IFF_BONDING))
3030                 return NOTIFY_DONE;
3031
3032         if (event_dev->flags & IFF_MASTER) {
3033                 netdev_dbg(event_dev, "IFF_MASTER\n");
3034                 return bond_master_netdev_event(event, event_dev);
3035         }
3036
3037         if (event_dev->flags & IFF_SLAVE) {
3038                 netdev_dbg(event_dev, "IFF_SLAVE\n");
3039                 return bond_slave_netdev_event(event, event_dev);
3040         }
3041
3042         return NOTIFY_DONE;
3043 }
3044
3045 static struct notifier_block bond_netdev_notifier = {
3046         .notifier_call = bond_netdev_event,
3047 };
3048
3049 /*---------------------------- Hashing Policies -----------------------------*/
3050
3051 /* L2 hash helper */
3052 static inline u32 bond_eth_hash(struct sk_buff *skb)
3053 {
3054         struct ethhdr *ep, hdr_tmp;
3055
3056         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3057         if (ep)
3058                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3059         return 0;
3060 }
3061
3062 /* Extract the appropriate headers based on bond's xmit policy */
3063 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3064                               struct flow_keys *fk)
3065 {
3066         const struct ipv6hdr *iph6;
3067         const struct iphdr *iph;
3068         int noff, proto = -1;
3069
3070         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3071                 return skb_flow_dissect_flow_keys(skb, fk);
3072
3073         fk->ports.ports = 0;
3074         noff = skb_network_offset(skb);
3075         if (skb->protocol == htons(ETH_P_IP)) {
3076                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3077                         return false;
3078                 iph = ip_hdr(skb);
3079                 iph_to_flow_copy_v4addrs(fk, iph);
3080                 noff += iph->ihl << 2;
3081                 if (!ip_is_fragment(iph))
3082                         proto = iph->protocol;
3083         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3084                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3085                         return false;
3086                 iph6 = ipv6_hdr(skb);
3087                 iph_to_flow_copy_v6addrs(fk, iph6);
3088                 noff += sizeof(*iph6);
3089                 proto = iph6->nexthdr;
3090         } else {
3091                 return false;
3092         }
3093         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3094                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3095
3096         return true;
3097 }
3098
3099 /**
3100  * bond_xmit_hash - generate a hash value based on the xmit policy
3101  * @bond: bonding device
3102  * @skb: buffer to use for headers
3103  *
3104  * This function will extract the necessary headers from the skb buffer and use
3105  * them to generate a hash based on the xmit_policy set in the bonding device
3106  */
3107 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3108 {
3109         struct flow_keys flow;
3110         u32 hash;
3111
3112         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3113             !bond_flow_dissect(bond, skb, &flow))
3114                 return bond_eth_hash(skb);
3115
3116         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3117             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3118                 hash = bond_eth_hash(skb);
3119         else
3120                 hash = (__force u32)flow.ports.ports;
3121         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3122                 (__force u32)flow_get_u32_src(&flow);
3123         hash ^= (hash >> 16);
3124         hash ^= (hash >> 8);
3125
3126         return hash;
3127 }
3128
3129 /*-------------------------- Device entry points ----------------------------*/
3130
3131 static void bond_work_init_all(struct bonding *bond)
3132 {
3133         INIT_DELAYED_WORK(&bond->mcast_work,
3134                           bond_resend_igmp_join_requests_delayed);
3135         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3136         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3137         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3138                 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3139         else
3140                 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3141         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3142         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3143 }
3144
3145 static void bond_work_cancel_all(struct bonding *bond)
3146 {
3147         cancel_delayed_work_sync(&bond->mii_work);
3148         cancel_delayed_work_sync(&bond->arp_work);
3149         cancel_delayed_work_sync(&bond->alb_work);
3150         cancel_delayed_work_sync(&bond->ad_work);
3151         cancel_delayed_work_sync(&bond->mcast_work);
3152         cancel_delayed_work_sync(&bond->slave_arr_work);
3153 }
3154
3155 static int bond_open(struct net_device *bond_dev)
3156 {
3157         struct bonding *bond = netdev_priv(bond_dev);
3158         struct list_head *iter;
3159         struct slave *slave;
3160
3161         /* reset slave->backup and slave->inactive */
3162         if (bond_has_slaves(bond)) {
3163                 bond_for_each_slave(bond, slave, iter) {
3164                         if (bond_uses_primary(bond) &&
3165                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3166                                 bond_set_slave_inactive_flags(slave,
3167                                                               BOND_SLAVE_NOTIFY_NOW);
3168                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3169                                 bond_set_slave_active_flags(slave,
3170                                                             BOND_SLAVE_NOTIFY_NOW);
3171                         }
3172                 }
3173         }
3174
3175         bond_work_init_all(bond);
3176
3177         if (bond_is_lb(bond)) {
3178                 /* bond_alb_initialize must be called before the timer
3179                  * is started.
3180                  */
3181                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3182                         return -ENOMEM;
3183                 if (bond->params.tlb_dynamic_lb)
3184                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3185         }
3186
3187         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3188                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3189
3190         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3191                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3192                 bond->recv_probe = bond_arp_rcv;
3193         }
3194
3195         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3196                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3197                 /* register to receive LACPDUs */
3198                 bond->recv_probe = bond_3ad_lacpdu_recv;
3199                 bond_3ad_initiate_agg_selection(bond, 1);
3200         }
3201
3202         if (bond_mode_uses_xmit_hash(bond))
3203                 bond_update_slave_arr(bond, NULL);
3204
3205         return 0;
3206 }
3207
3208 static int bond_close(struct net_device *bond_dev)
3209 {
3210         struct bonding *bond = netdev_priv(bond_dev);
3211
3212         bond_work_cancel_all(bond);
3213         bond->send_peer_notif = 0;
3214         if (bond_is_lb(bond))
3215                 bond_alb_deinitialize(bond);
3216         bond->recv_probe = NULL;
3217
3218         return 0;
3219 }
3220
3221 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3222                                                 struct rtnl_link_stats64 *stats)
3223 {
3224         struct bonding *bond = netdev_priv(bond_dev);
3225         struct rtnl_link_stats64 temp;
3226         struct list_head *iter;
3227         struct slave *slave;
3228
3229         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3230
3231         bond_for_each_slave(bond, slave, iter) {
3232                 const struct rtnl_link_stats64 *sstats =
3233                         dev_get_stats(slave->dev, &temp);
3234                 struct rtnl_link_stats64 *pstats = &slave->slave_stats;
3235
3236                 stats->rx_packets +=  sstats->rx_packets - pstats->rx_packets;
3237                 stats->rx_bytes += sstats->rx_bytes - pstats->rx_bytes;
3238                 stats->rx_errors += sstats->rx_errors - pstats->rx_errors;
3239                 stats->rx_dropped += sstats->rx_dropped - pstats->rx_dropped;
3240
3241                 stats->tx_packets += sstats->tx_packets - pstats->tx_packets;;
3242                 stats->tx_bytes += sstats->tx_bytes - pstats->tx_bytes;
3243                 stats->tx_errors += sstats->tx_errors - pstats->tx_errors;
3244                 stats->tx_dropped += sstats->tx_dropped - pstats->tx_dropped;
3245
3246                 stats->multicast += sstats->multicast - pstats->multicast;
3247                 stats->collisions += sstats->collisions - pstats->collisions;
3248
3249                 stats->rx_length_errors += sstats->rx_length_errors - pstats->rx_length_errors;
3250                 stats->rx_over_errors += sstats->rx_over_errors - pstats->rx_over_errors;
3251                 stats->rx_crc_errors += sstats->rx_crc_errors - pstats->rx_crc_errors;
3252                 stats->rx_frame_errors += sstats->rx_frame_errors - pstats->rx_frame_errors;
3253                 stats->rx_fifo_errors += sstats->rx_fifo_errors - pstats->rx_fifo_errors;
3254                 stats->rx_missed_errors += sstats->rx_missed_errors - pstats->rx_missed_errors;
3255
3256                 stats->tx_aborted_errors += sstats->tx_aborted_errors - pstats->tx_aborted_errors;
3257                 stats->tx_carrier_errors += sstats->tx_carrier_errors - pstats->tx_carrier_errors;
3258                 stats->tx_fifo_errors += sstats->tx_fifo_errors - pstats->tx_fifo_errors;
3259                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors - pstats->tx_heartbeat_errors;
3260                 stats->tx_window_errors += sstats->tx_window_errors - pstats->tx_window_errors;
3261
3262                 /* save off the slave stats for the next run */
3263                 memcpy(pstats, sstats, sizeof(*sstats));
3264         }
3265         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3266
3267         return stats;
3268 }
3269
3270 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3271 {
3272         struct bonding *bond = netdev_priv(bond_dev);
3273         struct net_device *slave_dev = NULL;
3274         struct ifbond k_binfo;
3275         struct ifbond __user *u_binfo = NULL;
3276         struct ifslave k_sinfo;
3277         struct ifslave __user *u_sinfo = NULL;
3278         struct mii_ioctl_data *mii = NULL;
3279         struct bond_opt_value newval;
3280         struct net *net;
3281         int res = 0;
3282
3283         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3284
3285         switch (cmd) {
3286         case SIOCGMIIPHY:
3287                 mii = if_mii(ifr);
3288                 if (!mii)
3289                         return -EINVAL;
3290
3291                 mii->phy_id = 0;
3292                 /* Fall Through */
3293         case SIOCGMIIREG:
3294                 /* We do this again just in case we were called by SIOCGMIIREG
3295                  * instead of SIOCGMIIPHY.
3296                  */
3297                 mii = if_mii(ifr);
3298                 if (!mii)
3299                         return -EINVAL;
3300
3301                 if (mii->reg_num == 1) {
3302                         mii->val_out = 0;
3303                         if (netif_carrier_ok(bond->dev))
3304                                 mii->val_out = BMSR_LSTATUS;
3305                 }
3306
3307                 return 0;
3308         case BOND_INFO_QUERY_OLD:
3309         case SIOCBONDINFOQUERY:
3310                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3311
3312                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3313                         return -EFAULT;
3314
3315                 res = bond_info_query(bond_dev, &k_binfo);
3316                 if (res == 0 &&
3317                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3318                         return -EFAULT;
3319
3320                 return res;
3321         case BOND_SLAVE_INFO_QUERY_OLD:
3322         case SIOCBONDSLAVEINFOQUERY:
3323                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3324
3325                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3326                         return -EFAULT;
3327
3328                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3329                 if (res == 0 &&
3330                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3331                         return -EFAULT;
3332
3333                 return res;
3334         default:
3335                 break;
3336         }
3337
3338         net = dev_net(bond_dev);
3339
3340         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3341                 return -EPERM;
3342
3343         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3344
3345         netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3346
3347         if (!slave_dev)
3348                 return -ENODEV;
3349
3350         netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3351         switch (cmd) {
3352         case BOND_ENSLAVE_OLD:
3353         case SIOCBONDENSLAVE:
3354                 res = bond_enslave(bond_dev, slave_dev);
3355                 break;
3356         case BOND_RELEASE_OLD:
3357         case SIOCBONDRELEASE:
3358                 res = bond_release(bond_dev, slave_dev);
3359                 break;
3360         case BOND_SETHWADDR_OLD:
3361         case SIOCBONDSETHWADDR:
3362                 bond_set_dev_addr(bond_dev, slave_dev);
3363                 res = 0;
3364                 break;
3365         case BOND_CHANGE_ACTIVE_OLD:
3366         case SIOCBONDCHANGEACTIVE:
3367                 bond_opt_initstr(&newval, slave_dev->name);
3368                 res = __bond_opt_set(bond, BOND_OPT_ACTIVE_SLAVE, &newval);
3369                 break;
3370         default:
3371                 res = -EOPNOTSUPP;
3372         }
3373
3374         return res;
3375 }
3376
3377 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3378 {
3379         struct bonding *bond = netdev_priv(bond_dev);
3380
3381         if (change & IFF_PROMISC)
3382                 bond_set_promiscuity(bond,
3383                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3384
3385         if (change & IFF_ALLMULTI)
3386                 bond_set_allmulti(bond,
3387                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3388 }
3389
3390 static void bond_set_rx_mode(struct net_device *bond_dev)
3391 {
3392         struct bonding *bond = netdev_priv(bond_dev);
3393         struct list_head *iter;
3394         struct slave *slave;
3395
3396         rcu_read_lock();
3397         if (bond_uses_primary(bond)) {
3398                 slave = rcu_dereference(bond->curr_active_slave);
3399                 if (slave) {
3400                         dev_uc_sync(slave->dev, bond_dev);
3401                         dev_mc_sync(slave->dev, bond_dev);
3402                 }
3403         } else {
3404                 bond_for_each_slave_rcu(bond, slave, iter) {
3405                         dev_uc_sync_multiple(slave->dev, bond_dev);
3406                         dev_mc_sync_multiple(slave->dev, bond_dev);
3407                 }
3408         }
3409         rcu_read_unlock();
3410 }
3411
3412 static int bond_neigh_init(struct neighbour *n)
3413 {
3414         struct bonding *bond = netdev_priv(n->dev);
3415         const struct net_device_ops *slave_ops;
3416         struct neigh_parms parms;
3417         struct slave *slave;
3418         int ret;
3419
3420         slave = bond_first_slave(bond);
3421         if (!slave)
3422                 return 0;
3423         slave_ops = slave->dev->netdev_ops;
3424         if (!slave_ops->ndo_neigh_setup)
3425                 return 0;
3426
3427         parms.neigh_setup = NULL;
3428         parms.neigh_cleanup = NULL;
3429         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3430         if (ret)
3431                 return ret;
3432
3433         /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3434          * after the last slave has been detached.  Assumes that all slaves
3435          * utilize the same neigh_cleanup (true at this writing as only user
3436          * is ipoib).
3437          */
3438         n->parms->neigh_cleanup = parms.neigh_cleanup;
3439
3440         if (!parms.neigh_setup)
3441                 return 0;
3442
3443         return parms.neigh_setup(n);
3444 }
3445
3446 /* The bonding ndo_neigh_setup is called at init time beofre any
3447  * slave exists. So we must declare proxy setup function which will
3448  * be used at run time to resolve the actual slave neigh param setup.
3449  *
3450  * It's also called by master devices (such as vlans) to setup their
3451  * underlying devices. In that case - do nothing, we're already set up from
3452  * our init.
3453  */
3454 static int bond_neigh_setup(struct net_device *dev,
3455                             struct neigh_parms *parms)
3456 {
3457         /* modify only our neigh_parms */
3458         if (parms->dev == dev)
3459                 parms->neigh_setup = bond_neigh_init;
3460
3461         return 0;
3462 }
3463
3464 /* Change the MTU of all of a master's slaves to match the master */
3465 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3466 {
3467         struct bonding *bond = netdev_priv(bond_dev);
3468         struct slave *slave, *rollback_slave;
3469         struct list_head *iter;
3470         int res = 0;
3471
3472         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3473
3474         bond_for_each_slave(bond, slave, iter) {
3475                 netdev_dbg(bond_dev, "s %p c_m %p\n",
3476                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3477
3478                 res = dev_set_mtu(slave->dev, new_mtu);
3479
3480                 if (res) {
3481                         /* If we failed to set the slave's mtu to the new value
3482                          * we must abort the operation even in ACTIVE_BACKUP
3483                          * mode, because if we allow the backup slaves to have
3484                          * different mtu values than the active slave we'll
3485                          * need to change their mtu when doing a failover. That
3486                          * means changing their mtu from timer context, which
3487                          * is probably not a good idea.
3488                          */
3489                         netdev_dbg(bond_dev, "err %d %s\n", res,
3490                                    slave->dev->name);
3491                         goto unwind;
3492                 }
3493         }
3494
3495         bond_dev->mtu = new_mtu;
3496
3497         return 0;
3498
3499 unwind:
3500         /* unwind from head to the slave that failed */
3501         bond_for_each_slave(bond, rollback_slave, iter) {
3502                 int tmp_res;
3503
3504                 if (rollback_slave == slave)
3505                         break;
3506
3507                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3508                 if (tmp_res) {
3509                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3510                                    tmp_res, rollback_slave->dev->name);
3511                 }
3512         }
3513
3514         return res;
3515 }
3516
3517 /* Change HW address
3518  *
3519  * Note that many devices must be down to change the HW address, and
3520  * downing the master releases all slaves.  We can make bonds full of
3521  * bonding devices to test this, however.
3522  */
3523 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3524 {
3525         struct bonding *bond = netdev_priv(bond_dev);
3526         struct slave *slave, *rollback_slave;
3527         struct sockaddr *sa = addr, tmp_sa;
3528         struct list_head *iter;
3529         int res = 0;
3530
3531         if (BOND_MODE(bond) == BOND_MODE_ALB)
3532                 return bond_alb_set_mac_address(bond_dev, addr);
3533
3534
3535         netdev_dbg(bond_dev, "bond=%p\n", bond);
3536
3537         /* If fail_over_mac is enabled, do nothing and return success.
3538          * Returning an error causes ifenslave to fail.
3539          */
3540         if (bond->params.fail_over_mac &&
3541             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3542                 return 0;
3543
3544         if (!is_valid_ether_addr(sa->sa_data))
3545                 return -EADDRNOTAVAIL;
3546
3547         bond_for_each_slave(bond, slave, iter) {
3548                 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3549                 res = dev_set_mac_address(slave->dev, addr);
3550                 if (res) {
3551                         /* TODO: consider downing the slave
3552                          * and retry ?
3553                          * User should expect communications
3554                          * breakage anyway until ARP finish
3555                          * updating, so...
3556                          */
3557                         netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3558                         goto unwind;
3559                 }
3560         }
3561
3562         /* success */
3563         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3564         return 0;
3565
3566 unwind:
3567         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3568         tmp_sa.sa_family = bond_dev->type;
3569
3570         /* unwind from head to the slave that failed */
3571         bond_for_each_slave(bond, rollback_slave, iter) {
3572                 int tmp_res;
3573
3574                 if (rollback_slave == slave)
3575                         break;
3576
3577                 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3578                 if (tmp_res) {
3579                         netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3580                                    tmp_res, rollback_slave->dev->name);
3581                 }
3582         }
3583
3584         return res;
3585 }
3586
3587 /**
3588  * bond_xmit_slave_id - transmit skb through slave with slave_id
3589  * @bond: bonding device that is transmitting
3590  * @skb: buffer to transmit
3591  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3592  *
3593  * This function tries to transmit through slave with slave_id but in case
3594  * it fails, it tries to find the first available slave for transmission.
3595  * The skb is consumed in all cases, thus the function is void.
3596  */
3597 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3598 {
3599         struct list_head *iter;
3600         struct slave *slave;
3601         int i = slave_id;
3602
3603         /* Here we start from the slave with slave_id */
3604         bond_for_each_slave_rcu(bond, slave, iter) {
3605                 if (--i < 0) {
3606                         if (bond_slave_can_tx(slave)) {
3607                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3608                                 return;
3609                         }
3610                 }
3611         }
3612
3613         /* Here we start from the first slave up to slave_id */
3614         i = slave_id;
3615         bond_for_each_slave_rcu(bond, slave, iter) {
3616                 if (--i < 0)
3617                         break;
3618                 if (bond_slave_can_tx(slave)) {
3619                         bond_dev_queue_xmit(bond, skb, slave->dev);
3620                         return;
3621                 }
3622         }
3623         /* no slave that can tx has been found */
3624         bond_tx_drop(bond->dev, skb);
3625 }
3626
3627 /**
3628  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3629  * @bond: bonding device to use
3630  *
3631  * Based on the value of the bonding device's packets_per_slave parameter
3632  * this function generates a slave id, which is usually used as the next
3633  * slave to transmit through.
3634  */
3635 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3636 {
3637         u32 slave_id;
3638         struct reciprocal_value reciprocal_packets_per_slave;
3639         int packets_per_slave = bond->params.packets_per_slave;
3640
3641         switch (packets_per_slave) {
3642         case 0:
3643                 slave_id = prandom_u32();
3644                 break;
3645         case 1:
3646                 slave_id = bond->rr_tx_counter;
3647                 break;
3648         default:
3649                 reciprocal_packets_per_slave =
3650                         bond->params.reciprocal_packets_per_slave;
3651                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3652                                              reciprocal_packets_per_slave);
3653                 break;
3654         }
3655         bond->rr_tx_counter++;
3656
3657         return slave_id;
3658 }
3659
3660 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3661 {
3662         struct bonding *bond = netdev_priv(bond_dev);
3663         struct iphdr *iph = ip_hdr(skb);
3664         struct slave *slave;
3665         u32 slave_id;
3666
3667         /* Start with the curr_active_slave that joined the bond as the
3668          * default for sending IGMP traffic.  For failover purposes one
3669          * needs to maintain some consistency for the interface that will
3670          * send the join/membership reports.  The curr_active_slave found
3671          * will send all of this type of traffic.
3672          */
3673         if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3674                 slave = rcu_dereference(bond->curr_active_slave);
3675                 if (slave)
3676                         bond_dev_queue_xmit(bond, skb, slave->dev);
3677                 else
3678                         bond_xmit_slave_id(bond, skb, 0);
3679         } else {
3680                 int slave_cnt = ACCESS_ONCE(bond->slave_cnt);
3681
3682                 if (likely(slave_cnt)) {
3683                         slave_id = bond_rr_gen_slave_id(bond);
3684                         bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3685                 } else {
3686                         bond_tx_drop(bond_dev, skb);
3687                 }
3688         }
3689
3690         return NETDEV_TX_OK;
3691 }
3692
3693 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3694  * the bond has a usable interface.
3695  */
3696 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3697 {
3698         struct bonding *bond = netdev_priv(bond_dev);
3699         struct slave *slave;
3700
3701         slave = rcu_dereference(bond->curr_active_slave);
3702         if (slave)
3703                 bond_dev_queue_xmit(bond, skb, slave->dev);
3704         else
3705                 bond_tx_drop(bond_dev, skb);
3706
3707         return NETDEV_TX_OK;
3708 }
3709
3710 /* Use this to update slave_array when (a) it's not appropriate to update
3711  * slave_array right away (note that update_slave_array() may sleep)
3712  * and / or (b) RTNL is not held.
3713  */
3714 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3715 {
3716         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3717 }
3718
3719 /* Slave array work handler. Holds only RTNL */
3720 static void bond_slave_arr_handler(struct work_struct *work)
3721 {
3722         struct bonding *bond = container_of(work, struct bonding,
3723                                             slave_arr_work.work);
3724         int ret;
3725
3726         if (!rtnl_trylock())
3727                 goto err;
3728
3729         ret = bond_update_slave_arr(bond, NULL);
3730         rtnl_unlock();
3731         if (ret) {
3732                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3733                 goto err;
3734         }
3735         return;
3736
3737 err:
3738         bond_slave_arr_work_rearm(bond, 1);
3739 }
3740
3741 /* Build the usable slaves array in control path for modes that use xmit-hash
3742  * to determine the slave interface -
3743  * (a) BOND_MODE_8023AD
3744  * (b) BOND_MODE_XOR
3745  * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3746  *
3747  * The caller is expected to hold RTNL only and NO other lock!
3748  */
3749 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3750 {
3751         struct slave *slave;
3752         struct list_head *iter;
3753         struct bond_up_slave *new_arr, *old_arr;
3754         int slaves_in_agg;
3755         int agg_id = 0;
3756         int ret = 0;
3757
3758 #ifdef CONFIG_LOCKDEP
3759         WARN_ON(lockdep_is_held(&bond->mode_lock));
3760 #endif
3761
3762         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3763                           GFP_KERNEL);
3764         if (!new_arr) {
3765                 ret = -ENOMEM;
3766                 pr_err("Failed to build slave-array.\n");
3767                 goto out;
3768         }
3769         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3770                 struct ad_info ad_info;
3771
3772                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3773                         pr_debug("bond_3ad_get_active_agg_info failed\n");
3774                         kfree_rcu(new_arr, rcu);
3775                         /* No active aggragator means it's not safe to use
3776                          * the previous array.
3777                          */
3778                         old_arr = rtnl_dereference(bond->slave_arr);
3779                         if (old_arr) {
3780                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
3781                                 kfree_rcu(old_arr, rcu);
3782                         }
3783                         goto out;
3784                 }
3785                 slaves_in_agg = ad_info.ports;
3786                 agg_id = ad_info.aggregator_id;
3787         }
3788         bond_for_each_slave(bond, slave, iter) {
3789                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3790                         struct aggregator *agg;
3791
3792                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
3793                         if (!agg || agg->aggregator_identifier != agg_id)
3794                                 continue;
3795                 }
3796                 if (!bond_slave_can_tx(slave))
3797                         continue;
3798                 if (skipslave == slave)
3799                         continue;
3800                 new_arr->arr[new_arr->count++] = slave;
3801         }
3802
3803         old_arr = rtnl_dereference(bond->slave_arr);
3804         rcu_assign_pointer(bond->slave_arr, new_arr);
3805         if (old_arr)
3806                 kfree_rcu(old_arr, rcu);
3807 out:
3808         if (ret != 0 && skipslave) {
3809                 int idx;
3810
3811                 /* Rare situation where caller has asked to skip a specific
3812                  * slave but allocation failed (most likely!). BTW this is
3813                  * only possible when the call is initiated from
3814                  * __bond_release_one(). In this situation; overwrite the
3815                  * skipslave entry in the array with the last entry from the
3816                  * array to avoid a situation where the xmit path may choose
3817                  * this to-be-skipped slave to send a packet out.
3818                  */
3819                 old_arr = rtnl_dereference(bond->slave_arr);
3820                 for (idx = 0; idx < old_arr->count; idx++) {
3821                         if (skipslave == old_arr->arr[idx]) {
3822                                 old_arr->arr[idx] =
3823                                     old_arr->arr[old_arr->count-1];
3824                                 old_arr->count--;
3825                                 break;
3826                         }
3827                 }
3828         }
3829         return ret;
3830 }
3831
3832 /* Use this Xmit function for 3AD as well as XOR modes. The current
3833  * usable slave array is formed in the control path. The xmit function
3834  * just calculates hash and sends the packet out.
3835  */
3836 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3837 {
3838         struct bonding *bond = netdev_priv(dev);
3839         struct slave *slave;
3840         struct bond_up_slave *slaves;
3841         unsigned int count;
3842
3843         slaves = rcu_dereference(bond->slave_arr);
3844         count = slaves ? ACCESS_ONCE(slaves->count) : 0;
3845         if (likely(count)) {
3846                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3847                 bond_dev_queue_xmit(bond, skb, slave->dev);
3848         } else {
3849                 bond_tx_drop(dev, skb);
3850         }
3851
3852         return NETDEV_TX_OK;
3853 }
3854
3855 /* in broadcast mode, we send everything to all usable interfaces. */
3856 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3857 {
3858         struct bonding *bond = netdev_priv(bond_dev);
3859         struct slave *slave = NULL;
3860         struct list_head *iter;
3861
3862         bond_for_each_slave_rcu(bond, slave, iter) {
3863                 if (bond_is_last_slave(bond, slave))
3864                         break;
3865                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3866                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3867
3868                         if (!skb2) {
3869                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
3870                                                     bond_dev->name, __func__);
3871                                 continue;
3872                         }
3873                         bond_dev_queue_xmit(bond, skb2, slave->dev);
3874                 }
3875         }
3876         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
3877                 bond_dev_queue_xmit(bond, skb, slave->dev);
3878         else
3879                 bond_tx_drop(bond_dev, skb);
3880
3881         return NETDEV_TX_OK;
3882 }
3883
3884 /*------------------------- Device initialization ---------------------------*/
3885
3886 /* Lookup the slave that corresponds to a qid */
3887 static inline int bond_slave_override(struct bonding *bond,
3888                                       struct sk_buff *skb)
3889 {
3890         struct slave *slave = NULL;
3891         struct list_head *iter;
3892
3893         if (!skb->queue_mapping)
3894                 return 1;
3895
3896         /* Find out if any slaves have the same mapping as this skb. */
3897         bond_for_each_slave_rcu(bond, slave, iter) {
3898                 if (slave->queue_id == skb->queue_mapping) {
3899                         if (bond_slave_is_up(slave) &&
3900                             slave->link == BOND_LINK_UP) {
3901                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3902                                 return 0;
3903                         }
3904                         /* If the slave isn't UP, use default transmit policy. */
3905                         break;
3906                 }
3907         }
3908
3909         return 1;
3910 }
3911
3912
3913 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3914                              void *accel_priv, select_queue_fallback_t fallback)
3915 {
3916         /* This helper function exists to help dev_pick_tx get the correct
3917          * destination queue.  Using a helper function skips a call to
3918          * skb_tx_hash and will put the skbs in the queue we expect on their
3919          * way down to the bonding driver.
3920          */
3921         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3922
3923         /* Save the original txq to restore before passing to the driver */
3924         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3925
3926         if (unlikely(txq >= dev->real_num_tx_queues)) {
3927                 do {
3928                         txq -= dev->real_num_tx_queues;
3929                 } while (txq >= dev->real_num_tx_queues);
3930         }
3931         return txq;
3932 }
3933
3934 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3935 {
3936         struct bonding *bond = netdev_priv(dev);
3937
3938         if (bond_should_override_tx_queue(bond) &&
3939             !bond_slave_override(bond, skb))
3940                 return NETDEV_TX_OK;
3941
3942         switch (BOND_MODE(bond)) {
3943         case BOND_MODE_ROUNDROBIN:
3944                 return bond_xmit_roundrobin(skb, dev);
3945         case BOND_MODE_ACTIVEBACKUP:
3946                 return bond_xmit_activebackup(skb, dev);
3947         case BOND_MODE_8023AD:
3948         case BOND_MODE_XOR:
3949                 return bond_3ad_xor_xmit(skb, dev);
3950         case BOND_MODE_BROADCAST:
3951                 return bond_xmit_broadcast(skb, dev);
3952         case BOND_MODE_ALB:
3953                 return bond_alb_xmit(skb, dev);
3954         case BOND_MODE_TLB:
3955                 return bond_tlb_xmit(skb, dev);
3956         default:
3957                 /* Should never happen, mode already checked */
3958                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
3959                 WARN_ON_ONCE(1);
3960                 bond_tx_drop(dev, skb);
3961                 return NETDEV_TX_OK;
3962         }
3963 }
3964
3965 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3966 {
3967         struct bonding *bond = netdev_priv(dev);
3968         netdev_tx_t ret = NETDEV_TX_OK;
3969
3970         /* If we risk deadlock from transmitting this in the
3971          * netpoll path, tell netpoll to queue the frame for later tx
3972          */
3973         if (unlikely(is_netpoll_tx_blocked(dev)))
3974                 return NETDEV_TX_BUSY;
3975
3976         rcu_read_lock();
3977         if (bond_has_slaves(bond))
3978                 ret = __bond_start_xmit(skb, dev);
3979         else
3980                 bond_tx_drop(dev, skb);
3981         rcu_read_unlock();
3982
3983         return ret;
3984 }
3985
3986 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3987                                      struct ethtool_cmd *ecmd)
3988 {
3989         struct bonding *bond = netdev_priv(bond_dev);
3990         unsigned long speed = 0;
3991         struct list_head *iter;
3992         struct slave *slave;
3993
3994         ecmd->duplex = DUPLEX_UNKNOWN;
3995         ecmd->port = PORT_OTHER;
3996
3997         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
3998          * do not need to check mode.  Though link speed might not represent
3999          * the true receive or transmit bandwidth (not all modes are symmetric)
4000          * this is an accurate maximum.
4001          */
4002         bond_for_each_slave(bond, slave, iter) {
4003                 if (bond_slave_can_tx(slave)) {
4004                         if (slave->speed != SPEED_UNKNOWN)
4005                                 speed += slave->speed;
4006                         if (ecmd->duplex == DUPLEX_UNKNOWN &&
4007                             slave->duplex != DUPLEX_UNKNOWN)
4008                                 ecmd->duplex = slave->duplex;
4009                 }
4010         }
4011         ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
4012
4013         return 0;
4014 }
4015
4016 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4017                                      struct ethtool_drvinfo *drvinfo)
4018 {
4019         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4020         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4021         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4022                  BOND_ABI_VERSION);
4023 }
4024
4025 static const struct ethtool_ops bond_ethtool_ops = {
4026         .get_drvinfo            = bond_ethtool_get_drvinfo,
4027         .get_settings           = bond_ethtool_get_settings,
4028         .get_link               = ethtool_op_get_link,
4029 };
4030
4031 static const struct net_device_ops bond_netdev_ops = {
4032         .ndo_init               = bond_init,
4033         .ndo_uninit             = bond_uninit,
4034         .ndo_open               = bond_open,
4035         .ndo_stop               = bond_close,
4036         .ndo_start_xmit         = bond_start_xmit,
4037         .ndo_select_queue       = bond_select_queue,
4038         .ndo_get_stats64        = bond_get_stats,
4039         .ndo_do_ioctl           = bond_do_ioctl,
4040         .ndo_change_rx_flags    = bond_change_rx_flags,
4041         .ndo_set_rx_mode        = bond_set_rx_mode,
4042         .ndo_change_mtu         = bond_change_mtu,
4043         .ndo_set_mac_address    = bond_set_mac_address,
4044         .ndo_neigh_setup        = bond_neigh_setup,
4045         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4046         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4047 #ifdef CONFIG_NET_POLL_CONTROLLER
4048         .ndo_netpoll_setup      = bond_netpoll_setup,
4049         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4050         .ndo_poll_controller    = bond_poll_controller,
4051 #endif
4052         .ndo_add_slave          = bond_enslave,
4053         .ndo_del_slave          = bond_release,
4054         .ndo_fix_features       = bond_fix_features,
4055         .ndo_bridge_setlink     = switchdev_port_bridge_setlink,
4056         .ndo_bridge_getlink     = switchdev_port_bridge_getlink,
4057         .ndo_bridge_dellink     = switchdev_port_bridge_dellink,
4058         .ndo_fdb_add            = switchdev_port_fdb_add,
4059         .ndo_fdb_del            = switchdev_port_fdb_del,
4060         .ndo_fdb_dump           = switchdev_port_fdb_dump,
4061         .ndo_features_check     = passthru_features_check,
4062 };
4063
4064 static const struct device_type bond_type = {
4065         .name = "bond",
4066 };
4067
4068 static void bond_destructor(struct net_device *bond_dev)
4069 {
4070         struct bonding *bond = netdev_priv(bond_dev);
4071         if (bond->wq)
4072                 destroy_workqueue(bond->wq);
4073         free_netdev(bond_dev);
4074 }
4075
4076 void bond_setup(struct net_device *bond_dev)
4077 {
4078         struct bonding *bond = netdev_priv(bond_dev);
4079
4080         spin_lock_init(&bond->mode_lock);
4081         bond->params = bonding_defaults;
4082
4083         /* Initialize pointers */
4084         bond->dev = bond_dev;
4085
4086         /* Initialize the device entry points */
4087         ether_setup(bond_dev);
4088         bond_dev->netdev_ops = &bond_netdev_ops;
4089         bond_dev->ethtool_ops = &bond_ethtool_ops;
4090
4091         bond_dev->destructor = bond_destructor;
4092
4093         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4094
4095         /* Initialize the device options */
4096         bond_dev->tx_queue_len = 0;
4097         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4098         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT;
4099         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4100
4101         /* don't acquire bond device's netif_tx_lock when transmitting */
4102         bond_dev->features |= NETIF_F_LLTX;
4103
4104         /* By default, we declare the bond to be fully
4105          * VLAN hardware accelerated capable. Special
4106          * care is taken in the various xmit functions
4107          * when there are slaves that are not hw accel
4108          * capable
4109          */
4110
4111         /* Don't allow bond devices to change network namespaces. */
4112         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4113
4114         bond_dev->hw_features = BOND_VLAN_FEATURES |
4115                                 NETIF_F_HW_VLAN_CTAG_TX |
4116                                 NETIF_F_HW_VLAN_CTAG_RX |
4117                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4118
4119         bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4120         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4121         bond_dev->features |= bond_dev->hw_features;
4122 }
4123
4124 /* Destroy a bonding device.
4125  * Must be under rtnl_lock when this function is called.
4126  */
4127 static void bond_uninit(struct net_device *bond_dev)
4128 {
4129         struct bonding *bond = netdev_priv(bond_dev);
4130         struct list_head *iter;
4131         struct slave *slave;
4132         struct bond_up_slave *arr;
4133
4134         bond_netpoll_cleanup(bond_dev);
4135
4136         /* Release the bonded slaves */
4137         bond_for_each_slave(bond, slave, iter)
4138                 __bond_release_one(bond_dev, slave->dev, true);
4139         netdev_info(bond_dev, "Released all slaves\n");
4140
4141         arr = rtnl_dereference(bond->slave_arr);
4142         if (arr) {
4143                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4144                 kfree_rcu(arr, rcu);
4145         }
4146
4147         list_del(&bond->bond_list);
4148
4149         bond_debug_unregister(bond);
4150 }
4151
4152 /*------------------------- Module initialization ---------------------------*/
4153
4154 static int bond_check_params(struct bond_params *params)
4155 {
4156         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4157         struct bond_opt_value newval;
4158         const struct bond_opt_value *valptr;
4159         int arp_all_targets_value;
4160         u16 ad_actor_sys_prio = 0;
4161         u16 ad_user_port_key = 0;
4162
4163         /* Convert string parameters. */
4164         if (mode) {
4165                 bond_opt_initstr(&newval, mode);
4166                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4167                 if (!valptr) {
4168                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4169                         return -EINVAL;
4170                 }
4171                 bond_mode = valptr->value;
4172         }
4173
4174         if (xmit_hash_policy) {
4175                 if ((bond_mode != BOND_MODE_XOR) &&
4176                     (bond_mode != BOND_MODE_8023AD) &&
4177                     (bond_mode != BOND_MODE_TLB)) {
4178                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4179                                 bond_mode_name(bond_mode));
4180                 } else {
4181                         bond_opt_initstr(&newval, xmit_hash_policy);
4182                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4183                                                 &newval);
4184                         if (!valptr) {
4185                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4186                                        xmit_hash_policy);
4187                                 return -EINVAL;
4188                         }
4189                         xmit_hashtype = valptr->value;
4190                 }
4191         }
4192
4193         if (lacp_rate) {
4194                 if (bond_mode != BOND_MODE_8023AD) {
4195                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4196                                 bond_mode_name(bond_mode));
4197                 } else {
4198                         bond_opt_initstr(&newval, lacp_rate);
4199                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4200                                                 &newval);
4201                         if (!valptr) {
4202                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4203                                        lacp_rate);
4204                                 return -EINVAL;
4205                         }
4206                         lacp_fast = valptr->value;
4207                 }
4208         }
4209
4210         if (ad_select) {
4211                 bond_opt_initstr(&newval, ad_select);
4212                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4213                                         &newval);
4214                 if (!valptr) {
4215                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4216                         return -EINVAL;
4217                 }
4218                 params->ad_select = valptr->value;
4219                 if (bond_mode != BOND_MODE_8023AD)
4220                         pr_warn("ad_select param only affects 802.3ad mode\n");
4221         } else {
4222                 params->ad_select = BOND_AD_STABLE;
4223         }
4224
4225         if (max_bonds < 0) {
4226                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4227                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4228                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4229         }
4230
4231         if (miimon < 0) {
4232                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4233                         miimon, INT_MAX);
4234                 miimon = 0;
4235         }
4236
4237         if (updelay < 0) {
4238                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4239                         updelay, INT_MAX);
4240                 updelay = 0;
4241         }
4242
4243         if (downdelay < 0) {
4244                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4245                         downdelay, INT_MAX);
4246                 downdelay = 0;
4247         }
4248
4249         if ((use_carrier != 0) && (use_carrier != 1)) {
4250                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4251                         use_carrier);
4252                 use_carrier = 1;
4253         }
4254
4255         if (num_peer_notif < 0 || num_peer_notif > 255) {
4256                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4257                         num_peer_notif);
4258                 num_peer_notif = 1;
4259         }
4260
4261         /* reset values for 802.3ad/TLB/ALB */
4262         if (!bond_mode_uses_arp(bond_mode)) {
4263                 if (!miimon) {
4264                         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");
4265                         pr_warn("Forcing miimon to 100msec\n");
4266                         miimon = BOND_DEFAULT_MIIMON;
4267                 }
4268         }
4269
4270         if (tx_queues < 1 || tx_queues > 255) {
4271                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4272                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4273                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4274         }
4275
4276         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4277                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4278                         all_slaves_active);
4279                 all_slaves_active = 0;
4280         }
4281
4282         if (resend_igmp < 0 || resend_igmp > 255) {
4283                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4284                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4285                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4286         }
4287
4288         bond_opt_initval(&newval, packets_per_slave);
4289         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4290                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4291                         packets_per_slave, USHRT_MAX);
4292                 packets_per_slave = 1;
4293         }
4294
4295         if (bond_mode == BOND_MODE_ALB) {
4296                 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",
4297                           updelay);
4298         }
4299
4300         if (!miimon) {
4301                 if (updelay || downdelay) {
4302                         /* just warn the user the up/down delay will have
4303                          * no effect since miimon is zero...
4304                          */
4305                         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",
4306                                 updelay, downdelay);
4307                 }
4308         } else {
4309                 /* don't allow arp monitoring */
4310                 if (arp_interval) {
4311                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4312                                 miimon, arp_interval);
4313                         arp_interval = 0;
4314                 }
4315
4316                 if ((updelay % miimon) != 0) {
4317                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4318                                 updelay, miimon, (updelay / miimon) * miimon);
4319                 }
4320
4321                 updelay /= miimon;
4322
4323                 if ((downdelay % miimon) != 0) {
4324                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4325                                 downdelay, miimon,
4326                                 (downdelay / miimon) * miimon);
4327                 }
4328
4329                 downdelay /= miimon;
4330         }
4331
4332         if (arp_interval < 0) {
4333                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4334                         arp_interval, INT_MAX);
4335                 arp_interval = 0;
4336         }
4337
4338         for (arp_ip_count = 0, i = 0;
4339              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4340                 __be32 ip;
4341
4342                 /* not a complete check, but good enough to catch mistakes */
4343                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4344                     !bond_is_ip_target_ok(ip)) {
4345                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4346                                 arp_ip_target[i]);
4347                         arp_interval = 0;
4348                 } else {
4349                         if (bond_get_targets_ip(arp_target, ip) == -1)
4350                                 arp_target[arp_ip_count++] = ip;
4351                         else
4352                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4353                                         &ip);
4354                 }
4355         }
4356
4357         if (arp_interval && !arp_ip_count) {
4358                 /* don't allow arping if no arp_ip_target given... */
4359                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4360                         arp_interval);
4361                 arp_interval = 0;
4362         }
4363
4364         if (arp_validate) {
4365                 if (!arp_interval) {
4366                         pr_err("arp_validate requires arp_interval\n");
4367                         return -EINVAL;
4368                 }
4369
4370                 bond_opt_initstr(&newval, arp_validate);
4371                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4372                                         &newval);
4373                 if (!valptr) {
4374                         pr_err("Error: invalid arp_validate \"%s\"\n",
4375                                arp_validate);
4376                         return -EINVAL;
4377                 }
4378                 arp_validate_value = valptr->value;
4379         } else {
4380                 arp_validate_value = 0;
4381         }
4382
4383         arp_all_targets_value = 0;
4384         if (arp_all_targets) {
4385                 bond_opt_initstr(&newval, arp_all_targets);
4386                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4387                                         &newval);
4388                 if (!valptr) {
4389                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4390                                arp_all_targets);
4391                         arp_all_targets_value = 0;
4392                 } else {
4393                         arp_all_targets_value = valptr->value;
4394                 }
4395         }
4396
4397         if (miimon) {
4398                 pr_info("MII link monitoring set to %d ms\n", miimon);
4399         } else if (arp_interval) {
4400                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4401                                           arp_validate_value);
4402                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4403                         arp_interval, valptr->string, arp_ip_count);
4404
4405                 for (i = 0; i < arp_ip_count; i++)
4406                         pr_cont(" %s", arp_ip_target[i]);
4407
4408                 pr_cont("\n");
4409
4410         } else if (max_bonds) {
4411                 /* miimon and arp_interval not set, we need one so things
4412                  * work as expected, see bonding.txt for details
4413                  */
4414                 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");
4415         }
4416
4417         if (primary && !bond_mode_uses_primary(bond_mode)) {
4418                 /* currently, using a primary only makes sense
4419                  * in active backup, TLB or ALB modes
4420                  */
4421                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4422                         primary, bond_mode_name(bond_mode));
4423                 primary = NULL;
4424         }
4425
4426         if (primary && primary_reselect) {
4427                 bond_opt_initstr(&newval, primary_reselect);
4428                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4429                                         &newval);
4430                 if (!valptr) {
4431                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4432                                primary_reselect);
4433                         return -EINVAL;
4434                 }
4435                 primary_reselect_value = valptr->value;
4436         } else {
4437                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4438         }
4439
4440         if (fail_over_mac) {
4441                 bond_opt_initstr(&newval, fail_over_mac);
4442                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4443                                         &newval);
4444                 if (!valptr) {
4445                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4446                                fail_over_mac);
4447                         return -EINVAL;
4448                 }
4449                 fail_over_mac_value = valptr->value;
4450                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4451                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4452         } else {
4453                 fail_over_mac_value = BOND_FOM_NONE;
4454         }
4455
4456         bond_opt_initstr(&newval, "default");
4457         valptr = bond_opt_parse(
4458                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4459                                      &newval);
4460         if (!valptr) {
4461                 pr_err("Error: No ad_actor_sys_prio default value");
4462                 return -EINVAL;
4463         }
4464         ad_actor_sys_prio = valptr->value;
4465
4466         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4467                                 &newval);
4468         if (!valptr) {
4469                 pr_err("Error: No ad_user_port_key default value");
4470                 return -EINVAL;
4471         }
4472         ad_user_port_key = valptr->value;
4473
4474         if (lp_interval == 0) {
4475                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4476                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4477                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4478         }
4479
4480         /* fill params struct with the proper values */
4481         params->mode = bond_mode;
4482         params->xmit_policy = xmit_hashtype;
4483         params->miimon = miimon;
4484         params->num_peer_notif = num_peer_notif;
4485         params->arp_interval = arp_interval;
4486         params->arp_validate = arp_validate_value;
4487         params->arp_all_targets = arp_all_targets_value;
4488         params->updelay = updelay;
4489         params->downdelay = downdelay;
4490         params->use_carrier = use_carrier;
4491         params->lacp_fast = lacp_fast;
4492         params->primary[0] = 0;
4493         params->primary_reselect = primary_reselect_value;
4494         params->fail_over_mac = fail_over_mac_value;
4495         params->tx_queues = tx_queues;
4496         params->all_slaves_active = all_slaves_active;
4497         params->resend_igmp = resend_igmp;
4498         params->min_links = min_links;
4499         params->lp_interval = lp_interval;
4500         params->packets_per_slave = packets_per_slave;
4501         params->tlb_dynamic_lb = 1; /* Default value */
4502         params->ad_actor_sys_prio = ad_actor_sys_prio;
4503         eth_zero_addr(params->ad_actor_system);
4504         params->ad_user_port_key = ad_user_port_key;
4505         if (packets_per_slave > 0) {
4506                 params->reciprocal_packets_per_slave =
4507                         reciprocal_value(packets_per_slave);
4508         } else {
4509                 /* reciprocal_packets_per_slave is unused if
4510                  * packets_per_slave is 0 or 1, just initialize it
4511                  */
4512                 params->reciprocal_packets_per_slave =
4513                         (struct reciprocal_value) { 0 };
4514         }
4515
4516         if (primary) {
4517                 strncpy(params->primary, primary, IFNAMSIZ);
4518                 params->primary[IFNAMSIZ - 1] = 0;
4519         }
4520
4521         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4522
4523         return 0;
4524 }
4525
4526 static struct lock_class_key bonding_netdev_xmit_lock_key;
4527 static struct lock_class_key bonding_netdev_addr_lock_key;
4528 static struct lock_class_key bonding_tx_busylock_key;
4529
4530 static void bond_set_lockdep_class_one(struct net_device *dev,
4531                                        struct netdev_queue *txq,
4532                                        void *_unused)
4533 {
4534         lockdep_set_class(&txq->_xmit_lock,
4535                           &bonding_netdev_xmit_lock_key);
4536 }
4537
4538 static void bond_set_lockdep_class(struct net_device *dev)
4539 {
4540         lockdep_set_class(&dev->addr_list_lock,
4541                           &bonding_netdev_addr_lock_key);
4542         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4543         dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4544 }
4545
4546 /* Called from registration process */
4547 static int bond_init(struct net_device *bond_dev)
4548 {
4549         struct bonding *bond = netdev_priv(bond_dev);
4550         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4551
4552         netdev_dbg(bond_dev, "Begin bond_init\n");
4553
4554         bond->wq = create_singlethread_workqueue(bond_dev->name);
4555         if (!bond->wq)
4556                 return -ENOMEM;
4557
4558         bond_set_lockdep_class(bond_dev);
4559
4560         list_add_tail(&bond->bond_list, &bn->dev_list);
4561
4562         bond_prepare_sysfs_group(bond);
4563
4564         bond_debug_register(bond);
4565
4566         /* Ensure valid dev_addr */
4567         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4568             bond_dev->addr_assign_type == NET_ADDR_PERM)
4569                 eth_hw_addr_random(bond_dev);
4570
4571         return 0;
4572 }
4573
4574 unsigned int bond_get_num_tx_queues(void)
4575 {
4576         return tx_queues;
4577 }
4578
4579 /* Create a new bond based on the specified name and bonding parameters.
4580  * If name is NULL, obtain a suitable "bond%d" name for us.
4581  * Caller must NOT hold rtnl_lock; we need to release it here before we
4582  * set up our sysfs entries.
4583  */
4584 int bond_create(struct net *net, const char *name)
4585 {
4586         struct net_device *bond_dev;
4587         struct bonding *bond;
4588         struct alb_bond_info *bond_info;
4589         int res;
4590
4591         rtnl_lock();
4592
4593         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4594                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4595                                    bond_setup, tx_queues);
4596         if (!bond_dev) {
4597                 pr_err("%s: eek! can't alloc netdev!\n", name);
4598                 rtnl_unlock();
4599                 return -ENOMEM;
4600         }
4601
4602         /*
4603          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4604          * It is set to 0 by default which is wrong.
4605          */
4606         bond = netdev_priv(bond_dev);
4607         bond_info = &(BOND_ALB_INFO(bond));
4608         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4609
4610         dev_net_set(bond_dev, net);
4611         bond_dev->rtnl_link_ops = &bond_link_ops;
4612
4613         res = register_netdevice(bond_dev);
4614
4615         netif_carrier_off(bond_dev);
4616
4617         rtnl_unlock();
4618         if (res < 0)
4619                 bond_destructor(bond_dev);
4620         return res;
4621 }
4622
4623 static int __net_init bond_net_init(struct net *net)
4624 {
4625         struct bond_net *bn = net_generic(net, bond_net_id);
4626
4627         bn->net = net;
4628         INIT_LIST_HEAD(&bn->dev_list);
4629
4630         bond_create_proc_dir(bn);
4631         bond_create_sysfs(bn);
4632
4633         return 0;
4634 }
4635
4636 static void __net_exit bond_net_exit(struct net *net)
4637 {
4638         struct bond_net *bn = net_generic(net, bond_net_id);
4639         struct bonding *bond, *tmp_bond;
4640         LIST_HEAD(list);
4641
4642         bond_destroy_sysfs(bn);
4643
4644         /* Kill off any bonds created after unregistering bond rtnl ops */
4645         rtnl_lock();
4646         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4647                 unregister_netdevice_queue(bond->dev, &list);
4648         unregister_netdevice_many(&list);
4649         rtnl_unlock();
4650
4651         bond_destroy_proc_dir(bn);
4652 }
4653
4654 static struct pernet_operations bond_net_ops = {
4655         .init = bond_net_init,
4656         .exit = bond_net_exit,
4657         .id   = &bond_net_id,
4658         .size = sizeof(struct bond_net),
4659 };
4660
4661 static int __init bonding_init(void)
4662 {
4663         int i;
4664         int res;
4665
4666         pr_info("%s", bond_version);
4667
4668         res = bond_check_params(&bonding_defaults);
4669         if (res)
4670                 goto out;
4671
4672         res = register_pernet_subsys(&bond_net_ops);
4673         if (res)
4674                 goto out;
4675
4676         res = bond_netlink_init();
4677         if (res)
4678                 goto err_link;
4679
4680         bond_create_debugfs();
4681
4682         for (i = 0; i < max_bonds; i++) {
4683                 res = bond_create(&init_net, NULL);
4684                 if (res)
4685                         goto err;
4686         }
4687
4688         register_netdevice_notifier(&bond_netdev_notifier);
4689 out:
4690         return res;
4691 err:
4692         bond_destroy_debugfs();
4693         bond_netlink_fini();
4694 err_link:
4695         unregister_pernet_subsys(&bond_net_ops);
4696         goto out;
4697
4698 }
4699
4700 static void __exit bonding_exit(void)
4701 {
4702         unregister_netdevice_notifier(&bond_netdev_notifier);
4703
4704         bond_destroy_debugfs();
4705
4706         bond_netlink_fini();
4707         unregister_pernet_subsys(&bond_net_ops);
4708
4709 #ifdef CONFIG_NET_POLL_CONTROLLER
4710         /* Make sure we don't have an imbalance on our netpoll blocking */
4711         WARN_ON(atomic_read(&netpoll_block_tx));
4712 #endif
4713 }
4714
4715 module_init(bonding_init);
4716 module_exit(bonding_exit);
4717 MODULE_LICENSE("GPL");
4718 MODULE_VERSION(DRV_VERSION);
4719 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4720 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");