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/icmp.h>
  45 #include <linux/icmpv6.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include <linux/rculist.h>
  81 #include <net/flow_dissector.h>
  82 #include <net/xfrm.h>
  83 #include <net/bonding.h>
  84 #include <net/bond_3ad.h>
  85 #include <net/bond_alb.h>
  86
  87 #include "bonding_priv.h"
  88
  89 /*---------------------------- Module parameters ----------------------------*/
  90
  91 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  92
  93 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  94 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  95 static int num_peer_notif = 1;
  96 static int miimon;
  97 static int updelay;
  98 static int downdelay;
  99 static int use_carrier  = 1;
 100 static char *mode;
 101 static char *primary;
 102 static char *primary_reselect;
 103 static char *lacp_rate;
 104 static int min_links;
 105 static char *ad_select;
 106 static char *xmit_hash_policy;
 107 static int arp_interval;
 108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 109 static char *arp_validate;
 110 static char *arp_all_targets;
 111 static char *fail_over_mac;
 112 static int all_slaves_active;
 113 static struct bond_params bonding_defaults;
 114 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 115 static int packets_per_slave = 1;
 116 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 117
 118 module_param(max_bonds, int, 0);
 119 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 120 module_param(tx_queues, int, 0);
 121 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 122 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 123 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 124                                "failover event (alias of num_unsol_na)");
 125 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 126 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 127                                "failover event (alias of num_grat_arp)");
 128 module_param(miimon, int, 0);
 129 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 130 module_param(updelay, int, 0);
 131 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 132 module_param(downdelay, int, 0);
 133 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 134                             "in milliseconds");
 135 module_param(use_carrier, int, 0);
 136 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 137                               "0 for off, 1 for on (default)");
 138 module_param(mode, charp, 0);
 139 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 140                        "1 for active-backup, 2 for balance-xor, "
 141                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 142                        "6 for balance-alb");
 143 module_param(primary, charp, 0);
 144 MODULE_PARM_DESC(primary, "Primary network device to use");
 145 module_param(primary_reselect, charp, 0);
 146 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 147                                    "once it comes up; "
 148                                    "0 for always (default), "
 149                                    "1 for only if speed of primary is "
 150                                    "better, "
 151                                    "2 for only on active slave "
 152                                    "failure");
 153 module_param(lacp_rate, charp, 0);
 154 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 155                             "0 for slow, 1 for fast");
 156 module_param(ad_select, charp, 0);
 157 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
 158                             "0 for stable (default), 1 for bandwidth, "
 159                             "2 for count");
 160 module_param(min_links, int, 0);
 161 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 162
 163 module_param(xmit_hash_policy, charp, 0);
 164 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
 165                                    "0 for layer 2 (default), 1 for layer 3+4, "
 166                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 167                                    "4 for encap layer 3+4");
 168 module_param(arp_interval, int, 0);
 169 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 170 module_param_array(arp_ip_target, charp, NULL, 0);
 171 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 172 module_param(arp_validate, charp, 0);
 173 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 174                                "0 for none (default), 1 for active, "
 175                                "2 for backup, 3 for all");
 176 module_param(arp_all_targets, charp, 0);
 177 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 178 module_param(fail_over_mac, charp, 0);
 179 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 180                                 "the same MAC; 0 for none (default), "
 181                                 "1 for active, 2 for follow");
 182 module_param(all_slaves_active, int, 0);
 183 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 184                                      "by setting active flag for all slaves; "
 185                                      "0 for never (default), 1 for always.");
 186 module_param(resend_igmp, int, 0);
 187 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 188                               "link failure");
 189 module_param(packets_per_slave, int, 0);
 190 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 191                                     "mode; 0 for a random slave, 1 packet per "
 192                                     "slave (default), >1 packets per slave.");
 193 module_param(lp_interval, uint, 0);
 194 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 195                               "the bonding driver sends learning packets to "
 196                               "each slaves peer switch. The default is 1.");
 197
 198 /*----------------------------- Global variables ----------------------------*/
 199
 200 #ifdef CONFIG_NET_POLL_CONTROLLER
 201 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 202 #endif
 203
 204 unsigned int bond_net_id __read_mostly;
 205
 206 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
 207         {
 208                 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
 209                 .offset = offsetof(struct flow_keys, control),
 210         },
 211         {
 212                 .key_id = FLOW_DISSECTOR_KEY_BASIC,
 213                 .offset = offsetof(struct flow_keys, basic),
 214         },
 215         {
 216                 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 217                 .offset = offsetof(struct flow_keys, addrs.v4addrs),
 218         },
 219         {
 220                 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 221                 .offset = offsetof(struct flow_keys, addrs.v6addrs),
 222         },
 223         {
 224                 .key_id = FLOW_DISSECTOR_KEY_TIPC,
 225                 .offset = offsetof(struct flow_keys, addrs.tipckey),
 226         },
 227         {
 228                 .key_id = FLOW_DISSECTOR_KEY_PORTS,
 229                 .offset = offsetof(struct flow_keys, ports),
 230         },
 231         {
 232                 .key_id = FLOW_DISSECTOR_KEY_ICMP,
 233                 .offset = offsetof(struct flow_keys, icmp),
 234         },
 235         {
 236                 .key_id = FLOW_DISSECTOR_KEY_VLAN,
 237                 .offset = offsetof(struct flow_keys, vlan),
 238         },
 239         {
 240                 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
 241                 .offset = offsetof(struct flow_keys, tags),
 242         },
 243         {
 244                 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
 245                 .offset = offsetof(struct flow_keys, keyid),
 246         },
 247 };
 248
 249 static struct flow_dissector flow_keys_bonding __read_mostly;
 250
 251 /*-------------------------- Forward declarations ---------------------------*/
 252
 253 static int bond_init(struct net_device *bond_dev);
 254 static void bond_uninit(struct net_device *bond_dev);
 255 static void bond_get_stats(struct net_device *bond_dev,
 256                            struct rtnl_link_stats64 *stats);
 257 static void bond_slave_arr_handler(struct work_struct *work);
 258 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 259                                   int mod);
 260 static void bond_netdev_notify_work(struct work_struct *work);
 261
 262 /*---------------------------- General routines -----------------------------*/
 263
 264 const char *bond_mode_name(int mode)
 265 {
 266         static const char *names[] = {
 267                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 268                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 269                 [BOND_MODE_XOR] = "load balancing (xor)",
 270                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 271                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 272                 [BOND_MODE_TLB] = "transmit load balancing",
 273                 [BOND_MODE_ALB] = "adaptive load balancing",
 274         };
 275
 276         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 277                 return "unknown";
 278
 279         return names[mode];
 280 }
 281
 282 /**
 283  * bond_dev_queue_xmit - Prepare skb for xmit.
 284  *
 285  * @bond: bond device that got this skb for tx.
 286  * @skb: hw accel VLAN tagged skb to transmit
 287  * @slave_dev: slave that is supposed to xmit this skbuff
 288  */
 289 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 290                         struct net_device *slave_dev)
 291 {
 292         skb->dev = slave_dev;
 293
 294         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 295                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 296         skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 297
 298         if (unlikely(netpoll_tx_running(bond->dev)))
 299                 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 300
 301         return dev_queue_xmit(skb);
 302 }
 303
 304 /*---------------------------------- VLAN -----------------------------------*/
 305
 306 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 307  * We don't protect the slave list iteration with a lock because:
 308  * a. This operation is performed in IOCTL context,
 309  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 310  * c. Holding a lock with BH disabled while directly calling a base driver
 311  *    entry point is generally a BAD idea.
 312  *
 313  * The design of synchronization/protection for this operation in the 8021q
 314  * module is good for one or more VLAN devices over a single physical device
 315  * and cannot be extended for a teaming solution like bonding, so there is a
 316  * potential race condition here where a net device from the vlan group might
 317  * be referenced (either by a base driver or the 8021q code) while it is being
 318  * removed from the system. However, it turns out we're not making matters
 319  * worse, and if it works for regular VLAN usage it will work here too.
 320 */
 321
 322 /**
 323  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 324  * @bond_dev: bonding net device that got called
 325  * @proto: network protocol ID
 326  * @vid: vlan id being added
 327  */
 328 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 329                                 __be16 proto, u16 vid)
 330 {
 331         struct bonding *bond = netdev_priv(bond_dev);
 332         struct slave *slave, *rollback_slave;
 333         struct list_head *iter;
 334         int res;
 335
 336         bond_for_each_slave(bond, slave, iter) {
 337                 res = vlan_vid_add(slave->dev, proto, vid);
 338                 if (res)
 339                         goto unwind;
 340         }
 341
 342         return 0;
 343
 344 unwind:
 345         /* unwind to the slave that failed */
 346         bond_for_each_slave(bond, rollback_slave, iter) {
 347                 if (rollback_slave == slave)
 348                         break;
 349
 350                 vlan_vid_del(rollback_slave->dev, proto, vid);
 351         }
 352
 353         return res;
 354 }
 355
 356 /**
 357  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 358  * @bond_dev: bonding net device that got called
 359  * @proto: network protocol ID
 360  * @vid: vlan id being removed
 361  */
 362 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 363                                  __be16 proto, u16 vid)
 364 {
 365         struct bonding *bond = netdev_priv(bond_dev);
 366         struct list_head *iter;
 367         struct slave *slave;
 368
 369         bond_for_each_slave(bond, slave, iter)
 370                 vlan_vid_del(slave->dev, proto, vid);
 371
 372         if (bond_is_lb(bond))
 373                 bond_alb_clear_vlan(bond, vid);
 374
 375         return 0;
 376 }
 377
 378 /*---------------------------------- XFRM -----------------------------------*/
 379
 380 #ifdef CONFIG_XFRM_OFFLOAD
 381 /**
 382  * bond_ipsec_add_sa - program device with a security association
 383  * @xs: pointer to transformer state struct
 384  **/
 385 static int bond_ipsec_add_sa(struct xfrm_state *xs)
 386 {
 387         struct net_device *bond_dev = xs->xso.dev;
 388         struct bonding *bond;
 389         struct slave *slave;
 390
 391         if (!bond_dev)
 392                 return -EINVAL;
 393
 394         bond = netdev_priv(bond_dev);
 395         slave = rcu_dereference(bond->curr_active_slave);
 396         xs->xso.real_dev = slave->dev;
 397         bond->xs = xs;
 398
 399         if (!(slave->dev->xfrmdev_ops
 400               && slave->dev->xfrmdev_ops->xdo_dev_state_add)) {
 401                 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
 402                 return -EINVAL;
 403         }
 404
 405         return slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
 406 }
 407
 408 /**
 409  * bond_ipsec_del_sa - clear out this specific SA
 410  * @xs: pointer to transformer state struct
 411  **/
 412 static void bond_ipsec_del_sa(struct xfrm_state *xs)
 413 {
 414         struct net_device *bond_dev = xs->xso.dev;
 415         struct bonding *bond;
 416         struct slave *slave;
 417
 418         if (!bond_dev)
 419                 return;
 420
 421         bond = netdev_priv(bond_dev);
 422         slave = rcu_dereference(bond->curr_active_slave);
 423
 424         if (!slave)
 425                 return;
 426
 427         xs->xso.real_dev = slave->dev;
 428
 429         if (!(slave->dev->xfrmdev_ops
 430               && slave->dev->xfrmdev_ops->xdo_dev_state_delete)) {
 431                 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
 432                 return;
 433         }
 434
 435         slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
 436 }
 437
 438 /**
 439  * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
 440  * @skb: current data packet
 441  * @xs: pointer to transformer state struct
 442  **/
 443 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
 444 {
 445         struct net_device *bond_dev = xs->xso.dev;
 446         struct bonding *bond = netdev_priv(bond_dev);
 447         struct slave *curr_active = rcu_dereference(bond->curr_active_slave);
 448         struct net_device *slave_dev = curr_active->dev;
 449
 450         if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
 451                 return true;
 452
 453         if (!(slave_dev->xfrmdev_ops
 454               && slave_dev->xfrmdev_ops->xdo_dev_offload_ok)) {
 455                 slave_warn(bond_dev, slave_dev, "%s: no slave xdo_dev_offload_ok\n", __func__);
 456                 return false;
 457         }
 458
 459         xs->xso.real_dev = slave_dev;
 460         return slave_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
 461 }
 462
 463 static const struct xfrmdev_ops bond_xfrmdev_ops = {
 464         .xdo_dev_state_add = bond_ipsec_add_sa,
 465         .xdo_dev_state_delete = bond_ipsec_del_sa,
 466         .xdo_dev_offload_ok = bond_ipsec_offload_ok,
 467 };
 468 #endif /* CONFIG_XFRM_OFFLOAD */
 469
 470 /*------------------------------- Link status -------------------------------*/
 471
 472 /* Set the carrier state for the master according to the state of its
 473  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 474  * do special 802.3ad magic.
 475  *
 476  * Returns zero if carrier state does not change, nonzero if it does.
 477  */
 478 int bond_set_carrier(struct bonding *bond)
 479 {
 480         struct list_head *iter;
 481         struct slave *slave;
 482
 483         if (!bond_has_slaves(bond))
 484                 goto down;
 485
 486         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 487                 return bond_3ad_set_carrier(bond);
 488
 489         bond_for_each_slave(bond, slave, iter) {
 490                 if (slave->link == BOND_LINK_UP) {
 491                         if (!netif_carrier_ok(bond->dev)) {
 492                                 netif_carrier_on(bond->dev);
 493                                 return 1;
 494                         }
 495                         return 0;
 496                 }
 497         }
 498
 499 down:
 500         if (netif_carrier_ok(bond->dev)) {
 501                 netif_carrier_off(bond->dev);
 502                 return 1;
 503         }
 504         return 0;
 505 }
 506
 507 /* Get link speed and duplex from the slave's base driver
 508  * using ethtool. If for some reason the call fails or the
 509  * values are invalid, set speed and duplex to -1,
 510  * and return. Return 1 if speed or duplex settings are
 511  * UNKNOWN; 0 otherwise.
 512  */
 513 static int bond_update_speed_duplex(struct slave *slave)
 514 {
 515         struct net_device *slave_dev = slave->dev;
 516         struct ethtool_link_ksettings ecmd;
 517         int res;
 518
 519         slave->speed = SPEED_UNKNOWN;
 520         slave->duplex = DUPLEX_UNKNOWN;
 521
 522         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 523         if (res < 0)
 524                 return 1;
 525         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 526                 return 1;
 527         switch (ecmd.base.duplex) {
 528         case DUPLEX_FULL:
 529         case DUPLEX_HALF:
 530                 break;
 531         default:
 532                 return 1;
 533         }
 534
 535         slave->speed = ecmd.base.speed;
 536         slave->duplex = ecmd.base.duplex;
 537
 538         return 0;
 539 }
 540
 541 const char *bond_slave_link_status(s8 link)
 542 {
 543         switch (link) {
 544         case BOND_LINK_UP:
 545                 return "up";
 546         case BOND_LINK_FAIL:
 547                 return "going down";
 548         case BOND_LINK_DOWN:
 549                 return "down";
 550         case BOND_LINK_BACK:
 551                 return "going back";
 552         default:
 553                 return "unknown";
 554         }
 555 }
 556
 557 /* if <dev> supports MII link status reporting, check its link status.
 558  *
 559  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 560  * depending upon the setting of the use_carrier parameter.
 561  *
 562  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 563  * can't tell and just pretend it is), or 0, meaning that the link is
 564  * down.
 565  *
 566  * If reporting is non-zero, instead of faking link up, return -1 if
 567  * both ETHTOOL and MII ioctls fail (meaning the device does not
 568  * support them).  If use_carrier is set, return whatever it says.
 569  * It'd be nice if there was a good way to tell if a driver supports
 570  * netif_carrier, but there really isn't.
 571  */
 572 static int bond_check_dev_link(struct bonding *bond,
 573                                struct net_device *slave_dev, int reporting)
 574 {
 575         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 576         int (*ioctl)(struct net_device *, struct ifreq *, int);
 577         struct ifreq ifr;
 578         struct mii_ioctl_data *mii;
 579
 580         if (!reporting && !netif_running(slave_dev))
 581                 return 0;
 582
 583         if (bond->params.use_carrier)
 584                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 585
 586         /* Try to get link status using Ethtool first. */
 587         if (slave_dev->ethtool_ops->get_link)
 588                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 589                         BMSR_LSTATUS : 0;
 590
 591         /* Ethtool can't be used, fallback to MII ioctls. */
 592         ioctl = slave_ops->ndo_do_ioctl;
 593         if (ioctl) {
 594                 /* TODO: set pointer to correct ioctl on a per team member
 595                  *       bases to make this more efficient. that is, once
 596                  *       we determine the correct ioctl, we will always
 597                  *       call it and not the others for that team
 598                  *       member.
 599                  */
 600
 601                 /* We cannot assume that SIOCGMIIPHY will also read a
 602                  * register; not all network drivers (e.g., e100)
 603                  * support that.
 604                  */
 605
 606                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 607                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 608                 mii = if_mii(&ifr);
 609                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 610                         mii->reg_num = MII_BMSR;
 611                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 612                                 return mii->val_out & BMSR_LSTATUS;
 613                 }
 614         }
 615
 616         /* If reporting, report that either there's no dev->do_ioctl,
 617          * or both SIOCGMIIREG and get_link failed (meaning that we
 618          * cannot report link status).  If not reporting, pretend
 619          * we're ok.
 620          */
 621         return reporting ? -1 : BMSR_LSTATUS;
 622 }
 623
 624 /*----------------------------- Multicast list ------------------------------*/
 625
 626 /* Push the promiscuity flag down to appropriate slaves */
 627 static int bond_set_promiscuity(struct bonding *bond, int inc)
 628 {
 629         struct list_head *iter;
 630         int err = 0;
 631
 632         if (bond_uses_primary(bond)) {
 633                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 634
 635                 if (curr_active)
 636                         err = dev_set_promiscuity(curr_active->dev, inc);
 637         } else {
 638                 struct slave *slave;
 639
 640                 bond_for_each_slave(bond, slave, iter) {
 641                         err = dev_set_promiscuity(slave->dev, inc);
 642                         if (err)
 643                                 return err;
 644                 }
 645         }
 646         return err;
 647 }
 648
 649 /* Push the allmulti flag down to all slaves */
 650 static int bond_set_allmulti(struct bonding *bond, int inc)
 651 {
 652         struct list_head *iter;
 653         int err = 0;
 654
 655         if (bond_uses_primary(bond)) {
 656                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 657
 658                 if (curr_active)
 659                         err = dev_set_allmulti(curr_active->dev, inc);
 660         } else {
 661                 struct slave *slave;
 662
 663                 bond_for_each_slave(bond, slave, iter) {
 664                         err = dev_set_allmulti(slave->dev, inc);
 665                         if (err)
 666                                 return err;
 667                 }
 668         }
 669         return err;
 670 }
 671
 672 /* Retrieve the list of registered multicast addresses for the bonding
 673  * device and retransmit an IGMP JOIN request to the current active
 674  * slave.
 675  */
 676 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 677 {
 678         struct bonding *bond = container_of(work, struct bonding,
 679                                             mcast_work.work);
 680
 681         if (!rtnl_trylock()) {
 682                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 683                 return;
 684         }
 685         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 686
 687         if (bond->igmp_retrans > 1) {
 688                 bond->igmp_retrans--;
 689                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 690         }
 691         rtnl_unlock();
 692 }
 693
 694 /* Flush bond's hardware addresses from slave */
 695 static void bond_hw_addr_flush(struct net_device *bond_dev,
 696                                struct net_device *slave_dev)
 697 {
 698         struct bonding *bond = netdev_priv(bond_dev);
 699
 700         dev_uc_unsync(slave_dev, bond_dev);
 701         dev_mc_unsync(slave_dev, bond_dev);
 702
 703         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 704                 /* del lacpdu mc addr from mc list */
 705                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 706
 707                 dev_mc_del(slave_dev, lacpdu_multicast);
 708         }
 709 }
 710
 711 /*--------------------------- Active slave change ---------------------------*/
 712
 713 /* Update the hardware address list and promisc/allmulti for the new and
 714  * old active slaves (if any).  Modes that are not using primary keep all
 715  * slaves up date at all times; only the modes that use primary need to call
 716  * this function to swap these settings during a failover.
 717  */
 718 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 719                               struct slave *old_active)
 720 {
 721         if (old_active) {
 722                 if (bond->dev->flags & IFF_PROMISC)
 723                         dev_set_promiscuity(old_active->dev, -1);
 724
 725                 if (bond->dev->flags & IFF_ALLMULTI)
 726                         dev_set_allmulti(old_active->dev, -1);
 727
 728                 bond_hw_addr_flush(bond->dev, old_active->dev);
 729         }
 730
 731         if (new_active) {
 732                 /* FIXME: Signal errors upstream. */
 733                 if (bond->dev->flags & IFF_PROMISC)
 734                         dev_set_promiscuity(new_active->dev, 1);
 735
 736                 if (bond->dev->flags & IFF_ALLMULTI)
 737                         dev_set_allmulti(new_active->dev, 1);
 738
 739                 netif_addr_lock_bh(bond->dev);
 740                 dev_uc_sync(new_active->dev, bond->dev);
 741                 dev_mc_sync(new_active->dev, bond->dev);
 742                 netif_addr_unlock_bh(bond->dev);
 743         }
 744 }
 745
 746 /**
 747  * bond_set_dev_addr - clone slave's address to bond
 748  * @bond_dev: bond net device
 749  * @slave_dev: slave net device
 750  *
 751  * Should be called with RTNL held.
 752  */
 753 static int bond_set_dev_addr(struct net_device *bond_dev,
 754                              struct net_device *slave_dev)
 755 {
 756         int err;
 757
 758         slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 759                   bond_dev, slave_dev, slave_dev->addr_len);
 760         err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
 761         if (err)
 762                 return err;
 763
 764         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 765         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 766         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 767         return 0;
 768 }
 769
 770 static struct slave *bond_get_old_active(struct bonding *bond,
 771                                          struct slave *new_active)
 772 {
 773         struct slave *slave;
 774         struct list_head *iter;
 775
 776         bond_for_each_slave(bond, slave, iter) {
 777                 if (slave == new_active)
 778                         continue;
 779
 780                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 781                         return slave;
 782         }
 783
 784         return NULL;
 785 }
 786
 787 /* bond_do_fail_over_mac
 788  *
 789  * Perform special MAC address swapping for fail_over_mac settings
 790  *
 791  * Called with RTNL
 792  */
 793 static void bond_do_fail_over_mac(struct bonding *bond,
 794                                   struct slave *new_active,
 795                                   struct slave *old_active)
 796 {
 797         u8 tmp_mac[MAX_ADDR_LEN];
 798         struct sockaddr_storage ss;
 799         int rv;
 800
 801         switch (bond->params.fail_over_mac) {
 802         case BOND_FOM_ACTIVE:
 803                 if (new_active) {
 804                         rv = bond_set_dev_addr(bond->dev, new_active->dev);
 805                         if (rv)
 806                                 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
 807                                           -rv);
 808                 }
 809                 break;
 810         case BOND_FOM_FOLLOW:
 811                 /* if new_active && old_active, swap them
 812                  * if just old_active, do nothing (going to no active slave)
 813                  * if just new_active, set new_active to bond's MAC
 814                  */
 815                 if (!new_active)
 816                         return;
 817
 818                 if (!old_active)
 819                         old_active = bond_get_old_active(bond, new_active);
 820
 821                 if (old_active) {
 822                         bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 823                                           new_active->dev->addr_len);
 824                         bond_hw_addr_copy(ss.__data,
 825                                           old_active->dev->dev_addr,
 826                                           old_active->dev->addr_len);
 827                         ss.ss_family = new_active->dev->type;
 828                 } else {
 829                         bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 830                                           bond->dev->addr_len);
 831                         ss.ss_family = bond->dev->type;
 832                 }
 833
 834                 rv = dev_set_mac_address(new_active->dev,
 835                                          (struct sockaddr *)&ss, NULL);
 836                 if (rv) {
 837                         slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
 838                                   -rv);
 839                         goto out;
 840                 }
 841
 842                 if (!old_active)
 843                         goto out;
 844
 845                 bond_hw_addr_copy(ss.__data, tmp_mac,
 846                                   new_active->dev->addr_len);
 847                 ss.ss_family = old_active->dev->type;
 848
 849                 rv = dev_set_mac_address(old_active->dev,
 850                                          (struct sockaddr *)&ss, NULL);
 851                 if (rv)
 852                         slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
 853                                   -rv);
 854 out:
 855                 break;
 856         default:
 857                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 858                            bond->params.fail_over_mac);
 859                 break;
 860         }
 861
 862 }
 863
 864 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 865 {
 866         struct slave *prim = rtnl_dereference(bond->primary_slave);
 867         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 868
 869         if (!prim || prim->link != BOND_LINK_UP) {
 870                 if (!curr || curr->link != BOND_LINK_UP)
 871                         return NULL;
 872                 return curr;
 873         }
 874
 875         if (bond->force_primary) {
 876                 bond->force_primary = false;
 877                 return prim;
 878         }
 879
 880         if (!curr || curr->link != BOND_LINK_UP)
 881                 return prim;
 882
 883         /* At this point, prim and curr are both up */
 884         switch (bond->params.primary_reselect) {
 885         case BOND_PRI_RESELECT_ALWAYS:
 886                 return prim;
 887         case BOND_PRI_RESELECT_BETTER:
 888                 if (prim->speed < curr->speed)
 889                         return curr;
 890                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 891                         return curr;
 892                 return prim;
 893         case BOND_PRI_RESELECT_FAILURE:
 894                 return curr;
 895         default:
 896                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 897                            bond->params.primary_reselect);
 898                 return curr;
 899         }
 900 }
 901
 902 /**
 903  * bond_find_best_slave - select the best available slave to be the active one
 904  * @bond: our bonding struct
 905  */
 906 static struct slave *bond_find_best_slave(struct bonding *bond)
 907 {
 908         struct slave *slave, *bestslave = NULL;
 909         struct list_head *iter;
 910         int mintime = bond->params.updelay;
 911
 912         slave = bond_choose_primary_or_current(bond);
 913         if (slave)
 914                 return slave;
 915
 916         bond_for_each_slave(bond, slave, iter) {
 917                 if (slave->link == BOND_LINK_UP)
 918                         return slave;
 919                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 920                     slave->delay < mintime) {
 921                         mintime = slave->delay;
 922                         bestslave = slave;
 923                 }
 924         }
 925
 926         return bestslave;
 927 }
 928
 929 static bool bond_should_notify_peers(struct bonding *bond)
 930 {
 931         struct slave *slave;
 932
 933         rcu_read_lock();
 934         slave = rcu_dereference(bond->curr_active_slave);
 935         rcu_read_unlock();
 936
 937         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 938                    slave ? slave->dev->name : "NULL");
 939
 940         if (!slave || !bond->send_peer_notif ||
 941             bond->send_peer_notif %
 942             max(1, bond->params.peer_notif_delay) != 0 ||
 943             !netif_carrier_ok(bond->dev) ||
 944             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 945                 return false;
 946
 947         return true;
 948 }
 949
 950 /**
 951  * change_active_interface - change the active slave into the specified one
 952  * @bond: our bonding struct
 953  * @new_active: the new slave to make the active one
 954  *
 955  * Set the new slave to the bond's settings and unset them on the old
 956  * curr_active_slave.
 957  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 958  *
 959  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 960  * because it is apparently the best available slave we have, even though its
 961  * updelay hasn't timed out yet.
 962  *
 963  * Caller must hold RTNL.
 964  */
 965 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 966 {
 967         struct slave *old_active;
 968
 969         ASSERT_RTNL();
 970
 971         old_active = rtnl_dereference(bond->curr_active_slave);
 972
 973         if (old_active == new_active)
 974                 return;
 975
 976 #ifdef CONFIG_XFRM_OFFLOAD
 977         if (old_active && bond->xs)
 978                 bond_ipsec_del_sa(bond->xs);
 979 #endif /* CONFIG_XFRM_OFFLOAD */
 980
 981         if (new_active) {
 982                 new_active->last_link_up = jiffies;
 983
 984                 if (new_active->link == BOND_LINK_BACK) {
 985                         if (bond_uses_primary(bond)) {
 986                                 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
 987                                            (bond->params.updelay - new_active->delay) * bond->params.miimon);
 988                         }
 989
 990                         new_active->delay = 0;
 991                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 992                                                   BOND_SLAVE_NOTIFY_NOW);
 993
 994                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 995                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 996
 997                         if (bond_is_lb(bond))
 998                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 999                 } else {
1000                         if (bond_uses_primary(bond)) {
1001                                 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1002                         }
1003                 }
1004         }
1005
1006         if (bond_uses_primary(bond))
1007                 bond_hw_addr_swap(bond, new_active, old_active);
1008
1009         if (bond_is_lb(bond)) {
1010                 bond_alb_handle_active_change(bond, new_active);
1011                 if (old_active)
1012                         bond_set_slave_inactive_flags(old_active,
1013                                                       BOND_SLAVE_NOTIFY_NOW);
1014                 if (new_active)
1015                         bond_set_slave_active_flags(new_active,
1016                                                     BOND_SLAVE_NOTIFY_NOW);
1017         } else {
1018                 rcu_assign_pointer(bond->curr_active_slave, new_active);
1019         }
1020
1021         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1022                 if (old_active)
1023                         bond_set_slave_inactive_flags(old_active,
1024                                                       BOND_SLAVE_NOTIFY_NOW);
1025
1026                 if (new_active) {
1027                         bool should_notify_peers = false;
1028
1029                         bond_set_slave_active_flags(new_active,
1030                                                     BOND_SLAVE_NOTIFY_NOW);
1031
1032                         if (bond->params.fail_over_mac)
1033                                 bond_do_fail_over_mac(bond, new_active,
1034                                                       old_active);
1035
1036                         if (netif_running(bond->dev)) {
1037                                 bond->send_peer_notif =
1038                                         bond->params.num_peer_notif *
1039                                         max(1, bond->params.peer_notif_delay);
1040                                 should_notify_peers =
1041                                         bond_should_notify_peers(bond);
1042                         }
1043
1044                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1045                         if (should_notify_peers) {
1046                                 bond->send_peer_notif--;
1047                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1048                                                          bond->dev);
1049                         }
1050                 }
1051         }
1052
1053 #ifdef CONFIG_XFRM_OFFLOAD
1054         if (new_active && bond->xs) {
1055                 xfrm_dev_state_flush(dev_net(bond->dev), bond->dev, true);
1056                 bond_ipsec_add_sa(bond->xs);
1057         }
1058 #endif /* CONFIG_XFRM_OFFLOAD */
1059
1060         /* resend IGMP joins since active slave has changed or
1061          * all were sent on curr_active_slave.
1062          * resend only if bond is brought up with the affected
1063          * bonding modes and the retransmission is enabled
1064          */
1065         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1066             ((bond_uses_primary(bond) && new_active) ||
1067              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1068                 bond->igmp_retrans = bond->params.resend_igmp;
1069                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1070         }
1071 }
1072
1073 /**
1074  * bond_select_active_slave - select a new active slave, if needed
1075  * @bond: our bonding struct
1076  *
1077  * This functions should be called when one of the following occurs:
1078  * - The old curr_active_slave has been released or lost its link.
1079  * - The primary_slave has got its link back.
1080  * - A slave has got its link back and there's no old curr_active_slave.
1081  *
1082  * Caller must hold RTNL.
1083  */
1084 void bond_select_active_slave(struct bonding *bond)
1085 {
1086         struct slave *best_slave;
1087         int rv;
1088
1089         ASSERT_RTNL();
1090
1091         best_slave = bond_find_best_slave(bond);
1092         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1093                 bond_change_active_slave(bond, best_slave);
1094                 rv = bond_set_carrier(bond);
1095                 if (!rv)
1096                         return;
1097
1098                 if (netif_carrier_ok(bond->dev))
1099                         netdev_info(bond->dev, "active interface up!\n");
1100                 else
1101                         netdev_info(bond->dev, "now running without any active interface!\n");
1102         }
1103 }
1104
1105 #ifdef CONFIG_NET_POLL_CONTROLLER
1106 static inline int slave_enable_netpoll(struct slave *slave)
1107 {
1108         struct netpoll *np;
1109         int err = 0;
1110
1111         np = kzalloc(sizeof(*np), GFP_KERNEL);
1112         err = -ENOMEM;
1113         if (!np)
1114                 goto out;
1115
1116         err = __netpoll_setup(np, slave->dev);
1117         if (err) {
1118                 kfree(np);
1119                 goto out;
1120         }
1121         slave->np = np;
1122 out:
1123         return err;
1124 }
1125 static inline void slave_disable_netpoll(struct slave *slave)
1126 {
1127         struct netpoll *np = slave->np;
1128
1129         if (!np)
1130                 return;
1131
1132         slave->np = NULL;
1133
1134         __netpoll_free(np);
1135 }
1136
1137 static void bond_poll_controller(struct net_device *bond_dev)
1138 {
1139         struct bonding *bond = netdev_priv(bond_dev);
1140         struct slave *slave = NULL;
1141         struct list_head *iter;
1142         struct ad_info ad_info;
1143
1144         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1145                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1146                         return;
1147
1148         bond_for_each_slave_rcu(bond, slave, iter) {
1149                 if (!bond_slave_is_up(slave))
1150                         continue;
1151
1152                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1153                         struct aggregator *agg =
1154                             SLAVE_AD_INFO(slave)->port.aggregator;
1155
1156                         if (agg &&
1157                             agg->aggregator_identifier != ad_info.aggregator_id)
1158                                 continue;
1159                 }
1160
1161                 netpoll_poll_dev(slave->dev);
1162         }
1163 }
1164
1165 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1166 {
1167         struct bonding *bond = netdev_priv(bond_dev);
1168         struct list_head *iter;
1169         struct slave *slave;
1170
1171         bond_for_each_slave(bond, slave, iter)
1172                 if (bond_slave_is_up(slave))
1173                         slave_disable_netpoll(slave);
1174 }
1175
1176 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1177 {
1178         struct bonding *bond = netdev_priv(dev);
1179         struct list_head *iter;
1180         struct slave *slave;
1181         int err = 0;
1182
1183         bond_for_each_slave(bond, slave, iter) {
1184                 err = slave_enable_netpoll(slave);
1185                 if (err) {
1186                         bond_netpoll_cleanup(dev);
1187                         break;
1188                 }
1189         }
1190         return err;
1191 }
1192 #else
1193 static inline int slave_enable_netpoll(struct slave *slave)
1194 {
1195         return 0;
1196 }
1197 static inline void slave_disable_netpoll(struct slave *slave)
1198 {
1199 }
1200 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1201 {
1202 }
1203 #endif
1204
1205 /*---------------------------------- IOCTL ----------------------------------*/
1206
1207 static netdev_features_t bond_fix_features(struct net_device *dev,
1208                                            netdev_features_t features)
1209 {
1210         struct bonding *bond = netdev_priv(dev);
1211         struct list_head *iter;
1212         netdev_features_t mask;
1213         struct slave *slave;
1214
1215         mask = features;
1216
1217         features &= ~NETIF_F_ONE_FOR_ALL;
1218         features |= NETIF_F_ALL_FOR_ALL;
1219
1220         bond_for_each_slave(bond, slave, iter) {
1221                 features = netdev_increment_features(features,
1222                                                      slave->dev->features,
1223                                                      mask);
1224         }
1225         features = netdev_add_tso_features(features, mask);
1226
1227         return features;
1228 }
1229
1230 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1231                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1232                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1233
1234 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1235                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1236
1237 #define BOND_MPLS_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1238                                  NETIF_F_ALL_TSO)
1239
1240
1241 static void bond_compute_features(struct bonding *bond)
1242 {
1243         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1244                                         IFF_XMIT_DST_RELEASE_PERM;
1245         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1246         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1247 #ifdef CONFIG_XFRM_OFFLOAD
1248         netdev_features_t xfrm_features  = BOND_XFRM_FEATURES;
1249 #endif /* CONFIG_XFRM_OFFLOAD */
1250         netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
1251         struct net_device *bond_dev = bond->dev;
1252         struct list_head *iter;
1253         struct slave *slave;
1254         unsigned short max_hard_header_len = ETH_HLEN;
1255         unsigned int gso_max_size = GSO_MAX_SIZE;
1256         u16 gso_max_segs = GSO_MAX_SEGS;
1257
1258         if (!bond_has_slaves(bond))
1259                 goto done;
1260         vlan_features &= NETIF_F_ALL_FOR_ALL;
1261         mpls_features &= NETIF_F_ALL_FOR_ALL;
1262
1263         bond_for_each_slave(bond, slave, iter) {
1264                 vlan_features = netdev_increment_features(vlan_features,
1265                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1266
1267                 enc_features = netdev_increment_features(enc_features,
1268                                                          slave->dev->hw_enc_features,
1269                                                          BOND_ENC_FEATURES);
1270
1271 #ifdef CONFIG_XFRM_OFFLOAD
1272                 xfrm_features = netdev_increment_features(xfrm_features,
1273                                                           slave->dev->hw_enc_features,
1274                                                           BOND_XFRM_FEATURES);
1275 #endif /* CONFIG_XFRM_OFFLOAD */
1276
1277                 mpls_features = netdev_increment_features(mpls_features,
1278                                                           slave->dev->mpls_features,
1279                                                           BOND_MPLS_FEATURES);
1280
1281                 dst_release_flag &= slave->dev->priv_flags;
1282                 if (slave->dev->hard_header_len > max_hard_header_len)
1283                         max_hard_header_len = slave->dev->hard_header_len;
1284
1285                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1286                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1287         }
1288         bond_dev->hard_header_len = max_hard_header_len;
1289
1290 done:
1291         bond_dev->vlan_features = vlan_features;
1292         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1293                                     NETIF_F_HW_VLAN_CTAG_TX |
1294                                     NETIF_F_HW_VLAN_STAG_TX |
1295                                     NETIF_F_GSO_UDP_L4;
1296 #ifdef CONFIG_XFRM_OFFLOAD
1297         bond_dev->hw_enc_features |= xfrm_features;
1298 #endif /* CONFIG_XFRM_OFFLOAD */
1299         bond_dev->mpls_features = mpls_features;
1300         bond_dev->gso_max_segs = gso_max_segs;
1301         netif_set_gso_max_size(bond_dev, gso_max_size);
1302
1303         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1304         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1305             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1306                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1307
1308         netdev_change_features(bond_dev);
1309 }
1310
1311 static void bond_setup_by_slave(struct net_device *bond_dev,
1312                                 struct net_device *slave_dev)
1313 {
1314         bond_dev->header_ops        = slave_dev->header_ops;
1315
1316         bond_dev->type              = slave_dev->type;
1317         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1318         bond_dev->needed_headroom   = slave_dev->needed_headroom;
1319         bond_dev->addr_len          = slave_dev->addr_len;
1320
1321         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1322                 slave_dev->addr_len);
1323 }
1324
1325 /* On bonding slaves other than the currently active slave, suppress
1326  * duplicates except for alb non-mcast/bcast.
1327  */
1328 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1329                                             struct slave *slave,
1330                                             struct bonding *bond)
1331 {
1332         if (bond_is_slave_inactive(slave)) {
1333                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1334                     skb->pkt_type != PACKET_BROADCAST &&
1335                     skb->pkt_type != PACKET_MULTICAST)
1336                         return false;
1337                 return true;
1338         }
1339         return false;
1340 }
1341
1342 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1343 {
1344         struct sk_buff *skb = *pskb;
1345         struct slave *slave;
1346         struct bonding *bond;
1347         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1348                           struct slave *);
1349         int ret = RX_HANDLER_ANOTHER;
1350
1351         skb = skb_share_check(skb, GFP_ATOMIC);
1352         if (unlikely(!skb))
1353                 return RX_HANDLER_CONSUMED;
1354
1355         *pskb = skb;
1356
1357         slave = bond_slave_get_rcu(skb->dev);
1358         bond = slave->bond;
1359
1360         recv_probe = READ_ONCE(bond->recv_probe);
1361         if (recv_probe) {
1362                 ret = recv_probe(skb, bond, slave);
1363                 if (ret == RX_HANDLER_CONSUMED) {
1364                         consume_skb(skb);
1365                         return ret;
1366                 }
1367         }
1368
1369         /*
1370          * For packets determined by bond_should_deliver_exact_match() call to
1371          * be suppressed we want to make an exception for link-local packets.
1372          * This is necessary for e.g. LLDP daemons to be able to monitor
1373          * inactive slave links without being forced to bind to them
1374          * explicitly.
1375          *
1376          * At the same time, packets that are passed to the bonding master
1377          * (including link-local ones) can have their originating interface
1378          * determined via PACKET_ORIGDEV socket option.
1379          */
1380         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1381                 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1382                         return RX_HANDLER_PASS;
1383                 return RX_HANDLER_EXACT;
1384         }
1385
1386         skb->dev = bond->dev;
1387
1388         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1389             netif_is_bridge_port(bond->dev) &&
1390             skb->pkt_type == PACKET_HOST) {
1391
1392                 if (unlikely(skb_cow_head(skb,
1393                                           skb->data - skb_mac_header(skb)))) {
1394                         kfree_skb(skb);
1395                         return RX_HANDLER_CONSUMED;
1396                 }
1397                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1398                                   bond->dev->addr_len);
1399         }
1400
1401         return ret;
1402 }
1403
1404 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1405 {
1406         switch (BOND_MODE(bond)) {
1407         case BOND_MODE_ROUNDROBIN:
1408                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1409         case BOND_MODE_ACTIVEBACKUP:
1410                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1411         case BOND_MODE_BROADCAST:
1412                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1413         case BOND_MODE_XOR:
1414         case BOND_MODE_8023AD:
1415                 return NETDEV_LAG_TX_TYPE_HASH;
1416         default:
1417                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1418         }
1419 }
1420
1421 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1422                                                enum netdev_lag_tx_type type)
1423 {
1424         if (type != NETDEV_LAG_TX_TYPE_HASH)
1425                 return NETDEV_LAG_HASH_NONE;
1426
1427         switch (bond->params.xmit_policy) {
1428         case BOND_XMIT_POLICY_LAYER2:
1429                 return NETDEV_LAG_HASH_L2;
1430         case BOND_XMIT_POLICY_LAYER34:
1431                 return NETDEV_LAG_HASH_L34;
1432         case BOND_XMIT_POLICY_LAYER23:
1433                 return NETDEV_LAG_HASH_L23;
1434         case BOND_XMIT_POLICY_ENCAP23:
1435                 return NETDEV_LAG_HASH_E23;
1436         case BOND_XMIT_POLICY_ENCAP34:
1437                 return NETDEV_LAG_HASH_E34;
1438         default:
1439                 return NETDEV_LAG_HASH_UNKNOWN;
1440         }
1441 }
1442
1443 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1444                                       struct netlink_ext_ack *extack)
1445 {
1446         struct netdev_lag_upper_info lag_upper_info;
1447         enum netdev_lag_tx_type type;
1448
1449         type = bond_lag_tx_type(bond);
1450         lag_upper_info.tx_type = type;
1451         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1452
1453         return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1454                                             &lag_upper_info, extack);
1455 }
1456
1457 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1458 {
1459         netdev_upper_dev_unlink(slave->dev, bond->dev);
1460         slave->dev->flags &= ~IFF_SLAVE;
1461 }
1462
1463 static void slave_kobj_release(struct kobject *kobj)
1464 {
1465         struct slave *slave = to_slave(kobj);
1466         struct bonding *bond = bond_get_bond_by_slave(slave);
1467
1468         cancel_delayed_work_sync(&slave->notify_work);
1469         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1470                 kfree(SLAVE_AD_INFO(slave));
1471
1472         kfree(slave);
1473 }
1474
1475 static struct kobj_type slave_ktype = {
1476         .release = slave_kobj_release,
1477 #ifdef CONFIG_SYSFS
1478         .sysfs_ops = &slave_sysfs_ops,
1479 #endif
1480 };
1481
1482 static int bond_kobj_init(struct slave *slave)
1483 {
1484         int err;
1485
1486         err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1487                                    &(slave->dev->dev.kobj), "bonding_slave");
1488         if (err)
1489                 kobject_put(&slave->kobj);
1490
1491         return err;
1492 }
1493
1494 static struct slave *bond_alloc_slave(struct bonding *bond,
1495                                       struct net_device *slave_dev)
1496 {
1497         struct slave *slave = NULL;
1498
1499         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1500         if (!slave)
1501                 return NULL;
1502
1503         slave->bond = bond;
1504         slave->dev = slave_dev;
1505
1506         if (bond_kobj_init(slave))
1507                 return NULL;
1508
1509         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1510                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1511                                                GFP_KERNEL);
1512                 if (!SLAVE_AD_INFO(slave)) {
1513                         kobject_put(&slave->kobj);
1514                         return NULL;
1515                 }
1516         }
1517         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1518
1519         return slave;
1520 }
1521
1522 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1523 {
1524         info->bond_mode = BOND_MODE(bond);
1525         info->miimon = bond->params.miimon;
1526         info->num_slaves = bond->slave_cnt;
1527 }
1528
1529 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1530 {
1531         strcpy(info->slave_name, slave->dev->name);
1532         info->link = slave->link;
1533         info->state = bond_slave_state(slave);
1534         info->link_failure_count = slave->link_failure_count;
1535 }
1536
1537 static void bond_netdev_notify_work(struct work_struct *_work)
1538 {
1539         struct slave *slave = container_of(_work, struct slave,
1540                                            notify_work.work);
1541
1542         if (rtnl_trylock()) {
1543                 struct netdev_bonding_info binfo;
1544
1545                 bond_fill_ifslave(slave, &binfo.slave);
1546                 bond_fill_ifbond(slave->bond, &binfo.master);
1547                 netdev_bonding_info_change(slave->dev, &binfo);
1548                 rtnl_unlock();
1549         } else {
1550                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1551         }
1552 }
1553
1554 void bond_queue_slave_event(struct slave *slave)
1555 {
1556         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1557 }
1558
1559 void bond_lower_state_changed(struct slave *slave)
1560 {
1561         struct netdev_lag_lower_state_info info;
1562
1563         info.link_up = slave->link == BOND_LINK_UP ||
1564                        slave->link == BOND_LINK_FAIL;
1565         info.tx_enabled = bond_is_active_slave(slave);
1566         netdev_lower_state_changed(slave->dev, &info);
1567 }
1568
1569 /* enslave device <slave> to bond device <master> */
1570 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1571                  struct netlink_ext_ack *extack)
1572 {
1573         struct bonding *bond = netdev_priv(bond_dev);
1574         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1575         struct slave *new_slave = NULL, *prev_slave;
1576         struct sockaddr_storage ss;
1577         int link_reporting;
1578         int res = 0, i;
1579
1580         if (!bond->params.use_carrier &&
1581             slave_dev->ethtool_ops->get_link == NULL &&
1582             slave_ops->ndo_do_ioctl == NULL) {
1583                 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1584         }
1585
1586         /* already in-use? */
1587         if (netdev_is_rx_handler_busy(slave_dev)) {
1588                 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1589                 slave_err(bond_dev, slave_dev,
1590                           "Error: Device is in use and cannot be enslaved\n");
1591                 return -EBUSY;
1592         }
1593
1594         if (bond_dev == slave_dev) {
1595                 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1596                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1597                 return -EPERM;
1598         }
1599
1600         /* vlan challenged mutual exclusion */
1601         /* no need to lock since we're protected by rtnl_lock */
1602         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1603                 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1604                 if (vlan_uses_dev(bond_dev)) {
1605                         NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1606                         slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1607                         return -EPERM;
1608                 } else {
1609                         slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1610                 }
1611         } else {
1612                 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1613         }
1614
1615         if (slave_dev->features & NETIF_F_HW_ESP)
1616                 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1617
1618         /* Old ifenslave binaries are no longer supported.  These can
1619          * be identified with moderate accuracy by the state of the slave:
1620          * the current ifenslave will set the interface down prior to
1621          * enslaving it; the old ifenslave will not.
1622          */
1623         if (slave_dev->flags & IFF_UP) {
1624                 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1625                 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1626                 return -EPERM;
1627         }
1628
1629         /* set bonding device ether type by slave - bonding netdevices are
1630          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1631          * there is a need to override some of the type dependent attribs/funcs.
1632          *
1633          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1634          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1635          */
1636         if (!bond_has_slaves(bond)) {
1637                 if (bond_dev->type != slave_dev->type) {
1638                         slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1639                                   bond_dev->type, slave_dev->type);
1640
1641                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1642                                                        bond_dev);
1643                         res = notifier_to_errno(res);
1644                         if (res) {
1645                                 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1646                                 return -EBUSY;
1647                         }
1648
1649                         /* Flush unicast and multicast addresses */
1650                         dev_uc_flush(bond_dev);
1651                         dev_mc_flush(bond_dev);
1652
1653                         if (slave_dev->type != ARPHRD_ETHER)
1654                                 bond_setup_by_slave(bond_dev, slave_dev);
1655                         else {
1656                                 ether_setup(bond_dev);
1657                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1658                         }
1659
1660                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1661                                                  bond_dev);
1662                 }
1663         } else if (bond_dev->type != slave_dev->type) {
1664                 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1665                 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1666                           slave_dev->type, bond_dev->type);
1667                 return -EINVAL;
1668         }
1669
1670         if (slave_dev->type == ARPHRD_INFINIBAND &&
1671             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1672                 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1673                 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1674                            slave_dev->type);
1675                 res = -EOPNOTSUPP;
1676                 goto err_undo_flags;
1677         }
1678
1679         if (!slave_ops->ndo_set_mac_address ||
1680             slave_dev->type == ARPHRD_INFINIBAND) {
1681                 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1682                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1683                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1684                         if (!bond_has_slaves(bond)) {
1685                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1686                                 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1687                         } else {
1688                                 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1689                                 slave_err(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1690                                 res = -EOPNOTSUPP;
1691                                 goto err_undo_flags;
1692                         }
1693                 }
1694         }
1695
1696         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1697
1698         /* If this is the first slave, then we need to set the master's hardware
1699          * address to be the same as the slave's.
1700          */
1701         if (!bond_has_slaves(bond) &&
1702             bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1703                 res = bond_set_dev_addr(bond->dev, slave_dev);
1704                 if (res)
1705                         goto err_undo_flags;
1706         }
1707
1708         new_slave = bond_alloc_slave(bond, slave_dev);
1709         if (!new_slave) {
1710                 res = -ENOMEM;
1711                 goto err_undo_flags;
1712         }
1713
1714         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1715          * is set via sysfs or module option if desired.
1716          */
1717         new_slave->queue_id = 0;
1718
1719         /* Save slave's original mtu and then set it to match the bond */
1720         new_slave->original_mtu = slave_dev->mtu;
1721         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1722         if (res) {
1723                 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1724                 goto err_free;
1725         }
1726
1727         /* Save slave's original ("permanent") mac address for modes
1728          * that need it, and for restoring it upon release, and then
1729          * set it to the master's address
1730          */
1731         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1732                           slave_dev->addr_len);
1733
1734         if (!bond->params.fail_over_mac ||
1735             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1736                 /* Set slave to master's mac address.  The application already
1737                  * set the master's mac address to that of the first slave
1738                  */
1739                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1740                 ss.ss_family = slave_dev->type;
1741                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1742                                           extack);
1743                 if (res) {
1744                         slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1745                         goto err_restore_mtu;
1746                 }
1747         }
1748
1749         /* set slave flag before open to prevent IPv6 addrconf */
1750         slave_dev->flags |= IFF_SLAVE;
1751
1752         /* open the slave since the application closed it */
1753         res = dev_open(slave_dev, extack);
1754         if (res) {
1755                 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1756                 goto err_restore_mac;
1757         }
1758
1759         slave_dev->priv_flags |= IFF_BONDING;
1760         /* initialize slave stats */
1761         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1762
1763         if (bond_is_lb(bond)) {
1764                 /* bond_alb_init_slave() must be called before all other stages since
1765                  * it might fail and we do not want to have to undo everything
1766                  */
1767                 res = bond_alb_init_slave(bond, new_slave);
1768                 if (res)
1769                         goto err_close;
1770         }
1771
1772         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1773         if (res) {
1774                 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1775                 goto err_close;
1776         }
1777
1778         prev_slave = bond_last_slave(bond);
1779
1780         new_slave->delay = 0;
1781         new_slave->link_failure_count = 0;
1782
1783         if (bond_update_speed_duplex(new_slave) &&
1784             bond_needs_speed_duplex(bond))
1785                 new_slave->link = BOND_LINK_DOWN;
1786
1787         new_slave->last_rx = jiffies -
1788                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1789         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1790                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1791
1792         if (bond->params.miimon && !bond->params.use_carrier) {
1793                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1794
1795                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1796                         /* miimon is set but a bonded network driver
1797                          * does not support ETHTOOL/MII and
1798                          * arp_interval is not set.  Note: if
1799                          * use_carrier is enabled, we will never go
1800                          * here (because netif_carrier is always
1801                          * supported); thus, we don't need to change
1802                          * the messages for netif_carrier.
1803                          */
1804                         slave_warn(bond_dev, slave_dev, "MII and ETHTOOL support not available for slave, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details\n");
1805                 } else if (link_reporting == -1) {
1806                         /* unable get link status using mii/ethtool */
1807                         slave_warn(bond_dev, slave_dev, "can't get link status from slave; 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");
1808                 }
1809         }
1810
1811         /* check for initial state */
1812         new_slave->link = BOND_LINK_NOCHANGE;
1813         if (bond->params.miimon) {
1814                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1815                         if (bond->params.updelay) {
1816                                 bond_set_slave_link_state(new_slave,
1817                                                           BOND_LINK_BACK,
1818                                                           BOND_SLAVE_NOTIFY_NOW);
1819                                 new_slave->delay = bond->params.updelay;
1820                         } else {
1821                                 bond_set_slave_link_state(new_slave,
1822                                                           BOND_LINK_UP,
1823                                                           BOND_SLAVE_NOTIFY_NOW);
1824                         }
1825                 } else {
1826                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1827                                                   BOND_SLAVE_NOTIFY_NOW);
1828                 }
1829         } else if (bond->params.arp_interval) {
1830                 bond_set_slave_link_state(new_slave,
1831                                           (netif_carrier_ok(slave_dev) ?
1832                                           BOND_LINK_UP : BOND_LINK_DOWN),
1833                                           BOND_SLAVE_NOTIFY_NOW);
1834         } else {
1835                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1836                                           BOND_SLAVE_NOTIFY_NOW);
1837         }
1838
1839         if (new_slave->link != BOND_LINK_DOWN)
1840                 new_slave->last_link_up = jiffies;
1841         slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1842                   new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1843                   (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1844
1845         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1846                 /* if there is a primary slave, remember it */
1847                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1848                         rcu_assign_pointer(bond->primary_slave, new_slave);
1849                         bond->force_primary = true;
1850                 }
1851         }
1852
1853         switch (BOND_MODE(bond)) {
1854         case BOND_MODE_ACTIVEBACKUP:
1855                 bond_set_slave_inactive_flags(new_slave,
1856                                               BOND_SLAVE_NOTIFY_NOW);
1857                 break;
1858         case BOND_MODE_8023AD:
1859                 /* in 802.3ad mode, the internal mechanism
1860                  * will activate the slaves in the selected
1861                  * aggregator
1862                  */
1863                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1864                 /* if this is the first slave */
1865                 if (!prev_slave) {
1866                         SLAVE_AD_INFO(new_slave)->id = 1;
1867                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1868                          * can be called only after the mac address of the bond is set
1869                          */
1870                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1871                 } else {
1872                         SLAVE_AD_INFO(new_slave)->id =
1873                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1874                 }
1875
1876                 bond_3ad_bind_slave(new_slave);
1877                 break;
1878         case BOND_MODE_TLB:
1879         case BOND_MODE_ALB:
1880                 bond_set_active_slave(new_slave);
1881                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1882                 break;
1883         default:
1884                 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
1885
1886                 /* always active in trunk mode */
1887                 bond_set_active_slave(new_slave);
1888
1889                 /* In trunking mode there is little meaning to curr_active_slave
1890                  * anyway (it holds no special properties of the bond device),
1891                  * so we can change it without calling change_active_interface()
1892                  */
1893                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1894                     new_slave->link == BOND_LINK_UP)
1895                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1896
1897                 break;
1898         } /* switch(bond_mode) */
1899
1900 #ifdef CONFIG_NET_POLL_CONTROLLER
1901         if (bond->dev->npinfo) {
1902                 if (slave_enable_netpoll(new_slave)) {
1903                         slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1904                         res = -EBUSY;
1905                         goto err_detach;
1906                 }
1907         }
1908 #endif
1909
1910         if (!(bond_dev->features & NETIF_F_LRO))
1911                 dev_disable_lro(slave_dev);
1912
1913         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1914                                          new_slave);
1915         if (res) {
1916                 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
1917                 goto err_detach;
1918         }
1919
1920         res = bond_master_upper_dev_link(bond, new_slave, extack);
1921         if (res) {
1922                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1923                 goto err_unregister;
1924         }
1925
1926         res = bond_sysfs_slave_add(new_slave);
1927         if (res) {
1928                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1929                 goto err_upper_unlink;
1930         }
1931
1932         /* If the mode uses primary, then the following is handled by
1933          * bond_change_active_slave().
1934          */
1935         if (!bond_uses_primary(bond)) {
1936                 /* set promiscuity level to new slave */
1937                 if (bond_dev->flags & IFF_PROMISC) {
1938                         res = dev_set_promiscuity(slave_dev, 1);
1939                         if (res)
1940                                 goto err_sysfs_del;
1941                 }
1942
1943                 /* set allmulti level to new slave */
1944                 if (bond_dev->flags & IFF_ALLMULTI) {
1945                         res = dev_set_allmulti(slave_dev, 1);
1946                         if (res) {
1947                                 if (bond_dev->flags & IFF_PROMISC)
1948                                         dev_set_promiscuity(slave_dev, -1);
1949                                 goto err_sysfs_del;
1950                         }
1951                 }
1952
1953                 netif_addr_lock_bh(bond_dev);
1954                 dev_mc_sync_multiple(slave_dev, bond_dev);
1955                 dev_uc_sync_multiple(slave_dev, bond_dev);
1956                 netif_addr_unlock_bh(bond_dev);
1957
1958                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1959                         /* add lacpdu mc addr to mc list */
1960                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1961
1962                         dev_mc_add(slave_dev, lacpdu_multicast);
1963                 }
1964         }
1965
1966         bond->slave_cnt++;
1967         bond_compute_features(bond);
1968         bond_set_carrier(bond);
1969
1970         if (bond_uses_primary(bond)) {
1971                 block_netpoll_tx();
1972                 bond_select_active_slave(bond);
1973                 unblock_netpoll_tx();
1974         }
1975
1976         if (bond_mode_can_use_xmit_hash(bond))
1977                 bond_update_slave_arr(bond, NULL);
1978
1979
1980         slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
1981                    bond_is_active_slave(new_slave) ? "an active" : "a backup",
1982                    new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1983
1984         /* enslave is successful */
1985         bond_queue_slave_event(new_slave);
1986         return 0;
1987
1988 /* Undo stages on error */
1989 err_sysfs_del:
1990         bond_sysfs_slave_del(new_slave);
1991
1992 err_upper_unlink:
1993         bond_upper_dev_unlink(bond, new_slave);
1994
1995 err_unregister:
1996         netdev_rx_handler_unregister(slave_dev);
1997
1998 err_detach:
1999         vlan_vids_del_by_dev(slave_dev, bond_dev);
2000         if (rcu_access_pointer(bond->primary_slave) == new_slave)
2001                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2002         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2003                 block_netpoll_tx();
2004                 bond_change_active_slave(bond, NULL);
2005                 bond_select_active_slave(bond);
2006                 unblock_netpoll_tx();
2007         }
2008         /* either primary_slave or curr_active_slave might've changed */
2009         synchronize_rcu();
2010         slave_disable_netpoll(new_slave);
2011
2012 err_close:
2013         if (!netif_is_bond_master(slave_dev))
2014                 slave_dev->priv_flags &= ~IFF_BONDING;
2015         dev_close(slave_dev);
2016
2017 err_restore_mac:
2018         slave_dev->flags &= ~IFF_SLAVE;
2019         if (!bond->params.fail_over_mac ||
2020             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2021                 /* XXX TODO - fom follow mode needs to change master's
2022                  * MAC if this slave's MAC is in use by the bond, or at
2023                  * least print a warning.
2024                  */
2025                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2026                                   new_slave->dev->addr_len);
2027                 ss.ss_family = slave_dev->type;
2028                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2029         }
2030
2031 err_restore_mtu:
2032         dev_set_mtu(slave_dev, new_slave->original_mtu);
2033
2034 err_free:
2035         kobject_put(&new_slave->kobj);
2036
2037 err_undo_flags:
2038         /* Enslave of first slave has failed and we need to fix master's mac */
2039         if (!bond_has_slaves(bond)) {
2040                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2041                                             slave_dev->dev_addr))
2042                         eth_hw_addr_random(bond_dev);
2043                 if (bond_dev->type != ARPHRD_ETHER) {
2044                         dev_close(bond_dev);
2045                         ether_setup(bond_dev);
2046                         bond_dev->flags |= IFF_MASTER;
2047                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2048                 }
2049         }
2050
2051         return res;
2052 }
2053
2054 /* Try to release the slave device <slave> from the bond device <master>
2055  * It is legal to access curr_active_slave without a lock because all the function
2056  * is RTNL-locked. If "all" is true it means that the function is being called
2057  * while destroying a bond interface and all slaves are being released.
2058  *
2059  * The rules for slave state should be:
2060  *   for Active/Backup:
2061  *     Active stays on all backups go down
2062  *   for Bonded connections:
2063  *     The first up interface should be left on and all others downed.
2064  */
2065 static int __bond_release_one(struct net_device *bond_dev,
2066                               struct net_device *slave_dev,
2067                               bool all, bool unregister)
2068 {
2069         struct bonding *bond = netdev_priv(bond_dev);
2070         struct slave *slave, *oldcurrent;
2071         struct sockaddr_storage ss;
2072         int old_flags = bond_dev->flags;
2073         netdev_features_t old_features = bond_dev->features;
2074
2075         /* slave is not a slave or master is not master of this slave */
2076         if (!(slave_dev->flags & IFF_SLAVE) ||
2077             !netdev_has_upper_dev(slave_dev, bond_dev)) {
2078                 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2079                 return -EINVAL;
2080         }
2081
2082         block_netpoll_tx();
2083
2084         slave = bond_get_slave_by_dev(bond, slave_dev);
2085         if (!slave) {
2086                 /* not a slave of this bond */
2087                 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2088                 unblock_netpoll_tx();
2089                 return -EINVAL;
2090         }
2091
2092         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2093
2094         bond_sysfs_slave_del(slave);
2095
2096         /* recompute stats just before removing the slave */
2097         bond_get_stats(bond->dev, &bond->bond_stats);
2098
2099         bond_upper_dev_unlink(bond, slave);
2100         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2101          * for this slave anymore.
2102          */
2103         netdev_rx_handler_unregister(slave_dev);
2104
2105         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2106                 bond_3ad_unbind_slave(slave);
2107
2108         if (bond_mode_can_use_xmit_hash(bond))
2109                 bond_update_slave_arr(bond, slave);
2110
2111         slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2112                     bond_is_active_slave(slave) ? "active" : "backup");
2113
2114         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2115
2116         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2117
2118         if (!all && (!bond->params.fail_over_mac ||
2119                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2120                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2121                     bond_has_slaves(bond))
2122                         slave_warn(bond_dev, slave_dev, "the permanent HWaddr of slave - %pM - is still in use by bond - set the HWaddr of slave to a different address to avoid conflicts\n",
2123                                    slave->perm_hwaddr);
2124         }
2125
2126         if (rtnl_dereference(bond->primary_slave) == slave)
2127                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2128
2129         if (oldcurrent == slave)
2130                 bond_change_active_slave(bond, NULL);
2131
2132         if (bond_is_lb(bond)) {
2133                 /* Must be called only after the slave has been
2134                  * detached from the list and the curr_active_slave
2135                  * has been cleared (if our_slave == old_current),
2136                  * but before a new active slave is selected.
2137                  */
2138                 bond_alb_deinit_slave(bond, slave);
2139         }
2140
2141         if (all) {
2142                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2143         } else if (oldcurrent == slave) {
2144                 /* Note that we hold RTNL over this sequence, so there
2145                  * is no concern that another slave add/remove event
2146                  * will interfere.
2147                  */
2148                 bond_select_active_slave(bond);
2149         }
2150
2151         if (!bond_has_slaves(bond)) {
2152                 bond_set_carrier(bond);
2153                 eth_hw_addr_random(bond_dev);
2154         }
2155
2156         unblock_netpoll_tx();
2157         synchronize_rcu();
2158         bond->slave_cnt--;
2159
2160         if (!bond_has_slaves(bond)) {
2161                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2162                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2163         }
2164
2165         bond_compute_features(bond);
2166         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2167             (old_features & NETIF_F_VLAN_CHALLENGED))
2168                 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2169
2170         vlan_vids_del_by_dev(slave_dev, bond_dev);
2171
2172         /* If the mode uses primary, then this case was handled above by
2173          * bond_change_active_slave(..., NULL)
2174          */
2175         if (!bond_uses_primary(bond)) {
2176                 /* unset promiscuity level from slave
2177                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
2178                  * of the IFF_PROMISC flag in the bond_dev, but we need the
2179                  * value of that flag before that change, as that was the value
2180                  * when this slave was attached, so we cache at the start of the
2181                  * function and use it here. Same goes for ALLMULTI below
2182                  */
2183                 if (old_flags & IFF_PROMISC)
2184                         dev_set_promiscuity(slave_dev, -1);
2185
2186                 /* unset allmulti level from slave */
2187                 if (old_flags & IFF_ALLMULTI)
2188                         dev_set_allmulti(slave_dev, -1);
2189
2190                 bond_hw_addr_flush(bond_dev, slave_dev);
2191         }
2192
2193         slave_disable_netpoll(slave);
2194
2195         /* close slave before restoring its mac address */
2196         dev_close(slave_dev);
2197
2198         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2199             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2200                 /* restore original ("permanent") mac address */
2201                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2202                                   slave->dev->addr_len);
2203                 ss.ss_family = slave_dev->type;
2204                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2205         }
2206
2207         if (unregister)
2208                 __dev_set_mtu(slave_dev, slave->original_mtu);
2209         else
2210                 dev_set_mtu(slave_dev, slave->original_mtu);
2211
2212         if (!netif_is_bond_master(slave_dev))
2213                 slave_dev->priv_flags &= ~IFF_BONDING;
2214
2215         kobject_put(&slave->kobj);
2216
2217         return 0;
2218 }
2219
2220 /* A wrapper used because of ndo_del_link */
2221 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2222 {
2223         return __bond_release_one(bond_dev, slave_dev, false, false);
2224 }
2225
2226 /* First release a slave and then destroy the bond if no more slaves are left.
2227  * Must be under rtnl_lock when this function is called.
2228  */
2229 static int bond_release_and_destroy(struct net_device *bond_dev,
2230                                     struct net_device *slave_dev)
2231 {
2232         struct bonding *bond = netdev_priv(bond_dev);
2233         int ret;
2234
2235         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2236         if (ret == 0 && !bond_has_slaves(bond) &&
2237             bond_dev->reg_state != NETREG_UNREGISTERING) {
2238                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2239                 netdev_info(bond_dev, "Destroying bond\n");
2240                 bond_remove_proc_entry(bond);
2241                 unregister_netdevice(bond_dev);
2242         }
2243         return ret;
2244 }
2245
2246 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2247 {
2248         struct bonding *bond = netdev_priv(bond_dev);
2249         bond_fill_ifbond(bond, info);
2250 }
2251
2252 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2253 {
2254         struct bonding *bond = netdev_priv(bond_dev);
2255         struct list_head *iter;
2256         int i = 0, res = -ENODEV;
2257         struct slave *slave;
2258
2259         bond_for_each_slave(bond, slave, iter) {
2260                 if (i++ == (int)info->slave_id) {
2261                         res = 0;
2262                         bond_fill_ifslave(slave, info);
2263                         break;
2264                 }
2265         }
2266
2267         return res;
2268 }
2269
2270 /*-------------------------------- Monitoring -------------------------------*/
2271
2272 /* called with rcu_read_lock() */
2273 static int bond_miimon_inspect(struct bonding *bond)
2274 {
2275         int link_state, commit = 0;
2276         struct list_head *iter;
2277         struct slave *slave;
2278         bool ignore_updelay;
2279
2280         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2281
2282         bond_for_each_slave_rcu(bond, slave, iter) {
2283                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2284
2285                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2286
2287                 switch (slave->link) {
2288                 case BOND_LINK_UP:
2289                         if (link_state)
2290                                 continue;
2291
2292                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2293                         commit++;
2294                         slave->delay = bond->params.downdelay;
2295                         if (slave->delay) {
2296                                 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2297                                            (BOND_MODE(bond) ==
2298                                             BOND_MODE_ACTIVEBACKUP) ?
2299                                             (bond_is_active_slave(slave) ?
2300                                              "active " : "backup ") : "",
2301                                            bond->params.downdelay * bond->params.miimon);
2302                         }
2303                         fallthrough;
2304                 case BOND_LINK_FAIL:
2305                         if (link_state) {
2306                                 /* recovered before downdelay expired */
2307                                 bond_propose_link_state(slave, BOND_LINK_UP);
2308                                 slave->last_link_up = jiffies;
2309                                 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2310                                            (bond->params.downdelay - slave->delay) *
2311                                            bond->params.miimon);
2312                                 commit++;
2313                                 continue;
2314                         }
2315
2316                         if (slave->delay <= 0) {
2317                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2318                                 commit++;
2319                                 continue;
2320                         }
2321
2322                         slave->delay--;
2323                         break;
2324
2325                 case BOND_LINK_DOWN:
2326                         if (!link_state)
2327                                 continue;
2328
2329                         bond_propose_link_state(slave, BOND_LINK_BACK);
2330                         commit++;
2331                         slave->delay = bond->params.updelay;
2332
2333                         if (slave->delay) {
2334                                 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2335                                            ignore_updelay ? 0 :
2336                                            bond->params.updelay *
2337                                            bond->params.miimon);
2338                         }
2339                         fallthrough;
2340                 case BOND_LINK_BACK:
2341                         if (!link_state) {
2342                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2343                                 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2344                                            (bond->params.updelay - slave->delay) *
2345                                            bond->params.miimon);
2346                                 commit++;
2347                                 continue;
2348                         }
2349
2350                         if (ignore_updelay)
2351                                 slave->delay = 0;
2352
2353                         if (slave->delay <= 0) {
2354                                 bond_propose_link_state(slave, BOND_LINK_UP);
2355                                 commit++;
2356                                 ignore_updelay = false;
2357                                 continue;
2358                         }
2359
2360                         slave->delay--;
2361                         break;
2362                 }
2363         }
2364
2365         return commit;
2366 }
2367
2368 static void bond_miimon_link_change(struct bonding *bond,
2369                                     struct slave *slave,
2370                                     char link)
2371 {
2372         switch (BOND_MODE(bond)) {
2373         case BOND_MODE_8023AD:
2374                 bond_3ad_handle_link_change(slave, link);
2375                 break;
2376         case BOND_MODE_TLB:
2377         case BOND_MODE_ALB:
2378                 bond_alb_handle_link_change(bond, slave, link);
2379                 break;
2380         case BOND_MODE_XOR:
2381                 bond_update_slave_arr(bond, NULL);
2382                 break;
2383         }
2384 }
2385
2386 static void bond_miimon_commit(struct bonding *bond)
2387 {
2388         struct list_head *iter;
2389         struct slave *slave, *primary;
2390
2391         bond_for_each_slave(bond, slave, iter) {
2392                 switch (slave->link_new_state) {
2393                 case BOND_LINK_NOCHANGE:
2394                         /* For 802.3ad mode, check current slave speed and
2395                          * duplex again in case its port was disabled after
2396                          * invalid speed/duplex reporting but recovered before
2397                          * link monitoring could make a decision on the actual
2398                          * link status
2399                          */
2400                         if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2401                             slave->link == BOND_LINK_UP)
2402                                 bond_3ad_adapter_speed_duplex_changed(slave);
2403                         continue;
2404
2405                 case BOND_LINK_UP:
2406                         if (bond_update_speed_duplex(slave) &&
2407                             bond_needs_speed_duplex(bond)) {
2408                                 slave->link = BOND_LINK_DOWN;
2409                                 if (net_ratelimit())
2410                                         slave_warn(bond->dev, slave->dev,
2411                                                    "failed to get link speed/duplex\n");
2412                                 continue;
2413                         }
2414                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2415                                                   BOND_SLAVE_NOTIFY_NOW);
2416                         slave->last_link_up = jiffies;
2417
2418                         primary = rtnl_dereference(bond->primary_slave);
2419                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2420                                 /* prevent it from being the active one */
2421                                 bond_set_backup_slave(slave);
2422                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2423                                 /* make it immediately active */
2424                                 bond_set_active_slave(slave);
2425                         }
2426
2427                         slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2428                                    slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2429                                    slave->duplex ? "full" : "half");
2430
2431                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2432
2433                         if (!bond->curr_active_slave || slave == primary)
2434                                 goto do_failover;
2435
2436                         continue;
2437
2438                 case BOND_LINK_DOWN:
2439                         if (slave->link_failure_count < UINT_MAX)
2440                                 slave->link_failure_count++;
2441
2442                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2443                                                   BOND_SLAVE_NOTIFY_NOW);
2444
2445                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2446                             BOND_MODE(bond) == BOND_MODE_8023AD)
2447                                 bond_set_slave_inactive_flags(slave,
2448                                                               BOND_SLAVE_NOTIFY_NOW);
2449
2450                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2451
2452                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2453
2454                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2455                                 goto do_failover;
2456
2457                         continue;
2458
2459                 default:
2460                         slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2461                                   slave->link_new_state);
2462                         bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2463
2464                         continue;
2465                 }
2466
2467 do_failover:
2468                 block_netpoll_tx();
2469                 bond_select_active_slave(bond);
2470                 unblock_netpoll_tx();
2471         }
2472
2473         bond_set_carrier(bond);
2474 }
2475
2476 /* bond_mii_monitor
2477  *
2478  * Really a wrapper that splits the mii monitor into two phases: an
2479  * inspection, then (if inspection indicates something needs to be done)
2480  * an acquisition of appropriate locks followed by a commit phase to
2481  * implement whatever link state changes are indicated.
2482  */
2483 static void bond_mii_monitor(struct work_struct *work)
2484 {
2485         struct bonding *bond = container_of(work, struct bonding,
2486                                             mii_work.work);
2487         bool should_notify_peers = false;
2488         bool commit;
2489         unsigned long delay;
2490         struct slave *slave;
2491         struct list_head *iter;
2492
2493         delay = msecs_to_jiffies(bond->params.miimon);
2494
2495         if (!bond_has_slaves(bond))
2496                 goto re_arm;
2497
2498         rcu_read_lock();
2499         should_notify_peers = bond_should_notify_peers(bond);
2500         commit = !!bond_miimon_inspect(bond);
2501         if (bond->send_peer_notif) {
2502                 rcu_read_unlock();
2503                 if (rtnl_trylock()) {
2504                         bond->send_peer_notif--;
2505                         rtnl_unlock();
2506                 }
2507         } else {
2508                 rcu_read_unlock();
2509         }
2510
2511         if (commit) {
2512                 /* Race avoidance with bond_close cancel of workqueue */
2513                 if (!rtnl_trylock()) {
2514                         delay = 1;
2515                         should_notify_peers = false;
2516                         goto re_arm;
2517                 }
2518
2519                 bond_for_each_slave(bond, slave, iter) {
2520                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2521                 }
2522                 bond_miimon_commit(bond);
2523
2524                 rtnl_unlock();  /* might sleep, hold no other locks */
2525         }
2526
2527 re_arm:
2528         if (bond->params.miimon)
2529                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2530
2531         if (should_notify_peers) {
2532                 if (!rtnl_trylock())
2533                         return;
2534                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2535                 rtnl_unlock();
2536         }
2537 }
2538
2539 static int bond_upper_dev_walk(struct net_device *upper,
2540                                struct netdev_nested_priv *priv)
2541 {
2542         __be32 ip = *(__be32 *)priv->data;
2543
2544         return ip == bond_confirm_addr(upper, 0, ip);
2545 }
2546
2547 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2548 {
2549         struct netdev_nested_priv priv = {
2550                 .data = (void *)&ip,
2551         };
2552         bool ret = false;
2553
2554         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2555                 return true;
2556
2557         rcu_read_lock();
2558         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2559                 ret = true;
2560         rcu_read_unlock();
2561
2562         return ret;
2563 }
2564
2565 /* We go to the (large) trouble of VLAN tagging ARP frames because
2566  * switches in VLAN mode (especially if ports are configured as
2567  * "native" to a VLAN) might not pass non-tagged frames.
2568  */
2569 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2570                           __be32 src_ip, struct bond_vlan_tag *tags)
2571 {
2572         struct sk_buff *skb;
2573         struct bond_vlan_tag *outer_tag = tags;
2574         struct net_device *slave_dev = slave->dev;
2575         struct net_device *bond_dev = slave->bond->dev;
2576
2577         slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2578                   arp_op, &dest_ip, &src_ip);
2579
2580         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2581                          NULL, slave_dev->dev_addr, NULL);
2582
2583         if (!skb) {
2584                 net_err_ratelimited("ARP packet allocation failed\n");
2585                 return;
2586         }
2587
2588         if (!tags || tags->vlan_proto == VLAN_N_VID)
2589                 goto xmit;
2590
2591         tags++;
2592
2593         /* Go through all the tags backwards and add them to the packet */
2594         while (tags->vlan_proto != VLAN_N_VID) {
2595                 if (!tags->vlan_id) {
2596                         tags++;
2597                         continue;
2598                 }
2599
2600                 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2601                           ntohs(outer_tag->vlan_proto), tags->vlan_id);
2602                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2603                                                 tags->vlan_id);
2604                 if (!skb) {
2605                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2606                         return;
2607                 }
2608
2609                 tags++;
2610         }
2611         /* Set the outer tag */
2612         if (outer_tag->vlan_id) {
2613                 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2614                           ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2615                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2616                                        outer_tag->vlan_id);
2617         }
2618
2619 xmit:
2620         arp_xmit(skb);
2621 }
2622
2623 /* Validate the device path between the @start_dev and the @end_dev.
2624  * The path is valid if the @end_dev is reachable through device
2625  * stacking.
2626  * When the path is validated, collect any vlan information in the
2627  * path.
2628  */
2629 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2630                                               struct net_device *end_dev,
2631                                               int level)
2632 {
2633         struct bond_vlan_tag *tags;
2634         struct net_device *upper;
2635         struct list_head  *iter;
2636
2637         if (start_dev == end_dev) {
2638                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2639                 if (!tags)
2640                         return ERR_PTR(-ENOMEM);
2641                 tags[level].vlan_proto = VLAN_N_VID;
2642                 return tags;
2643         }
2644
2645         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2646                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2647                 if (IS_ERR_OR_NULL(tags)) {
2648                         if (IS_ERR(tags))
2649                                 return tags;
2650                         continue;
2651                 }
2652                 if (is_vlan_dev(upper)) {
2653                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2654                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2655                 }
2656
2657                 return tags;
2658         }
2659
2660         return NULL;
2661 }
2662
2663 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2664 {
2665         struct rtable *rt;
2666         struct bond_vlan_tag *tags;
2667         __be32 *targets = bond->params.arp_targets, addr;
2668         int i;
2669
2670         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2671                 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2672                           __func__, &targets[i]);
2673                 tags = NULL;
2674
2675                 /* Find out through which dev should the packet go */
2676                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2677                                      RTO_ONLINK, 0);
2678                 if (IS_ERR(rt)) {
2679                         /* there's no route to target - try to send arp
2680                          * probe to generate any traffic (arp_validate=0)
2681                          */
2682                         if (bond->params.arp_validate)
2683                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2684                                                      bond->dev->name,
2685                                                      &targets[i]);
2686                         bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2687                                       0, tags);
2688                         continue;
2689                 }
2690
2691                 /* bond device itself */
2692                 if (rt->dst.dev == bond->dev)
2693                         goto found;
2694
2695                 rcu_read_lock();
2696                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2697                 rcu_read_unlock();
2698
2699                 if (!IS_ERR_OR_NULL(tags))
2700                         goto found;
2701
2702                 /* Not our device - skip */
2703                 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2704                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2705
2706                 ip_rt_put(rt);
2707                 continue;
2708
2709 found:
2710                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2711                 ip_rt_put(rt);
2712                 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2713                 kfree(tags);
2714         }
2715 }
2716
2717 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2718 {
2719         int i;
2720
2721         if (!sip || !bond_has_this_ip(bond, tip)) {
2722                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2723                            __func__, &sip, &tip);
2724                 return;
2725         }
2726
2727         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2728         if (i == -1) {
2729                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2730                            __func__, &sip);
2731                 return;
2732         }
2733         slave->last_rx = jiffies;
2734         slave->target_last_arp_rx[i] = jiffies;
2735 }
2736
2737 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2738                  struct slave *slave)
2739 {
2740         struct arphdr *arp = (struct arphdr *)skb->data;
2741         struct slave *curr_active_slave, *curr_arp_slave;
2742         unsigned char *arp_ptr;
2743         __be32 sip, tip;
2744         int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2745         unsigned int alen;
2746
2747         if (!slave_do_arp_validate(bond, slave)) {
2748                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2749                     !slave_do_arp_validate_only(bond))
2750                         slave->last_rx = jiffies;
2751                 return RX_HANDLER_ANOTHER;
2752         } else if (!is_arp) {
2753                 return RX_HANDLER_ANOTHER;
2754         }
2755
2756         alen = arp_hdr_len(bond->dev);
2757
2758         slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2759                    __func__, skb->dev->name);
2760
2761         if (alen > skb_headlen(skb)) {
2762                 arp = kmalloc(alen, GFP_ATOMIC);
2763                 if (!arp)
2764                         goto out_unlock;
2765                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2766                         goto out_unlock;
2767         }
2768
2769         if (arp->ar_hln != bond->dev->addr_len ||
2770             skb->pkt_type == PACKET_OTHERHOST ||
2771             skb->pkt_type == PACKET_LOOPBACK ||
2772             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2773             arp->ar_pro != htons(ETH_P_IP) ||
2774             arp->ar_pln != 4)
2775                 goto out_unlock;
2776
2777         arp_ptr = (unsigned char *)(arp + 1);
2778         arp_ptr += bond->dev->addr_len;
2779         memcpy(&sip, arp_ptr, 4);
2780         arp_ptr += 4 + bond->dev->addr_len;
2781         memcpy(&tip, arp_ptr, 4);
2782
2783         slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2784                   __func__, slave->dev->name, bond_slave_state(slave),
2785                   bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2786                   &sip, &tip);
2787
2788         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2789         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2790
2791         /* We 'trust' the received ARP enough to validate it if:
2792          *
2793          * (a) the slave receiving the ARP is active (which includes the
2794          * current ARP slave, if any), or
2795          *
2796          * (b) the receiving slave isn't active, but there is a currently
2797          * active slave and it received valid arp reply(s) after it became
2798          * the currently active slave, or
2799          *
2800          * (c) there is an ARP slave that sent an ARP during the prior ARP
2801          * interval, and we receive an ARP reply on any slave.  We accept
2802          * these because switch FDB update delays may deliver the ARP
2803          * reply to a slave other than the sender of the ARP request.
2804          *
2805          * Note: for (b), backup slaves are receiving the broadcast ARP
2806          * request, not a reply.  This request passes from the sending
2807          * slave through the L2 switch(es) to the receiving slave.  Since
2808          * this is checking the request, sip/tip are swapped for
2809          * validation.
2810          *
2811          * This is done to avoid endless looping when we can't reach the
2812          * arp_ip_target and fool ourselves with our own arp requests.
2813          */
2814         if (bond_is_active_slave(slave))
2815                 bond_validate_arp(bond, slave, sip, tip);
2816         else if (curr_active_slave &&
2817                  time_after(slave_last_rx(bond, curr_active_slave),
2818                             curr_active_slave->last_link_up))
2819                 bond_validate_arp(bond, slave, tip, sip);
2820         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2821                  bond_time_in_interval(bond,
2822                                        dev_trans_start(curr_arp_slave->dev), 1))
2823                 bond_validate_arp(bond, slave, sip, tip);
2824
2825 out_unlock:
2826         if (arp != (struct arphdr *)skb->data)
2827                 kfree(arp);
2828         return RX_HANDLER_ANOTHER;
2829 }
2830
2831 /* function to verify if we're in the arp_interval timeslice, returns true if
2832  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2833  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2834  */
2835 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2836                                   int mod)
2837 {
2838         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2839
2840         return time_in_range(jiffies,
2841                              last_act - delta_in_ticks,
2842                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2843 }
2844
2845 /* This function is called regularly to monitor each slave's link
2846  * ensuring that traffic is being sent and received when arp monitoring
2847  * is used in load-balancing mode. if the adapter has been dormant, then an
2848  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2849  * arp monitoring in active backup mode.
2850  */
2851 static void bond_loadbalance_arp_mon(struct bonding *bond)
2852 {
2853         struct slave *slave, *oldcurrent;
2854         struct list_head *iter;
2855         int do_failover = 0, slave_state_changed = 0;
2856
2857         if (!bond_has_slaves(bond))
2858                 goto re_arm;
2859
2860         rcu_read_lock();
2861
2862         oldcurrent = rcu_dereference(bond->curr_active_slave);
2863         /* see if any of the previous devices are up now (i.e. they have
2864          * xmt and rcv traffic). the curr_active_slave does not come into
2865          * the picture unless it is null. also, slave->last_link_up is not
2866          * needed here because we send an arp on each slave and give a slave
2867          * as long as it needs to get the tx/rx within the delta.
2868          * TODO: what about up/down delay in arp mode? it wasn't here before
2869          *       so it can wait
2870          */
2871         bond_for_each_slave_rcu(bond, slave, iter) {
2872                 unsigned long trans_start = dev_trans_start(slave->dev);
2873
2874                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2875
2876                 if (slave->link != BOND_LINK_UP) {
2877                         if (bond_time_in_interval(bond, trans_start, 1) &&
2878                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2879
2880                                 bond_propose_link_state(slave, BOND_LINK_UP);
2881                                 slave_state_changed = 1;
2882
2883                                 /* primary_slave has no meaning in round-robin
2884                                  * mode. the window of a slave being up and
2885                                  * curr_active_slave being null after enslaving
2886                                  * is closed.
2887                                  */
2888                                 if (!oldcurrent) {
2889                                         slave_info(bond->dev, slave->dev, "link status definitely up\n");
2890                                         do_failover = 1;
2891                                 } else {
2892                                         slave_info(bond->dev, slave->dev, "interface is now up\n");
2893                                 }
2894                         }
2895                 } else {
2896                         /* slave->link == BOND_LINK_UP */
2897
2898                         /* not all switches will respond to an arp request
2899                          * when the source ip is 0, so don't take the link down
2900                          * if we don't know our ip yet
2901                          */
2902                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2903                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2904
2905                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2906                                 slave_state_changed = 1;
2907
2908                                 if (slave->link_failure_count < UINT_MAX)
2909                                         slave->link_failure_count++;
2910
2911                                 slave_info(bond->dev, slave->dev, "interface is now down\n");
2912
2913                                 if (slave == oldcurrent)
2914                                         do_failover = 1;
2915                         }
2916                 }
2917
2918                 /* note: if switch is in round-robin mode, all links
2919                  * must tx arp to ensure all links rx an arp - otherwise
2920                  * links may oscillate or not come up at all; if switch is
2921                  * in something like xor mode, there is nothing we can
2922                  * do - all replies will be rx'ed on same link causing slaves
2923                  * to be unstable during low/no traffic periods
2924                  */
2925                 if (bond_slave_is_up(slave))
2926                         bond_arp_send_all(bond, slave);
2927         }
2928
2929         rcu_read_unlock();
2930
2931         if (do_failover || slave_state_changed) {
2932                 if (!rtnl_trylock())
2933                         goto re_arm;
2934
2935                 bond_for_each_slave(bond, slave, iter) {
2936                         if (slave->link_new_state != BOND_LINK_NOCHANGE)
2937                                 slave->link = slave->link_new_state;
2938                 }
2939
2940                 if (slave_state_changed) {
2941                         bond_slave_state_change(bond);
2942                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2943                                 bond_update_slave_arr(bond, NULL);
2944                 }
2945                 if (do_failover) {
2946                         block_netpoll_tx();
2947                         bond_select_active_slave(bond);
2948                         unblock_netpoll_tx();
2949                 }
2950                 rtnl_unlock();
2951         }
2952
2953 re_arm:
2954         if (bond->params.arp_interval)
2955                 queue_delayed_work(bond->wq, &bond->arp_work,
2956                                    msecs_to_jiffies(bond->params.arp_interval));
2957 }
2958
2959 /* Called to inspect slaves for active-backup mode ARP monitor link state
2960  * changes.  Sets proposed link state in slaves to specify what action
2961  * should take place for the slave.  Returns 0 if no changes are found, >0
2962  * if changes to link states must be committed.
2963  *
2964  * Called with rcu_read_lock held.
2965  */
2966 static int bond_ab_arp_inspect(struct bonding *bond)
2967 {
2968         unsigned long trans_start, last_rx;
2969         struct list_head *iter;
2970         struct slave *slave;
2971         int commit = 0;
2972
2973         bond_for_each_slave_rcu(bond, slave, iter) {
2974                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2975                 last_rx = slave_last_rx(bond, slave);
2976
2977                 if (slave->link != BOND_LINK_UP) {
2978                         if (bond_time_in_interval(bond, last_rx, 1)) {
2979                                 bond_propose_link_state(slave, BOND_LINK_UP);
2980                                 commit++;
2981                         } else if (slave->link == BOND_LINK_BACK) {
2982                                 bond_propose_link_state(slave, BOND_LINK_FAIL);
2983                                 commit++;
2984                         }
2985                         continue;
2986                 }
2987
2988                 /* Give slaves 2*delta after being enslaved or made
2989                  * active.  This avoids bouncing, as the last receive
2990                  * times need a full ARP monitor cycle to be updated.
2991                  */
2992                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2993                         continue;
2994
2995                 /* Backup slave is down if:
2996                  * - No current_arp_slave AND
2997                  * - more than 3*delta since last receive AND
2998                  * - the bond has an IP address
2999                  *
3000                  * Note: a non-null current_arp_slave indicates
3001                  * the curr_active_slave went down and we are
3002                  * searching for a new one; under this condition
3003                  * we only take the curr_active_slave down - this
3004                  * gives each slave a chance to tx/rx traffic
3005                  * before being taken out
3006                  */
3007                 if (!bond_is_active_slave(slave) &&
3008                     !rcu_access_pointer(bond->current_arp_slave) &&
3009                     !bond_time_in_interval(bond, last_rx, 3)) {
3010                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3011                         commit++;
3012                 }
3013
3014                 /* Active slave is down if:
3015                  * - more than 2*delta since transmitting OR
3016                  * - (more than 2*delta since receive AND
3017                  *    the bond has an IP address)
3018                  */
3019                 trans_start = dev_trans_start(slave->dev);
3020                 if (bond_is_active_slave(slave) &&
3021                     (!bond_time_in_interval(bond, trans_start, 2) ||
3022                      !bond_time_in_interval(bond, last_rx, 2))) {
3023                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3024                         commit++;
3025                 }
3026         }
3027
3028         return commit;
3029 }
3030
3031 /* Called to commit link state changes noted by inspection step of
3032  * active-backup mode ARP monitor.
3033  *
3034  * Called with RTNL hold.
3035  */
3036 static void bond_ab_arp_commit(struct bonding *bond)
3037 {
3038         unsigned long trans_start;
3039         struct list_head *iter;
3040         struct slave *slave;
3041
3042         bond_for_each_slave(bond, slave, iter) {
3043                 switch (slave->link_new_state) {
3044                 case BOND_LINK_NOCHANGE:
3045                         continue;
3046
3047                 case BOND_LINK_UP:
3048                         trans_start = dev_trans_start(slave->dev);
3049                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
3050                             (!rtnl_dereference(bond->curr_active_slave) &&
3051                              bond_time_in_interval(bond, trans_start, 1))) {
3052                                 struct slave *current_arp_slave;
3053
3054                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3055                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
3056                                                           BOND_SLAVE_NOTIFY_NOW);
3057                                 if (current_arp_slave) {
3058                                         bond_set_slave_inactive_flags(
3059                                                 current_arp_slave,
3060                                                 BOND_SLAVE_NOTIFY_NOW);
3061                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3062                                 }
3063
3064                                 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3065
3066                                 if (!rtnl_dereference(bond->curr_active_slave) ||
3067                                     slave == rtnl_dereference(bond->primary_slave))
3068                                         goto do_failover;
3069
3070                         }
3071
3072                         continue;
3073
3074                 case BOND_LINK_DOWN:
3075                         if (slave->link_failure_count < UINT_MAX)
3076                                 slave->link_failure_count++;
3077
3078                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3079                                                   BOND_SLAVE_NOTIFY_NOW);
3080                         bond_set_slave_inactive_flags(slave,
3081                                                       BOND_SLAVE_NOTIFY_NOW);
3082
3083                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3084
3085                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
3086                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3087                                 goto do_failover;
3088                         }
3089
3090                         continue;
3091
3092                 case BOND_LINK_FAIL:
3093                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3094                                                   BOND_SLAVE_NOTIFY_NOW);
3095                         bond_set_slave_inactive_flags(slave,
3096                                                       BOND_SLAVE_NOTIFY_NOW);
3097
3098                         /* A slave has just been enslaved and has become
3099                          * the current active slave.
3100                          */
3101                         if (rtnl_dereference(bond->curr_active_slave))
3102                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3103                         continue;
3104
3105                 default:
3106                         slave_err(bond->dev, slave->dev,
3107                                   "impossible: link_new_state %d on slave\n",
3108                                   slave->link_new_state);
3109                         continue;
3110                 }
3111
3112 do_failover:
3113                 block_netpoll_tx();
3114                 bond_select_active_slave(bond);
3115                 unblock_netpoll_tx();
3116         }
3117
3118         bond_set_carrier(bond);
3119 }
3120
3121 /* Send ARP probes for active-backup mode ARP monitor.
3122  *
3123  * Called with rcu_read_lock held.
3124  */
3125 static bool bond_ab_arp_probe(struct bonding *bond)
3126 {
3127         struct slave *slave, *before = NULL, *new_slave = NULL,
3128                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3129                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3130         struct list_head *iter;
3131         bool found = false;
3132         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3133
3134         if (curr_arp_slave && curr_active_slave)
3135                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3136                             curr_arp_slave->dev->name,
3137                             curr_active_slave->dev->name);
3138
3139         if (curr_active_slave) {
3140                 bond_arp_send_all(bond, curr_active_slave);
3141                 return should_notify_rtnl;
3142         }
3143
3144         /* if we don't have a curr_active_slave, search for the next available
3145          * backup slave from the current_arp_slave and make it the candidate
3146          * for becoming the curr_active_slave
3147          */
3148
3149         if (!curr_arp_slave) {
3150                 curr_arp_slave = bond_first_slave_rcu(bond);
3151                 if (!curr_arp_slave)
3152                         return should_notify_rtnl;
3153         }
3154
3155         bond_for_each_slave_rcu(bond, slave, iter) {
3156                 if (!found && !before && bond_slave_is_up(slave))
3157                         before = slave;
3158
3159                 if (found && !new_slave && bond_slave_is_up(slave))
3160                         new_slave = slave;
3161                 /* if the link state is up at this point, we
3162                  * mark it down - this can happen if we have
3163                  * simultaneous link failures and
3164                  * reselect_active_interface doesn't make this
3165                  * one the current slave so it is still marked
3166                  * up when it is actually down
3167                  */
3168                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3169                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3170                                                   BOND_SLAVE_NOTIFY_LATER);
3171                         if (slave->link_failure_count < UINT_MAX)
3172                                 slave->link_failure_count++;
3173
3174                         bond_set_slave_inactive_flags(slave,
3175                                                       BOND_SLAVE_NOTIFY_LATER);
3176
3177                         slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3178                 }
3179                 if (slave == curr_arp_slave)
3180                         found = true;
3181         }
3182
3183         if (!new_slave && before)
3184                 new_slave = before;
3185
3186         if (!new_slave)
3187                 goto check_state;
3188
3189         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3190                                   BOND_SLAVE_NOTIFY_LATER);
3191         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3192         bond_arp_send_all(bond, new_slave);
3193         new_slave->last_link_up = jiffies;
3194         rcu_assign_pointer(bond->current_arp_slave, new_slave);
3195
3196 check_state:
3197         bond_for_each_slave_rcu(bond, slave, iter) {
3198                 if (slave->should_notify || slave->should_notify_link) {
3199                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3200                         break;
3201                 }
3202         }
3203         return should_notify_rtnl;
3204 }
3205
3206 static void bond_activebackup_arp_mon(struct bonding *bond)
3207 {
3208         bool should_notify_peers = false;
3209         bool should_notify_rtnl = false;
3210         int delta_in_ticks;
3211
3212         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3213
3214         if (!bond_has_slaves(bond))
3215                 goto re_arm;
3216
3217         rcu_read_lock();
3218
3219         should_notify_peers = bond_should_notify_peers(bond);
3220
3221         if (bond_ab_arp_inspect(bond)) {
3222                 rcu_read_unlock();
3223
3224                 /* Race avoidance with bond_close flush of workqueue */
3225                 if (!rtnl_trylock()) {
3226                         delta_in_ticks = 1;
3227                         should_notify_peers = false;
3228                         goto re_arm;
3229                 }
3230
3231                 bond_ab_arp_commit(bond);
3232
3233                 rtnl_unlock();
3234                 rcu_read_lock();
3235         }
3236
3237         should_notify_rtnl = bond_ab_arp_probe(bond);
3238         rcu_read_unlock();
3239
3240 re_arm:
3241         if (bond->params.arp_interval)
3242                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3243
3244         if (should_notify_peers || should_notify_rtnl) {
3245                 if (!rtnl_trylock())
3246                         return;
3247
3248                 if (should_notify_peers)
3249                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3250                                                  bond->dev);
3251                 if (should_notify_rtnl) {
3252                         bond_slave_state_notify(bond);
3253                         bond_slave_link_notify(bond);
3254                 }
3255
3256                 rtnl_unlock();
3257         }
3258 }
3259
3260 static void bond_arp_monitor(struct work_struct *work)
3261 {
3262         struct bonding *bond = container_of(work, struct bonding,
3263                                             arp_work.work);
3264
3265         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3266                 bond_activebackup_arp_mon(bond);
3267         else
3268                 bond_loadbalance_arp_mon(bond);
3269 }
3270
3271 /*-------------------------- netdev event handling --------------------------*/
3272
3273 /* Change device name */
3274 static int bond_event_changename(struct bonding *bond)
3275 {
3276         bond_remove_proc_entry(bond);
3277         bond_create_proc_entry(bond);
3278
3279         bond_debug_reregister(bond);
3280
3281         return NOTIFY_DONE;
3282 }
3283
3284 static int bond_master_netdev_event(unsigned long event,
3285                                     struct net_device *bond_dev)
3286 {
3287         struct bonding *event_bond = netdev_priv(bond_dev);
3288
3289         netdev_dbg(bond_dev, "%s called\n", __func__);
3290
3291         switch (event) {
3292         case NETDEV_CHANGENAME:
3293                 return bond_event_changename(event_bond);
3294         case NETDEV_UNREGISTER:
3295                 bond_remove_proc_entry(event_bond);
3296                 break;
3297         case NETDEV_REGISTER:
3298                 bond_create_proc_entry(event_bond);
3299                 break;
3300         default:
3301                 break;
3302         }
3303
3304         return NOTIFY_DONE;
3305 }
3306
3307 static int bond_slave_netdev_event(unsigned long event,
3308                                    struct net_device *slave_dev)
3309 {
3310         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3311         struct bonding *bond;
3312         struct net_device *bond_dev;
3313
3314         /* A netdev event can be generated while enslaving a device
3315          * before netdev_rx_handler_register is called in which case
3316          * slave will be NULL
3317          */
3318         if (!slave) {
3319                 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3320                 return NOTIFY_DONE;
3321         }
3322
3323         bond_dev = slave->bond->dev;
3324         bond = slave->bond;
3325         primary = rtnl_dereference(bond->primary_slave);
3326
3327         slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3328
3329         switch (event) {
3330         case NETDEV_UNREGISTER:
3331                 if (bond_dev->type != ARPHRD_ETHER)
3332                         bond_release_and_destroy(bond_dev, slave_dev);
3333                 else
3334                         __bond_release_one(bond_dev, slave_dev, false, true);
3335                 break;
3336         case NETDEV_UP:
3337         case NETDEV_CHANGE:
3338                 /* For 802.3ad mode only:
3339                  * Getting invalid Speed/Duplex values here will put slave
3340                  * in weird state. Mark it as link-fail if the link was
3341                  * previously up or link-down if it hasn't yet come up, and
3342                  * let link-monitoring (miimon) set it right when correct
3343                  * speeds/duplex are available.
3344                  */
3345                 if (bond_update_speed_duplex(slave) &&
3346                     BOND_MODE(bond) == BOND_MODE_8023AD) {
3347                         if (slave->last_link_up)
3348                                 slave->link = BOND_LINK_FAIL;
3349                         else
3350                                 slave->link = BOND_LINK_DOWN;
3351                 }
3352
3353                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3354                         bond_3ad_adapter_speed_duplex_changed(slave);
3355                 fallthrough;
3356         case NETDEV_DOWN:
3357                 /* Refresh slave-array if applicable!
3358                  * If the setup does not use miimon or arpmon (mode-specific!),
3359                  * then these events will not cause the slave-array to be
3360                  * refreshed. This will cause xmit to use a slave that is not
3361                  * usable. Avoid such situation by refeshing the array at these
3362                  * events. If these (miimon/arpmon) parameters are configured
3363                  * then array gets refreshed twice and that should be fine!
3364                  */
3365                 if (bond_mode_can_use_xmit_hash(bond))
3366                         bond_update_slave_arr(bond, NULL);
3367                 break;
3368         case NETDEV_CHANGEMTU:
3369                 /* TODO: Should slaves be allowed to
3370                  * independently alter their MTU?  For
3371                  * an active-backup bond, slaves need
3372                  * not be the same type of device, so
3373                  * MTUs may vary.  For other modes,
3374                  * slaves arguably should have the
3375                  * same MTUs. To do this, we'd need to
3376                  * take over the slave's change_mtu
3377                  * function for the duration of their
3378                  * servitude.
3379                  */
3380                 break;
3381         case NETDEV_CHANGENAME:
3382                 /* we don't care if we don't have primary set */
3383                 if (!bond_uses_primary(bond) ||
3384                     !bond->params.primary[0])
3385                         break;
3386
3387                 if (slave == primary) {
3388                         /* slave's name changed - he's no longer primary */
3389                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3390                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3391                         /* we have a new primary slave */
3392                         rcu_assign_pointer(bond->primary_slave, slave);
3393                 } else { /* we didn't change primary - exit */
3394                         break;
3395                 }
3396
3397                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3398                             primary ? slave_dev->name : "none");
3399
3400                 block_netpoll_tx();
3401                 bond_select_active_slave(bond);
3402                 unblock_netpoll_tx();
3403                 break;
3404         case NETDEV_FEAT_CHANGE:
3405                 bond_compute_features(bond);
3406                 break;
3407         case NETDEV_RESEND_IGMP:
3408                 /* Propagate to master device */
3409                 call_netdevice_notifiers(event, slave->bond->dev);
3410                 break;
3411         default:
3412                 break;
3413         }
3414
3415         return NOTIFY_DONE;
3416 }
3417
3418 /* bond_netdev_event: handle netdev notifier chain events.
3419  *
3420  * This function receives events for the netdev chain.  The caller (an
3421  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3422  * locks for us to safely manipulate the slave devices (RTNL lock,
3423  * dev_probe_lock).
3424  */
3425 static int bond_netdev_event(struct notifier_block *this,
3426                              unsigned long event, void *ptr)
3427 {
3428         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3429
3430         netdev_dbg(event_dev, "%s received %s\n",
3431                    __func__, netdev_cmd_to_name(event));
3432
3433         if (!(event_dev->priv_flags & IFF_BONDING))
3434                 return NOTIFY_DONE;
3435
3436         if (event_dev->flags & IFF_MASTER) {
3437                 int ret;
3438
3439                 ret = bond_master_netdev_event(event, event_dev);
3440                 if (ret != NOTIFY_DONE)
3441                         return ret;
3442         }
3443
3444         if (event_dev->flags & IFF_SLAVE)
3445                 return bond_slave_netdev_event(event, event_dev);
3446
3447         return NOTIFY_DONE;
3448 }
3449
3450 static struct notifier_block bond_netdev_notifier = {
3451         .notifier_call = bond_netdev_event,
3452 };
3453
3454 /*---------------------------- Hashing Policies -----------------------------*/
3455
3456 /* L2 hash helper */
3457 static inline u32 bond_eth_hash(struct sk_buff *skb)
3458 {
3459         struct ethhdr *ep, hdr_tmp;
3460
3461         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3462         if (ep)
3463                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3464         return 0;
3465 }
3466
3467 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk,
3468                          int *noff, int *proto, bool l34)
3469 {
3470         const struct ipv6hdr *iph6;
3471         const struct iphdr *iph;
3472
3473         if (skb->protocol == htons(ETH_P_IP)) {
3474                 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph))))
3475                         return false;
3476                 iph = (const struct iphdr *)(skb->data + *noff);
3477                 iph_to_flow_copy_v4addrs(fk, iph);
3478                 *noff += iph->ihl << 2;
3479                 if (!ip_is_fragment(iph))
3480                         *proto = iph->protocol;
3481         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3482                 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6))))
3483                         return false;
3484                 iph6 = (const struct ipv6hdr *)(skb->data + *noff);
3485                 iph_to_flow_copy_v6addrs(fk, iph6);
3486                 *noff += sizeof(*iph6);
3487                 *proto = iph6->nexthdr;
3488         } else {
3489                 return false;
3490         }
3491
3492         if (l34 && *proto >= 0)
3493                 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto);
3494
3495         return true;
3496 }
3497
3498 /* Extract the appropriate headers based on bond's xmit policy */
3499 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3500                               struct flow_keys *fk)
3501 {
3502         bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
3503         int noff, proto = -1;
3504
3505         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) {
3506                 memset(fk, 0, sizeof(*fk));
3507                 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3508                                           fk, NULL, 0, 0, 0, 0);
3509         }
3510
3511         fk->ports.ports = 0;
3512         memset(&fk->icmp, 0, sizeof(fk->icmp));
3513         noff = skb_network_offset(skb);
3514         if (!bond_flow_ip(skb, fk, &noff, &proto, l34))
3515                 return false;
3516
3517         /* ICMP error packets contains at least 8 bytes of the header
3518          * of the packet which generated the error. Use this information
3519          * to correlate ICMP error packets within the same flow which
3520          * generated the error.
3521          */
3522         if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
3523                 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data,
3524                                       skb_transport_offset(skb),
3525                                       skb_headlen(skb));
3526                 if (proto == IPPROTO_ICMP) {
3527                         if (!icmp_is_err(fk->icmp.type))
3528                                 return true;
3529
3530                         noff += sizeof(struct icmphdr);
3531                 } else if (proto == IPPROTO_ICMPV6) {
3532                         if (!icmpv6_is_err(fk->icmp.type))
3533                                 return true;
3534
3535                         noff += sizeof(struct icmp6hdr);
3536                 }
3537                 return bond_flow_ip(skb, fk, &noff, &proto, l34);
3538         }
3539
3540         return true;
3541 }
3542
3543 /**
3544  * bond_xmit_hash - generate a hash value based on the xmit policy
3545  * @bond: bonding device
3546  * @skb: buffer to use for headers
3547  *
3548  * This function will extract the necessary headers from the skb buffer and use
3549  * them to generate a hash based on the xmit_policy set in the bonding device
3550  */
3551 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3552 {
3553         struct flow_keys flow;
3554         u32 hash;
3555
3556         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3557             skb->l4_hash)
3558                 return skb->hash;
3559
3560         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3561             !bond_flow_dissect(bond, skb, &flow))
3562                 return bond_eth_hash(skb);
3563
3564         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3565             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
3566                 hash = bond_eth_hash(skb);
3567         } else {
3568                 if (flow.icmp.id)
3569                         memcpy(&hash, &flow.icmp, sizeof(hash));
3570                 else
3571                         memcpy(&hash, &flow.ports.ports, sizeof(hash));
3572         }
3573         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3574                 (__force u32)flow_get_u32_src(&flow);
3575         hash ^= (hash >> 16);
3576         hash ^= (hash >> 8);
3577
3578         return hash >> 1;
3579 }
3580
3581 /*-------------------------- Device entry points ----------------------------*/
3582
3583 void bond_work_init_all(struct bonding *bond)
3584 {
3585         INIT_DELAYED_WORK(&bond->mcast_work,
3586                           bond_resend_igmp_join_requests_delayed);
3587         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3588         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3589         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3590         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3591         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3592 }
3593
3594 static void bond_work_cancel_all(struct bonding *bond)
3595 {
3596         cancel_delayed_work_sync(&bond->mii_work);
3597         cancel_delayed_work_sync(&bond->arp_work);
3598         cancel_delayed_work_sync(&bond->alb_work);
3599         cancel_delayed_work_sync(&bond->ad_work);
3600         cancel_delayed_work_sync(&bond->mcast_work);
3601         cancel_delayed_work_sync(&bond->slave_arr_work);
3602 }
3603
3604 static int bond_open(struct net_device *bond_dev)
3605 {
3606         struct bonding *bond = netdev_priv(bond_dev);
3607         struct list_head *iter;
3608         struct slave *slave;
3609
3610         /* reset slave->backup and slave->inactive */
3611         if (bond_has_slaves(bond)) {
3612                 bond_for_each_slave(bond, slave, iter) {
3613                         if (bond_uses_primary(bond) &&
3614                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3615                                 bond_set_slave_inactive_flags(slave,
3616                                                               BOND_SLAVE_NOTIFY_NOW);
3617                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3618                                 bond_set_slave_active_flags(slave,
3619                                                             BOND_SLAVE_NOTIFY_NOW);
3620                         }
3621                 }
3622         }
3623
3624         if (bond_is_lb(bond)) {
3625                 /* bond_alb_initialize must be called before the timer
3626                  * is started.
3627                  */
3628                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3629                         return -ENOMEM;
3630                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3631                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3632         }
3633
3634         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3635                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3636
3637         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3638                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3639                 bond->recv_probe = bond_arp_rcv;
3640         }
3641
3642         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3643                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3644                 /* register to receive LACPDUs */
3645                 bond->recv_probe = bond_3ad_lacpdu_recv;
3646                 bond_3ad_initiate_agg_selection(bond, 1);
3647         }
3648
3649         if (bond_mode_can_use_xmit_hash(bond))
3650                 bond_update_slave_arr(bond, NULL);
3651
3652         return 0;
3653 }
3654
3655 static int bond_close(struct net_device *bond_dev)
3656 {
3657         struct bonding *bond = netdev_priv(bond_dev);
3658
3659         bond_work_cancel_all(bond);
3660         bond->send_peer_notif = 0;
3661         if (bond_is_lb(bond))
3662                 bond_alb_deinitialize(bond);
3663         bond->recv_probe = NULL;
3664
3665         return 0;
3666 }
3667
3668 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3669  * that some drivers can provide 32bit values only.
3670  */
3671 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3672                             const struct rtnl_link_stats64 *_new,
3673                             const struct rtnl_link_stats64 *_old)
3674 {
3675         const u64 *new = (const u64 *)_new;
3676         const u64 *old = (const u64 *)_old;
3677         u64 *res = (u64 *)_res;
3678         int i;
3679
3680         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3681                 u64 nv = new[i];
3682                 u64 ov = old[i];
3683                 s64 delta = nv - ov;
3684
3685                 /* detects if this particular field is 32bit only */
3686                 if (((nv | ov) >> 32) == 0)
3687                         delta = (s64)(s32)((u32)nv - (u32)ov);
3688
3689                 /* filter anomalies, some drivers reset their stats
3690                  * at down/up events.
3691                  */
3692                 if (delta > 0)
3693                         res[i] += delta;
3694         }
3695 }
3696
3697 #ifdef CONFIG_LOCKDEP
3698 static int bond_get_lowest_level_rcu(struct net_device *dev)
3699 {
3700         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
3701         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
3702         int cur = 0, max = 0;
3703
3704         now = dev;
3705         iter = &dev->adj_list.lower;
3706
3707         while (1) {
3708                 next = NULL;
3709                 while (1) {
3710                         ldev = netdev_next_lower_dev_rcu(now, &iter);
3711                         if (!ldev)
3712                                 break;
3713
3714                         next = ldev;
3715                         niter = &ldev->adj_list.lower;
3716                         dev_stack[cur] = now;
3717                         iter_stack[cur++] = iter;
3718                         if (max <= cur)
3719                                 max = cur;
3720                         break;
3721                 }
3722
3723                 if (!next) {
3724                         if (!cur)
3725                                 return max;
3726                         next = dev_stack[--cur];
3727                         niter = iter_stack[cur];
3728                 }
3729
3730                 now = next;
3731                 iter = niter;
3732         }
3733
3734         return max;
3735 }
3736 #endif
3737
3738 static void bond_get_stats(struct net_device *bond_dev,
3739                            struct rtnl_link_stats64 *stats)
3740 {
3741         struct bonding *bond = netdev_priv(bond_dev);
3742         struct rtnl_link_stats64 temp;
3743         struct list_head *iter;
3744         struct slave *slave;
3745         int nest_level = 0;
3746
3747
3748         rcu_read_lock();
3749 #ifdef CONFIG_LOCKDEP
3750         nest_level = bond_get_lowest_level_rcu(bond_dev);
3751 #endif
3752
3753         spin_lock_nested(&bond->stats_lock, nest_level);
3754         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3755
3756         bond_for_each_slave_rcu(bond, slave, iter) {
3757                 const struct rtnl_link_stats64 *new =
3758                         dev_get_stats(slave->dev, &temp);
3759
3760                 bond_fold_stats(stats, new, &slave->slave_stats);
3761
3762                 /* save off the slave stats for the next run */
3763                 memcpy(&slave->slave_stats, new, sizeof(*new));
3764         }
3765
3766         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3767         spin_unlock(&bond->stats_lock);
3768         rcu_read_unlock();
3769 }
3770
3771 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3772 {
3773         struct bonding *bond = netdev_priv(bond_dev);
3774         struct net_device *slave_dev = NULL;
3775         struct ifbond k_binfo;
3776         struct ifbond __user *u_binfo = NULL;
3777         struct ifslave k_sinfo;
3778         struct ifslave __user *u_sinfo = NULL;
3779         struct mii_ioctl_data *mii = NULL;
3780         struct bond_opt_value newval;
3781         struct net *net;
3782         int res = 0;
3783
3784         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3785
3786         switch (cmd) {
3787         case SIOCGMIIPHY:
3788                 mii = if_mii(ifr);
3789                 if (!mii)
3790                         return -EINVAL;
3791
3792                 mii->phy_id = 0;
3793                 fallthrough;
3794         case SIOCGMIIREG:
3795                 /* We do this again just in case we were called by SIOCGMIIREG
3796                  * instead of SIOCGMIIPHY.
3797                  */
3798                 mii = if_mii(ifr);
3799                 if (!mii)
3800                         return -EINVAL;
3801
3802                 if (mii->reg_num == 1) {
3803                         mii->val_out = 0;
3804                         if (netif_carrier_ok(bond->dev))
3805                                 mii->val_out = BMSR_LSTATUS;
3806                 }
3807
3808                 return 0;
3809         case BOND_INFO_QUERY_OLD:
3810         case SIOCBONDINFOQUERY:
3811                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3812
3813                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3814                         return -EFAULT;
3815
3816                 bond_info_query(bond_dev, &k_binfo);
3817                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3818                         return -EFAULT;
3819
3820                 return 0;
3821         case BOND_SLAVE_INFO_QUERY_OLD:
3822         case SIOCBONDSLAVEINFOQUERY:
3823                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3824
3825                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3826                         return -EFAULT;
3827
3828                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3829                 if (res == 0 &&
3830                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3831                         return -EFAULT;
3832
3833                 return res;
3834         default:
3835                 break;
3836         }
3837
3838         net = dev_net(bond_dev);
3839
3840         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3841                 return -EPERM;
3842
3843         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3844
3845         slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
3846
3847         if (!slave_dev)
3848                 return -ENODEV;
3849
3850         switch (cmd) {
3851         case BOND_ENSLAVE_OLD:
3852         case SIOCBONDENSLAVE:
3853                 res = bond_enslave(bond_dev, slave_dev, NULL);
3854                 break;
3855         case BOND_RELEASE_OLD:
3856         case SIOCBONDRELEASE:
3857                 res = bond_release(bond_dev, slave_dev);
3858                 break;
3859         case BOND_SETHWADDR_OLD:
3860         case SIOCBONDSETHWADDR:
3861                 res = bond_set_dev_addr(bond_dev, slave_dev);
3862                 break;
3863         case BOND_CHANGE_ACTIVE_OLD:
3864         case SIOCBONDCHANGEACTIVE:
3865                 bond_opt_initstr(&newval, slave_dev->name);
3866                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3867                                             &newval);
3868                 break;
3869         default:
3870                 res = -EOPNOTSUPP;
3871         }
3872
3873         return res;
3874 }
3875
3876 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3877 {
3878         struct bonding *bond = netdev_priv(bond_dev);
3879
3880         if (change & IFF_PROMISC)
3881                 bond_set_promiscuity(bond,
3882                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3883
3884         if (change & IFF_ALLMULTI)
3885                 bond_set_allmulti(bond,
3886                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3887 }
3888
3889 static void bond_set_rx_mode(struct net_device *bond_dev)
3890 {
3891         struct bonding *bond = netdev_priv(bond_dev);
3892         struct list_head *iter;
3893         struct slave *slave;
3894
3895         rcu_read_lock();
3896         if (bond_uses_primary(bond)) {
3897                 slave = rcu_dereference(bond->curr_active_slave);
3898                 if (slave) {
3899                         dev_uc_sync(slave->dev, bond_dev);
3900                         dev_mc_sync(slave->dev, bond_dev);
3901                 }
3902         } else {
3903                 bond_for_each_slave_rcu(bond, slave, iter) {
3904                         dev_uc_sync_multiple(slave->dev, bond_dev);
3905                         dev_mc_sync_multiple(slave->dev, bond_dev);
3906                 }
3907         }
3908         rcu_read_unlock();
3909 }
3910
3911 static int bond_neigh_init(struct neighbour *n)
3912 {
3913         struct bonding *bond = netdev_priv(n->dev);
3914         const struct net_device_ops *slave_ops;
3915         struct neigh_parms parms;
3916         struct slave *slave;
3917         int ret = 0;
3918
3919         rcu_read_lock();
3920         slave = bond_first_slave_rcu(bond);
3921         if (!slave)
3922                 goto out;
3923         slave_ops = slave->dev->netdev_ops;
3924         if (!slave_ops->ndo_neigh_setup)
3925                 goto out;
3926
3927         /* TODO: find another way [1] to implement this.
3928          * Passing a zeroed structure is fragile,
3929          * but at least we do not pass garbage.
3930          *
3931          * [1] One way would be that ndo_neigh_setup() never touch
3932          *     struct neigh_parms, but propagate the new neigh_setup()
3933          *     back to ___neigh_create() / neigh_parms_alloc()
3934          */
3935         memset(&parms, 0, sizeof(parms));
3936         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3937
3938         if (ret)
3939                 goto out;
3940
3941         if (parms.neigh_setup)
3942                 ret = parms.neigh_setup(n);
3943 out:
3944         rcu_read_unlock();
3945         return ret;
3946 }
3947
3948 /* The bonding ndo_neigh_setup is called at init time beofre any
3949  * slave exists. So we must declare proxy setup function which will
3950  * be used at run time to resolve the actual slave neigh param setup.
3951  *
3952  * It's also called by master devices (such as vlans) to setup their
3953  * underlying devices. In that case - do nothing, we're already set up from
3954  * our init.
3955  */
3956 static int bond_neigh_setup(struct net_device *dev,
3957                             struct neigh_parms *parms)
3958 {
3959         /* modify only our neigh_parms */
3960         if (parms->dev == dev)
3961                 parms->neigh_setup = bond_neigh_init;
3962
3963         return 0;
3964 }
3965
3966 /* Change the MTU of all of a master's slaves to match the master */
3967 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3968 {
3969         struct bonding *bond = netdev_priv(bond_dev);
3970         struct slave *slave, *rollback_slave;
3971         struct list_head *iter;
3972         int res = 0;
3973
3974         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3975
3976         bond_for_each_slave(bond, slave, iter) {
3977                 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
3978                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3979
3980                 res = dev_set_mtu(slave->dev, new_mtu);
3981
3982                 if (res) {
3983                         /* If we failed to set the slave's mtu to the new value
3984                          * we must abort the operation even in ACTIVE_BACKUP
3985                          * mode, because if we allow the backup slaves to have
3986                          * different mtu values than the active slave we'll
3987                          * need to change their mtu when doing a failover. That
3988                          * means changing their mtu from timer context, which
3989                          * is probably not a good idea.
3990                          */
3991                         slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
3992                                   res, new_mtu);
3993                         goto unwind;
3994                 }
3995         }
3996
3997         bond_dev->mtu = new_mtu;
3998
3999         return 0;
4000
4001 unwind:
4002         /* unwind from head to the slave that failed */
4003         bond_for_each_slave(bond, rollback_slave, iter) {
4004                 int tmp_res;
4005
4006                 if (rollback_slave == slave)
4007                         break;
4008
4009                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4010                 if (tmp_res)
4011                         slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4012                                   tmp_res);
4013         }
4014
4015         return res;
4016 }
4017
4018 /* Change HW address
4019  *
4020  * Note that many devices must be down to change the HW address, and
4021  * downing the master releases all slaves.  We can make bonds full of
4022  * bonding devices to test this, however.
4023  */
4024 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4025 {
4026         struct bonding *bond = netdev_priv(bond_dev);
4027         struct slave *slave, *rollback_slave;
4028         struct sockaddr_storage *ss = addr, tmp_ss;
4029         struct list_head *iter;
4030         int res = 0;
4031
4032         if (BOND_MODE(bond) == BOND_MODE_ALB)
4033                 return bond_alb_set_mac_address(bond_dev, addr);
4034
4035
4036         netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4037
4038         /* If fail_over_mac is enabled, do nothing and return success.
4039          * Returning an error causes ifenslave to fail.
4040          */
4041         if (bond->params.fail_over_mac &&
4042             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4043                 return 0;
4044
4045         if (!is_valid_ether_addr(ss->__data))
4046                 return -EADDRNOTAVAIL;
4047
4048         bond_for_each_slave(bond, slave, iter) {
4049                 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4050                           __func__, slave);
4051                 res = dev_set_mac_address(slave->dev, addr, NULL);
4052                 if (res) {
4053                         /* TODO: consider downing the slave
4054                          * and retry ?
4055                          * User should expect communications
4056                          * breakage anyway until ARP finish
4057                          * updating, so...
4058                          */
4059                         slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4060                                   __func__, res);
4061                         goto unwind;
4062                 }
4063         }
4064
4065         /* success */
4066         memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
4067         return 0;
4068
4069 unwind:
4070         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4071         tmp_ss.ss_family = bond_dev->type;
4072
4073         /* unwind from head to the slave that failed */
4074         bond_for_each_slave(bond, rollback_slave, iter) {
4075                 int tmp_res;
4076
4077                 if (rollback_slave == slave)
4078                         break;
4079
4080                 tmp_res = dev_set_mac_address(rollback_slave->dev,
4081                                               (struct sockaddr *)&tmp_ss, NULL);
4082                 if (tmp_res) {
4083                         slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4084                                    __func__, tmp_res);
4085                 }
4086         }
4087
4088         return res;
4089 }
4090
4091 /**
4092  * bond_get_slave_by_id - get xmit slave with slave_id
4093  * @bond: bonding device that is transmitting
4094  * @slave_id: slave id up to slave_cnt-1 through which to transmit
4095  *
4096  * This function tries to get slave with slave_id but in case
4097  * it fails, it tries to find the first available slave for transmission.
4098  */
4099 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4100                                           int slave_id)
4101 {
4102         struct list_head *iter;
4103         struct slave *slave;
4104         int i = slave_id;
4105
4106         /* Here we start from the slave with slave_id */
4107         bond_for_each_slave_rcu(bond, slave, iter) {
4108                 if (--i < 0) {
4109                         if (bond_slave_can_tx(slave))
4110                                 return slave;
4111                 }
4112         }
4113
4114         /* Here we start from the first slave up to slave_id */
4115         i = slave_id;
4116         bond_for_each_slave_rcu(bond, slave, iter) {
4117                 if (--i < 0)
4118                         break;
4119                 if (bond_slave_can_tx(slave))
4120                         return slave;
4121         }
4122         /* no slave that can tx has been found */
4123         return NULL;
4124 }
4125
4126 /**
4127  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4128  * @bond: bonding device to use
4129  *
4130  * Based on the value of the bonding device's packets_per_slave parameter
4131  * this function generates a slave id, which is usually used as the next
4132  * slave to transmit through.
4133  */
4134 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4135 {
4136         u32 slave_id;
4137         struct reciprocal_value reciprocal_packets_per_slave;
4138         int packets_per_slave = bond->params.packets_per_slave;
4139
4140         switch (packets_per_slave) {
4141         case 0:
4142                 slave_id = prandom_u32();
4143                 break;
4144         case 1:
4145                 slave_id = bond->rr_tx_counter;
4146                 break;
4147         default:
4148                 reciprocal_packets_per_slave =
4149                         bond->params.reciprocal_packets_per_slave;
4150                 slave_id = reciprocal_divide(bond->rr_tx_counter,
4151                                              reciprocal_packets_per_slave);
4152                 break;
4153         }
4154         bond->rr_tx_counter++;
4155
4156         return slave_id;
4157 }
4158
4159 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4160                                                     struct sk_buff *skb)
4161 {
4162         struct slave *slave;
4163         int slave_cnt;
4164         u32 slave_id;
4165
4166         /* Start with the curr_active_slave that joined the bond as the
4167          * default for sending IGMP traffic.  For failover purposes one
4168          * needs to maintain some consistency for the interface that will
4169          * send the join/membership reports.  The curr_active_slave found
4170          * will send all of this type of traffic.
4171          */
4172         if (skb->protocol == htons(ETH_P_IP)) {
4173                 int noff = skb_network_offset(skb);
4174                 struct iphdr *iph;
4175
4176                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4177                         goto non_igmp;
4178
4179                 iph = ip_hdr(skb);
4180                 if (iph->protocol == IPPROTO_IGMP) {
4181                         slave = rcu_dereference(bond->curr_active_slave);
4182                         if (slave)
4183                                 return slave;
4184                         return bond_get_slave_by_id(bond, 0);
4185                 }
4186         }
4187
4188 non_igmp:
4189         slave_cnt = READ_ONCE(bond->slave_cnt);
4190         if (likely(slave_cnt)) {
4191                 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4192                 return bond_get_slave_by_id(bond, slave_id);
4193         }
4194         return NULL;
4195 }
4196
4197 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4198                                         struct net_device *bond_dev)
4199 {
4200         struct bonding *bond = netdev_priv(bond_dev);
4201         struct slave *slave;
4202
4203         slave = bond_xmit_roundrobin_slave_get(bond, skb);
4204         if (likely(slave))
4205                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4206
4207         return bond_tx_drop(bond_dev, skb);
4208 }
4209
4210 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond,
4211                                                       struct sk_buff *skb)
4212 {
4213         return rcu_dereference(bond->curr_active_slave);
4214 }
4215
4216 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4217  * the bond has a usable interface.
4218  */
4219 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4220                                           struct net_device *bond_dev)
4221 {
4222         struct bonding *bond = netdev_priv(bond_dev);
4223         struct slave *slave;
4224
4225         slave = bond_xmit_activebackup_slave_get(bond, skb);
4226         if (slave)
4227                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4228
4229         return bond_tx_drop(bond_dev, skb);
4230 }
4231
4232 /* Use this to update slave_array when (a) it's not appropriate to update
4233  * slave_array right away (note that update_slave_array() may sleep)
4234  * and / or (b) RTNL is not held.
4235  */
4236 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4237 {
4238         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4239 }
4240
4241 /* Slave array work handler. Holds only RTNL */
4242 static void bond_slave_arr_handler(struct work_struct *work)
4243 {
4244         struct bonding *bond = container_of(work, struct bonding,
4245                                             slave_arr_work.work);
4246         int ret;
4247
4248         if (!rtnl_trylock())
4249                 goto err;
4250
4251         ret = bond_update_slave_arr(bond, NULL);
4252         rtnl_unlock();
4253         if (ret) {
4254                 pr_warn_ratelimited("Failed to update slave array from WT\n");
4255                 goto err;
4256         }
4257         return;
4258
4259 err:
4260         bond_slave_arr_work_rearm(bond, 1);
4261 }
4262
4263 static void bond_skip_slave(struct bond_up_slave *slaves,
4264                             struct slave *skipslave)
4265 {
4266         int idx;
4267
4268         /* Rare situation where caller has asked to skip a specific
4269          * slave but allocation failed (most likely!). BTW this is
4270          * only possible when the call is initiated from
4271          * __bond_release_one(). In this situation; overwrite the
4272          * skipslave entry in the array with the last entry from the
4273          * array to avoid a situation where the xmit path may choose
4274          * this to-be-skipped slave to send a packet out.
4275          */
4276         for (idx = 0; slaves && idx < slaves->count; idx++) {
4277                 if (skipslave == slaves->arr[idx]) {
4278                         slaves->arr[idx] =
4279                                 slaves->arr[slaves->count - 1];
4280                         slaves->count--;
4281                         break;
4282                 }
4283         }
4284 }
4285
4286 static void bond_set_slave_arr(struct bonding *bond,
4287                                struct bond_up_slave *usable_slaves,
4288                                struct bond_up_slave *all_slaves)
4289 {
4290         struct bond_up_slave *usable, *all;
4291
4292         usable = rtnl_dereference(bond->usable_slaves);
4293         rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4294         kfree_rcu(usable, rcu);
4295
4296         all = rtnl_dereference(bond->all_slaves);
4297         rcu_assign_pointer(bond->all_slaves, all_slaves);
4298         kfree_rcu(all, rcu);
4299 }
4300
4301 static void bond_reset_slave_arr(struct bonding *bond)
4302 {
4303         struct bond_up_slave *usable, *all;
4304
4305         usable = rtnl_dereference(bond->usable_slaves);
4306         if (usable) {
4307                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4308                 kfree_rcu(usable, rcu);
4309         }
4310
4311         all = rtnl_dereference(bond->all_slaves);
4312         if (all) {
4313                 RCU_INIT_POINTER(bond->all_slaves, NULL);
4314                 kfree_rcu(all, rcu);
4315         }
4316 }
4317
4318 /* Build the usable slaves array in control path for modes that use xmit-hash
4319  * to determine the slave interface -
4320  * (a) BOND_MODE_8023AD
4321  * (b) BOND_MODE_XOR
4322  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4323  *
4324  * The caller is expected to hold RTNL only and NO other lock!
4325  */
4326 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4327 {
4328         struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
4329         struct slave *slave;
4330         struct list_head *iter;
4331         int agg_id = 0;
4332         int ret = 0;
4333
4334 #ifdef CONFIG_LOCKDEP
4335         WARN_ON(lockdep_is_held(&bond->mode_lock));
4336 #endif
4337
4338         usable_slaves = kzalloc(struct_size(usable_slaves, arr,
4339                                             bond->slave_cnt), GFP_KERNEL);
4340         all_slaves = kzalloc(struct_size(all_slaves, arr,
4341                                          bond->slave_cnt), GFP_KERNEL);
4342         if (!usable_slaves || !all_slaves) {
4343                 ret = -ENOMEM;
4344                 goto out;
4345         }
4346         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4347                 struct ad_info ad_info;
4348
4349                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4350                         pr_debug("bond_3ad_get_active_agg_info failed\n");
4351                         /* No active aggragator means it's not safe to use
4352                          * the previous array.
4353                          */
4354                         bond_reset_slave_arr(bond);
4355                         goto out;
4356                 }
4357                 agg_id = ad_info.aggregator_id;
4358         }
4359         bond_for_each_slave(bond, slave, iter) {
4360                 if (skipslave == slave)
4361                         continue;
4362
4363                 all_slaves->arr[all_slaves->count++] = slave;
4364                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4365                         struct aggregator *agg;
4366
4367                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
4368                         if (!agg || agg->aggregator_identifier != agg_id)
4369                                 continue;
4370                 }
4371                 if (!bond_slave_can_tx(slave))
4372                         continue;
4373
4374                 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4375                           usable_slaves->count);
4376
4377                 usable_slaves->arr[usable_slaves->count++] = slave;
4378         }
4379
4380         bond_set_slave_arr(bond, usable_slaves, all_slaves);
4381         return ret;
4382 out:
4383         if (ret != 0 && skipslave) {
4384                 bond_skip_slave(rtnl_dereference(bond->all_slaves),
4385                                 skipslave);
4386                 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
4387                                 skipslave);
4388         }
4389         kfree_rcu(all_slaves, rcu);
4390         kfree_rcu(usable_slaves, rcu);
4391
4392         return ret;
4393 }
4394
4395 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
4396                                                  struct sk_buff *skb,
4397                                                  struct bond_up_slave *slaves)
4398 {
4399         struct slave *slave;
4400         unsigned int count;
4401         u32 hash;
4402
4403         hash = bond_xmit_hash(bond, skb);
4404         count = slaves ? READ_ONCE(slaves->count) : 0;
4405         if (unlikely(!count))
4406                 return NULL;
4407
4408         slave = slaves->arr[hash % count];
4409         return slave;
4410 }
4411
4412 /* Use this Xmit function for 3AD as well as XOR modes. The current
4413  * usable slave array is formed in the control path. The xmit function
4414  * just calculates hash and sends the packet out.
4415  */
4416 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4417                                      struct net_device *dev)
4418 {
4419         struct bonding *bond = netdev_priv(dev);
4420         struct bond_up_slave *slaves;
4421         struct slave *slave;
4422
4423         slaves = rcu_dereference(bond->usable_slaves);
4424         slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4425         if (likely(slave))
4426                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4427
4428         return bond_tx_drop(dev, skb);
4429 }
4430
4431 /* in broadcast mode, we send everything to all usable interfaces. */
4432 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4433                                        struct net_device *bond_dev)
4434 {
4435         struct bonding *bond = netdev_priv(bond_dev);
4436         struct slave *slave = NULL;
4437         struct list_head *iter;
4438
4439         bond_for_each_slave_rcu(bond, slave, iter) {
4440                 if (bond_is_last_slave(bond, slave))
4441                         break;
4442                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4443                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4444
4445                         if (!skb2) {
4446                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4447                                                     bond_dev->name, __func__);
4448                                 continue;
4449                         }
4450                         bond_dev_queue_xmit(bond, skb2, slave->dev);
4451                 }
4452         }
4453         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4454                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4455
4456         return bond_tx_drop(bond_dev, skb);
4457 }
4458
4459 /*------------------------- Device initialization ---------------------------*/
4460
4461 /* Lookup the slave that corresponds to a qid */
4462 static inline int bond_slave_override(struct bonding *bond,
4463                                       struct sk_buff *skb)
4464 {
4465         struct slave *slave = NULL;
4466         struct list_head *iter;
4467
4468         if (!skb_rx_queue_recorded(skb))
4469                 return 1;
4470
4471         /* Find out if any slaves have the same mapping as this skb. */
4472         bond_for_each_slave_rcu(bond, slave, iter) {
4473                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4474                         if (bond_slave_is_up(slave) &&
4475                             slave->link == BOND_LINK_UP) {
4476                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4477                                 return 0;
4478                         }
4479                         /* If the slave isn't UP, use default transmit policy. */
4480                         break;
4481                 }
4482         }
4483
4484         return 1;
4485 }
4486
4487
4488 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4489                              struct net_device *sb_dev)
4490 {
4491         /* This helper function exists to help dev_pick_tx get the correct
4492          * destination queue.  Using a helper function skips a call to
4493          * skb_tx_hash and will put the skbs in the queue we expect on their
4494          * way down to the bonding driver.
4495          */
4496         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4497
4498         /* Save the original txq to restore before passing to the driver */
4499         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4500
4501         if (unlikely(txq >= dev->real_num_tx_queues)) {
4502                 do {
4503                         txq -= dev->real_num_tx_queues;
4504                 } while (txq >= dev->real_num_tx_queues);
4505         }
4506         return txq;
4507 }
4508
4509 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
4510                                               struct sk_buff *skb,
4511                                               bool all_slaves)
4512 {
4513         struct bonding *bond = netdev_priv(master_dev);
4514         struct bond_up_slave *slaves;
4515         struct slave *slave = NULL;
4516
4517         switch (BOND_MODE(bond)) {
4518         case BOND_MODE_ROUNDROBIN:
4519                 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4520                 break;
4521         case BOND_MODE_ACTIVEBACKUP:
4522                 slave = bond_xmit_activebackup_slave_get(bond, skb);
4523                 break;
4524         case BOND_MODE_8023AD:
4525         case BOND_MODE_XOR:
4526                 if (all_slaves)
4527                         slaves = rcu_dereference(bond->all_slaves);
4528                 else
4529                         slaves = rcu_dereference(bond->usable_slaves);
4530                 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4531                 break;
4532         case BOND_MODE_BROADCAST:
4533                 break;
4534         case BOND_MODE_ALB:
4535                 slave = bond_xmit_alb_slave_get(bond, skb);
4536                 break;
4537         case BOND_MODE_TLB:
4538                 slave = bond_xmit_tlb_slave_get(bond, skb);
4539                 break;
4540         default:
4541                 /* Should never happen, mode already checked */
4542                 WARN_ONCE(true, "Unknown bonding mode");
4543                 break;
4544         }
4545
4546         if (slave)
4547                 return slave->dev;
4548         return NULL;
4549 }
4550
4551 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4552 {
4553         struct bonding *bond = netdev_priv(dev);
4554
4555         if (bond_should_override_tx_queue(bond) &&
4556             !bond_slave_override(bond, skb))
4557                 return NETDEV_TX_OK;
4558
4559         switch (BOND_MODE(bond)) {
4560         case BOND_MODE_ROUNDROBIN:
4561                 return bond_xmit_roundrobin(skb, dev);
4562         case BOND_MODE_ACTIVEBACKUP:
4563                 return bond_xmit_activebackup(skb, dev);
4564         case BOND_MODE_8023AD:
4565         case BOND_MODE_XOR:
4566                 return bond_3ad_xor_xmit(skb, dev);
4567         case BOND_MODE_BROADCAST:
4568                 return bond_xmit_broadcast(skb, dev);
4569         case BOND_MODE_ALB:
4570                 return bond_alb_xmit(skb, dev);
4571         case BOND_MODE_TLB:
4572                 return bond_tlb_xmit(skb, dev);
4573         default:
4574                 /* Should never happen, mode already checked */
4575                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4576                 WARN_ON_ONCE(1);
4577                 return bond_tx_drop(dev, skb);
4578         }
4579 }
4580
4581 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4582 {
4583         struct bonding *bond = netdev_priv(dev);
4584         netdev_tx_t ret = NETDEV_TX_OK;
4585
4586         /* If we risk deadlock from transmitting this in the
4587          * netpoll path, tell netpoll to queue the frame for later tx
4588          */
4589         if (unlikely(is_netpoll_tx_blocked(dev)))
4590                 return NETDEV_TX_BUSY;
4591
4592         rcu_read_lock();
4593         if (bond_has_slaves(bond))
4594                 ret = __bond_start_xmit(skb, dev);
4595         else
4596                 ret = bond_tx_drop(dev, skb);
4597         rcu_read_unlock();
4598
4599         return ret;
4600 }
4601
4602 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
4603 {
4604         if (speed == 0 || speed == SPEED_UNKNOWN)
4605                 speed = slave->speed;
4606         else
4607                 speed = min(speed, slave->speed);
4608
4609         return speed;
4610 }
4611
4612 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4613                                            struct ethtool_link_ksettings *cmd)
4614 {
4615         struct bonding *bond = netdev_priv(bond_dev);
4616         struct list_head *iter;
4617         struct slave *slave;
4618         u32 speed = 0;
4619
4620         cmd->base.duplex = DUPLEX_UNKNOWN;
4621         cmd->base.port = PORT_OTHER;
4622
4623         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4624          * do not need to check mode.  Though link speed might not represent
4625          * the true receive or transmit bandwidth (not all modes are symmetric)
4626          * this is an accurate maximum.
4627          */
4628         bond_for_each_slave(bond, slave, iter) {
4629                 if (bond_slave_can_tx(slave)) {
4630                         if (slave->speed != SPEED_UNKNOWN) {
4631                                 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
4632                                         speed = bond_mode_bcast_speed(slave,
4633                                                                       speed);
4634                                 else
4635                                         speed += slave->speed;
4636                         }
4637                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4638                             slave->duplex != DUPLEX_UNKNOWN)
4639                                 cmd->base.duplex = slave->duplex;
4640                 }
4641         }
4642         cmd->base.speed = speed ? : SPEED_UNKNOWN;
4643
4644         return 0;
4645 }
4646
4647 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4648                                      struct ethtool_drvinfo *drvinfo)
4649 {
4650         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4651         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4652                  BOND_ABI_VERSION);
4653 }
4654
4655 static const struct ethtool_ops bond_ethtool_ops = {
4656         .get_drvinfo            = bond_ethtool_get_drvinfo,
4657         .get_link               = ethtool_op_get_link,
4658         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
4659 };
4660
4661 static const struct net_device_ops bond_netdev_ops = {
4662         .ndo_init               = bond_init,
4663         .ndo_uninit             = bond_uninit,
4664         .ndo_open               = bond_open,
4665         .ndo_stop               = bond_close,
4666         .ndo_start_xmit         = bond_start_xmit,
4667         .ndo_select_queue       = bond_select_queue,
4668         .ndo_get_stats64        = bond_get_stats,
4669         .ndo_do_ioctl           = bond_do_ioctl,
4670         .ndo_change_rx_flags    = bond_change_rx_flags,
4671         .ndo_set_rx_mode        = bond_set_rx_mode,
4672         .ndo_change_mtu         = bond_change_mtu,
4673         .ndo_set_mac_address    = bond_set_mac_address,
4674         .ndo_neigh_setup        = bond_neigh_setup,
4675         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4676         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4677 #ifdef CONFIG_NET_POLL_CONTROLLER
4678         .ndo_netpoll_setup      = bond_netpoll_setup,
4679         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4680         .ndo_poll_controller    = bond_poll_controller,
4681 #endif
4682         .ndo_add_slave          = bond_enslave,
4683         .ndo_del_slave          = bond_release,
4684         .ndo_fix_features       = bond_fix_features,
4685         .ndo_features_check     = passthru_features_check,
4686         .ndo_get_xmit_slave     = bond_xmit_get_slave,
4687 };
4688
4689 static const struct device_type bond_type = {
4690         .name = "bond",
4691 };
4692
4693 static void bond_destructor(struct net_device *bond_dev)
4694 {
4695         struct bonding *bond = netdev_priv(bond_dev);
4696         if (bond->wq)
4697                 destroy_workqueue(bond->wq);
4698 }
4699
4700 void bond_setup(struct net_device *bond_dev)
4701 {
4702         struct bonding *bond = netdev_priv(bond_dev);
4703
4704         spin_lock_init(&bond->mode_lock);
4705         bond->params = bonding_defaults;
4706
4707         /* Initialize pointers */
4708         bond->dev = bond_dev;
4709
4710         /* Initialize the device entry points */
4711         ether_setup(bond_dev);
4712         bond_dev->max_mtu = ETH_MAX_MTU;
4713         bond_dev->netdev_ops = &bond_netdev_ops;
4714         bond_dev->ethtool_ops = &bond_ethtool_ops;
4715
4716         bond_dev->needs_free_netdev = true;
4717         bond_dev->priv_destructor = bond_destructor;
4718
4719         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4720
4721         /* Initialize the device options */
4722         bond_dev->flags |= IFF_MASTER;
4723         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4724         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4725
4726 #ifdef CONFIG_XFRM_OFFLOAD
4727         /* set up xfrm device ops (only supported in active-backup right now) */
4728         bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
4729         bond->xs = NULL;
4730 #endif /* CONFIG_XFRM_OFFLOAD */
4731
4732         /* don't acquire bond device's netif_tx_lock when transmitting */
4733         bond_dev->features |= NETIF_F_LLTX;
4734
4735         /* By default, we declare the bond to be fully
4736          * VLAN hardware accelerated capable. Special
4737          * care is taken in the various xmit functions
4738          * when there are slaves that are not hw accel
4739          * capable
4740          */
4741
4742         /* Don't allow bond devices to change network namespaces. */
4743         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4744
4745         bond_dev->hw_features = BOND_VLAN_FEATURES |
4746                                 NETIF_F_HW_VLAN_CTAG_RX |
4747                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4748
4749         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4750 #ifdef CONFIG_XFRM_OFFLOAD
4751         bond_dev->hw_features |= BOND_XFRM_FEATURES;
4752 #endif /* CONFIG_XFRM_OFFLOAD */
4753         bond_dev->features |= bond_dev->hw_features;
4754         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4755 #ifdef CONFIG_XFRM_OFFLOAD
4756         /* Disable XFRM features if this isn't an active-backup config */
4757         if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
4758                 bond_dev->features &= ~BOND_XFRM_FEATURES;
4759 #endif /* CONFIG_XFRM_OFFLOAD */
4760 }
4761
4762 /* Destroy a bonding device.
4763  * Must be under rtnl_lock when this function is called.
4764  */
4765 static void bond_uninit(struct net_device *bond_dev)
4766 {
4767         struct bonding *bond = netdev_priv(bond_dev);
4768         struct bond_up_slave *usable, *all;
4769         struct list_head *iter;
4770         struct slave *slave;
4771
4772         bond_netpoll_cleanup(bond_dev);
4773
4774         /* Release the bonded slaves */
4775         bond_for_each_slave(bond, slave, iter)
4776                 __bond_release_one(bond_dev, slave->dev, true, true);
4777         netdev_info(bond_dev, "Released all slaves\n");
4778
4779         usable = rtnl_dereference(bond->usable_slaves);
4780         if (usable) {
4781                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4782                 kfree_rcu(usable, rcu);
4783         }
4784
4785         all = rtnl_dereference(bond->all_slaves);
4786         if (all) {
4787                 RCU_INIT_POINTER(bond->all_slaves, NULL);
4788                 kfree_rcu(all, rcu);
4789         }
4790
4791         list_del(&bond->bond_list);
4792
4793         bond_debug_unregister(bond);
4794 }
4795
4796 /*------------------------- Module initialization ---------------------------*/
4797
4798 static int bond_check_params(struct bond_params *params)
4799 {
4800         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4801         struct bond_opt_value newval;
4802         const struct bond_opt_value *valptr;
4803         int arp_all_targets_value = 0;
4804         u16 ad_actor_sys_prio = 0;
4805         u16 ad_user_port_key = 0;
4806         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4807         int arp_ip_count;
4808         int bond_mode   = BOND_MODE_ROUNDROBIN;
4809         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4810         int lacp_fast = 0;
4811         int tlb_dynamic_lb;
4812
4813         /* Convert string parameters. */
4814         if (mode) {
4815                 bond_opt_initstr(&newval, mode);
4816                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4817                 if (!valptr) {
4818                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4819                         return -EINVAL;
4820                 }
4821                 bond_mode = valptr->value;
4822         }
4823
4824         if (xmit_hash_policy) {
4825                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4826                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
4827                     bond_mode == BOND_MODE_BROADCAST) {
4828                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4829                                 bond_mode_name(bond_mode));
4830                 } else {
4831                         bond_opt_initstr(&newval, xmit_hash_policy);
4832                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4833                                                 &newval);
4834                         if (!valptr) {
4835                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4836                                        xmit_hash_policy);
4837                                 return -EINVAL;
4838                         }
4839                         xmit_hashtype = valptr->value;
4840                 }
4841         }
4842
4843         if (lacp_rate) {
4844                 if (bond_mode != BOND_MODE_8023AD) {
4845                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4846                                 bond_mode_name(bond_mode));
4847                 } else {
4848                         bond_opt_initstr(&newval, lacp_rate);
4849                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4850                                                 &newval);
4851                         if (!valptr) {
4852                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4853                                        lacp_rate);
4854                                 return -EINVAL;
4855                         }
4856                         lacp_fast = valptr->value;
4857                 }
4858         }
4859
4860         if (ad_select) {
4861                 bond_opt_initstr(&newval, ad_select);
4862                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4863                                         &newval);
4864                 if (!valptr) {
4865                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4866                         return -EINVAL;
4867                 }
4868                 params->ad_select = valptr->value;
4869                 if (bond_mode != BOND_MODE_8023AD)
4870                         pr_warn("ad_select param only affects 802.3ad mode\n");
4871         } else {
4872                 params->ad_select = BOND_AD_STABLE;
4873         }
4874
4875         if (max_bonds < 0) {
4876                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4877                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4878                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4879         }
4880
4881         if (miimon < 0) {
4882                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4883                         miimon, INT_MAX);
4884                 miimon = 0;
4885         }
4886
4887         if (updelay < 0) {
4888                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4889                         updelay, INT_MAX);
4890                 updelay = 0;
4891         }
4892
4893         if (downdelay < 0) {
4894                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4895                         downdelay, INT_MAX);
4896                 downdelay = 0;
4897         }
4898
4899         if ((use_carrier != 0) && (use_carrier != 1)) {
4900                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4901                         use_carrier);
4902                 use_carrier = 1;
4903         }
4904
4905         if (num_peer_notif < 0 || num_peer_notif > 255) {
4906                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4907                         num_peer_notif);
4908                 num_peer_notif = 1;
4909         }
4910
4911         /* reset values for 802.3ad/TLB/ALB */
4912         if (!bond_mode_uses_arp(bond_mode)) {
4913                 if (!miimon) {
4914                         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");
4915                         pr_warn("Forcing miimon to 100msec\n");
4916                         miimon = BOND_DEFAULT_MIIMON;
4917                 }
4918         }
4919
4920         if (tx_queues < 1 || tx_queues > 255) {
4921                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4922                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4923                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4924         }
4925
4926         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4927                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4928                         all_slaves_active);
4929                 all_slaves_active = 0;
4930         }
4931
4932         if (resend_igmp < 0 || resend_igmp > 255) {
4933                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4934                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4935                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4936         }
4937
4938         bond_opt_initval(&newval, packets_per_slave);
4939         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4940                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4941                         packets_per_slave, USHRT_MAX);
4942                 packets_per_slave = 1;
4943         }
4944
4945         if (bond_mode == BOND_MODE_ALB) {
4946                 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",
4947                           updelay);
4948         }
4949
4950         if (!miimon) {
4951                 if (updelay || downdelay) {
4952                         /* just warn the user the up/down delay will have
4953                          * no effect since miimon is zero...
4954                          */
4955                         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",
4956                                 updelay, downdelay);
4957                 }
4958         } else {
4959                 /* don't allow arp monitoring */
4960                 if (arp_interval) {
4961                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4962                                 miimon, arp_interval);
4963                         arp_interval = 0;
4964                 }
4965
4966                 if ((updelay % miimon) != 0) {
4967                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4968                                 updelay, miimon, (updelay / miimon) * miimon);
4969                 }
4970
4971                 updelay /= miimon;
4972
4973                 if ((downdelay % miimon) != 0) {
4974                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4975                                 downdelay, miimon,
4976                                 (downdelay / miimon) * miimon);
4977                 }
4978
4979                 downdelay /= miimon;
4980         }
4981
4982         if (arp_interval < 0) {
4983                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4984                         arp_interval, INT_MAX);
4985                 arp_interval = 0;
4986         }
4987
4988         for (arp_ip_count = 0, i = 0;
4989              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4990                 __be32 ip;
4991
4992                 /* not a complete check, but good enough to catch mistakes */
4993                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4994                     !bond_is_ip_target_ok(ip)) {
4995                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4996                                 arp_ip_target[i]);
4997                         arp_interval = 0;
4998                 } else {
4999                         if (bond_get_targets_ip(arp_target, ip) == -1)
5000                                 arp_target[arp_ip_count++] = ip;
5001                         else
5002                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
5003                                         &ip);
5004                 }
5005         }
5006
5007         if (arp_interval && !arp_ip_count) {
5008                 /* don't allow arping if no arp_ip_target given... */
5009                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5010                         arp_interval);
5011                 arp_interval = 0;
5012         }
5013
5014         if (arp_validate) {
5015                 if (!arp_interval) {
5016                         pr_err("arp_validate requires arp_interval\n");
5017                         return -EINVAL;
5018                 }
5019
5020                 bond_opt_initstr(&newval, arp_validate);
5021                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
5022                                         &newval);
5023                 if (!valptr) {
5024                         pr_err("Error: invalid arp_validate \"%s\"\n",
5025                                arp_validate);
5026                         return -EINVAL;
5027                 }
5028                 arp_validate_value = valptr->value;
5029         } else {
5030                 arp_validate_value = 0;
5031         }
5032
5033         if (arp_all_targets) {
5034                 bond_opt_initstr(&newval, arp_all_targets);
5035                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
5036                                         &newval);
5037                 if (!valptr) {
5038                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
5039                                arp_all_targets);
5040                         arp_all_targets_value = 0;
5041                 } else {
5042                         arp_all_targets_value = valptr->value;
5043                 }
5044         }
5045
5046         if (miimon) {
5047                 pr_info("MII link monitoring set to %d ms\n", miimon);
5048         } else if (arp_interval) {
5049                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
5050                                           arp_validate_value);
5051                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5052                         arp_interval, valptr->string, arp_ip_count);
5053
5054                 for (i = 0; i < arp_ip_count; i++)
5055                         pr_cont(" %s", arp_ip_target[i]);
5056
5057                 pr_cont("\n");
5058
5059         } else if (max_bonds) {
5060                 /* miimon and arp_interval not set, we need one so things
5061                  * work as expected, see bonding.txt for details
5062                  */
5063                 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");
5064         }
5065
5066         if (primary && !bond_mode_uses_primary(bond_mode)) {
5067                 /* currently, using a primary only makes sense
5068                  * in active backup, TLB or ALB modes
5069                  */
5070                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
5071                         primary, bond_mode_name(bond_mode));
5072                 primary = NULL;
5073         }
5074
5075         if (primary && primary_reselect) {
5076                 bond_opt_initstr(&newval, primary_reselect);
5077                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
5078                                         &newval);
5079                 if (!valptr) {
5080                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5081                                primary_reselect);
5082                         return -EINVAL;
5083                 }
5084                 primary_reselect_value = valptr->value;
5085         } else {
5086                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5087         }
5088
5089         if (fail_over_mac) {
5090                 bond_opt_initstr(&newval, fail_over_mac);
5091                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
5092                                         &newval);
5093                 if (!valptr) {
5094                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5095                                fail_over_mac);
5096                         return -EINVAL;
5097                 }
5098                 fail_over_mac_value = valptr->value;
5099                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5100                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
5101         } else {
5102                 fail_over_mac_value = BOND_FOM_NONE;
5103         }
5104
5105         bond_opt_initstr(&newval, "default");
5106         valptr = bond_opt_parse(
5107                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
5108                                      &newval);
5109         if (!valptr) {
5110                 pr_err("Error: No ad_actor_sys_prio default value");
5111                 return -EINVAL;
5112         }
5113         ad_actor_sys_prio = valptr->value;
5114
5115         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
5116                                 &newval);
5117         if (!valptr) {
5118                 pr_err("Error: No ad_user_port_key default value");
5119                 return -EINVAL;
5120         }
5121         ad_user_port_key = valptr->value;
5122
5123         bond_opt_initstr(&newval, "default");
5124         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
5125         if (!valptr) {
5126                 pr_err("Error: No tlb_dynamic_lb default value");
5127                 return -EINVAL;
5128         }
5129         tlb_dynamic_lb = valptr->value;
5130
5131         if (lp_interval == 0) {
5132                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
5133                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
5134                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
5135         }
5136
5137         /* fill params struct with the proper values */
5138         params->mode = bond_mode;
5139         params->xmit_policy = xmit_hashtype;
5140         params->miimon = miimon;
5141         params->num_peer_notif = num_peer_notif;
5142         params->arp_interval = arp_interval;
5143         params->arp_validate = arp_validate_value;
5144         params->arp_all_targets = arp_all_targets_value;
5145         params->updelay = updelay;
5146         params->downdelay = downdelay;
5147         params->peer_notif_delay = 0;
5148         params->use_carrier = use_carrier;
5149         params->lacp_fast = lacp_fast;
5150         params->primary[0] = 0;
5151         params->primary_reselect = primary_reselect_value;
5152         params->fail_over_mac = fail_over_mac_value;
5153         params->tx_queues = tx_queues;
5154         params->all_slaves_active = all_slaves_active;
5155         params->resend_igmp = resend_igmp;
5156         params->min_links = min_links;
5157         params->lp_interval = lp_interval;
5158         params->packets_per_slave = packets_per_slave;
5159         params->tlb_dynamic_lb = tlb_dynamic_lb;
5160         params->ad_actor_sys_prio = ad_actor_sys_prio;
5161         eth_zero_addr(params->ad_actor_system);
5162         params->ad_user_port_key = ad_user_port_key;
5163         if (packets_per_slave > 0) {
5164                 params->reciprocal_packets_per_slave =
5165                         reciprocal_value(packets_per_slave);
5166         } else {
5167                 /* reciprocal_packets_per_slave is unused if
5168                  * packets_per_slave is 0 or 1, just initialize it
5169                  */
5170                 params->reciprocal_packets_per_slave =
5171                         (struct reciprocal_value) { 0 };
5172         }
5173
5174         if (primary) {
5175                 strncpy(params->primary, primary, IFNAMSIZ);
5176                 params->primary[IFNAMSIZ - 1] = 0;
5177         }
5178
5179         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5180
5181         return 0;
5182 }
5183
5184 /* Called from registration process */
5185 static int bond_init(struct net_device *bond_dev)
5186 {
5187         struct bonding *bond = netdev_priv(bond_dev);
5188         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5189
5190         netdev_dbg(bond_dev, "Begin bond_init\n");
5191
5192         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
5193         if (!bond->wq)
5194                 return -ENOMEM;
5195
5196         spin_lock_init(&bond->stats_lock);
5197         netdev_lockdep_set_classes(bond_dev);
5198
5199         list_add_tail(&bond->bond_list, &bn->dev_list);
5200
5201         bond_prepare_sysfs_group(bond);
5202
5203         bond_debug_register(bond);
5204
5205         /* Ensure valid dev_addr */
5206         if (is_zero_ether_addr(bond_dev->dev_addr) &&
5207             bond_dev->addr_assign_type == NET_ADDR_PERM)
5208                 eth_hw_addr_random(bond_dev);
5209
5210         return 0;
5211 }
5212
5213 unsigned int bond_get_num_tx_queues(void)
5214 {
5215         return tx_queues;
5216 }
5217
5218 /* Create a new bond based on the specified name and bonding parameters.
5219  * If name is NULL, obtain a suitable "bond%d" name for us.
5220  * Caller must NOT hold rtnl_lock; we need to release it here before we
5221  * set up our sysfs entries.
5222  */
5223 int bond_create(struct net *net, const char *name)
5224 {
5225         struct net_device *bond_dev;
5226         struct bonding *bond;
5227         struct alb_bond_info *bond_info;
5228         int res;
5229
5230         rtnl_lock();
5231
5232         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
5233                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
5234                                    bond_setup, tx_queues);
5235         if (!bond_dev) {
5236                 pr_err("%s: eek! can't alloc netdev!\n", name);
5237                 rtnl_unlock();
5238                 return -ENOMEM;
5239         }
5240
5241         /*
5242          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
5243          * It is set to 0 by default which is wrong.
5244          */
5245         bond = netdev_priv(bond_dev);
5246         bond_info = &(BOND_ALB_INFO(bond));
5247         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
5248
5249         dev_net_set(bond_dev, net);
5250         bond_dev->rtnl_link_ops = &bond_link_ops;
5251
5252         res = register_netdevice(bond_dev);
5253         if (res < 0) {
5254                 free_netdev(bond_dev);
5255                 rtnl_unlock();
5256
5257                 return res;
5258         }
5259
5260         netif_carrier_off(bond_dev);
5261
5262         bond_work_init_all(bond);
5263
5264         rtnl_unlock();
5265         return 0;
5266 }
5267
5268 static int __net_init bond_net_init(struct net *net)
5269 {
5270         struct bond_net *bn = net_generic(net, bond_net_id);
5271
5272         bn->net = net;
5273         INIT_LIST_HEAD(&bn->dev_list);
5274
5275         bond_create_proc_dir(bn);
5276         bond_create_sysfs(bn);
5277
5278         return 0;
5279 }
5280
5281 static void __net_exit bond_net_exit(struct net *net)
5282 {
5283         struct bond_net *bn = net_generic(net, bond_net_id);
5284         struct bonding *bond, *tmp_bond;
5285         LIST_HEAD(list);
5286
5287         bond_destroy_sysfs(bn);
5288
5289         /* Kill off any bonds created after unregistering bond rtnl ops */
5290         rtnl_lock();
5291         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
5292                 unregister_netdevice_queue(bond->dev, &list);
5293         unregister_netdevice_many(&list);
5294         rtnl_unlock();
5295
5296         bond_destroy_proc_dir(bn);
5297 }
5298
5299 static struct pernet_operations bond_net_ops = {
5300         .init = bond_net_init,
5301         .exit = bond_net_exit,
5302         .id   = &bond_net_id,
5303         .size = sizeof(struct bond_net),
5304 };
5305
5306 static int __init bonding_init(void)
5307 {
5308         int i;
5309         int res;
5310
5311         res = bond_check_params(&bonding_defaults);
5312         if (res)
5313                 goto out;
5314
5315         res = register_pernet_subsys(&bond_net_ops);
5316         if (res)
5317                 goto out;
5318
5319         res = bond_netlink_init();
5320         if (res)
5321                 goto err_link;
5322
5323         bond_create_debugfs();
5324
5325         for (i = 0; i < max_bonds; i++) {
5326                 res = bond_create(&init_net, NULL);
5327                 if (res)
5328                         goto err;
5329         }
5330
5331         skb_flow_dissector_init(&flow_keys_bonding,
5332                                 flow_keys_bonding_keys,
5333                                 ARRAY_SIZE(flow_keys_bonding_keys));
5334
5335         register_netdevice_notifier(&bond_netdev_notifier);
5336 out:
5337         return res;
5338 err:
5339         bond_destroy_debugfs();
5340         bond_netlink_fini();
5341 err_link:
5342         unregister_pernet_subsys(&bond_net_ops);
5343         goto out;
5344
5345 }
5346
5347 static void __exit bonding_exit(void)
5348 {
5349         unregister_netdevice_notifier(&bond_netdev_notifier);
5350
5351         bond_destroy_debugfs();
5352
5353         bond_netlink_fini();
5354         unregister_pernet_subsys(&bond_net_ops);
5355
5356 #ifdef CONFIG_NET_POLL_CONTROLLER
5357         /* Make sure we don't have an imbalance on our netpoll blocking */
5358         WARN_ON(atomic_read(&netpoll_block_tx));
5359 #endif
5360 }
5361
5362 module_init(bonding_init);
5363 module_exit(bonding_exit);
5364 MODULE_LICENSE("GPL");
5365 MODULE_DESCRIPTION(DRV_DESCRIPTION);
5366 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");