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_GSO_SOFTWARE | \
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_GSO_SOFTWARE)
1236
1237 #define BOND_MPLS_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1238                                  NETIF_F_GSO_SOFTWARE)
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 #ifdef CONFIG_XFRM_OFFLOAD
1296         bond_dev->hw_enc_features |= xfrm_features;
1297 #endif /* CONFIG_XFRM_OFFLOAD */
1298         bond_dev->mpls_features = mpls_features;
1299         bond_dev->gso_max_segs = gso_max_segs;
1300         netif_set_gso_max_size(bond_dev, gso_max_size);
1301
1302         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1303         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1304             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1305                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1306
1307         netdev_change_features(bond_dev);
1308 }
1309
1310 static void bond_setup_by_slave(struct net_device *bond_dev,
1311                                 struct net_device *slave_dev)
1312 {
1313         bond_dev->header_ops        = slave_dev->header_ops;
1314
1315         bond_dev->type              = slave_dev->type;
1316         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1317         bond_dev->needed_headroom   = slave_dev->needed_headroom;
1318         bond_dev->addr_len          = slave_dev->addr_len;
1319
1320         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1321                 slave_dev->addr_len);
1322 }
1323
1324 /* On bonding slaves other than the currently active slave, suppress
1325  * duplicates except for alb non-mcast/bcast.
1326  */
1327 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1328                                             struct slave *slave,
1329                                             struct bonding *bond)
1330 {
1331         if (bond_is_slave_inactive(slave)) {
1332                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1333                     skb->pkt_type != PACKET_BROADCAST &&
1334                     skb->pkt_type != PACKET_MULTICAST)
1335                         return false;
1336                 return true;
1337         }
1338         return false;
1339 }
1340
1341 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1342 {
1343         struct sk_buff *skb = *pskb;
1344         struct slave *slave;
1345         struct bonding *bond;
1346         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1347                           struct slave *);
1348         int ret = RX_HANDLER_ANOTHER;
1349
1350         skb = skb_share_check(skb, GFP_ATOMIC);
1351         if (unlikely(!skb))
1352                 return RX_HANDLER_CONSUMED;
1353
1354         *pskb = skb;
1355
1356         slave = bond_slave_get_rcu(skb->dev);
1357         bond = slave->bond;
1358
1359         recv_probe = READ_ONCE(bond->recv_probe);
1360         if (recv_probe) {
1361                 ret = recv_probe(skb, bond, slave);
1362                 if (ret == RX_HANDLER_CONSUMED) {
1363                         consume_skb(skb);
1364                         return ret;
1365                 }
1366         }
1367
1368         /*
1369          * For packets determined by bond_should_deliver_exact_match() call to
1370          * be suppressed we want to make an exception for link-local packets.
1371          * This is necessary for e.g. LLDP daemons to be able to monitor
1372          * inactive slave links without being forced to bind to them
1373          * explicitly.
1374          *
1375          * At the same time, packets that are passed to the bonding master
1376          * (including link-local ones) can have their originating interface
1377          * determined via PACKET_ORIGDEV socket option.
1378          */
1379         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1380                 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1381                         return RX_HANDLER_PASS;
1382                 return RX_HANDLER_EXACT;
1383         }
1384
1385         skb->dev = bond->dev;
1386
1387         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1388             netif_is_bridge_port(bond->dev) &&
1389             skb->pkt_type == PACKET_HOST) {
1390
1391                 if (unlikely(skb_cow_head(skb,
1392                                           skb->data - skb_mac_header(skb)))) {
1393                         kfree_skb(skb);
1394                         return RX_HANDLER_CONSUMED;
1395                 }
1396                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1397                                   bond->dev->addr_len);
1398         }
1399
1400         return ret;
1401 }
1402
1403 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1404 {
1405         switch (BOND_MODE(bond)) {
1406         case BOND_MODE_ROUNDROBIN:
1407                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1408         case BOND_MODE_ACTIVEBACKUP:
1409                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1410         case BOND_MODE_BROADCAST:
1411                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1412         case BOND_MODE_XOR:
1413         case BOND_MODE_8023AD:
1414                 return NETDEV_LAG_TX_TYPE_HASH;
1415         default:
1416                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1417         }
1418 }
1419
1420 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1421                                                enum netdev_lag_tx_type type)
1422 {
1423         if (type != NETDEV_LAG_TX_TYPE_HASH)
1424                 return NETDEV_LAG_HASH_NONE;
1425
1426         switch (bond->params.xmit_policy) {
1427         case BOND_XMIT_POLICY_LAYER2:
1428                 return NETDEV_LAG_HASH_L2;
1429         case BOND_XMIT_POLICY_LAYER34:
1430                 return NETDEV_LAG_HASH_L34;
1431         case BOND_XMIT_POLICY_LAYER23:
1432                 return NETDEV_LAG_HASH_L23;
1433         case BOND_XMIT_POLICY_ENCAP23:
1434                 return NETDEV_LAG_HASH_E23;
1435         case BOND_XMIT_POLICY_ENCAP34:
1436                 return NETDEV_LAG_HASH_E34;
1437         default:
1438                 return NETDEV_LAG_HASH_UNKNOWN;
1439         }
1440 }
1441
1442 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1443                                       struct netlink_ext_ack *extack)
1444 {
1445         struct netdev_lag_upper_info lag_upper_info;
1446         enum netdev_lag_tx_type type;
1447
1448         type = bond_lag_tx_type(bond);
1449         lag_upper_info.tx_type = type;
1450         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1451
1452         return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1453                                             &lag_upper_info, extack);
1454 }
1455
1456 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1457 {
1458         netdev_upper_dev_unlink(slave->dev, bond->dev);
1459         slave->dev->flags &= ~IFF_SLAVE;
1460 }
1461
1462 static void slave_kobj_release(struct kobject *kobj)
1463 {
1464         struct slave *slave = to_slave(kobj);
1465         struct bonding *bond = bond_get_bond_by_slave(slave);
1466
1467         cancel_delayed_work_sync(&slave->notify_work);
1468         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1469                 kfree(SLAVE_AD_INFO(slave));
1470
1471         kfree(slave);
1472 }
1473
1474 static struct kobj_type slave_ktype = {
1475         .release = slave_kobj_release,
1476 #ifdef CONFIG_SYSFS
1477         .sysfs_ops = &slave_sysfs_ops,
1478 #endif
1479 };
1480
1481 static int bond_kobj_init(struct slave *slave)
1482 {
1483         int err;
1484
1485         err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1486                                    &(slave->dev->dev.kobj), "bonding_slave");
1487         if (err)
1488                 kobject_put(&slave->kobj);
1489
1490         return err;
1491 }
1492
1493 static struct slave *bond_alloc_slave(struct bonding *bond,
1494                                       struct net_device *slave_dev)
1495 {
1496         struct slave *slave = NULL;
1497
1498         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1499         if (!slave)
1500                 return NULL;
1501
1502         slave->bond = bond;
1503         slave->dev = slave_dev;
1504
1505         if (bond_kobj_init(slave))
1506                 return NULL;
1507
1508         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1509                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1510                                                GFP_KERNEL);
1511                 if (!SLAVE_AD_INFO(slave)) {
1512                         kobject_put(&slave->kobj);
1513                         return NULL;
1514                 }
1515         }
1516         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1517
1518         return slave;
1519 }
1520
1521 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1522 {
1523         info->bond_mode = BOND_MODE(bond);
1524         info->miimon = bond->params.miimon;
1525         info->num_slaves = bond->slave_cnt;
1526 }
1527
1528 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1529 {
1530         strcpy(info->slave_name, slave->dev->name);
1531         info->link = slave->link;
1532         info->state = bond_slave_state(slave);
1533         info->link_failure_count = slave->link_failure_count;
1534 }
1535
1536 static void bond_netdev_notify_work(struct work_struct *_work)
1537 {
1538         struct slave *slave = container_of(_work, struct slave,
1539                                            notify_work.work);
1540
1541         if (rtnl_trylock()) {
1542                 struct netdev_bonding_info binfo;
1543
1544                 bond_fill_ifslave(slave, &binfo.slave);
1545                 bond_fill_ifbond(slave->bond, &binfo.master);
1546                 netdev_bonding_info_change(slave->dev, &binfo);
1547                 rtnl_unlock();
1548         } else {
1549                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1550         }
1551 }
1552
1553 void bond_queue_slave_event(struct slave *slave)
1554 {
1555         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1556 }
1557
1558 void bond_lower_state_changed(struct slave *slave)
1559 {
1560         struct netdev_lag_lower_state_info info;
1561
1562         info.link_up = slave->link == BOND_LINK_UP ||
1563                        slave->link == BOND_LINK_FAIL;
1564         info.tx_enabled = bond_is_active_slave(slave);
1565         netdev_lower_state_changed(slave->dev, &info);
1566 }
1567
1568 /* enslave device <slave> to bond device <master> */
1569 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1570                  struct netlink_ext_ack *extack)
1571 {
1572         struct bonding *bond = netdev_priv(bond_dev);
1573         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1574         struct slave *new_slave = NULL, *prev_slave;
1575         struct sockaddr_storage ss;
1576         int link_reporting;
1577         int res = 0, i;
1578
1579         if (!bond->params.use_carrier &&
1580             slave_dev->ethtool_ops->get_link == NULL &&
1581             slave_ops->ndo_do_ioctl == NULL) {
1582                 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1583         }
1584
1585         /* already in-use? */
1586         if (netdev_is_rx_handler_busy(slave_dev)) {
1587                 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1588                 slave_err(bond_dev, slave_dev,
1589                           "Error: Device is in use and cannot be enslaved\n");
1590                 return -EBUSY;
1591         }
1592
1593         if (bond_dev == slave_dev) {
1594                 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1595                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1596                 return -EPERM;
1597         }
1598
1599         /* vlan challenged mutual exclusion */
1600         /* no need to lock since we're protected by rtnl_lock */
1601         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1602                 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1603                 if (vlan_uses_dev(bond_dev)) {
1604                         NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1605                         slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1606                         return -EPERM;
1607                 } else {
1608                         slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1609                 }
1610         } else {
1611                 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1612         }
1613
1614         if (slave_dev->features & NETIF_F_HW_ESP)
1615                 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1616
1617         /* Old ifenslave binaries are no longer supported.  These can
1618          * be identified with moderate accuracy by the state of the slave:
1619          * the current ifenslave will set the interface down prior to
1620          * enslaving it; the old ifenslave will not.
1621          */
1622         if (slave_dev->flags & IFF_UP) {
1623                 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1624                 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1625                 return -EPERM;
1626         }
1627
1628         /* set bonding device ether type by slave - bonding netdevices are
1629          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1630          * there is a need to override some of the type dependent attribs/funcs.
1631          *
1632          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1633          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1634          */
1635         if (!bond_has_slaves(bond)) {
1636                 if (bond_dev->type != slave_dev->type) {
1637                         slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1638                                   bond_dev->type, slave_dev->type);
1639
1640                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1641                                                        bond_dev);
1642                         res = notifier_to_errno(res);
1643                         if (res) {
1644                                 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1645                                 return -EBUSY;
1646                         }
1647
1648                         /* Flush unicast and multicast addresses */
1649                         dev_uc_flush(bond_dev);
1650                         dev_mc_flush(bond_dev);
1651
1652                         if (slave_dev->type != ARPHRD_ETHER)
1653                                 bond_setup_by_slave(bond_dev, slave_dev);
1654                         else {
1655                                 ether_setup(bond_dev);
1656                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1657                         }
1658
1659                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1660                                                  bond_dev);
1661                 }
1662         } else if (bond_dev->type != slave_dev->type) {
1663                 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1664                 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1665                           slave_dev->type, bond_dev->type);
1666                 return -EINVAL;
1667         }
1668
1669         if (slave_dev->type == ARPHRD_INFINIBAND &&
1670             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1671                 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1672                 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1673                            slave_dev->type);
1674                 res = -EOPNOTSUPP;
1675                 goto err_undo_flags;
1676         }
1677
1678         if (!slave_ops->ndo_set_mac_address ||
1679             slave_dev->type == ARPHRD_INFINIBAND) {
1680                 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1681                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1682                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1683                         if (!bond_has_slaves(bond)) {
1684                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1685                                 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1686                         } else {
1687                                 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1688                                 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");
1689                                 res = -EOPNOTSUPP;
1690                                 goto err_undo_flags;
1691                         }
1692                 }
1693         }
1694
1695         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1696
1697         /* If this is the first slave, then we need to set the master's hardware
1698          * address to be the same as the slave's.
1699          */
1700         if (!bond_has_slaves(bond) &&
1701             bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1702                 res = bond_set_dev_addr(bond->dev, slave_dev);
1703                 if (res)
1704                         goto err_undo_flags;
1705         }
1706
1707         new_slave = bond_alloc_slave(bond, slave_dev);
1708         if (!new_slave) {
1709                 res = -ENOMEM;
1710                 goto err_undo_flags;
1711         }
1712
1713         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1714          * is set via sysfs or module option if desired.
1715          */
1716         new_slave->queue_id = 0;
1717
1718         /* Save slave's original mtu and then set it to match the bond */
1719         new_slave->original_mtu = slave_dev->mtu;
1720         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1721         if (res) {
1722                 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1723                 goto err_free;
1724         }
1725
1726         /* Save slave's original ("permanent") mac address for modes
1727          * that need it, and for restoring it upon release, and then
1728          * set it to the master's address
1729          */
1730         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1731                           slave_dev->addr_len);
1732
1733         if (!bond->params.fail_over_mac ||
1734             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1735                 /* Set slave to master's mac address.  The application already
1736                  * set the master's mac address to that of the first slave
1737                  */
1738                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1739                 ss.ss_family = slave_dev->type;
1740                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1741                                           extack);
1742                 if (res) {
1743                         slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1744                         goto err_restore_mtu;
1745                 }
1746         }
1747
1748         /* set slave flag before open to prevent IPv6 addrconf */
1749         slave_dev->flags |= IFF_SLAVE;
1750
1751         /* open the slave since the application closed it */
1752         res = dev_open(slave_dev, extack);
1753         if (res) {
1754                 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1755                 goto err_restore_mac;
1756         }
1757
1758         slave_dev->priv_flags |= IFF_BONDING;
1759         /* initialize slave stats */
1760         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1761
1762         if (bond_is_lb(bond)) {
1763                 /* bond_alb_init_slave() must be called before all other stages since
1764                  * it might fail and we do not want to have to undo everything
1765                  */
1766                 res = bond_alb_init_slave(bond, new_slave);
1767                 if (res)
1768                         goto err_close;
1769         }
1770
1771         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1772         if (res) {
1773                 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1774                 goto err_close;
1775         }
1776
1777         prev_slave = bond_last_slave(bond);
1778
1779         new_slave->delay = 0;
1780         new_slave->link_failure_count = 0;
1781
1782         if (bond_update_speed_duplex(new_slave) &&
1783             bond_needs_speed_duplex(bond))
1784                 new_slave->link = BOND_LINK_DOWN;
1785
1786         new_slave->last_rx = jiffies -
1787                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1788         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1789                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1790
1791         if (bond->params.miimon && !bond->params.use_carrier) {
1792                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1793
1794                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1795                         /* miimon is set but a bonded network driver
1796                          * does not support ETHTOOL/MII and
1797                          * arp_interval is not set.  Note: if
1798                          * use_carrier is enabled, we will never go
1799                          * here (because netif_carrier is always
1800                          * supported); thus, we don't need to change
1801                          * the messages for netif_carrier.
1802                          */
1803                         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");
1804                 } else if (link_reporting == -1) {
1805                         /* unable get link status using mii/ethtool */
1806                         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");
1807                 }
1808         }
1809
1810         /* check for initial state */
1811         new_slave->link = BOND_LINK_NOCHANGE;
1812         if (bond->params.miimon) {
1813                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1814                         if (bond->params.updelay) {
1815                                 bond_set_slave_link_state(new_slave,
1816                                                           BOND_LINK_BACK,
1817                                                           BOND_SLAVE_NOTIFY_NOW);
1818                                 new_slave->delay = bond->params.updelay;
1819                         } else {
1820                                 bond_set_slave_link_state(new_slave,
1821                                                           BOND_LINK_UP,
1822                                                           BOND_SLAVE_NOTIFY_NOW);
1823                         }
1824                 } else {
1825                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1826                                                   BOND_SLAVE_NOTIFY_NOW);
1827                 }
1828         } else if (bond->params.arp_interval) {
1829                 bond_set_slave_link_state(new_slave,
1830                                           (netif_carrier_ok(slave_dev) ?
1831                                           BOND_LINK_UP : BOND_LINK_DOWN),
1832                                           BOND_SLAVE_NOTIFY_NOW);
1833         } else {
1834                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1835                                           BOND_SLAVE_NOTIFY_NOW);
1836         }
1837
1838         if (new_slave->link != BOND_LINK_DOWN)
1839                 new_slave->last_link_up = jiffies;
1840         slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1841                   new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1842                   (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1843
1844         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1845                 /* if there is a primary slave, remember it */
1846                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1847                         rcu_assign_pointer(bond->primary_slave, new_slave);
1848                         bond->force_primary = true;
1849                 }
1850         }
1851
1852         switch (BOND_MODE(bond)) {
1853         case BOND_MODE_ACTIVEBACKUP:
1854                 bond_set_slave_inactive_flags(new_slave,
1855                                               BOND_SLAVE_NOTIFY_NOW);
1856                 break;
1857         case BOND_MODE_8023AD:
1858                 /* in 802.3ad mode, the internal mechanism
1859                  * will activate the slaves in the selected
1860                  * aggregator
1861                  */
1862                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1863                 /* if this is the first slave */
1864                 if (!prev_slave) {
1865                         SLAVE_AD_INFO(new_slave)->id = 1;
1866                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1867                          * can be called only after the mac address of the bond is set
1868                          */
1869                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1870                 } else {
1871                         SLAVE_AD_INFO(new_slave)->id =
1872                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1873                 }
1874
1875                 bond_3ad_bind_slave(new_slave);
1876                 break;
1877         case BOND_MODE_TLB:
1878         case BOND_MODE_ALB:
1879                 bond_set_active_slave(new_slave);
1880                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1881                 break;
1882         default:
1883                 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
1884
1885                 /* always active in trunk mode */
1886                 bond_set_active_slave(new_slave);
1887
1888                 /* In trunking mode there is little meaning to curr_active_slave
1889                  * anyway (it holds no special properties of the bond device),
1890                  * so we can change it without calling change_active_interface()
1891                  */
1892                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1893                     new_slave->link == BOND_LINK_UP)
1894                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1895
1896                 break;
1897         } /* switch(bond_mode) */
1898
1899 #ifdef CONFIG_NET_POLL_CONTROLLER
1900         if (bond->dev->npinfo) {
1901                 if (slave_enable_netpoll(new_slave)) {
1902                         slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1903                         res = -EBUSY;
1904                         goto err_detach;
1905                 }
1906         }
1907 #endif
1908
1909         if (!(bond_dev->features & NETIF_F_LRO))
1910                 dev_disable_lro(slave_dev);
1911
1912         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1913                                          new_slave);
1914         if (res) {
1915                 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
1916                 goto err_detach;
1917         }
1918
1919         res = bond_master_upper_dev_link(bond, new_slave, extack);
1920         if (res) {
1921                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1922                 goto err_unregister;
1923         }
1924
1925         res = bond_sysfs_slave_add(new_slave);
1926         if (res) {
1927                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1928                 goto err_upper_unlink;
1929         }
1930
1931         /* If the mode uses primary, then the following is handled by
1932          * bond_change_active_slave().
1933          */
1934         if (!bond_uses_primary(bond)) {
1935                 /* set promiscuity level to new slave */
1936                 if (bond_dev->flags & IFF_PROMISC) {
1937                         res = dev_set_promiscuity(slave_dev, 1);
1938                         if (res)
1939                                 goto err_sysfs_del;
1940                 }
1941
1942                 /* set allmulti level to new slave */
1943                 if (bond_dev->flags & IFF_ALLMULTI) {
1944                         res = dev_set_allmulti(slave_dev, 1);
1945                         if (res) {
1946                                 if (bond_dev->flags & IFF_PROMISC)
1947                                         dev_set_promiscuity(slave_dev, -1);
1948                                 goto err_sysfs_del;
1949                         }
1950                 }
1951
1952                 netif_addr_lock_bh(bond_dev);
1953                 dev_mc_sync_multiple(slave_dev, bond_dev);
1954                 dev_uc_sync_multiple(slave_dev, bond_dev);
1955                 netif_addr_unlock_bh(bond_dev);
1956
1957                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1958                         /* add lacpdu mc addr to mc list */
1959                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1960
1961                         dev_mc_add(slave_dev, lacpdu_multicast);
1962                 }
1963         }
1964
1965         bond->slave_cnt++;
1966         bond_compute_features(bond);
1967         bond_set_carrier(bond);
1968
1969         if (bond_uses_primary(bond)) {
1970                 block_netpoll_tx();
1971                 bond_select_active_slave(bond);
1972                 unblock_netpoll_tx();
1973         }
1974
1975         if (bond_mode_can_use_xmit_hash(bond))
1976                 bond_update_slave_arr(bond, NULL);
1977
1978
1979         slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
1980                    bond_is_active_slave(new_slave) ? "an active" : "a backup",
1981                    new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1982
1983         /* enslave is successful */
1984         bond_queue_slave_event(new_slave);
1985         return 0;
1986
1987 /* Undo stages on error */
1988 err_sysfs_del:
1989         bond_sysfs_slave_del(new_slave);
1990
1991 err_upper_unlink:
1992         bond_upper_dev_unlink(bond, new_slave);
1993
1994 err_unregister:
1995         netdev_rx_handler_unregister(slave_dev);
1996
1997 err_detach:
1998         vlan_vids_del_by_dev(slave_dev, bond_dev);
1999         if (rcu_access_pointer(bond->primary_slave) == new_slave)
2000                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2001         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2002                 block_netpoll_tx();
2003                 bond_change_active_slave(bond, NULL);
2004                 bond_select_active_slave(bond);
2005                 unblock_netpoll_tx();
2006         }
2007         /* either primary_slave or curr_active_slave might've changed */
2008         synchronize_rcu();
2009         slave_disable_netpoll(new_slave);
2010
2011 err_close:
2012         if (!netif_is_bond_master(slave_dev))
2013                 slave_dev->priv_flags &= ~IFF_BONDING;
2014         dev_close(slave_dev);
2015
2016 err_restore_mac:
2017         slave_dev->flags &= ~IFF_SLAVE;
2018         if (!bond->params.fail_over_mac ||
2019             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2020                 /* XXX TODO - fom follow mode needs to change master's
2021                  * MAC if this slave's MAC is in use by the bond, or at
2022                  * least print a warning.
2023                  */
2024                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2025                                   new_slave->dev->addr_len);
2026                 ss.ss_family = slave_dev->type;
2027                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2028         }
2029
2030 err_restore_mtu:
2031         dev_set_mtu(slave_dev, new_slave->original_mtu);
2032
2033 err_free:
2034         kobject_put(&new_slave->kobj);
2035
2036 err_undo_flags:
2037         /* Enslave of first slave has failed and we need to fix master's mac */
2038         if (!bond_has_slaves(bond)) {
2039                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2040                                             slave_dev->dev_addr))
2041                         eth_hw_addr_random(bond_dev);
2042                 if (bond_dev->type != ARPHRD_ETHER) {
2043                         dev_close(bond_dev);
2044                         ether_setup(bond_dev);
2045                         bond_dev->flags |= IFF_MASTER;
2046                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2047                 }
2048         }
2049
2050         return res;
2051 }
2052
2053 /* Try to release the slave device <slave> from the bond device <master>
2054  * It is legal to access curr_active_slave without a lock because all the function
2055  * is RTNL-locked. If "all" is true it means that the function is being called
2056  * while destroying a bond interface and all slaves are being released.
2057  *
2058  * The rules for slave state should be:
2059  *   for Active/Backup:
2060  *     Active stays on all backups go down
2061  *   for Bonded connections:
2062  *     The first up interface should be left on and all others downed.
2063  */
2064 static int __bond_release_one(struct net_device *bond_dev,
2065                               struct net_device *slave_dev,
2066                               bool all, bool unregister)
2067 {
2068         struct bonding *bond = netdev_priv(bond_dev);
2069         struct slave *slave, *oldcurrent;
2070         struct sockaddr_storage ss;
2071         int old_flags = bond_dev->flags;
2072         netdev_features_t old_features = bond_dev->features;
2073
2074         /* slave is not a slave or master is not master of this slave */
2075         if (!(slave_dev->flags & IFF_SLAVE) ||
2076             !netdev_has_upper_dev(slave_dev, bond_dev)) {
2077                 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2078                 return -EINVAL;
2079         }
2080
2081         block_netpoll_tx();
2082
2083         slave = bond_get_slave_by_dev(bond, slave_dev);
2084         if (!slave) {
2085                 /* not a slave of this bond */
2086                 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2087                 unblock_netpoll_tx();
2088                 return -EINVAL;
2089         }
2090
2091         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2092
2093         bond_sysfs_slave_del(slave);
2094
2095         /* recompute stats just before removing the slave */
2096         bond_get_stats(bond->dev, &bond->bond_stats);
2097
2098         bond_upper_dev_unlink(bond, slave);
2099         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2100          * for this slave anymore.
2101          */
2102         netdev_rx_handler_unregister(slave_dev);
2103
2104         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2105                 bond_3ad_unbind_slave(slave);
2106
2107         if (bond_mode_can_use_xmit_hash(bond))
2108                 bond_update_slave_arr(bond, slave);
2109
2110         slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2111                     bond_is_active_slave(slave) ? "active" : "backup");
2112
2113         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2114
2115         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2116
2117         if (!all && (!bond->params.fail_over_mac ||
2118                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2119                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2120                     bond_has_slaves(bond))
2121                         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",
2122                                    slave->perm_hwaddr);
2123         }
2124
2125         if (rtnl_dereference(bond->primary_slave) == slave)
2126                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2127
2128         if (oldcurrent == slave)
2129                 bond_change_active_slave(bond, NULL);
2130
2131         if (bond_is_lb(bond)) {
2132                 /* Must be called only after the slave has been
2133                  * detached from the list and the curr_active_slave
2134                  * has been cleared (if our_slave == old_current),
2135                  * but before a new active slave is selected.
2136                  */
2137                 bond_alb_deinit_slave(bond, slave);
2138         }
2139
2140         if (all) {
2141                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2142         } else if (oldcurrent == slave) {
2143                 /* Note that we hold RTNL over this sequence, so there
2144                  * is no concern that another slave add/remove event
2145                  * will interfere.
2146                  */
2147                 bond_select_active_slave(bond);
2148         }
2149
2150         if (!bond_has_slaves(bond)) {
2151                 bond_set_carrier(bond);
2152                 eth_hw_addr_random(bond_dev);
2153         }
2154
2155         unblock_netpoll_tx();
2156         synchronize_rcu();
2157         bond->slave_cnt--;
2158
2159         if (!bond_has_slaves(bond)) {
2160                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2161                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2162         }
2163
2164         bond_compute_features(bond);
2165         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2166             (old_features & NETIF_F_VLAN_CHALLENGED))
2167                 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2168
2169         vlan_vids_del_by_dev(slave_dev, bond_dev);
2170
2171         /* If the mode uses primary, then this case was handled above by
2172          * bond_change_active_slave(..., NULL)
2173          */
2174         if (!bond_uses_primary(bond)) {
2175                 /* unset promiscuity level from slave
2176                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
2177                  * of the IFF_PROMISC flag in the bond_dev, but we need the
2178                  * value of that flag before that change, as that was the value
2179                  * when this slave was attached, so we cache at the start of the
2180                  * function and use it here. Same goes for ALLMULTI below
2181                  */
2182                 if (old_flags & IFF_PROMISC)
2183                         dev_set_promiscuity(slave_dev, -1);
2184
2185                 /* unset allmulti level from slave */
2186                 if (old_flags & IFF_ALLMULTI)
2187                         dev_set_allmulti(slave_dev, -1);
2188
2189                 bond_hw_addr_flush(bond_dev, slave_dev);
2190         }
2191
2192         slave_disable_netpoll(slave);
2193
2194         /* close slave before restoring its mac address */
2195         dev_close(slave_dev);
2196
2197         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2198             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2199                 /* restore original ("permanent") mac address */
2200                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2201                                   slave->dev->addr_len);
2202                 ss.ss_family = slave_dev->type;
2203                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2204         }
2205
2206         if (unregister)
2207                 __dev_set_mtu(slave_dev, slave->original_mtu);
2208         else
2209                 dev_set_mtu(slave_dev, slave->original_mtu);
2210
2211         if (!netif_is_bond_master(slave_dev))
2212                 slave_dev->priv_flags &= ~IFF_BONDING;
2213
2214         kobject_put(&slave->kobj);
2215
2216         return 0;
2217 }
2218
2219 /* A wrapper used because of ndo_del_link */
2220 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2221 {
2222         return __bond_release_one(bond_dev, slave_dev, false, false);
2223 }
2224
2225 /* First release a slave and then destroy the bond if no more slaves are left.
2226  * Must be under rtnl_lock when this function is called.
2227  */
2228 static int bond_release_and_destroy(struct net_device *bond_dev,
2229                                     struct net_device *slave_dev)
2230 {
2231         struct bonding *bond = netdev_priv(bond_dev);
2232         int ret;
2233
2234         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2235         if (ret == 0 && !bond_has_slaves(bond) &&
2236             bond_dev->reg_state != NETREG_UNREGISTERING) {
2237                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2238                 netdev_info(bond_dev, "Destroying bond\n");
2239                 bond_remove_proc_entry(bond);
2240                 unregister_netdevice(bond_dev);
2241         }
2242         return ret;
2243 }
2244
2245 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2246 {
2247         struct bonding *bond = netdev_priv(bond_dev);
2248         bond_fill_ifbond(bond, info);
2249 }
2250
2251 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2252 {
2253         struct bonding *bond = netdev_priv(bond_dev);
2254         struct list_head *iter;
2255         int i = 0, res = -ENODEV;
2256         struct slave *slave;
2257
2258         bond_for_each_slave(bond, slave, iter) {
2259                 if (i++ == (int)info->slave_id) {
2260                         res = 0;
2261                         bond_fill_ifslave(slave, info);
2262                         break;
2263                 }
2264         }
2265
2266         return res;
2267 }
2268
2269 /*-------------------------------- Monitoring -------------------------------*/
2270
2271 /* called with rcu_read_lock() */
2272 static int bond_miimon_inspect(struct bonding *bond)
2273 {
2274         int link_state, commit = 0;
2275         struct list_head *iter;
2276         struct slave *slave;
2277         bool ignore_updelay;
2278
2279         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2280
2281         bond_for_each_slave_rcu(bond, slave, iter) {
2282                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2283
2284                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2285
2286                 switch (slave->link) {
2287                 case BOND_LINK_UP:
2288                         if (link_state)
2289                                 continue;
2290
2291                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2292                         commit++;
2293                         slave->delay = bond->params.downdelay;
2294                         if (slave->delay) {
2295                                 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2296                                            (BOND_MODE(bond) ==
2297                                             BOND_MODE_ACTIVEBACKUP) ?
2298                                             (bond_is_active_slave(slave) ?
2299                                              "active " : "backup ") : "",
2300                                            bond->params.downdelay * bond->params.miimon);
2301                         }
2302                         fallthrough;
2303                 case BOND_LINK_FAIL:
2304                         if (link_state) {
2305                                 /* recovered before downdelay expired */
2306                                 bond_propose_link_state(slave, BOND_LINK_UP);
2307                                 slave->last_link_up = jiffies;
2308                                 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2309                                            (bond->params.downdelay - slave->delay) *
2310                                            bond->params.miimon);
2311                                 commit++;
2312                                 continue;
2313                         }
2314
2315                         if (slave->delay <= 0) {
2316                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2317                                 commit++;
2318                                 continue;
2319                         }
2320
2321                         slave->delay--;
2322                         break;
2323
2324                 case BOND_LINK_DOWN:
2325                         if (!link_state)
2326                                 continue;
2327
2328                         bond_propose_link_state(slave, BOND_LINK_BACK);
2329                         commit++;
2330                         slave->delay = bond->params.updelay;
2331
2332                         if (slave->delay) {
2333                                 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2334                                            ignore_updelay ? 0 :
2335                                            bond->params.updelay *
2336                                            bond->params.miimon);
2337                         }
2338                         fallthrough;
2339                 case BOND_LINK_BACK:
2340                         if (!link_state) {
2341                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2342                                 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2343                                            (bond->params.updelay - slave->delay) *
2344                                            bond->params.miimon);
2345                                 commit++;
2346                                 continue;
2347                         }
2348
2349                         if (ignore_updelay)
2350                                 slave->delay = 0;
2351
2352                         if (slave->delay <= 0) {
2353                                 bond_propose_link_state(slave, BOND_LINK_UP);
2354                                 commit++;
2355                                 ignore_updelay = false;
2356                                 continue;
2357                         }
2358
2359                         slave->delay--;
2360                         break;
2361                 }
2362         }
2363
2364         return commit;
2365 }
2366
2367 static void bond_miimon_link_change(struct bonding *bond,
2368                                     struct slave *slave,
2369                                     char link)
2370 {
2371         switch (BOND_MODE(bond)) {
2372         case BOND_MODE_8023AD:
2373                 bond_3ad_handle_link_change(slave, link);
2374                 break;
2375         case BOND_MODE_TLB:
2376         case BOND_MODE_ALB:
2377                 bond_alb_handle_link_change(bond, slave, link);
2378                 break;
2379         case BOND_MODE_XOR:
2380                 bond_update_slave_arr(bond, NULL);
2381                 break;
2382         }
2383 }
2384
2385 static void bond_miimon_commit(struct bonding *bond)
2386 {
2387         struct list_head *iter;
2388         struct slave *slave, *primary;
2389
2390         bond_for_each_slave(bond, slave, iter) {
2391                 switch (slave->link_new_state) {
2392                 case BOND_LINK_NOCHANGE:
2393                         /* For 802.3ad mode, check current slave speed and
2394                          * duplex again in case its port was disabled after
2395                          * invalid speed/duplex reporting but recovered before
2396                          * link monitoring could make a decision on the actual
2397                          * link status
2398                          */
2399                         if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2400                             slave->link == BOND_LINK_UP)
2401                                 bond_3ad_adapter_speed_duplex_changed(slave);
2402                         continue;
2403
2404                 case BOND_LINK_UP:
2405                         if (bond_update_speed_duplex(slave) &&
2406                             bond_needs_speed_duplex(bond)) {
2407                                 slave->link = BOND_LINK_DOWN;
2408                                 if (net_ratelimit())
2409                                         slave_warn(bond->dev, slave->dev,
2410                                                    "failed to get link speed/duplex\n");
2411                                 continue;
2412                         }
2413                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2414                                                   BOND_SLAVE_NOTIFY_NOW);
2415                         slave->last_link_up = jiffies;
2416
2417                         primary = rtnl_dereference(bond->primary_slave);
2418                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2419                                 /* prevent it from being the active one */
2420                                 bond_set_backup_slave(slave);
2421                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2422                                 /* make it immediately active */
2423                                 bond_set_active_slave(slave);
2424                         }
2425
2426                         slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2427                                    slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2428                                    slave->duplex ? "full" : "half");
2429
2430                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2431
2432                         if (!bond->curr_active_slave || slave == primary)
2433                                 goto do_failover;
2434
2435                         continue;
2436
2437                 case BOND_LINK_DOWN:
2438                         if (slave->link_failure_count < UINT_MAX)
2439                                 slave->link_failure_count++;
2440
2441                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2442                                                   BOND_SLAVE_NOTIFY_NOW);
2443
2444                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2445                             BOND_MODE(bond) == BOND_MODE_8023AD)
2446                                 bond_set_slave_inactive_flags(slave,
2447                                                               BOND_SLAVE_NOTIFY_NOW);
2448
2449                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2450
2451                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2452
2453                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2454                                 goto do_failover;
2455
2456                         continue;
2457
2458                 default:
2459                         slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2460                                   slave->link_new_state);
2461                         bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2462
2463                         continue;
2464                 }
2465
2466 do_failover:
2467                 block_netpoll_tx();
2468                 bond_select_active_slave(bond);
2469                 unblock_netpoll_tx();
2470         }
2471
2472         bond_set_carrier(bond);
2473 }
2474
2475 /* bond_mii_monitor
2476  *
2477  * Really a wrapper that splits the mii monitor into two phases: an
2478  * inspection, then (if inspection indicates something needs to be done)
2479  * an acquisition of appropriate locks followed by a commit phase to
2480  * implement whatever link state changes are indicated.
2481  */
2482 static void bond_mii_monitor(struct work_struct *work)
2483 {
2484         struct bonding *bond = container_of(work, struct bonding,
2485                                             mii_work.work);
2486         bool should_notify_peers = false;
2487         bool commit;
2488         unsigned long delay;
2489         struct slave *slave;
2490         struct list_head *iter;
2491
2492         delay = msecs_to_jiffies(bond->params.miimon);
2493
2494         if (!bond_has_slaves(bond))
2495                 goto re_arm;
2496
2497         rcu_read_lock();
2498         should_notify_peers = bond_should_notify_peers(bond);
2499         commit = !!bond_miimon_inspect(bond);
2500         if (bond->send_peer_notif) {
2501                 rcu_read_unlock();
2502                 if (rtnl_trylock()) {
2503                         bond->send_peer_notif--;
2504                         rtnl_unlock();
2505                 }
2506         } else {
2507                 rcu_read_unlock();
2508         }
2509
2510         if (commit) {
2511                 /* Race avoidance with bond_close cancel of workqueue */
2512                 if (!rtnl_trylock()) {
2513                         delay = 1;
2514                         should_notify_peers = false;
2515                         goto re_arm;
2516                 }
2517
2518                 bond_for_each_slave(bond, slave, iter) {
2519                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2520                 }
2521                 bond_miimon_commit(bond);
2522
2523                 rtnl_unlock();  /* might sleep, hold no other locks */
2524         }
2525
2526 re_arm:
2527         if (bond->params.miimon)
2528                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2529
2530         if (should_notify_peers) {
2531                 if (!rtnl_trylock())
2532                         return;
2533                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2534                 rtnl_unlock();
2535         }
2536 }
2537
2538 static int bond_upper_dev_walk(struct net_device *upper,
2539                                struct netdev_nested_priv *priv)
2540 {
2541         __be32 ip = *(__be32 *)priv->data;
2542
2543         return ip == bond_confirm_addr(upper, 0, ip);
2544 }
2545
2546 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2547 {
2548         struct netdev_nested_priv priv = {
2549                 .data = (void *)&ip,
2550         };
2551         bool ret = false;
2552
2553         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2554                 return true;
2555
2556         rcu_read_lock();
2557         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2558                 ret = true;
2559         rcu_read_unlock();
2560
2561         return ret;
2562 }
2563
2564 /* We go to the (large) trouble of VLAN tagging ARP frames because
2565  * switches in VLAN mode (especially if ports are configured as
2566  * "native" to a VLAN) might not pass non-tagged frames.
2567  */
2568 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2569                           __be32 src_ip, struct bond_vlan_tag *tags)
2570 {
2571         struct sk_buff *skb;
2572         struct bond_vlan_tag *outer_tag = tags;
2573         struct net_device *slave_dev = slave->dev;
2574         struct net_device *bond_dev = slave->bond->dev;
2575
2576         slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2577                   arp_op, &dest_ip, &src_ip);
2578
2579         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2580                          NULL, slave_dev->dev_addr, NULL);
2581
2582         if (!skb) {
2583                 net_err_ratelimited("ARP packet allocation failed\n");
2584                 return;
2585         }
2586
2587         if (!tags || tags->vlan_proto == VLAN_N_VID)
2588                 goto xmit;
2589
2590         tags++;
2591
2592         /* Go through all the tags backwards and add them to the packet */
2593         while (tags->vlan_proto != VLAN_N_VID) {
2594                 if (!tags->vlan_id) {
2595                         tags++;
2596                         continue;
2597                 }
2598
2599                 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2600                           ntohs(outer_tag->vlan_proto), tags->vlan_id);
2601                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2602                                                 tags->vlan_id);
2603                 if (!skb) {
2604                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2605                         return;
2606                 }
2607
2608                 tags++;
2609         }
2610         /* Set the outer tag */
2611         if (outer_tag->vlan_id) {
2612                 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2613                           ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2614                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2615                                        outer_tag->vlan_id);
2616         }
2617
2618 xmit:
2619         arp_xmit(skb);
2620 }
2621
2622 /* Validate the device path between the @start_dev and the @end_dev.
2623  * The path is valid if the @end_dev is reachable through device
2624  * stacking.
2625  * When the path is validated, collect any vlan information in the
2626  * path.
2627  */
2628 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2629                                               struct net_device *end_dev,
2630                                               int level)
2631 {
2632         struct bond_vlan_tag *tags;
2633         struct net_device *upper;
2634         struct list_head  *iter;
2635
2636         if (start_dev == end_dev) {
2637                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2638                 if (!tags)
2639                         return ERR_PTR(-ENOMEM);
2640                 tags[level].vlan_proto = VLAN_N_VID;
2641                 return tags;
2642         }
2643
2644         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2645                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2646                 if (IS_ERR_OR_NULL(tags)) {
2647                         if (IS_ERR(tags))
2648                                 return tags;
2649                         continue;
2650                 }
2651                 if (is_vlan_dev(upper)) {
2652                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2653                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2654                 }
2655
2656                 return tags;
2657         }
2658
2659         return NULL;
2660 }
2661
2662 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2663 {
2664         struct rtable *rt;
2665         struct bond_vlan_tag *tags;
2666         __be32 *targets = bond->params.arp_targets, addr;
2667         int i;
2668
2669         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2670                 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2671                           __func__, &targets[i]);
2672                 tags = NULL;
2673
2674                 /* Find out through which dev should the packet go */
2675                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2676                                      RTO_ONLINK, 0);
2677                 if (IS_ERR(rt)) {
2678                         /* there's no route to target - try to send arp
2679                          * probe to generate any traffic (arp_validate=0)
2680                          */
2681                         if (bond->params.arp_validate)
2682                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2683                                                      bond->dev->name,
2684                                                      &targets[i]);
2685                         bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2686                                       0, tags);
2687                         continue;
2688                 }
2689
2690                 /* bond device itself */
2691                 if (rt->dst.dev == bond->dev)
2692                         goto found;
2693
2694                 rcu_read_lock();
2695                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2696                 rcu_read_unlock();
2697
2698                 if (!IS_ERR_OR_NULL(tags))
2699                         goto found;
2700
2701                 /* Not our device - skip */
2702                 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2703                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2704
2705                 ip_rt_put(rt);
2706                 continue;
2707
2708 found:
2709                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2710                 ip_rt_put(rt);
2711                 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2712                 kfree(tags);
2713         }
2714 }
2715
2716 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2717 {
2718         int i;
2719
2720         if (!sip || !bond_has_this_ip(bond, tip)) {
2721                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2722                            __func__, &sip, &tip);
2723                 return;
2724         }
2725
2726         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2727         if (i == -1) {
2728                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2729                            __func__, &sip);
2730                 return;
2731         }
2732         slave->last_rx = jiffies;
2733         slave->target_last_arp_rx[i] = jiffies;
2734 }
2735
2736 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2737                  struct slave *slave)
2738 {
2739         struct arphdr *arp = (struct arphdr *)skb->data;
2740         struct slave *curr_active_slave, *curr_arp_slave;
2741         unsigned char *arp_ptr;
2742         __be32 sip, tip;
2743         int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2744         unsigned int alen;
2745
2746         if (!slave_do_arp_validate(bond, slave)) {
2747                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2748                     !slave_do_arp_validate_only(bond))
2749                         slave->last_rx = jiffies;
2750                 return RX_HANDLER_ANOTHER;
2751         } else if (!is_arp) {
2752                 return RX_HANDLER_ANOTHER;
2753         }
2754
2755         alen = arp_hdr_len(bond->dev);
2756
2757         slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2758                    __func__, skb->dev->name);
2759
2760         if (alen > skb_headlen(skb)) {
2761                 arp = kmalloc(alen, GFP_ATOMIC);
2762                 if (!arp)
2763                         goto out_unlock;
2764                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2765                         goto out_unlock;
2766         }
2767
2768         if (arp->ar_hln != bond->dev->addr_len ||
2769             skb->pkt_type == PACKET_OTHERHOST ||
2770             skb->pkt_type == PACKET_LOOPBACK ||
2771             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2772             arp->ar_pro != htons(ETH_P_IP) ||
2773             arp->ar_pln != 4)
2774                 goto out_unlock;
2775
2776         arp_ptr = (unsigned char *)(arp + 1);
2777         arp_ptr += bond->dev->addr_len;
2778         memcpy(&sip, arp_ptr, 4);
2779         arp_ptr += 4 + bond->dev->addr_len;
2780         memcpy(&tip, arp_ptr, 4);
2781
2782         slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2783                   __func__, slave->dev->name, bond_slave_state(slave),
2784                   bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2785                   &sip, &tip);
2786
2787         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2788         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2789
2790         /* We 'trust' the received ARP enough to validate it if:
2791          *
2792          * (a) the slave receiving the ARP is active (which includes the
2793          * current ARP slave, if any), or
2794          *
2795          * (b) the receiving slave isn't active, but there is a currently
2796          * active slave and it received valid arp reply(s) after it became
2797          * the currently active slave, or
2798          *
2799          * (c) there is an ARP slave that sent an ARP during the prior ARP
2800          * interval, and we receive an ARP reply on any slave.  We accept
2801          * these because switch FDB update delays may deliver the ARP
2802          * reply to a slave other than the sender of the ARP request.
2803          *
2804          * Note: for (b), backup slaves are receiving the broadcast ARP
2805          * request, not a reply.  This request passes from the sending
2806          * slave through the L2 switch(es) to the receiving slave.  Since
2807          * this is checking the request, sip/tip are swapped for
2808          * validation.
2809          *
2810          * This is done to avoid endless looping when we can't reach the
2811          * arp_ip_target and fool ourselves with our own arp requests.
2812          */
2813         if (bond_is_active_slave(slave))
2814                 bond_validate_arp(bond, slave, sip, tip);
2815         else if (curr_active_slave &&
2816                  time_after(slave_last_rx(bond, curr_active_slave),
2817                             curr_active_slave->last_link_up))
2818                 bond_validate_arp(bond, slave, tip, sip);
2819         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2820                  bond_time_in_interval(bond,
2821                                        dev_trans_start(curr_arp_slave->dev), 1))
2822                 bond_validate_arp(bond, slave, sip, tip);
2823
2824 out_unlock:
2825         if (arp != (struct arphdr *)skb->data)
2826                 kfree(arp);
2827         return RX_HANDLER_ANOTHER;
2828 }
2829
2830 /* function to verify if we're in the arp_interval timeslice, returns true if
2831  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2832  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2833  */
2834 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2835                                   int mod)
2836 {
2837         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2838
2839         return time_in_range(jiffies,
2840                              last_act - delta_in_ticks,
2841                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2842 }
2843
2844 /* This function is called regularly to monitor each slave's link
2845  * ensuring that traffic is being sent and received when arp monitoring
2846  * is used in load-balancing mode. if the adapter has been dormant, then an
2847  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2848  * arp monitoring in active backup mode.
2849  */
2850 static void bond_loadbalance_arp_mon(struct bonding *bond)
2851 {
2852         struct slave *slave, *oldcurrent;
2853         struct list_head *iter;
2854         int do_failover = 0, slave_state_changed = 0;
2855
2856         if (!bond_has_slaves(bond))
2857                 goto re_arm;
2858
2859         rcu_read_lock();
2860
2861         oldcurrent = rcu_dereference(bond->curr_active_slave);
2862         /* see if any of the previous devices are up now (i.e. they have
2863          * xmt and rcv traffic). the curr_active_slave does not come into
2864          * the picture unless it is null. also, slave->last_link_up is not
2865          * needed here because we send an arp on each slave and give a slave
2866          * as long as it needs to get the tx/rx within the delta.
2867          * TODO: what about up/down delay in arp mode? it wasn't here before
2868          *       so it can wait
2869          */
2870         bond_for_each_slave_rcu(bond, slave, iter) {
2871                 unsigned long trans_start = dev_trans_start(slave->dev);
2872
2873                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2874
2875                 if (slave->link != BOND_LINK_UP) {
2876                         if (bond_time_in_interval(bond, trans_start, 1) &&
2877                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2878
2879                                 bond_propose_link_state(slave, BOND_LINK_UP);
2880                                 slave_state_changed = 1;
2881
2882                                 /* primary_slave has no meaning in round-robin
2883                                  * mode. the window of a slave being up and
2884                                  * curr_active_slave being null after enslaving
2885                                  * is closed.
2886                                  */
2887                                 if (!oldcurrent) {
2888                                         slave_info(bond->dev, slave->dev, "link status definitely up\n");
2889                                         do_failover = 1;
2890                                 } else {
2891                                         slave_info(bond->dev, slave->dev, "interface is now up\n");
2892                                 }
2893                         }
2894                 } else {
2895                         /* slave->link == BOND_LINK_UP */
2896
2897                         /* not all switches will respond to an arp request
2898                          * when the source ip is 0, so don't take the link down
2899                          * if we don't know our ip yet
2900                          */
2901                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2902                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2903
2904                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2905                                 slave_state_changed = 1;
2906
2907                                 if (slave->link_failure_count < UINT_MAX)
2908                                         slave->link_failure_count++;
2909
2910                                 slave_info(bond->dev, slave->dev, "interface is now down\n");
2911
2912                                 if (slave == oldcurrent)
2913                                         do_failover = 1;
2914                         }
2915                 }
2916
2917                 /* note: if switch is in round-robin mode, all links
2918                  * must tx arp to ensure all links rx an arp - otherwise
2919                  * links may oscillate or not come up at all; if switch is
2920                  * in something like xor mode, there is nothing we can
2921                  * do - all replies will be rx'ed on same link causing slaves
2922                  * to be unstable during low/no traffic periods
2923                  */
2924                 if (bond_slave_is_up(slave))
2925                         bond_arp_send_all(bond, slave);
2926         }
2927
2928         rcu_read_unlock();
2929
2930         if (do_failover || slave_state_changed) {
2931                 if (!rtnl_trylock())
2932                         goto re_arm;
2933
2934                 bond_for_each_slave(bond, slave, iter) {
2935                         if (slave->link_new_state != BOND_LINK_NOCHANGE)
2936                                 slave->link = slave->link_new_state;
2937                 }
2938
2939                 if (slave_state_changed) {
2940                         bond_slave_state_change(bond);
2941                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2942                                 bond_update_slave_arr(bond, NULL);
2943                 }
2944                 if (do_failover) {
2945                         block_netpoll_tx();
2946                         bond_select_active_slave(bond);
2947                         unblock_netpoll_tx();
2948                 }
2949                 rtnl_unlock();
2950         }
2951
2952 re_arm:
2953         if (bond->params.arp_interval)
2954                 queue_delayed_work(bond->wq, &bond->arp_work,
2955                                    msecs_to_jiffies(bond->params.arp_interval));
2956 }
2957
2958 /* Called to inspect slaves for active-backup mode ARP monitor link state
2959  * changes.  Sets proposed link state in slaves to specify what action
2960  * should take place for the slave.  Returns 0 if no changes are found, >0
2961  * if changes to link states must be committed.
2962  *
2963  * Called with rcu_read_lock held.
2964  */
2965 static int bond_ab_arp_inspect(struct bonding *bond)
2966 {
2967         unsigned long trans_start, last_rx;
2968         struct list_head *iter;
2969         struct slave *slave;
2970         int commit = 0;
2971
2972         bond_for_each_slave_rcu(bond, slave, iter) {
2973                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2974                 last_rx = slave_last_rx(bond, slave);
2975
2976                 if (slave->link != BOND_LINK_UP) {
2977                         if (bond_time_in_interval(bond, last_rx, 1)) {
2978                                 bond_propose_link_state(slave, BOND_LINK_UP);
2979                                 commit++;
2980                         } else if (slave->link == BOND_LINK_BACK) {
2981                                 bond_propose_link_state(slave, BOND_LINK_FAIL);
2982                                 commit++;
2983                         }
2984                         continue;
2985                 }
2986
2987                 /* Give slaves 2*delta after being enslaved or made
2988                  * active.  This avoids bouncing, as the last receive
2989                  * times need a full ARP monitor cycle to be updated.
2990                  */
2991                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2992                         continue;
2993
2994                 /* Backup slave is down if:
2995                  * - No current_arp_slave AND
2996                  * - more than 3*delta since last receive AND
2997                  * - the bond has an IP address
2998                  *
2999                  * Note: a non-null current_arp_slave indicates
3000                  * the curr_active_slave went down and we are
3001                  * searching for a new one; under this condition
3002                  * we only take the curr_active_slave down - this
3003                  * gives each slave a chance to tx/rx traffic
3004                  * before being taken out
3005                  */
3006                 if (!bond_is_active_slave(slave) &&
3007                     !rcu_access_pointer(bond->current_arp_slave) &&
3008                     !bond_time_in_interval(bond, last_rx, 3)) {
3009                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3010                         commit++;
3011                 }
3012
3013                 /* Active slave is down if:
3014                  * - more than 2*delta since transmitting OR
3015                  * - (more than 2*delta since receive AND
3016                  *    the bond has an IP address)
3017                  */
3018                 trans_start = dev_trans_start(slave->dev);
3019                 if (bond_is_active_slave(slave) &&
3020                     (!bond_time_in_interval(bond, trans_start, 2) ||
3021                      !bond_time_in_interval(bond, last_rx, 2))) {
3022                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3023                         commit++;
3024                 }
3025         }
3026
3027         return commit;
3028 }
3029
3030 /* Called to commit link state changes noted by inspection step of
3031  * active-backup mode ARP monitor.
3032  *
3033  * Called with RTNL hold.
3034  */
3035 static void bond_ab_arp_commit(struct bonding *bond)
3036 {
3037         unsigned long trans_start;
3038         struct list_head *iter;
3039         struct slave *slave;
3040
3041         bond_for_each_slave(bond, slave, iter) {
3042                 switch (slave->link_new_state) {
3043                 case BOND_LINK_NOCHANGE:
3044                         continue;
3045
3046                 case BOND_LINK_UP:
3047                         trans_start = dev_trans_start(slave->dev);
3048                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
3049                             (!rtnl_dereference(bond->curr_active_slave) &&
3050                              bond_time_in_interval(bond, trans_start, 1))) {
3051                                 struct slave *current_arp_slave;
3052
3053                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3054                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
3055                                                           BOND_SLAVE_NOTIFY_NOW);
3056                                 if (current_arp_slave) {
3057                                         bond_set_slave_inactive_flags(
3058                                                 current_arp_slave,
3059                                                 BOND_SLAVE_NOTIFY_NOW);
3060                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3061                                 }
3062
3063                                 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3064
3065                                 if (!rtnl_dereference(bond->curr_active_slave) ||
3066                                     slave == rtnl_dereference(bond->primary_slave))
3067                                         goto do_failover;
3068
3069                         }
3070
3071                         continue;
3072
3073                 case BOND_LINK_DOWN:
3074                         if (slave->link_failure_count < UINT_MAX)
3075                                 slave->link_failure_count++;
3076
3077                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3078                                                   BOND_SLAVE_NOTIFY_NOW);
3079                         bond_set_slave_inactive_flags(slave,
3080                                                       BOND_SLAVE_NOTIFY_NOW);
3081
3082                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3083
3084                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
3085                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3086                                 goto do_failover;
3087                         }
3088
3089                         continue;
3090
3091                 case BOND_LINK_FAIL:
3092                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3093                                                   BOND_SLAVE_NOTIFY_NOW);
3094                         bond_set_slave_inactive_flags(slave,
3095                                                       BOND_SLAVE_NOTIFY_NOW);
3096
3097                         /* A slave has just been enslaved and has become
3098                          * the current active slave.
3099                          */
3100                         if (rtnl_dereference(bond->curr_active_slave))
3101                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3102                         continue;
3103
3104                 default:
3105                         slave_err(bond->dev, slave->dev,
3106                                   "impossible: link_new_state %d on slave\n",
3107                                   slave->link_new_state);
3108                         continue;
3109                 }
3110
3111 do_failover:
3112                 block_netpoll_tx();
3113                 bond_select_active_slave(bond);
3114                 unblock_netpoll_tx();
3115         }
3116
3117         bond_set_carrier(bond);
3118 }
3119
3120 /* Send ARP probes for active-backup mode ARP monitor.
3121  *
3122  * Called with rcu_read_lock held.
3123  */
3124 static bool bond_ab_arp_probe(struct bonding *bond)
3125 {
3126         struct slave *slave, *before = NULL, *new_slave = NULL,
3127                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3128                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3129         struct list_head *iter;
3130         bool found = false;
3131         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3132
3133         if (curr_arp_slave && curr_active_slave)
3134                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3135                             curr_arp_slave->dev->name,
3136                             curr_active_slave->dev->name);
3137
3138         if (curr_active_slave) {
3139                 bond_arp_send_all(bond, curr_active_slave);
3140                 return should_notify_rtnl;
3141         }
3142
3143         /* if we don't have a curr_active_slave, search for the next available
3144          * backup slave from the current_arp_slave and make it the candidate
3145          * for becoming the curr_active_slave
3146          */
3147
3148         if (!curr_arp_slave) {
3149                 curr_arp_slave = bond_first_slave_rcu(bond);
3150                 if (!curr_arp_slave)
3151                         return should_notify_rtnl;
3152         }
3153
3154         bond_for_each_slave_rcu(bond, slave, iter) {
3155                 if (!found && !before && bond_slave_is_up(slave))
3156                         before = slave;
3157
3158                 if (found && !new_slave && bond_slave_is_up(slave))
3159                         new_slave = slave;
3160                 /* if the link state is up at this point, we
3161                  * mark it down - this can happen if we have
3162                  * simultaneous link failures and
3163                  * reselect_active_interface doesn't make this
3164                  * one the current slave so it is still marked
3165                  * up when it is actually down
3166                  */
3167                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3168                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3169                                                   BOND_SLAVE_NOTIFY_LATER);
3170                         if (slave->link_failure_count < UINT_MAX)
3171                                 slave->link_failure_count++;
3172
3173                         bond_set_slave_inactive_flags(slave,
3174                                                       BOND_SLAVE_NOTIFY_LATER);
3175
3176                         slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3177                 }
3178                 if (slave == curr_arp_slave)
3179                         found = true;
3180         }
3181
3182         if (!new_slave && before)
3183                 new_slave = before;
3184
3185         if (!new_slave)
3186                 goto check_state;
3187
3188         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3189                                   BOND_SLAVE_NOTIFY_LATER);
3190         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3191         bond_arp_send_all(bond, new_slave);
3192         new_slave->last_link_up = jiffies;
3193         rcu_assign_pointer(bond->current_arp_slave, new_slave);
3194
3195 check_state:
3196         bond_for_each_slave_rcu(bond, slave, iter) {
3197                 if (slave->should_notify || slave->should_notify_link) {
3198                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3199                         break;
3200                 }
3201         }
3202         return should_notify_rtnl;
3203 }
3204
3205 static void bond_activebackup_arp_mon(struct bonding *bond)
3206 {
3207         bool should_notify_peers = false;
3208         bool should_notify_rtnl = false;
3209         int delta_in_ticks;
3210
3211         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3212
3213         if (!bond_has_slaves(bond))
3214                 goto re_arm;
3215
3216         rcu_read_lock();
3217
3218         should_notify_peers = bond_should_notify_peers(bond);
3219
3220         if (bond_ab_arp_inspect(bond)) {
3221                 rcu_read_unlock();
3222
3223                 /* Race avoidance with bond_close flush of workqueue */
3224                 if (!rtnl_trylock()) {
3225                         delta_in_ticks = 1;
3226                         should_notify_peers = false;
3227                         goto re_arm;
3228                 }
3229
3230                 bond_ab_arp_commit(bond);
3231
3232                 rtnl_unlock();
3233                 rcu_read_lock();
3234         }
3235
3236         should_notify_rtnl = bond_ab_arp_probe(bond);
3237         rcu_read_unlock();
3238
3239 re_arm:
3240         if (bond->params.arp_interval)
3241                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3242
3243         if (should_notify_peers || should_notify_rtnl) {
3244                 if (!rtnl_trylock())
3245                         return;
3246
3247                 if (should_notify_peers)
3248                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3249                                                  bond->dev);
3250                 if (should_notify_rtnl) {
3251                         bond_slave_state_notify(bond);
3252                         bond_slave_link_notify(bond);
3253                 }
3254
3255                 rtnl_unlock();
3256         }
3257 }
3258
3259 static void bond_arp_monitor(struct work_struct *work)
3260 {
3261         struct bonding *bond = container_of(work, struct bonding,
3262                                             arp_work.work);
3263
3264         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3265                 bond_activebackup_arp_mon(bond);
3266         else
3267                 bond_loadbalance_arp_mon(bond);
3268 }
3269
3270 /*-------------------------- netdev event handling --------------------------*/
3271
3272 /* Change device name */
3273 static int bond_event_changename(struct bonding *bond)
3274 {
3275         bond_remove_proc_entry(bond);
3276         bond_create_proc_entry(bond);
3277
3278         bond_debug_reregister(bond);
3279
3280         return NOTIFY_DONE;
3281 }
3282
3283 static int bond_master_netdev_event(unsigned long event,
3284                                     struct net_device *bond_dev)
3285 {
3286         struct bonding *event_bond = netdev_priv(bond_dev);
3287
3288         netdev_dbg(bond_dev, "%s called\n", __func__);
3289
3290         switch (event) {
3291         case NETDEV_CHANGENAME:
3292                 return bond_event_changename(event_bond);
3293         case NETDEV_UNREGISTER:
3294                 bond_remove_proc_entry(event_bond);
3295                 break;
3296         case NETDEV_REGISTER:
3297                 bond_create_proc_entry(event_bond);
3298                 break;
3299         default:
3300                 break;
3301         }
3302
3303         return NOTIFY_DONE;
3304 }
3305
3306 static int bond_slave_netdev_event(unsigned long event,
3307                                    struct net_device *slave_dev)
3308 {
3309         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3310         struct bonding *bond;
3311         struct net_device *bond_dev;
3312
3313         /* A netdev event can be generated while enslaving a device
3314          * before netdev_rx_handler_register is called in which case
3315          * slave will be NULL
3316          */
3317         if (!slave) {
3318                 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3319                 return NOTIFY_DONE;
3320         }
3321
3322         bond_dev = slave->bond->dev;
3323         bond = slave->bond;
3324         primary = rtnl_dereference(bond->primary_slave);
3325
3326         slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3327
3328         switch (event) {
3329         case NETDEV_UNREGISTER:
3330                 if (bond_dev->type != ARPHRD_ETHER)
3331                         bond_release_and_destroy(bond_dev, slave_dev);
3332                 else
3333                         __bond_release_one(bond_dev, slave_dev, false, true);
3334                 break;
3335         case NETDEV_UP:
3336         case NETDEV_CHANGE:
3337                 /* For 802.3ad mode only:
3338                  * Getting invalid Speed/Duplex values here will put slave
3339                  * in weird state. Mark it as link-fail if the link was
3340                  * previously up or link-down if it hasn't yet come up, and
3341                  * let link-monitoring (miimon) set it right when correct
3342                  * speeds/duplex are available.
3343                  */
3344                 if (bond_update_speed_duplex(slave) &&
3345                     BOND_MODE(bond) == BOND_MODE_8023AD) {
3346                         if (slave->last_link_up)
3347                                 slave->link = BOND_LINK_FAIL;
3348                         else
3349                                 slave->link = BOND_LINK_DOWN;
3350                 }
3351
3352                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3353                         bond_3ad_adapter_speed_duplex_changed(slave);
3354                 fallthrough;
3355         case NETDEV_DOWN:
3356                 /* Refresh slave-array if applicable!
3357                  * If the setup does not use miimon or arpmon (mode-specific!),
3358                  * then these events will not cause the slave-array to be
3359                  * refreshed. This will cause xmit to use a slave that is not
3360                  * usable. Avoid such situation by refeshing the array at these
3361                  * events. If these (miimon/arpmon) parameters are configured
3362                  * then array gets refreshed twice and that should be fine!
3363                  */
3364                 if (bond_mode_can_use_xmit_hash(bond))
3365                         bond_update_slave_arr(bond, NULL);
3366                 break;
3367         case NETDEV_CHANGEMTU:
3368                 /* TODO: Should slaves be allowed to
3369                  * independently alter their MTU?  For
3370                  * an active-backup bond, slaves need
3371                  * not be the same type of device, so
3372                  * MTUs may vary.  For other modes,
3373                  * slaves arguably should have the
3374                  * same MTUs. To do this, we'd need to
3375                  * take over the slave's change_mtu
3376                  * function for the duration of their
3377                  * servitude.
3378                  */
3379                 break;
3380         case NETDEV_CHANGENAME:
3381                 /* we don't care if we don't have primary set */
3382                 if (!bond_uses_primary(bond) ||
3383                     !bond->params.primary[0])
3384                         break;
3385
3386                 if (slave == primary) {
3387                         /* slave's name changed - he's no longer primary */
3388                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3389                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3390                         /* we have a new primary slave */
3391                         rcu_assign_pointer(bond->primary_slave, slave);
3392                 } else { /* we didn't change primary - exit */
3393                         break;
3394                 }
3395
3396                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3397                             primary ? slave_dev->name : "none");
3398
3399                 block_netpoll_tx();
3400                 bond_select_active_slave(bond);
3401                 unblock_netpoll_tx();
3402                 break;
3403         case NETDEV_FEAT_CHANGE:
3404                 bond_compute_features(bond);
3405                 break;
3406         case NETDEV_RESEND_IGMP:
3407                 /* Propagate to master device */
3408                 call_netdevice_notifiers(event, slave->bond->dev);
3409                 break;
3410         default:
3411                 break;
3412         }
3413
3414         return NOTIFY_DONE;
3415 }
3416
3417 /* bond_netdev_event: handle netdev notifier chain events.
3418  *
3419  * This function receives events for the netdev chain.  The caller (an
3420  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3421  * locks for us to safely manipulate the slave devices (RTNL lock,
3422  * dev_probe_lock).
3423  */
3424 static int bond_netdev_event(struct notifier_block *this,
3425                              unsigned long event, void *ptr)
3426 {
3427         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3428
3429         netdev_dbg(event_dev, "%s received %s\n",
3430                    __func__, netdev_cmd_to_name(event));
3431
3432         if (!(event_dev->priv_flags & IFF_BONDING))
3433                 return NOTIFY_DONE;
3434
3435         if (event_dev->flags & IFF_MASTER) {
3436                 int ret;
3437
3438                 ret = bond_master_netdev_event(event, event_dev);
3439                 if (ret != NOTIFY_DONE)
3440                         return ret;
3441         }
3442
3443         if (event_dev->flags & IFF_SLAVE)
3444                 return bond_slave_netdev_event(event, event_dev);
3445
3446         return NOTIFY_DONE;
3447 }
3448
3449 static struct notifier_block bond_netdev_notifier = {
3450         .notifier_call = bond_netdev_event,
3451 };
3452
3453 /*---------------------------- Hashing Policies -----------------------------*/
3454
3455 /* L2 hash helper */
3456 static inline u32 bond_eth_hash(struct sk_buff *skb)
3457 {
3458         struct ethhdr *ep, hdr_tmp;
3459
3460         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3461         if (ep)
3462                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3463         return 0;
3464 }
3465
3466 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk,
3467                          int *noff, int *proto, bool l34)
3468 {
3469         const struct ipv6hdr *iph6;
3470         const struct iphdr *iph;
3471
3472         if (skb->protocol == htons(ETH_P_IP)) {
3473                 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph))))
3474                         return false;
3475                 iph = (const struct iphdr *)(skb->data + *noff);
3476                 iph_to_flow_copy_v4addrs(fk, iph);
3477                 *noff += iph->ihl << 2;
3478                 if (!ip_is_fragment(iph))
3479                         *proto = iph->protocol;
3480         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3481                 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6))))
3482                         return false;
3483                 iph6 = (const struct ipv6hdr *)(skb->data + *noff);
3484                 iph_to_flow_copy_v6addrs(fk, iph6);
3485                 *noff += sizeof(*iph6);
3486                 *proto = iph6->nexthdr;
3487         } else {
3488                 return false;
3489         }
3490
3491         if (l34 && *proto >= 0)
3492                 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto);
3493
3494         return true;
3495 }
3496
3497 /* Extract the appropriate headers based on bond's xmit policy */
3498 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3499                               struct flow_keys *fk)
3500 {
3501         bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
3502         int noff, proto = -1;
3503
3504         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) {
3505                 memset(fk, 0, sizeof(*fk));
3506                 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3507                                           fk, NULL, 0, 0, 0, 0);
3508         }
3509
3510         fk->ports.ports = 0;
3511         memset(&fk->icmp, 0, sizeof(fk->icmp));
3512         noff = skb_network_offset(skb);
3513         if (!bond_flow_ip(skb, fk, &noff, &proto, l34))
3514                 return false;
3515
3516         /* ICMP error packets contains at least 8 bytes of the header
3517          * of the packet which generated the error. Use this information
3518          * to correlate ICMP error packets within the same flow which
3519          * generated the error.
3520          */
3521         if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
3522                 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data,
3523                                       skb_transport_offset(skb),
3524                                       skb_headlen(skb));
3525                 if (proto == IPPROTO_ICMP) {
3526                         if (!icmp_is_err(fk->icmp.type))
3527                                 return true;
3528
3529                         noff += sizeof(struct icmphdr);
3530                 } else if (proto == IPPROTO_ICMPV6) {
3531                         if (!icmpv6_is_err(fk->icmp.type))
3532                                 return true;
3533
3534                         noff += sizeof(struct icmp6hdr);
3535                 }
3536                 return bond_flow_ip(skb, fk, &noff, &proto, l34);
3537         }
3538
3539         return true;
3540 }
3541
3542 /**
3543  * bond_xmit_hash - generate a hash value based on the xmit policy
3544  * @bond: bonding device
3545  * @skb: buffer to use for headers
3546  *
3547  * This function will extract the necessary headers from the skb buffer and use
3548  * them to generate a hash based on the xmit_policy set in the bonding device
3549  */
3550 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3551 {
3552         struct flow_keys flow;
3553         u32 hash;
3554
3555         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3556             skb->l4_hash)
3557                 return skb->hash;
3558
3559         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3560             !bond_flow_dissect(bond, skb, &flow))
3561                 return bond_eth_hash(skb);
3562
3563         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3564             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
3565                 hash = bond_eth_hash(skb);
3566         } else {
3567                 if (flow.icmp.id)
3568                         memcpy(&hash, &flow.icmp, sizeof(hash));
3569                 else
3570                         memcpy(&hash, &flow.ports.ports, sizeof(hash));
3571         }
3572         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3573                 (__force u32)flow_get_u32_src(&flow);
3574         hash ^= (hash >> 16);
3575         hash ^= (hash >> 8);
3576
3577         return hash >> 1;
3578 }
3579
3580 /*-------------------------- Device entry points ----------------------------*/
3581
3582 void bond_work_init_all(struct bonding *bond)
3583 {
3584         INIT_DELAYED_WORK(&bond->mcast_work,
3585                           bond_resend_igmp_join_requests_delayed);
3586         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3587         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3588         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3589         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3590         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3591 }
3592
3593 static void bond_work_cancel_all(struct bonding *bond)
3594 {
3595         cancel_delayed_work_sync(&bond->mii_work);
3596         cancel_delayed_work_sync(&bond->arp_work);
3597         cancel_delayed_work_sync(&bond->alb_work);
3598         cancel_delayed_work_sync(&bond->ad_work);
3599         cancel_delayed_work_sync(&bond->mcast_work);
3600         cancel_delayed_work_sync(&bond->slave_arr_work);
3601 }
3602
3603 static int bond_open(struct net_device *bond_dev)
3604 {
3605         struct bonding *bond = netdev_priv(bond_dev);
3606         struct list_head *iter;
3607         struct slave *slave;
3608
3609         /* reset slave->backup and slave->inactive */
3610         if (bond_has_slaves(bond)) {
3611                 bond_for_each_slave(bond, slave, iter) {
3612                         if (bond_uses_primary(bond) &&
3613                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3614                                 bond_set_slave_inactive_flags(slave,
3615                                                               BOND_SLAVE_NOTIFY_NOW);
3616                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3617                                 bond_set_slave_active_flags(slave,
3618                                                             BOND_SLAVE_NOTIFY_NOW);
3619                         }
3620                 }
3621         }
3622
3623         if (bond_is_lb(bond)) {
3624                 /* bond_alb_initialize must be called before the timer
3625                  * is started.
3626                  */
3627                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3628                         return -ENOMEM;
3629                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3630                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3631         }
3632
3633         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3634                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3635
3636         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3637                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3638                 bond->recv_probe = bond_arp_rcv;
3639         }
3640
3641         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3642                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3643                 /* register to receive LACPDUs */
3644                 bond->recv_probe = bond_3ad_lacpdu_recv;
3645                 bond_3ad_initiate_agg_selection(bond, 1);
3646         }
3647
3648         if (bond_mode_can_use_xmit_hash(bond))
3649                 bond_update_slave_arr(bond, NULL);
3650
3651         return 0;
3652 }
3653
3654 static int bond_close(struct net_device *bond_dev)
3655 {
3656         struct bonding *bond = netdev_priv(bond_dev);
3657
3658         bond_work_cancel_all(bond);
3659         bond->send_peer_notif = 0;
3660         if (bond_is_lb(bond))
3661                 bond_alb_deinitialize(bond);
3662         bond->recv_probe = NULL;
3663
3664         return 0;
3665 }
3666
3667 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3668  * that some drivers can provide 32bit values only.
3669  */
3670 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3671                             const struct rtnl_link_stats64 *_new,
3672                             const struct rtnl_link_stats64 *_old)
3673 {
3674         const u64 *new = (const u64 *)_new;
3675         const u64 *old = (const u64 *)_old;
3676         u64 *res = (u64 *)_res;
3677         int i;
3678
3679         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3680                 u64 nv = new[i];
3681                 u64 ov = old[i];
3682                 s64 delta = nv - ov;
3683
3684                 /* detects if this particular field is 32bit only */
3685                 if (((nv | ov) >> 32) == 0)
3686                         delta = (s64)(s32)((u32)nv - (u32)ov);
3687
3688                 /* filter anomalies, some drivers reset their stats
3689                  * at down/up events.
3690                  */
3691                 if (delta > 0)
3692                         res[i] += delta;
3693         }
3694 }
3695
3696 #ifdef CONFIG_LOCKDEP
3697 static int bond_get_lowest_level_rcu(struct net_device *dev)
3698 {
3699         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
3700         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
3701         int cur = 0, max = 0;
3702
3703         now = dev;
3704         iter = &dev->adj_list.lower;
3705
3706         while (1) {
3707                 next = NULL;
3708                 while (1) {
3709                         ldev = netdev_next_lower_dev_rcu(now, &iter);
3710                         if (!ldev)
3711                                 break;
3712
3713                         next = ldev;
3714                         niter = &ldev->adj_list.lower;
3715                         dev_stack[cur] = now;
3716                         iter_stack[cur++] = iter;
3717                         if (max <= cur)
3718                                 max = cur;
3719                         break;
3720                 }
3721
3722                 if (!next) {
3723                         if (!cur)
3724                                 return max;
3725                         next = dev_stack[--cur];
3726                         niter = iter_stack[cur];
3727                 }
3728
3729                 now = next;
3730                 iter = niter;
3731         }
3732
3733         return max;
3734 }
3735 #endif
3736
3737 static void bond_get_stats(struct net_device *bond_dev,
3738                            struct rtnl_link_stats64 *stats)
3739 {
3740         struct bonding *bond = netdev_priv(bond_dev);
3741         struct rtnl_link_stats64 temp;
3742         struct list_head *iter;
3743         struct slave *slave;
3744         int nest_level = 0;
3745
3746
3747         rcu_read_lock();
3748 #ifdef CONFIG_LOCKDEP
3749         nest_level = bond_get_lowest_level_rcu(bond_dev);
3750 #endif
3751
3752         spin_lock_nested(&bond->stats_lock, nest_level);
3753         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3754
3755         bond_for_each_slave_rcu(bond, slave, iter) {
3756                 const struct rtnl_link_stats64 *new =
3757                         dev_get_stats(slave->dev, &temp);
3758
3759                 bond_fold_stats(stats, new, &slave->slave_stats);
3760
3761                 /* save off the slave stats for the next run */
3762                 memcpy(&slave->slave_stats, new, sizeof(*new));
3763         }
3764
3765         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3766         spin_unlock(&bond->stats_lock);
3767         rcu_read_unlock();
3768 }
3769
3770 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3771 {
3772         struct bonding *bond = netdev_priv(bond_dev);
3773         struct net_device *slave_dev = NULL;
3774         struct ifbond k_binfo;
3775         struct ifbond __user *u_binfo = NULL;
3776         struct ifslave k_sinfo;
3777         struct ifslave __user *u_sinfo = NULL;
3778         struct mii_ioctl_data *mii = NULL;
3779         struct bond_opt_value newval;
3780         struct net *net;
3781         int res = 0;
3782
3783         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3784
3785         switch (cmd) {
3786         case SIOCGMIIPHY:
3787                 mii = if_mii(ifr);
3788                 if (!mii)
3789                         return -EINVAL;
3790
3791                 mii->phy_id = 0;
3792                 fallthrough;
3793         case SIOCGMIIREG:
3794                 /* We do this again just in case we were called by SIOCGMIIREG
3795                  * instead of SIOCGMIIPHY.
3796                  */
3797                 mii = if_mii(ifr);
3798                 if (!mii)
3799                         return -EINVAL;
3800
3801                 if (mii->reg_num == 1) {
3802                         mii->val_out = 0;
3803                         if (netif_carrier_ok(bond->dev))
3804                                 mii->val_out = BMSR_LSTATUS;
3805                 }
3806
3807                 return 0;
3808         case BOND_INFO_QUERY_OLD:
3809         case SIOCBONDINFOQUERY:
3810                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3811
3812                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3813                         return -EFAULT;
3814
3815                 bond_info_query(bond_dev, &k_binfo);
3816                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3817                         return -EFAULT;
3818
3819                 return 0;
3820         case BOND_SLAVE_INFO_QUERY_OLD:
3821         case SIOCBONDSLAVEINFOQUERY:
3822                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3823
3824                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3825                         return -EFAULT;
3826
3827                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3828                 if (res == 0 &&
3829                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3830                         return -EFAULT;
3831
3832                 return res;
3833         default:
3834                 break;
3835         }
3836
3837         net = dev_net(bond_dev);
3838
3839         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3840                 return -EPERM;
3841
3842         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3843
3844         slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
3845
3846         if (!slave_dev)
3847                 return -ENODEV;
3848
3849         switch (cmd) {
3850         case BOND_ENSLAVE_OLD:
3851         case SIOCBONDENSLAVE:
3852                 res = bond_enslave(bond_dev, slave_dev, NULL);
3853                 break;
3854         case BOND_RELEASE_OLD:
3855         case SIOCBONDRELEASE:
3856                 res = bond_release(bond_dev, slave_dev);
3857                 break;
3858         case BOND_SETHWADDR_OLD:
3859         case SIOCBONDSETHWADDR:
3860                 res = bond_set_dev_addr(bond_dev, slave_dev);
3861                 break;
3862         case BOND_CHANGE_ACTIVE_OLD:
3863         case SIOCBONDCHANGEACTIVE:
3864                 bond_opt_initstr(&newval, slave_dev->name);
3865                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3866                                             &newval);
3867                 break;
3868         default:
3869                 res = -EOPNOTSUPP;
3870         }
3871
3872         return res;
3873 }
3874
3875 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3876 {
3877         struct bonding *bond = netdev_priv(bond_dev);
3878
3879         if (change & IFF_PROMISC)
3880                 bond_set_promiscuity(bond,
3881                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3882
3883         if (change & IFF_ALLMULTI)
3884                 bond_set_allmulti(bond,
3885                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3886 }
3887
3888 static void bond_set_rx_mode(struct net_device *bond_dev)
3889 {
3890         struct bonding *bond = netdev_priv(bond_dev);
3891         struct list_head *iter;
3892         struct slave *slave;
3893
3894         rcu_read_lock();
3895         if (bond_uses_primary(bond)) {
3896                 slave = rcu_dereference(bond->curr_active_slave);
3897                 if (slave) {
3898                         dev_uc_sync(slave->dev, bond_dev);
3899                         dev_mc_sync(slave->dev, bond_dev);
3900                 }
3901         } else {
3902                 bond_for_each_slave_rcu(bond, slave, iter) {
3903                         dev_uc_sync_multiple(slave->dev, bond_dev);
3904                         dev_mc_sync_multiple(slave->dev, bond_dev);
3905                 }
3906         }
3907         rcu_read_unlock();
3908 }
3909
3910 static int bond_neigh_init(struct neighbour *n)
3911 {
3912         struct bonding *bond = netdev_priv(n->dev);
3913         const struct net_device_ops *slave_ops;
3914         struct neigh_parms parms;
3915         struct slave *slave;
3916         int ret = 0;
3917
3918         rcu_read_lock();
3919         slave = bond_first_slave_rcu(bond);
3920         if (!slave)
3921                 goto out;
3922         slave_ops = slave->dev->netdev_ops;
3923         if (!slave_ops->ndo_neigh_setup)
3924                 goto out;
3925
3926         /* TODO: find another way [1] to implement this.
3927          * Passing a zeroed structure is fragile,
3928          * but at least we do not pass garbage.
3929          *
3930          * [1] One way would be that ndo_neigh_setup() never touch
3931          *     struct neigh_parms, but propagate the new neigh_setup()
3932          *     back to ___neigh_create() / neigh_parms_alloc()
3933          */
3934         memset(&parms, 0, sizeof(parms));
3935         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3936
3937         if (ret)
3938                 goto out;
3939
3940         if (parms.neigh_setup)
3941                 ret = parms.neigh_setup(n);
3942 out:
3943         rcu_read_unlock();
3944         return ret;
3945 }
3946
3947 /* The bonding ndo_neigh_setup is called at init time beofre any
3948  * slave exists. So we must declare proxy setup function which will
3949  * be used at run time to resolve the actual slave neigh param setup.
3950  *
3951  * It's also called by master devices (such as vlans) to setup their
3952  * underlying devices. In that case - do nothing, we're already set up from
3953  * our init.
3954  */
3955 static int bond_neigh_setup(struct net_device *dev,
3956                             struct neigh_parms *parms)
3957 {
3958         /* modify only our neigh_parms */
3959         if (parms->dev == dev)
3960                 parms->neigh_setup = bond_neigh_init;
3961
3962         return 0;
3963 }
3964
3965 /* Change the MTU of all of a master's slaves to match the master */
3966 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3967 {
3968         struct bonding *bond = netdev_priv(bond_dev);
3969         struct slave *slave, *rollback_slave;
3970         struct list_head *iter;
3971         int res = 0;
3972
3973         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3974
3975         bond_for_each_slave(bond, slave, iter) {
3976                 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
3977                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3978
3979                 res = dev_set_mtu(slave->dev, new_mtu);
3980
3981                 if (res) {
3982                         /* If we failed to set the slave's mtu to the new value
3983                          * we must abort the operation even in ACTIVE_BACKUP
3984                          * mode, because if we allow the backup slaves to have
3985                          * different mtu values than the active slave we'll
3986                          * need to change their mtu when doing a failover. That
3987                          * means changing their mtu from timer context, which
3988                          * is probably not a good idea.
3989                          */
3990                         slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
3991                                   res, new_mtu);
3992                         goto unwind;
3993                 }
3994         }
3995
3996         bond_dev->mtu = new_mtu;
3997
3998         return 0;
3999
4000 unwind:
4001         /* unwind from head to the slave that failed */
4002         bond_for_each_slave(bond, rollback_slave, iter) {
4003                 int tmp_res;
4004
4005                 if (rollback_slave == slave)
4006                         break;
4007
4008                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4009                 if (tmp_res)
4010                         slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4011                                   tmp_res);
4012         }
4013
4014         return res;
4015 }
4016
4017 /* Change HW address
4018  *
4019  * Note that many devices must be down to change the HW address, and
4020  * downing the master releases all slaves.  We can make bonds full of
4021  * bonding devices to test this, however.
4022  */
4023 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4024 {
4025         struct bonding *bond = netdev_priv(bond_dev);
4026         struct slave *slave, *rollback_slave;
4027         struct sockaddr_storage *ss = addr, tmp_ss;
4028         struct list_head *iter;
4029         int res = 0;
4030
4031         if (BOND_MODE(bond) == BOND_MODE_ALB)
4032                 return bond_alb_set_mac_address(bond_dev, addr);
4033
4034
4035         netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4036
4037         /* If fail_over_mac is enabled, do nothing and return success.
4038          * Returning an error causes ifenslave to fail.
4039          */
4040         if (bond->params.fail_over_mac &&
4041             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4042                 return 0;
4043
4044         if (!is_valid_ether_addr(ss->__data))
4045                 return -EADDRNOTAVAIL;
4046
4047         bond_for_each_slave(bond, slave, iter) {
4048                 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4049                           __func__, slave);
4050                 res = dev_set_mac_address(slave->dev, addr, NULL);
4051                 if (res) {
4052                         /* TODO: consider downing the slave
4053                          * and retry ?
4054                          * User should expect communications
4055                          * breakage anyway until ARP finish
4056                          * updating, so...
4057                          */
4058                         slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4059                                   __func__, res);
4060                         goto unwind;
4061                 }
4062         }
4063
4064         /* success */
4065         memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
4066         return 0;
4067
4068 unwind:
4069         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4070         tmp_ss.ss_family = bond_dev->type;
4071
4072         /* unwind from head to the slave that failed */
4073         bond_for_each_slave(bond, rollback_slave, iter) {
4074                 int tmp_res;
4075
4076                 if (rollback_slave == slave)
4077                         break;
4078
4079                 tmp_res = dev_set_mac_address(rollback_slave->dev,
4080                                               (struct sockaddr *)&tmp_ss, NULL);
4081                 if (tmp_res) {
4082                         slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4083                                    __func__, tmp_res);
4084                 }
4085         }
4086
4087         return res;
4088 }
4089
4090 /**
4091  * bond_get_slave_by_id - get xmit slave with slave_id
4092  * @bond: bonding device that is transmitting
4093  * @slave_id: slave id up to slave_cnt-1 through which to transmit
4094  *
4095  * This function tries to get slave with slave_id but in case
4096  * it fails, it tries to find the first available slave for transmission.
4097  */
4098 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4099                                           int slave_id)
4100 {
4101         struct list_head *iter;
4102         struct slave *slave;
4103         int i = slave_id;
4104
4105         /* Here we start from the slave with slave_id */
4106         bond_for_each_slave_rcu(bond, slave, iter) {
4107                 if (--i < 0) {
4108                         if (bond_slave_can_tx(slave))
4109                                 return slave;
4110                 }
4111         }
4112
4113         /* Here we start from the first slave up to slave_id */
4114         i = slave_id;
4115         bond_for_each_slave_rcu(bond, slave, iter) {
4116                 if (--i < 0)
4117                         break;
4118                 if (bond_slave_can_tx(slave))
4119                         return slave;
4120         }
4121         /* no slave that can tx has been found */
4122         return NULL;
4123 }
4124
4125 /**
4126  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4127  * @bond: bonding device to use
4128  *
4129  * Based on the value of the bonding device's packets_per_slave parameter
4130  * this function generates a slave id, which is usually used as the next
4131  * slave to transmit through.
4132  */
4133 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4134 {
4135         u32 slave_id;
4136         struct reciprocal_value reciprocal_packets_per_slave;
4137         int packets_per_slave = bond->params.packets_per_slave;
4138
4139         switch (packets_per_slave) {
4140         case 0:
4141                 slave_id = prandom_u32();
4142                 break;
4143         case 1:
4144                 slave_id = bond->rr_tx_counter;
4145                 break;
4146         default:
4147                 reciprocal_packets_per_slave =
4148                         bond->params.reciprocal_packets_per_slave;
4149                 slave_id = reciprocal_divide(bond->rr_tx_counter,
4150                                              reciprocal_packets_per_slave);
4151                 break;
4152         }
4153         bond->rr_tx_counter++;
4154
4155         return slave_id;
4156 }
4157
4158 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4159                                                     struct sk_buff *skb)
4160 {
4161         struct slave *slave;
4162         int slave_cnt;
4163         u32 slave_id;
4164
4165         /* Start with the curr_active_slave that joined the bond as the
4166          * default for sending IGMP traffic.  For failover purposes one
4167          * needs to maintain some consistency for the interface that will
4168          * send the join/membership reports.  The curr_active_slave found
4169          * will send all of this type of traffic.
4170          */
4171         if (skb->protocol == htons(ETH_P_IP)) {
4172                 int noff = skb_network_offset(skb);
4173                 struct iphdr *iph;
4174
4175                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4176                         goto non_igmp;
4177
4178                 iph = ip_hdr(skb);
4179                 if (iph->protocol == IPPROTO_IGMP) {
4180                         slave = rcu_dereference(bond->curr_active_slave);
4181                         if (slave)
4182                                 return slave;
4183                         return bond_get_slave_by_id(bond, 0);
4184                 }
4185         }
4186
4187 non_igmp:
4188         slave_cnt = READ_ONCE(bond->slave_cnt);
4189         if (likely(slave_cnt)) {
4190                 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4191                 return bond_get_slave_by_id(bond, slave_id);
4192         }
4193         return NULL;
4194 }
4195
4196 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4197                                         struct net_device *bond_dev)
4198 {
4199         struct bonding *bond = netdev_priv(bond_dev);
4200         struct slave *slave;
4201
4202         slave = bond_xmit_roundrobin_slave_get(bond, skb);
4203         if (likely(slave))
4204                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4205
4206         return bond_tx_drop(bond_dev, skb);
4207 }
4208
4209 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond,
4210                                                       struct sk_buff *skb)
4211 {
4212         return rcu_dereference(bond->curr_active_slave);
4213 }
4214
4215 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4216  * the bond has a usable interface.
4217  */
4218 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4219                                           struct net_device *bond_dev)
4220 {
4221         struct bonding *bond = netdev_priv(bond_dev);
4222         struct slave *slave;
4223
4224         slave = bond_xmit_activebackup_slave_get(bond, skb);
4225         if (slave)
4226                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4227
4228         return bond_tx_drop(bond_dev, skb);
4229 }
4230
4231 /* Use this to update slave_array when (a) it's not appropriate to update
4232  * slave_array right away (note that update_slave_array() may sleep)
4233  * and / or (b) RTNL is not held.
4234  */
4235 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4236 {
4237         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4238 }
4239
4240 /* Slave array work handler. Holds only RTNL */
4241 static void bond_slave_arr_handler(struct work_struct *work)
4242 {
4243         struct bonding *bond = container_of(work, struct bonding,
4244                                             slave_arr_work.work);
4245         int ret;
4246
4247         if (!rtnl_trylock())
4248                 goto err;
4249
4250         ret = bond_update_slave_arr(bond, NULL);
4251         rtnl_unlock();
4252         if (ret) {
4253                 pr_warn_ratelimited("Failed to update slave array from WT\n");
4254                 goto err;
4255         }
4256         return;
4257
4258 err:
4259         bond_slave_arr_work_rearm(bond, 1);
4260 }
4261
4262 static void bond_skip_slave(struct bond_up_slave *slaves,
4263                             struct slave *skipslave)
4264 {
4265         int idx;
4266
4267         /* Rare situation where caller has asked to skip a specific
4268          * slave but allocation failed (most likely!). BTW this is
4269          * only possible when the call is initiated from
4270          * __bond_release_one(). In this situation; overwrite the
4271          * skipslave entry in the array with the last entry from the
4272          * array to avoid a situation where the xmit path may choose
4273          * this to-be-skipped slave to send a packet out.
4274          */
4275         for (idx = 0; slaves && idx < slaves->count; idx++) {
4276                 if (skipslave == slaves->arr[idx]) {
4277                         slaves->arr[idx] =
4278                                 slaves->arr[slaves->count - 1];
4279                         slaves->count--;
4280                         break;
4281                 }
4282         }
4283 }
4284
4285 static void bond_set_slave_arr(struct bonding *bond,
4286                                struct bond_up_slave *usable_slaves,
4287                                struct bond_up_slave *all_slaves)
4288 {
4289         struct bond_up_slave *usable, *all;
4290
4291         usable = rtnl_dereference(bond->usable_slaves);
4292         rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4293         kfree_rcu(usable, rcu);
4294
4295         all = rtnl_dereference(bond->all_slaves);
4296         rcu_assign_pointer(bond->all_slaves, all_slaves);
4297         kfree_rcu(all, rcu);
4298 }
4299
4300 static void bond_reset_slave_arr(struct bonding *bond)
4301 {
4302         struct bond_up_slave *usable, *all;
4303
4304         usable = rtnl_dereference(bond->usable_slaves);
4305         if (usable) {
4306                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4307                 kfree_rcu(usable, rcu);
4308         }
4309
4310         all = rtnl_dereference(bond->all_slaves);
4311         if (all) {
4312                 RCU_INIT_POINTER(bond->all_slaves, NULL);
4313                 kfree_rcu(all, rcu);
4314         }
4315 }
4316
4317 /* Build the usable slaves array in control path for modes that use xmit-hash
4318  * to determine the slave interface -
4319  * (a) BOND_MODE_8023AD
4320  * (b) BOND_MODE_XOR
4321  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4322  *
4323  * The caller is expected to hold RTNL only and NO other lock!
4324  */
4325 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4326 {
4327         struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
4328         struct slave *slave;
4329         struct list_head *iter;
4330         int agg_id = 0;
4331         int ret = 0;
4332
4333 #ifdef CONFIG_LOCKDEP
4334         WARN_ON(lockdep_is_held(&bond->mode_lock));
4335 #endif
4336
4337         usable_slaves = kzalloc(struct_size(usable_slaves, arr,
4338                                             bond->slave_cnt), GFP_KERNEL);
4339         all_slaves = kzalloc(struct_size(all_slaves, arr,
4340                                          bond->slave_cnt), GFP_KERNEL);
4341         if (!usable_slaves || !all_slaves) {
4342                 ret = -ENOMEM;
4343                 goto out;
4344         }
4345         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4346                 struct ad_info ad_info;
4347
4348                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4349                         pr_debug("bond_3ad_get_active_agg_info failed\n");
4350                         /* No active aggragator means it's not safe to use
4351                          * the previous array.
4352                          */
4353                         bond_reset_slave_arr(bond);
4354                         goto out;
4355                 }
4356                 agg_id = ad_info.aggregator_id;
4357         }
4358         bond_for_each_slave(bond, slave, iter) {
4359                 if (skipslave == slave)
4360                         continue;
4361
4362                 all_slaves->arr[all_slaves->count++] = slave;
4363                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4364                         struct aggregator *agg;
4365
4366                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
4367                         if (!agg || agg->aggregator_identifier != agg_id)
4368                                 continue;
4369                 }
4370                 if (!bond_slave_can_tx(slave))
4371                         continue;
4372
4373                 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4374                           usable_slaves->count);
4375
4376                 usable_slaves->arr[usable_slaves->count++] = slave;
4377         }
4378
4379         bond_set_slave_arr(bond, usable_slaves, all_slaves);
4380         return ret;
4381 out:
4382         if (ret != 0 && skipslave) {
4383                 bond_skip_slave(rtnl_dereference(bond->all_slaves),
4384                                 skipslave);
4385                 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
4386                                 skipslave);
4387         }
4388         kfree_rcu(all_slaves, rcu);
4389         kfree_rcu(usable_slaves, rcu);
4390
4391         return ret;
4392 }
4393
4394 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
4395                                                  struct sk_buff *skb,
4396                                                  struct bond_up_slave *slaves)
4397 {
4398         struct slave *slave;
4399         unsigned int count;
4400         u32 hash;
4401
4402         hash = bond_xmit_hash(bond, skb);
4403         count = slaves ? READ_ONCE(slaves->count) : 0;
4404         if (unlikely(!count))
4405                 return NULL;
4406
4407         slave = slaves->arr[hash % count];
4408         return slave;
4409 }
4410
4411 /* Use this Xmit function for 3AD as well as XOR modes. The current
4412  * usable slave array is formed in the control path. The xmit function
4413  * just calculates hash and sends the packet out.
4414  */
4415 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4416                                      struct net_device *dev)
4417 {
4418         struct bonding *bond = netdev_priv(dev);
4419         struct bond_up_slave *slaves;
4420         struct slave *slave;
4421
4422         slaves = rcu_dereference(bond->usable_slaves);
4423         slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4424         if (likely(slave))
4425                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4426
4427         return bond_tx_drop(dev, skb);
4428 }
4429
4430 /* in broadcast mode, we send everything to all usable interfaces. */
4431 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4432                                        struct net_device *bond_dev)
4433 {
4434         struct bonding *bond = netdev_priv(bond_dev);
4435         struct slave *slave = NULL;
4436         struct list_head *iter;
4437
4438         bond_for_each_slave_rcu(bond, slave, iter) {
4439                 if (bond_is_last_slave(bond, slave))
4440                         break;
4441                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4442                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4443
4444                         if (!skb2) {
4445                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4446                                                     bond_dev->name, __func__);
4447                                 continue;
4448                         }
4449                         bond_dev_queue_xmit(bond, skb2, slave->dev);
4450                 }
4451         }
4452         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4453                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4454
4455         return bond_tx_drop(bond_dev, skb);
4456 }
4457
4458 /*------------------------- Device initialization ---------------------------*/
4459
4460 /* Lookup the slave that corresponds to a qid */
4461 static inline int bond_slave_override(struct bonding *bond,
4462                                       struct sk_buff *skb)
4463 {
4464         struct slave *slave = NULL;
4465         struct list_head *iter;
4466
4467         if (!skb_rx_queue_recorded(skb))
4468                 return 1;
4469
4470         /* Find out if any slaves have the same mapping as this skb. */
4471         bond_for_each_slave_rcu(bond, slave, iter) {
4472                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4473                         if (bond_slave_is_up(slave) &&
4474                             slave->link == BOND_LINK_UP) {
4475                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4476                                 return 0;
4477                         }
4478                         /* If the slave isn't UP, use default transmit policy. */
4479                         break;
4480                 }
4481         }
4482
4483         return 1;
4484 }
4485
4486
4487 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4488                              struct net_device *sb_dev)
4489 {
4490         /* This helper function exists to help dev_pick_tx get the correct
4491          * destination queue.  Using a helper function skips a call to
4492          * skb_tx_hash and will put the skbs in the queue we expect on their
4493          * way down to the bonding driver.
4494          */
4495         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4496
4497         /* Save the original txq to restore before passing to the driver */
4498         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4499
4500         if (unlikely(txq >= dev->real_num_tx_queues)) {
4501                 do {
4502                         txq -= dev->real_num_tx_queues;
4503                 } while (txq >= dev->real_num_tx_queues);
4504         }
4505         return txq;
4506 }
4507
4508 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
4509                                               struct sk_buff *skb,
4510                                               bool all_slaves)
4511 {
4512         struct bonding *bond = netdev_priv(master_dev);
4513         struct bond_up_slave *slaves;
4514         struct slave *slave = NULL;
4515
4516         switch (BOND_MODE(bond)) {
4517         case BOND_MODE_ROUNDROBIN:
4518                 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4519                 break;
4520         case BOND_MODE_ACTIVEBACKUP:
4521                 slave = bond_xmit_activebackup_slave_get(bond, skb);
4522                 break;
4523         case BOND_MODE_8023AD:
4524         case BOND_MODE_XOR:
4525                 if (all_slaves)
4526                         slaves = rcu_dereference(bond->all_slaves);
4527                 else
4528                         slaves = rcu_dereference(bond->usable_slaves);
4529                 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4530                 break;
4531         case BOND_MODE_BROADCAST:
4532                 break;
4533         case BOND_MODE_ALB:
4534                 slave = bond_xmit_alb_slave_get(bond, skb);
4535                 break;
4536         case BOND_MODE_TLB:
4537                 slave = bond_xmit_tlb_slave_get(bond, skb);
4538                 break;
4539         default:
4540                 /* Should never happen, mode already checked */
4541                 WARN_ONCE(true, "Unknown bonding mode");
4542                 break;
4543         }
4544
4545         if (slave)
4546                 return slave->dev;
4547         return NULL;
4548 }
4549
4550 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4551 {
4552         struct bonding *bond = netdev_priv(dev);
4553
4554         if (bond_should_override_tx_queue(bond) &&
4555             !bond_slave_override(bond, skb))
4556                 return NETDEV_TX_OK;
4557
4558         switch (BOND_MODE(bond)) {
4559         case BOND_MODE_ROUNDROBIN:
4560                 return bond_xmit_roundrobin(skb, dev);
4561         case BOND_MODE_ACTIVEBACKUP:
4562                 return bond_xmit_activebackup(skb, dev);
4563         case BOND_MODE_8023AD:
4564         case BOND_MODE_XOR:
4565                 return bond_3ad_xor_xmit(skb, dev);
4566         case BOND_MODE_BROADCAST:
4567                 return bond_xmit_broadcast(skb, dev);
4568         case BOND_MODE_ALB:
4569                 return bond_alb_xmit(skb, dev);
4570         case BOND_MODE_TLB:
4571                 return bond_tlb_xmit(skb, dev);
4572         default:
4573                 /* Should never happen, mode already checked */
4574                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4575                 WARN_ON_ONCE(1);
4576                 return bond_tx_drop(dev, skb);
4577         }
4578 }
4579
4580 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4581 {
4582         struct bonding *bond = netdev_priv(dev);
4583         netdev_tx_t ret = NETDEV_TX_OK;
4584
4585         /* If we risk deadlock from transmitting this in the
4586          * netpoll path, tell netpoll to queue the frame for later tx
4587          */
4588         if (unlikely(is_netpoll_tx_blocked(dev)))
4589                 return NETDEV_TX_BUSY;
4590
4591         rcu_read_lock();
4592         if (bond_has_slaves(bond))
4593                 ret = __bond_start_xmit(skb, dev);
4594         else
4595                 ret = bond_tx_drop(dev, skb);
4596         rcu_read_unlock();
4597
4598         return ret;
4599 }
4600
4601 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
4602 {
4603         if (speed == 0 || speed == SPEED_UNKNOWN)
4604                 speed = slave->speed;
4605         else
4606                 speed = min(speed, slave->speed);
4607
4608         return speed;
4609 }
4610
4611 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4612                                            struct ethtool_link_ksettings *cmd)
4613 {
4614         struct bonding *bond = netdev_priv(bond_dev);
4615         struct list_head *iter;
4616         struct slave *slave;
4617         u32 speed = 0;
4618
4619         cmd->base.duplex = DUPLEX_UNKNOWN;
4620         cmd->base.port = PORT_OTHER;
4621
4622         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4623          * do not need to check mode.  Though link speed might not represent
4624          * the true receive or transmit bandwidth (not all modes are symmetric)
4625          * this is an accurate maximum.
4626          */
4627         bond_for_each_slave(bond, slave, iter) {
4628                 if (bond_slave_can_tx(slave)) {
4629                         if (slave->speed != SPEED_UNKNOWN) {
4630                                 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
4631                                         speed = bond_mode_bcast_speed(slave,
4632                                                                       speed);
4633                                 else
4634                                         speed += slave->speed;
4635                         }
4636                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4637                             slave->duplex != DUPLEX_UNKNOWN)
4638                                 cmd->base.duplex = slave->duplex;
4639                 }
4640         }
4641         cmd->base.speed = speed ? : SPEED_UNKNOWN;
4642
4643         return 0;
4644 }
4645
4646 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4647                                      struct ethtool_drvinfo *drvinfo)
4648 {
4649         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4650         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4651                  BOND_ABI_VERSION);
4652 }
4653
4654 static const struct ethtool_ops bond_ethtool_ops = {
4655         .get_drvinfo            = bond_ethtool_get_drvinfo,
4656         .get_link               = ethtool_op_get_link,
4657         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
4658 };
4659
4660 static const struct net_device_ops bond_netdev_ops = {
4661         .ndo_init               = bond_init,
4662         .ndo_uninit             = bond_uninit,
4663         .ndo_open               = bond_open,
4664         .ndo_stop               = bond_close,
4665         .ndo_start_xmit         = bond_start_xmit,
4666         .ndo_select_queue       = bond_select_queue,
4667         .ndo_get_stats64        = bond_get_stats,
4668         .ndo_do_ioctl           = bond_do_ioctl,
4669         .ndo_change_rx_flags    = bond_change_rx_flags,
4670         .ndo_set_rx_mode        = bond_set_rx_mode,
4671         .ndo_change_mtu         = bond_change_mtu,
4672         .ndo_set_mac_address    = bond_set_mac_address,
4673         .ndo_neigh_setup        = bond_neigh_setup,
4674         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4675         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4676 #ifdef CONFIG_NET_POLL_CONTROLLER
4677         .ndo_netpoll_setup      = bond_netpoll_setup,
4678         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4679         .ndo_poll_controller    = bond_poll_controller,
4680 #endif
4681         .ndo_add_slave          = bond_enslave,
4682         .ndo_del_slave          = bond_release,
4683         .ndo_fix_features       = bond_fix_features,
4684         .ndo_features_check     = passthru_features_check,
4685         .ndo_get_xmit_slave     = bond_xmit_get_slave,
4686 };
4687
4688 static const struct device_type bond_type = {
4689         .name = "bond",
4690 };
4691
4692 static void bond_destructor(struct net_device *bond_dev)
4693 {
4694         struct bonding *bond = netdev_priv(bond_dev);
4695         if (bond->wq)
4696                 destroy_workqueue(bond->wq);
4697 }
4698
4699 void bond_setup(struct net_device *bond_dev)
4700 {
4701         struct bonding *bond = netdev_priv(bond_dev);
4702
4703         spin_lock_init(&bond->mode_lock);
4704         bond->params = bonding_defaults;
4705
4706         /* Initialize pointers */
4707         bond->dev = bond_dev;
4708
4709         /* Initialize the device entry points */
4710         ether_setup(bond_dev);
4711         bond_dev->max_mtu = ETH_MAX_MTU;
4712         bond_dev->netdev_ops = &bond_netdev_ops;
4713         bond_dev->ethtool_ops = &bond_ethtool_ops;
4714
4715         bond_dev->needs_free_netdev = true;
4716         bond_dev->priv_destructor = bond_destructor;
4717
4718         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4719
4720         /* Initialize the device options */
4721         bond_dev->flags |= IFF_MASTER;
4722         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4723         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4724
4725 #ifdef CONFIG_XFRM_OFFLOAD
4726         /* set up xfrm device ops (only supported in active-backup right now) */
4727         bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
4728         bond->xs = NULL;
4729 #endif /* CONFIG_XFRM_OFFLOAD */
4730
4731         /* don't acquire bond device's netif_tx_lock when transmitting */
4732         bond_dev->features |= NETIF_F_LLTX;
4733
4734         /* By default, we declare the bond to be fully
4735          * VLAN hardware accelerated capable. Special
4736          * care is taken in the various xmit functions
4737          * when there are slaves that are not hw accel
4738          * capable
4739          */
4740
4741         /* Don't allow bond devices to change network namespaces. */
4742         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4743
4744         bond_dev->hw_features = BOND_VLAN_FEATURES |
4745                                 NETIF_F_HW_VLAN_CTAG_RX |
4746                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4747
4748         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4749         bond_dev->features |= bond_dev->hw_features;
4750         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4751 #ifdef CONFIG_XFRM_OFFLOAD
4752         bond_dev->hw_features |= BOND_XFRM_FEATURES;
4753         /* Only enable XFRM features if this is an active-backup config */
4754         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4755                 bond_dev->features |= BOND_XFRM_FEATURES;
4756 #endif /* CONFIG_XFRM_OFFLOAD */
4757 }
4758
4759 /* Destroy a bonding device.
4760  * Must be under rtnl_lock when this function is called.
4761  */
4762 static void bond_uninit(struct net_device *bond_dev)
4763 {
4764         struct bonding *bond = netdev_priv(bond_dev);
4765         struct bond_up_slave *usable, *all;
4766         struct list_head *iter;
4767         struct slave *slave;
4768
4769         bond_netpoll_cleanup(bond_dev);
4770
4771         /* Release the bonded slaves */
4772         bond_for_each_slave(bond, slave, iter)
4773                 __bond_release_one(bond_dev, slave->dev, true, true);
4774         netdev_info(bond_dev, "Released all slaves\n");
4775
4776         usable = rtnl_dereference(bond->usable_slaves);
4777         if (usable) {
4778                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4779                 kfree_rcu(usable, rcu);
4780         }
4781
4782         all = rtnl_dereference(bond->all_slaves);
4783         if (all) {
4784                 RCU_INIT_POINTER(bond->all_slaves, NULL);
4785                 kfree_rcu(all, rcu);
4786         }
4787
4788         list_del(&bond->bond_list);
4789
4790         bond_debug_unregister(bond);
4791 }
4792
4793 /*------------------------- Module initialization ---------------------------*/
4794
4795 static int bond_check_params(struct bond_params *params)
4796 {
4797         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4798         struct bond_opt_value newval;
4799         const struct bond_opt_value *valptr;
4800         int arp_all_targets_value = 0;
4801         u16 ad_actor_sys_prio = 0;
4802         u16 ad_user_port_key = 0;
4803         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4804         int arp_ip_count;
4805         int bond_mode   = BOND_MODE_ROUNDROBIN;
4806         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4807         int lacp_fast = 0;
4808         int tlb_dynamic_lb;
4809
4810         /* Convert string parameters. */
4811         if (mode) {
4812                 bond_opt_initstr(&newval, mode);
4813                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4814                 if (!valptr) {
4815                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4816                         return -EINVAL;
4817                 }
4818                 bond_mode = valptr->value;
4819         }
4820
4821         if (xmit_hash_policy) {
4822                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4823                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
4824                     bond_mode == BOND_MODE_BROADCAST) {
4825                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4826                                 bond_mode_name(bond_mode));
4827                 } else {
4828                         bond_opt_initstr(&newval, xmit_hash_policy);
4829                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4830                                                 &newval);
4831                         if (!valptr) {
4832                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4833                                        xmit_hash_policy);
4834                                 return -EINVAL;
4835                         }
4836                         xmit_hashtype = valptr->value;
4837                 }
4838         }
4839
4840         if (lacp_rate) {
4841                 if (bond_mode != BOND_MODE_8023AD) {
4842                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4843                                 bond_mode_name(bond_mode));
4844                 } else {
4845                         bond_opt_initstr(&newval, lacp_rate);
4846                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4847                                                 &newval);
4848                         if (!valptr) {
4849                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4850                                        lacp_rate);
4851                                 return -EINVAL;
4852                         }
4853                         lacp_fast = valptr->value;
4854                 }
4855         }
4856
4857         if (ad_select) {
4858                 bond_opt_initstr(&newval, ad_select);
4859                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4860                                         &newval);
4861                 if (!valptr) {
4862                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4863                         return -EINVAL;
4864                 }
4865                 params->ad_select = valptr->value;
4866                 if (bond_mode != BOND_MODE_8023AD)
4867                         pr_warn("ad_select param only affects 802.3ad mode\n");
4868         } else {
4869                 params->ad_select = BOND_AD_STABLE;
4870         }
4871
4872         if (max_bonds < 0) {
4873                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4874                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4875                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4876         }
4877
4878         if (miimon < 0) {
4879                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4880                         miimon, INT_MAX);
4881                 miimon = 0;
4882         }
4883
4884         if (updelay < 0) {
4885                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4886                         updelay, INT_MAX);
4887                 updelay = 0;
4888         }
4889
4890         if (downdelay < 0) {
4891                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4892                         downdelay, INT_MAX);
4893                 downdelay = 0;
4894         }
4895
4896         if ((use_carrier != 0) && (use_carrier != 1)) {
4897                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4898                         use_carrier);
4899                 use_carrier = 1;
4900         }
4901
4902         if (num_peer_notif < 0 || num_peer_notif > 255) {
4903                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4904                         num_peer_notif);
4905                 num_peer_notif = 1;
4906         }
4907
4908         /* reset values for 802.3ad/TLB/ALB */
4909         if (!bond_mode_uses_arp(bond_mode)) {
4910                 if (!miimon) {
4911                         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");
4912                         pr_warn("Forcing miimon to 100msec\n");
4913                         miimon = BOND_DEFAULT_MIIMON;
4914                 }
4915         }
4916
4917         if (tx_queues < 1 || tx_queues > 255) {
4918                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4919                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4920                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4921         }
4922
4923         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4924                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4925                         all_slaves_active);
4926                 all_slaves_active = 0;
4927         }
4928
4929         if (resend_igmp < 0 || resend_igmp > 255) {
4930                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4931                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4932                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4933         }
4934
4935         bond_opt_initval(&newval, packets_per_slave);
4936         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4937                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4938                         packets_per_slave, USHRT_MAX);
4939                 packets_per_slave = 1;
4940         }
4941
4942         if (bond_mode == BOND_MODE_ALB) {
4943                 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",
4944                           updelay);
4945         }
4946
4947         if (!miimon) {
4948                 if (updelay || downdelay) {
4949                         /* just warn the user the up/down delay will have
4950                          * no effect since miimon is zero...
4951                          */
4952                         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",
4953                                 updelay, downdelay);
4954                 }
4955         } else {
4956                 /* don't allow arp monitoring */
4957                 if (arp_interval) {
4958                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4959                                 miimon, arp_interval);
4960                         arp_interval = 0;
4961                 }
4962
4963                 if ((updelay % miimon) != 0) {
4964                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4965                                 updelay, miimon, (updelay / miimon) * miimon);
4966                 }
4967
4968                 updelay /= miimon;
4969
4970                 if ((downdelay % miimon) != 0) {
4971                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4972                                 downdelay, miimon,
4973                                 (downdelay / miimon) * miimon);
4974                 }
4975
4976                 downdelay /= miimon;
4977         }
4978
4979         if (arp_interval < 0) {
4980                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4981                         arp_interval, INT_MAX);
4982                 arp_interval = 0;
4983         }
4984
4985         for (arp_ip_count = 0, i = 0;
4986              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4987                 __be32 ip;
4988
4989                 /* not a complete check, but good enough to catch mistakes */
4990                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4991                     !bond_is_ip_target_ok(ip)) {
4992                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4993                                 arp_ip_target[i]);
4994                         arp_interval = 0;
4995                 } else {
4996                         if (bond_get_targets_ip(arp_target, ip) == -1)
4997                                 arp_target[arp_ip_count++] = ip;
4998                         else
4999                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
5000                                         &ip);
5001                 }
5002         }
5003
5004         if (arp_interval && !arp_ip_count) {
5005                 /* don't allow arping if no arp_ip_target given... */
5006                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5007                         arp_interval);
5008                 arp_interval = 0;
5009         }
5010
5011         if (arp_validate) {
5012                 if (!arp_interval) {
5013                         pr_err("arp_validate requires arp_interval\n");
5014                         return -EINVAL;
5015                 }
5016
5017                 bond_opt_initstr(&newval, arp_validate);
5018                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
5019                                         &newval);
5020                 if (!valptr) {
5021                         pr_err("Error: invalid arp_validate \"%s\"\n",
5022                                arp_validate);
5023                         return -EINVAL;
5024                 }
5025                 arp_validate_value = valptr->value;
5026         } else {
5027                 arp_validate_value = 0;
5028         }
5029
5030         if (arp_all_targets) {
5031                 bond_opt_initstr(&newval, arp_all_targets);
5032                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
5033                                         &newval);
5034                 if (!valptr) {
5035                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
5036                                arp_all_targets);
5037                         arp_all_targets_value = 0;
5038                 } else {
5039                         arp_all_targets_value = valptr->value;
5040                 }
5041         }
5042
5043         if (miimon) {
5044                 pr_info("MII link monitoring set to %d ms\n", miimon);
5045         } else if (arp_interval) {
5046                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
5047                                           arp_validate_value);
5048                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5049                         arp_interval, valptr->string, arp_ip_count);
5050
5051                 for (i = 0; i < arp_ip_count; i++)
5052                         pr_cont(" %s", arp_ip_target[i]);
5053
5054                 pr_cont("\n");
5055
5056         } else if (max_bonds) {
5057                 /* miimon and arp_interval not set, we need one so things
5058                  * work as expected, see bonding.txt for details
5059                  */
5060                 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");
5061         }
5062
5063         if (primary && !bond_mode_uses_primary(bond_mode)) {
5064                 /* currently, using a primary only makes sense
5065                  * in active backup, TLB or ALB modes
5066                  */
5067                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
5068                         primary, bond_mode_name(bond_mode));
5069                 primary = NULL;
5070         }
5071
5072         if (primary && primary_reselect) {
5073                 bond_opt_initstr(&newval, primary_reselect);
5074                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
5075                                         &newval);
5076                 if (!valptr) {
5077                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5078                                primary_reselect);
5079                         return -EINVAL;
5080                 }
5081                 primary_reselect_value = valptr->value;
5082         } else {
5083                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5084         }
5085
5086         if (fail_over_mac) {
5087                 bond_opt_initstr(&newval, fail_over_mac);
5088                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
5089                                         &newval);
5090                 if (!valptr) {
5091                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5092                                fail_over_mac);
5093                         return -EINVAL;
5094                 }
5095                 fail_over_mac_value = valptr->value;
5096                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5097                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
5098         } else {
5099                 fail_over_mac_value = BOND_FOM_NONE;
5100         }
5101
5102         bond_opt_initstr(&newval, "default");
5103         valptr = bond_opt_parse(
5104                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
5105                                      &newval);
5106         if (!valptr) {
5107                 pr_err("Error: No ad_actor_sys_prio default value");
5108                 return -EINVAL;
5109         }
5110         ad_actor_sys_prio = valptr->value;
5111
5112         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
5113                                 &newval);
5114         if (!valptr) {
5115                 pr_err("Error: No ad_user_port_key default value");
5116                 return -EINVAL;
5117         }
5118         ad_user_port_key = valptr->value;
5119
5120         bond_opt_initstr(&newval, "default");
5121         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
5122         if (!valptr) {
5123                 pr_err("Error: No tlb_dynamic_lb default value");
5124                 return -EINVAL;
5125         }
5126         tlb_dynamic_lb = valptr->value;
5127
5128         if (lp_interval == 0) {
5129                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
5130                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
5131                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
5132         }
5133
5134         /* fill params struct with the proper values */
5135         params->mode = bond_mode;
5136         params->xmit_policy = xmit_hashtype;
5137         params->miimon = miimon;
5138         params->num_peer_notif = num_peer_notif;
5139         params->arp_interval = arp_interval;
5140         params->arp_validate = arp_validate_value;
5141         params->arp_all_targets = arp_all_targets_value;
5142         params->updelay = updelay;
5143         params->downdelay = downdelay;
5144         params->peer_notif_delay = 0;
5145         params->use_carrier = use_carrier;
5146         params->lacp_fast = lacp_fast;
5147         params->primary[0] = 0;
5148         params->primary_reselect = primary_reselect_value;
5149         params->fail_over_mac = fail_over_mac_value;
5150         params->tx_queues = tx_queues;
5151         params->all_slaves_active = all_slaves_active;
5152         params->resend_igmp = resend_igmp;
5153         params->min_links = min_links;
5154         params->lp_interval = lp_interval;
5155         params->packets_per_slave = packets_per_slave;
5156         params->tlb_dynamic_lb = tlb_dynamic_lb;
5157         params->ad_actor_sys_prio = ad_actor_sys_prio;
5158         eth_zero_addr(params->ad_actor_system);
5159         params->ad_user_port_key = ad_user_port_key;
5160         if (packets_per_slave > 0) {
5161                 params->reciprocal_packets_per_slave =
5162                         reciprocal_value(packets_per_slave);
5163         } else {
5164                 /* reciprocal_packets_per_slave is unused if
5165                  * packets_per_slave is 0 or 1, just initialize it
5166                  */
5167                 params->reciprocal_packets_per_slave =
5168                         (struct reciprocal_value) { 0 };
5169         }
5170
5171         if (primary) {
5172                 strncpy(params->primary, primary, IFNAMSIZ);
5173                 params->primary[IFNAMSIZ - 1] = 0;
5174         }
5175
5176         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5177
5178         return 0;
5179 }
5180
5181 /* Called from registration process */
5182 static int bond_init(struct net_device *bond_dev)
5183 {
5184         struct bonding *bond = netdev_priv(bond_dev);
5185         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5186
5187         netdev_dbg(bond_dev, "Begin bond_init\n");
5188
5189         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
5190         if (!bond->wq)
5191                 return -ENOMEM;
5192
5193         spin_lock_init(&bond->stats_lock);
5194         netdev_lockdep_set_classes(bond_dev);
5195
5196         list_add_tail(&bond->bond_list, &bn->dev_list);
5197
5198         bond_prepare_sysfs_group(bond);
5199
5200         bond_debug_register(bond);
5201
5202         /* Ensure valid dev_addr */
5203         if (is_zero_ether_addr(bond_dev->dev_addr) &&
5204             bond_dev->addr_assign_type == NET_ADDR_PERM)
5205                 eth_hw_addr_random(bond_dev);
5206
5207         return 0;
5208 }
5209
5210 unsigned int bond_get_num_tx_queues(void)
5211 {
5212         return tx_queues;
5213 }
5214
5215 /* Create a new bond based on the specified name and bonding parameters.
5216  * If name is NULL, obtain a suitable "bond%d" name for us.
5217  * Caller must NOT hold rtnl_lock; we need to release it here before we
5218  * set up our sysfs entries.
5219  */
5220 int bond_create(struct net *net, const char *name)
5221 {
5222         struct net_device *bond_dev;
5223         struct bonding *bond;
5224         struct alb_bond_info *bond_info;
5225         int res;
5226
5227         rtnl_lock();
5228
5229         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
5230                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
5231                                    bond_setup, tx_queues);
5232         if (!bond_dev) {
5233                 pr_err("%s: eek! can't alloc netdev!\n", name);
5234                 rtnl_unlock();
5235                 return -ENOMEM;
5236         }
5237
5238         /*
5239          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
5240          * It is set to 0 by default which is wrong.
5241          */
5242         bond = netdev_priv(bond_dev);
5243         bond_info = &(BOND_ALB_INFO(bond));
5244         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
5245
5246         dev_net_set(bond_dev, net);
5247         bond_dev->rtnl_link_ops = &bond_link_ops;
5248
5249         res = register_netdevice(bond_dev);
5250         if (res < 0) {
5251                 free_netdev(bond_dev);
5252                 rtnl_unlock();
5253
5254                 return res;
5255         }
5256
5257         netif_carrier_off(bond_dev);
5258
5259         bond_work_init_all(bond);
5260
5261         rtnl_unlock();
5262         return 0;
5263 }
5264
5265 static int __net_init bond_net_init(struct net *net)
5266 {
5267         struct bond_net *bn = net_generic(net, bond_net_id);
5268
5269         bn->net = net;
5270         INIT_LIST_HEAD(&bn->dev_list);
5271
5272         bond_create_proc_dir(bn);
5273         bond_create_sysfs(bn);
5274
5275         return 0;
5276 }
5277
5278 static void __net_exit bond_net_exit(struct net *net)
5279 {
5280         struct bond_net *bn = net_generic(net, bond_net_id);
5281         struct bonding *bond, *tmp_bond;
5282         LIST_HEAD(list);
5283
5284         bond_destroy_sysfs(bn);
5285
5286         /* Kill off any bonds created after unregistering bond rtnl ops */
5287         rtnl_lock();
5288         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
5289                 unregister_netdevice_queue(bond->dev, &list);
5290         unregister_netdevice_many(&list);
5291         rtnl_unlock();
5292
5293         bond_destroy_proc_dir(bn);
5294 }
5295
5296 static struct pernet_operations bond_net_ops = {
5297         .init = bond_net_init,
5298         .exit = bond_net_exit,
5299         .id   = &bond_net_id,
5300         .size = sizeof(struct bond_net),
5301 };
5302
5303 static int __init bonding_init(void)
5304 {
5305         int i;
5306         int res;
5307
5308         res = bond_check_params(&bonding_defaults);
5309         if (res)
5310                 goto out;
5311
5312         res = register_pernet_subsys(&bond_net_ops);
5313         if (res)
5314                 goto out;
5315
5316         res = bond_netlink_init();
5317         if (res)
5318                 goto err_link;
5319
5320         bond_create_debugfs();
5321
5322         for (i = 0; i < max_bonds; i++) {
5323                 res = bond_create(&init_net, NULL);
5324                 if (res)
5325                         goto err;
5326         }
5327
5328         skb_flow_dissector_init(&flow_keys_bonding,
5329                                 flow_keys_bonding_keys,
5330                                 ARRAY_SIZE(flow_keys_bonding_keys));
5331
5332         register_netdevice_notifier(&bond_netdev_notifier);
5333 out:
5334         return res;
5335 err:
5336         bond_destroy_debugfs();
5337         bond_netlink_fini();
5338 err_link:
5339         unregister_pernet_subsys(&bond_net_ops);
5340         goto out;
5341
5342 }
5343
5344 static void __exit bonding_exit(void)
5345 {
5346         unregister_netdevice_notifier(&bond_netdev_notifier);
5347
5348         bond_destroy_debugfs();
5349
5350         bond_netlink_fini();
5351         unregister_pernet_subsys(&bond_net_ops);
5352
5353 #ifdef CONFIG_NET_POLL_CONTROLLER
5354         /* Make sure we don't have an imbalance on our netpoll blocking */
5355         WARN_ON(atomic_read(&netpoll_block_tx));
5356 #endif
5357 }
5358
5359 module_init(bonding_init);
5360 module_exit(bonding_exit);
5361 MODULE_LICENSE("GPL");
5362 MODULE_DESCRIPTION(DRV_DESCRIPTION);
5363 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");