drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/icmp.h>
  45 #include <linux/icmpv6.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include <linux/rculist.h>
  81 #include <net/flow_dissector.h>
  82 #include <net/xfrm.h>
  83 #include <net/bonding.h>
  84 #include <net/bond_3ad.h>
  85 #include <net/bond_alb.h>
  86
  87 #include "bonding_priv.h"
  88
  89 /*---------------------------- Module parameters ----------------------------*/
  90
  91 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  92
  93 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  94 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  95 static int num_peer_notif = 1;
  96 static int miimon;
  97 static int updelay;
  98 static int downdelay;
  99 static int use_carrier  = 1;
 100 static char *mode;
 101 static char *primary;
 102 static char *primary_reselect;
 103 static char *lacp_rate;
 104 static int min_links;
 105 static char *ad_select;
 106 static char *xmit_hash_policy;
 107 static int arp_interval;
 108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 109 static char *arp_validate;
 110 static char *arp_all_targets;
 111 static char *fail_over_mac;
 112 static int all_slaves_active;
 113 static struct bond_params bonding_defaults;
 114 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 115 static int packets_per_slave = 1;
 116 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 117
 118 module_param(max_bonds, int, 0);
 119 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 120 module_param(tx_queues, int, 0);
 121 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 122 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 123 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 124                                "failover event (alias of num_unsol_na)");
 125 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 126 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 127                                "failover event (alias of num_grat_arp)");
 128 module_param(miimon, int, 0);
 129 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 130 module_param(updelay, int, 0);
 131 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 132 module_param(downdelay, int, 0);
 133 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 134                             "in milliseconds");
 135 module_param(use_carrier, int, 0);
 136 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 137                               "0 for off, 1 for on (default)");
 138 module_param(mode, charp, 0);
 139 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 140                        "1 for active-backup, 2 for balance-xor, "
 141                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 142                        "6 for balance-alb");
 143 module_param(primary, charp, 0);
 144 MODULE_PARM_DESC(primary, "Primary network device to use");
 145 module_param(primary_reselect, charp, 0);
 146 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 147                                    "once it comes up; "
 148                                    "0 for always (default), "
 149                                    "1 for only if speed of primary is "
 150                                    "better, "
 151                                    "2 for only on active slave "
 152                                    "failure");
 153 module_param(lacp_rate, charp, 0);
 154 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 155                             "0 for slow, 1 for fast");
 156 module_param(ad_select, charp, 0);
 157 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
 158                             "0 for stable (default), 1 for bandwidth, "
 159                             "2 for count");
 160 module_param(min_links, int, 0);
 161 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 162
 163 module_param(xmit_hash_policy, charp, 0);
 164 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
 165                                    "0 for layer 2 (default), 1 for layer 3+4, "
 166                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 167                                    "4 for encap layer 3+4");
 168 module_param(arp_interval, int, 0);
 169 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 170 module_param_array(arp_ip_target, charp, NULL, 0);
 171 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 172 module_param(arp_validate, charp, 0);
 173 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 174                                "0 for none (default), 1 for active, "
 175                                "2 for backup, 3 for all");
 176 module_param(arp_all_targets, charp, 0);
 177 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 178 module_param(fail_over_mac, charp, 0);
 179 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 180                                 "the same MAC; 0 for none (default), "
 181                                 "1 for active, 2 for follow");
 182 module_param(all_slaves_active, int, 0);
 183 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 184                                      "by setting active flag for all slaves; "
 185                                      "0 for never (default), 1 for always.");
 186 module_param(resend_igmp, int, 0);
 187 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 188                               "link failure");
 189 module_param(packets_per_slave, int, 0);
 190 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 191                                     "mode; 0 for a random slave, 1 packet per "
 192                                     "slave (default), >1 packets per slave.");
 193 module_param(lp_interval, uint, 0);
 194 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 195                               "the bonding driver sends learning packets to "
 196                               "each slaves peer switch. The default is 1.");
 197
 198 /*----------------------------- Global variables ----------------------------*/
 199
 200 #ifdef CONFIG_NET_POLL_CONTROLLER
 201 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 202 #endif
 203
 204 unsigned int bond_net_id __read_mostly;
 205
 206 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
 207         {
 208                 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
 209                 .offset = offsetof(struct flow_keys, control),
 210         },
 211         {
 212                 .key_id = FLOW_DISSECTOR_KEY_BASIC,
 213                 .offset = offsetof(struct flow_keys, basic),
 214         },
 215         {
 216                 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 217                 .offset = offsetof(struct flow_keys, addrs.v4addrs),
 218         },
 219         {
 220                 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 221                 .offset = offsetof(struct flow_keys, addrs.v6addrs),
 222         },
 223         {
 224                 .key_id = FLOW_DISSECTOR_KEY_TIPC,
 225                 .offset = offsetof(struct flow_keys, addrs.tipckey),
 226         },
 227         {
 228                 .key_id = FLOW_DISSECTOR_KEY_PORTS,
 229                 .offset = offsetof(struct flow_keys, ports),
 230         },
 231         {
 232                 .key_id = FLOW_DISSECTOR_KEY_ICMP,
 233                 .offset = offsetof(struct flow_keys, icmp),
 234         },
 235         {
 236                 .key_id = FLOW_DISSECTOR_KEY_VLAN,
 237                 .offset = offsetof(struct flow_keys, vlan),
 238         },
 239         {
 240                 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
 241                 .offset = offsetof(struct flow_keys, tags),
 242         },
 243         {
 244                 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
 245                 .offset = offsetof(struct flow_keys, keyid),
 246         },
 247 };
 248
 249 static struct flow_dissector flow_keys_bonding __read_mostly;
 250
 251 /*-------------------------- Forward declarations ---------------------------*/
 252
 253 static int bond_init(struct net_device *bond_dev);
 254 static void bond_uninit(struct net_device *bond_dev);
 255 static void bond_get_stats(struct net_device *bond_dev,
 256                            struct rtnl_link_stats64 *stats);
 257 static void bond_slave_arr_handler(struct work_struct *work);
 258 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 259                                   int mod);
 260 static void bond_netdev_notify_work(struct work_struct *work);
 261
 262 /*---------------------------- General routines -----------------------------*/
 263
 264 const char *bond_mode_name(int mode)
 265 {
 266         static const char *names[] = {
 267                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 268                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 269                 [BOND_MODE_XOR] = "load balancing (xor)",
 270                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 271                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 272                 [BOND_MODE_TLB] = "transmit load balancing",
 273                 [BOND_MODE_ALB] = "adaptive load balancing",
 274         };
 275
 276         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 277                 return "unknown";
 278
 279         return names[mode];
 280 }
 281
 282 /**
 283  * bond_dev_queue_xmit - Prepare skb for xmit.
 284  *
 285  * @bond: bond device that got this skb for tx.
 286  * @skb: hw accel VLAN tagged skb to transmit
 287  * @slave_dev: slave that is supposed to xmit this skbuff
 288  */
 289 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 290                         struct net_device *slave_dev)
 291 {
 292         skb->dev = slave_dev;
 293
 294         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 295                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 296         skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 297
 298         if (unlikely(netpoll_tx_running(bond->dev)))
 299                 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 300
 301         return dev_queue_xmit(skb);
 302 }
 303
 304 /*---------------------------------- VLAN -----------------------------------*/
 305
 306 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 307  * We don't protect the slave list iteration with a lock because:
 308  * a. This operation is performed in IOCTL context,
 309  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 310  * c. Holding a lock with BH disabled while directly calling a base driver
 311  *    entry point is generally a BAD idea.
 312  *
 313  * The design of synchronization/protection for this operation in the 8021q
 314  * module is good for one or more VLAN devices over a single physical device
 315  * and cannot be extended for a teaming solution like bonding, so there is a
 316  * potential race condition here where a net device from the vlan group might
 317  * be referenced (either by a base driver or the 8021q code) while it is being
 318  * removed from the system. However, it turns out we're not making matters
 319  * worse, and if it works for regular VLAN usage it will work here too.
 320 */
 321
 322 /**
 323  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 324  * @bond_dev: bonding net device that got called
 325  * @proto: network protocol ID
 326  * @vid: vlan id being added
 327  */
 328 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 329                                 __be16 proto, u16 vid)
 330 {
 331         struct bonding *bond = netdev_priv(bond_dev);
 332         struct slave *slave, *rollback_slave;
 333         struct list_head *iter;
 334         int res;
 335
 336         bond_for_each_slave(bond, slave, iter) {
 337                 res = vlan_vid_add(slave->dev, proto, vid);
 338                 if (res)
 339                         goto unwind;
 340         }
 341
 342         return 0;
 343
 344 unwind:
 345         /* unwind to the slave that failed */
 346         bond_for_each_slave(bond, rollback_slave, iter) {
 347                 if (rollback_slave == slave)
 348                         break;
 349
 350                 vlan_vid_del(rollback_slave->dev, proto, vid);
 351         }
 352
 353         return res;
 354 }
 355
 356 /**
 357  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 358  * @bond_dev: bonding net device that got called
 359  * @proto: network protocol ID
 360  * @vid: vlan id being removed
 361  */
 362 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 363                                  __be16 proto, u16 vid)
 364 {
 365         struct bonding *bond = netdev_priv(bond_dev);
 366         struct list_head *iter;
 367         struct slave *slave;
 368
 369         bond_for_each_slave(bond, slave, iter)
 370                 vlan_vid_del(slave->dev, proto, vid);
 371
 372         if (bond_is_lb(bond))
 373                 bond_alb_clear_vlan(bond, vid);
 374
 375         return 0;
 376 }
 377
 378 /*---------------------------------- XFRM -----------------------------------*/
 379
 380 #ifdef CONFIG_XFRM_OFFLOAD
 381 /**
 382  * bond_ipsec_add_sa - program device with a security association
 383  * @xs: pointer to transformer state struct
 384  **/
 385 static int bond_ipsec_add_sa(struct xfrm_state *xs)
 386 {
 387         struct net_device *bond_dev = xs->xso.dev;
 388         struct bonding *bond;
 389         struct slave *slave;
 390
 391         if (!bond_dev)
 392                 return -EINVAL;
 393
 394         bond = netdev_priv(bond_dev);
 395         slave = rcu_dereference(bond->curr_active_slave);
 396         xs->xso.real_dev = slave->dev;
 397         bond->xs = xs;
 398
 399         if (!(slave->dev->xfrmdev_ops
 400               && slave->dev->xfrmdev_ops->xdo_dev_state_add)) {
 401                 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
 402                 return -EINVAL;
 403         }
 404
 405         return slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
 406 }
 407
 408 /**
 409  * bond_ipsec_del_sa - clear out this specific SA
 410  * @xs: pointer to transformer state struct
 411  **/
 412 static void bond_ipsec_del_sa(struct xfrm_state *xs)
 413 {
 414         struct net_device *bond_dev = xs->xso.dev;
 415         struct bonding *bond;
 416         struct slave *slave;
 417
 418         if (!bond_dev)
 419                 return;
 420
 421         bond = netdev_priv(bond_dev);
 422         slave = rcu_dereference(bond->curr_active_slave);
 423
 424         if (!slave)
 425                 return;
 426
 427         xs->xso.real_dev = slave->dev;
 428
 429         if (!(slave->dev->xfrmdev_ops
 430               && slave->dev->xfrmdev_ops->xdo_dev_state_delete)) {
 431                 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
 432                 return;
 433         }
 434
 435         slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
 436 }
 437
 438 /**
 439  * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
 440  * @skb: current data packet
 441  * @xs: pointer to transformer state struct
 442  **/
 443 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
 444 {
 445         struct net_device *bond_dev = xs->xso.dev;
 446         struct bonding *bond = netdev_priv(bond_dev);
 447         struct slave *curr_active = rcu_dereference(bond->curr_active_slave);
 448         struct net_device *slave_dev = curr_active->dev;
 449
 450         if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
 451                 return true;
 452
 453         if (!(slave_dev->xfrmdev_ops
 454               && slave_dev->xfrmdev_ops->xdo_dev_offload_ok)) {
 455                 slave_warn(bond_dev, slave_dev, "%s: no slave xdo_dev_offload_ok\n", __func__);
 456                 return false;
 457         }
 458
 459         xs->xso.real_dev = slave_dev;
 460         return slave_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
 461 }
 462
 463 static const struct xfrmdev_ops bond_xfrmdev_ops = {
 464         .xdo_dev_state_add = bond_ipsec_add_sa,
 465         .xdo_dev_state_delete = bond_ipsec_del_sa,
 466         .xdo_dev_offload_ok = bond_ipsec_offload_ok,
 467 };
 468 #endif /* CONFIG_XFRM_OFFLOAD */
 469
 470 /*------------------------------- Link status -------------------------------*/
 471
 472 /* Set the carrier state for the master according to the state of its
 473  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 474  * do special 802.3ad magic.
 475  *
 476  * Returns zero if carrier state does not change, nonzero if it does.
 477  */
 478 int bond_set_carrier(struct bonding *bond)
 479 {
 480         struct list_head *iter;
 481         struct slave *slave;
 482
 483         if (!bond_has_slaves(bond))
 484                 goto down;
 485
 486         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 487                 return bond_3ad_set_carrier(bond);
 488
 489         bond_for_each_slave(bond, slave, iter) {
 490                 if (slave->link == BOND_LINK_UP) {
 491                         if (!netif_carrier_ok(bond->dev)) {
 492                                 netif_carrier_on(bond->dev);
 493                                 return 1;
 494                         }
 495                         return 0;
 496                 }
 497         }
 498
 499 down:
 500         if (netif_carrier_ok(bond->dev)) {
 501                 netif_carrier_off(bond->dev);
 502                 return 1;
 503         }
 504         return 0;
 505 }
 506
 507 /* Get link speed and duplex from the slave's base driver
 508  * using ethtool. If for some reason the call fails or the
 509  * values are invalid, set speed and duplex to -1,
 510  * and return. Return 1 if speed or duplex settings are
 511  * UNKNOWN; 0 otherwise.
 512  */
 513 static int bond_update_speed_duplex(struct slave *slave)
 514 {
 515         struct net_device *slave_dev = slave->dev;
 516         struct ethtool_link_ksettings ecmd;
 517         int res;
 518
 519         slave->speed = SPEED_UNKNOWN;
 520         slave->duplex = DUPLEX_UNKNOWN;
 521
 522         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 523         if (res < 0)
 524                 return 1;
 525         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 526                 return 1;
 527         switch (ecmd.base.duplex) {
 528         case DUPLEX_FULL:
 529         case DUPLEX_HALF:
 530                 break;
 531         default:
 532                 return 1;
 533         }
 534
 535         slave->speed = ecmd.base.speed;
 536         slave->duplex = ecmd.base.duplex;
 537
 538         return 0;
 539 }
 540
 541 const char *bond_slave_link_status(s8 link)
 542 {
 543         switch (link) {
 544         case BOND_LINK_UP:
 545                 return "up";
 546         case BOND_LINK_FAIL:
 547                 return "going down";
 548         case BOND_LINK_DOWN:
 549                 return "down";
 550         case BOND_LINK_BACK:
 551                 return "going back";
 552         default:
 553                 return "unknown";
 554         }
 555 }
 556
 557 /* if <dev> supports MII link status reporting, check its link status.
 558  *
 559  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 560  * depending upon the setting of the use_carrier parameter.
 561  *
 562  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 563  * can't tell and just pretend it is), or 0, meaning that the link is
 564  * down.
 565  *
 566  * If reporting is non-zero, instead of faking link up, return -1 if
 567  * both ETHTOOL and MII ioctls fail (meaning the device does not
 568  * support them).  If use_carrier is set, return whatever it says.
 569  * It'd be nice if there was a good way to tell if a driver supports
 570  * netif_carrier, but there really isn't.
 571  */
 572 static int bond_check_dev_link(struct bonding *bond,
 573                                struct net_device *slave_dev, int reporting)
 574 {
 575         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 576         int (*ioctl)(struct net_device *, struct ifreq *, int);
 577         struct ifreq ifr;
 578         struct mii_ioctl_data *mii;
 579
 580         if (!reporting && !netif_running(slave_dev))
 581                 return 0;
 582
 583         if (bond->params.use_carrier)
 584                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 585
 586         /* Try to get link status using Ethtool first. */
 587         if (slave_dev->ethtool_ops->get_link)
 588                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 589                         BMSR_LSTATUS : 0;
 590
 591         /* Ethtool can't be used, fallback to MII ioctls. */
 592         ioctl = slave_ops->ndo_do_ioctl;
 593         if (ioctl) {
 594                 /* TODO: set pointer to correct ioctl on a per team member
 595                  *       bases to make this more efficient. that is, once
 596                  *       we determine the correct ioctl, we will always
 597                  *       call it and not the others for that team
 598                  *       member.
 599                  */
 600
 601                 /* We cannot assume that SIOCGMIIPHY will also read a
 602                  * register; not all network drivers (e.g., e100)
 603                  * support that.
 604                  */
 605
 606                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 607                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 608                 mii = if_mii(&ifr);
 609                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 610                         mii->reg_num = MII_BMSR;
 611                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 612                                 return mii->val_out & BMSR_LSTATUS;
 613                 }
 614         }
 615
 616         /* If reporting, report that either there's no dev->do_ioctl,
 617          * or both SIOCGMIIREG and get_link failed (meaning that we
 618          * cannot report link status).  If not reporting, pretend
 619          * we're ok.
 620          */
 621         return reporting ? -1 : BMSR_LSTATUS;
 622 }
 623
 624 /*----------------------------- Multicast list ------------------------------*/
 625
 626 /* Push the promiscuity flag down to appropriate slaves */
 627 static int bond_set_promiscuity(struct bonding *bond, int inc)
 628 {
 629         struct list_head *iter;
 630         int err = 0;
 631
 632         if (bond_uses_primary(bond)) {
 633                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 634
 635                 if (curr_active)
 636                         err = dev_set_promiscuity(curr_active->dev, inc);
 637         } else {
 638                 struct slave *slave;
 639
 640                 bond_for_each_slave(bond, slave, iter) {
 641                         err = dev_set_promiscuity(slave->dev, inc);
 642                         if (err)
 643                                 return err;
 644                 }
 645         }
 646         return err;
 647 }
 648
 649 /* Push the allmulti flag down to all slaves */
 650 static int bond_set_allmulti(struct bonding *bond, int inc)
 651 {
 652         struct list_head *iter;
 653         int err = 0;
 654
 655         if (bond_uses_primary(bond)) {
 656                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 657
 658                 if (curr_active)
 659                         err = dev_set_allmulti(curr_active->dev, inc);
 660         } else {
 661                 struct slave *slave;
 662
 663                 bond_for_each_slave(bond, slave, iter) {
 664                         err = dev_set_allmulti(slave->dev, inc);
 665                         if (err)
 666                                 return err;
 667                 }
 668         }
 669         return err;
 670 }
 671
 672 /* Retrieve the list of registered multicast addresses for the bonding
 673  * device and retransmit an IGMP JOIN request to the current active
 674  * slave.
 675  */
 676 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 677 {
 678         struct bonding *bond = container_of(work, struct bonding,
 679                                             mcast_work.work);
 680
 681         if (!rtnl_trylock()) {
 682                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 683                 return;
 684         }
 685         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 686
 687         if (bond->igmp_retrans > 1) {
 688                 bond->igmp_retrans--;
 689                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 690         }
 691         rtnl_unlock();
 692 }
 693
 694 /* Flush bond's hardware addresses from slave */
 695 static void bond_hw_addr_flush(struct net_device *bond_dev,
 696                                struct net_device *slave_dev)
 697 {
 698         struct bonding *bond = netdev_priv(bond_dev);
 699
 700         dev_uc_unsync(slave_dev, bond_dev);
 701         dev_mc_unsync(slave_dev, bond_dev);
 702
 703         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 704                 /* del lacpdu mc addr from mc list */
 705                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 706
 707                 dev_mc_del(slave_dev, lacpdu_multicast);
 708         }
 709 }
 710
 711 /*--------------------------- Active slave change ---------------------------*/
 712
 713 /* Update the hardware address list and promisc/allmulti for the new and
 714  * old active slaves (if any).  Modes that are not using primary keep all
 715  * slaves up date at all times; only the modes that use primary need to call
 716  * this function to swap these settings during a failover.
 717  */
 718 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 719                               struct slave *old_active)
 720 {
 721         if (old_active) {
 722                 if (bond->dev->flags & IFF_PROMISC)
 723                         dev_set_promiscuity(old_active->dev, -1);
 724
 725                 if (bond->dev->flags & IFF_ALLMULTI)
 726                         dev_set_allmulti(old_active->dev, -1);
 727
 728                 bond_hw_addr_flush(bond->dev, old_active->dev);
 729         }
 730
 731         if (new_active) {
 732                 /* FIXME: Signal errors upstream. */
 733                 if (bond->dev->flags & IFF_PROMISC)
 734                         dev_set_promiscuity(new_active->dev, 1);
 735
 736                 if (bond->dev->flags & IFF_ALLMULTI)
 737                         dev_set_allmulti(new_active->dev, 1);
 738
 739                 netif_addr_lock_bh(bond->dev);
 740                 dev_uc_sync(new_active->dev, bond->dev);
 741                 dev_mc_sync(new_active->dev, bond->dev);
 742                 netif_addr_unlock_bh(bond->dev);
 743         }
 744 }
 745
 746 /**
 747  * bond_set_dev_addr - clone slave's address to bond
 748  * @bond_dev: bond net device
 749  * @slave_dev: slave net device
 750  *
 751  * Should be called with RTNL held.
 752  */
 753 static int bond_set_dev_addr(struct net_device *bond_dev,
 754                              struct net_device *slave_dev)
 755 {
 756         int err;
 757
 758         slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 759                   bond_dev, slave_dev, slave_dev->addr_len);
 760         err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
 761         if (err)
 762                 return err;
 763
 764         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 765         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 766         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 767         return 0;
 768 }
 769
 770 static struct slave *bond_get_old_active(struct bonding *bond,
 771                                          struct slave *new_active)
 772 {
 773         struct slave *slave;
 774         struct list_head *iter;
 775
 776         bond_for_each_slave(bond, slave, iter) {
 777                 if (slave == new_active)
 778                         continue;
 779
 780                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 781                         return slave;
 782         }
 783
 784         return NULL;
 785 }
 786
 787 /* bond_do_fail_over_mac
 788  *
 789  * Perform special MAC address swapping for fail_over_mac settings
 790  *
 791  * Called with RTNL
 792  */
 793 static void bond_do_fail_over_mac(struct bonding *bond,
 794                                   struct slave *new_active,
 795                                   struct slave *old_active)
 796 {
 797         u8 tmp_mac[MAX_ADDR_LEN];
 798         struct sockaddr_storage ss;
 799         int rv;
 800
 801         switch (bond->params.fail_over_mac) {
 802         case BOND_FOM_ACTIVE:
 803                 if (new_active) {
 804                         rv = bond_set_dev_addr(bond->dev, new_active->dev);
 805                         if (rv)
 806                                 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
 807                                           -rv);
 808                 }
 809                 break;
 810         case BOND_FOM_FOLLOW:
 811                 /* if new_active && old_active, swap them
 812                  * if just old_active, do nothing (going to no active slave)
 813                  * if just new_active, set new_active to bond's MAC
 814                  */
 815                 if (!new_active)
 816                         return;
 817
 818                 if (!old_active)
 819                         old_active = bond_get_old_active(bond, new_active);
 820
 821                 if (old_active) {
 822                         bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 823                                           new_active->dev->addr_len);
 824                         bond_hw_addr_copy(ss.__data,
 825                                           old_active->dev->dev_addr,
 826                                           old_active->dev->addr_len);
 827                         ss.ss_family = new_active->dev->type;
 828                 } else {
 829                         bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 830                                           bond->dev->addr_len);
 831                         ss.ss_family = bond->dev->type;
 832                 }
 833
 834                 rv = dev_set_mac_address(new_active->dev,
 835                                          (struct sockaddr *)&ss, NULL);
 836                 if (rv) {
 837                         slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
 838                                   -rv);
 839                         goto out;
 840                 }
 841
 842                 if (!old_active)
 843                         goto out;
 844
 845                 bond_hw_addr_copy(ss.__data, tmp_mac,
 846                                   new_active->dev->addr_len);
 847                 ss.ss_family = old_active->dev->type;
 848
 849                 rv = dev_set_mac_address(old_active->dev,
 850                                          (struct sockaddr *)&ss, NULL);
 851                 if (rv)
 852                         slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
 853                                   -rv);
 854 out:
 855                 break;
 856         default:
 857                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 858                            bond->params.fail_over_mac);
 859                 break;
 860         }
 861
 862 }
 863
 864 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 865 {
 866         struct slave *prim = rtnl_dereference(bond->primary_slave);
 867         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 868
 869         if (!prim || prim->link != BOND_LINK_UP) {
 870                 if (!curr || curr->link != BOND_LINK_UP)
 871                         return NULL;
 872                 return curr;
 873         }
 874
 875         if (bond->force_primary) {
 876                 bond->force_primary = false;
 877                 return prim;
 878         }
 879
 880         if (!curr || curr->link != BOND_LINK_UP)
 881                 return prim;
 882
 883         /* At this point, prim and curr are both up */
 884         switch (bond->params.primary_reselect) {
 885         case BOND_PRI_RESELECT_ALWAYS:
 886                 return prim;
 887         case BOND_PRI_RESELECT_BETTER:
 888                 if (prim->speed < curr->speed)
 889                         return curr;
 890                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 891                         return curr;
 892                 return prim;
 893         case BOND_PRI_RESELECT_FAILURE:
 894                 return curr;
 895         default:
 896                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 897                            bond->params.primary_reselect);
 898                 return curr;
 899         }
 900 }
 901
 902 /**
 903  * bond_find_best_slave - select the best available slave to be the active one
 904  * @bond: our bonding struct
 905  */
 906 static struct slave *bond_find_best_slave(struct bonding *bond)
 907 {
 908         struct slave *slave, *bestslave = NULL;
 909         struct list_head *iter;
 910         int mintime = bond->params.updelay;
 911
 912         slave = bond_choose_primary_or_current(bond);
 913         if (slave)
 914                 return slave;
 915
 916         bond_for_each_slave(bond, slave, iter) {
 917                 if (slave->link == BOND_LINK_UP)
 918                         return slave;
 919                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 920                     slave->delay < mintime) {
 921                         mintime = slave->delay;
 922                         bestslave = slave;
 923                 }
 924         }
 925
 926         return bestslave;
 927 }
 928
 929 static bool bond_should_notify_peers(struct bonding *bond)
 930 {
 931         struct slave *slave;
 932
 933         rcu_read_lock();
 934         slave = rcu_dereference(bond->curr_active_slave);
 935         rcu_read_unlock();
 936
 937         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 938                    slave ? slave->dev->name : "NULL");
 939
 940         if (!slave || !bond->send_peer_notif ||
 941             bond->send_peer_notif %
 942             max(1, bond->params.peer_notif_delay) != 0 ||
 943             !netif_carrier_ok(bond->dev) ||
 944             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 945                 return false;
 946
 947         return true;
 948 }
 949
 950 /**
 951  * change_active_interface - change the active slave into the specified one
 952  * @bond: our bonding struct
 953  * @new_active: the new slave to make the active one
 954  *
 955  * Set the new slave to the bond's settings and unset them on the old
 956  * curr_active_slave.
 957  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 958  *
 959  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 960  * because it is apparently the best available slave we have, even though its
 961  * updelay hasn't timed out yet.
 962  *
 963  * Caller must hold RTNL.
 964  */
 965 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 966 {
 967         struct slave *old_active;
 968
 969         ASSERT_RTNL();
 970
 971         old_active = rtnl_dereference(bond->curr_active_slave);
 972
 973         if (old_active == new_active)
 974                 return;
 975
 976 #ifdef CONFIG_XFRM_OFFLOAD
 977         if (old_active && bond->xs)
 978                 bond_ipsec_del_sa(bond->xs);
 979 #endif /* CONFIG_XFRM_OFFLOAD */
 980
 981         if (new_active) {
 982                 new_active->last_link_up = jiffies;
 983
 984                 if (new_active->link == BOND_LINK_BACK) {
 985                         if (bond_uses_primary(bond)) {
 986                                 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
 987                                            (bond->params.updelay - new_active->delay) * bond->params.miimon);
 988                         }
 989
 990                         new_active->delay = 0;
 991                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 992                                                   BOND_SLAVE_NOTIFY_NOW);
 993
 994                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 995                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 996
 997                         if (bond_is_lb(bond))
 998                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 999                 } else {
1000                         if (bond_uses_primary(bond)) {
1001                                 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1002                         }
1003                 }
1004         }
1005
1006         if (bond_uses_primary(bond))
1007                 bond_hw_addr_swap(bond, new_active, old_active);
1008
1009         if (bond_is_lb(bond)) {
1010                 bond_alb_handle_active_change(bond, new_active);
1011                 if (old_active)
1012                         bond_set_slave_inactive_flags(old_active,
1013                                                       BOND_SLAVE_NOTIFY_NOW);
1014                 if (new_active)
1015                         bond_set_slave_active_flags(new_active,
1016                                                     BOND_SLAVE_NOTIFY_NOW);
1017         } else {
1018                 rcu_assign_pointer(bond->curr_active_slave, new_active);
1019         }
1020
1021         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1022                 if (old_active)
1023                         bond_set_slave_inactive_flags(old_active,
1024                                                       BOND_SLAVE_NOTIFY_NOW);
1025
1026                 if (new_active) {
1027                         bool should_notify_peers = false;
1028
1029                         bond_set_slave_active_flags(new_active,
1030                                                     BOND_SLAVE_NOTIFY_NOW);
1031
1032                         if (bond->params.fail_over_mac)
1033                                 bond_do_fail_over_mac(bond, new_active,
1034                                                       old_active);
1035
1036                         if (netif_running(bond->dev)) {
1037                                 bond->send_peer_notif =
1038                                         bond->params.num_peer_notif *
1039                                         max(1, bond->params.peer_notif_delay);
1040                                 should_notify_peers =
1041                                         bond_should_notify_peers(bond);
1042                         }
1043
1044                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1045                         if (should_notify_peers) {
1046                                 bond->send_peer_notif--;
1047                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1048                                                          bond->dev);
1049                         }
1050                 }
1051         }
1052
1053 #ifdef CONFIG_XFRM_OFFLOAD
1054         if (new_active && bond->xs) {
1055                 xfrm_dev_state_flush(dev_net(bond->dev), bond->dev, true);
1056                 bond_ipsec_add_sa(bond->xs);
1057         }
1058 #endif /* CONFIG_XFRM_OFFLOAD */
1059
1060         /* resend IGMP joins since active slave has changed or
1061          * all were sent on curr_active_slave.
1062          * resend only if bond is brought up with the affected
1063          * bonding modes and the retransmission is enabled
1064          */
1065         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1066             ((bond_uses_primary(bond) && new_active) ||
1067              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1068                 bond->igmp_retrans = bond->params.resend_igmp;
1069                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1070         }
1071 }
1072
1073 /**
1074  * bond_select_active_slave - select a new active slave, if needed
1075  * @bond: our bonding struct
1076  *
1077  * This functions should be called when one of the following occurs:
1078  * - The old curr_active_slave has been released or lost its link.
1079  * - The primary_slave has got its link back.
1080  * - A slave has got its link back and there's no old curr_active_slave.
1081  *
1082  * Caller must hold RTNL.
1083  */
1084 void bond_select_active_slave(struct bonding *bond)
1085 {
1086         struct slave *best_slave;
1087         int rv;
1088
1089         ASSERT_RTNL();
1090
1091         best_slave = bond_find_best_slave(bond);
1092         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1093                 bond_change_active_slave(bond, best_slave);
1094                 rv = bond_set_carrier(bond);
1095                 if (!rv)
1096                         return;
1097
1098                 if (netif_carrier_ok(bond->dev))
1099                         netdev_info(bond->dev, "active interface up!\n");
1100                 else
1101                         netdev_info(bond->dev, "now running without any active interface!\n");
1102         }
1103 }
1104
1105 #ifdef CONFIG_NET_POLL_CONTROLLER
1106 static inline int slave_enable_netpoll(struct slave *slave)
1107 {
1108         struct netpoll *np;
1109         int err = 0;
1110
1111         np = kzalloc(sizeof(*np), GFP_KERNEL);
1112         err = -ENOMEM;
1113         if (!np)
1114                 goto out;
1115
1116         err = __netpoll_setup(np, slave->dev);
1117         if (err) {
1118                 kfree(np);
1119                 goto out;
1120         }
1121         slave->np = np;
1122 out:
1123         return err;
1124 }
1125 static inline void slave_disable_netpoll(struct slave *slave)
1126 {
1127         struct netpoll *np = slave->np;
1128
1129         if (!np)
1130                 return;
1131
1132         slave->np = NULL;
1133
1134         __netpoll_free(np);
1135 }
1136
1137 static void bond_poll_controller(struct net_device *bond_dev)
1138 {
1139         struct bonding *bond = netdev_priv(bond_dev);
1140         struct slave *slave = NULL;
1141         struct list_head *iter;
1142         struct ad_info ad_info;
1143
1144         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1145                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1146                         return;
1147
1148         bond_for_each_slave_rcu(bond, slave, iter) {
1149                 if (!bond_slave_is_up(slave))
1150                         continue;
1151
1152                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1153                         struct aggregator *agg =
1154                             SLAVE_AD_INFO(slave)->port.aggregator;
1155
1156                         if (agg &&
1157                             agg->aggregator_identifier != ad_info.aggregator_id)
1158                                 continue;
1159                 }
1160
1161                 netpoll_poll_dev(slave->dev);
1162         }
1163 }
1164
1165 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1166 {
1167         struct bonding *bond = netdev_priv(bond_dev);
1168         struct list_head *iter;
1169         struct slave *slave;
1170
1171         bond_for_each_slave(bond, slave, iter)
1172                 if (bond_slave_is_up(slave))
1173                         slave_disable_netpoll(slave);
1174 }
1175
1176 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1177 {
1178         struct bonding *bond = netdev_priv(dev);
1179         struct list_head *iter;
1180         struct slave *slave;
1181         int err = 0;
1182
1183         bond_for_each_slave(bond, slave, iter) {
1184                 err = slave_enable_netpoll(slave);
1185                 if (err) {
1186                         bond_netpoll_cleanup(dev);
1187                         break;
1188                 }
1189         }
1190         return err;
1191 }
1192 #else
1193 static inline int slave_enable_netpoll(struct slave *slave)
1194 {
1195         return 0;
1196 }
1197 static inline void slave_disable_netpoll(struct slave *slave)
1198 {
1199 }
1200 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1201 {
1202 }
1203 #endif
1204
1205 /*---------------------------------- IOCTL ----------------------------------*/
1206
1207 static netdev_features_t bond_fix_features(struct net_device *dev,
1208                                            netdev_features_t features)
1209 {
1210         struct bonding *bond = netdev_priv(dev);
1211         struct list_head *iter;
1212         netdev_features_t mask;
1213         struct slave *slave;
1214
1215         mask = features;
1216
1217         features &= ~NETIF_F_ONE_FOR_ALL;
1218         features |= NETIF_F_ALL_FOR_ALL;
1219
1220         bond_for_each_slave(bond, slave, iter) {
1221                 features = netdev_increment_features(features,
1222                                                      slave->dev->features,
1223                                                      mask);
1224         }
1225         features = netdev_add_tso_features(features, mask);
1226
1227         return features;
1228 }
1229
1230 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1231                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1232                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1233
1234 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1235                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1236
1237 #define BOND_MPLS_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1238                                  NETIF_F_ALL_TSO)
1239
1240
1241 static void bond_compute_features(struct bonding *bond)
1242 {
1243         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1244                                         IFF_XMIT_DST_RELEASE_PERM;
1245         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1246         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1247 #ifdef CONFIG_XFRM_OFFLOAD
1248         netdev_features_t xfrm_features  = BOND_XFRM_FEATURES;
1249 #endif /* CONFIG_XFRM_OFFLOAD */
1250         netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
1251         struct net_device *bond_dev = bond->dev;
1252         struct list_head *iter;
1253         struct slave *slave;
1254         unsigned short max_hard_header_len = ETH_HLEN;
1255         unsigned int gso_max_size = GSO_MAX_SIZE;
1256         u16 gso_max_segs = GSO_MAX_SEGS;
1257
1258         if (!bond_has_slaves(bond))
1259                 goto done;
1260         vlan_features &= NETIF_F_ALL_FOR_ALL;
1261         mpls_features &= NETIF_F_ALL_FOR_ALL;
1262
1263         bond_for_each_slave(bond, slave, iter) {
1264                 vlan_features = netdev_increment_features(vlan_features,
1265                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1266
1267                 enc_features = netdev_increment_features(enc_features,
1268                                                          slave->dev->hw_enc_features,
1269                                                          BOND_ENC_FEATURES);
1270
1271 #ifdef CONFIG_XFRM_OFFLOAD
1272                 xfrm_features = netdev_increment_features(xfrm_features,
1273                                                           slave->dev->hw_enc_features,
1274                                                           BOND_XFRM_FEATURES);
1275 #endif /* CONFIG_XFRM_OFFLOAD */
1276
1277                 mpls_features = netdev_increment_features(mpls_features,
1278                                                           slave->dev->mpls_features,
1279                                                           BOND_MPLS_FEATURES);
1280
1281                 dst_release_flag &= slave->dev->priv_flags;
1282                 if (slave->dev->hard_header_len > max_hard_header_len)
1283                         max_hard_header_len = slave->dev->hard_header_len;
1284
1285                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1286                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1287         }
1288         bond_dev->hard_header_len = max_hard_header_len;
1289
1290 done:
1291         bond_dev->vlan_features = vlan_features;
1292         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1293                                     NETIF_F_HW_VLAN_CTAG_TX |
1294                                     NETIF_F_HW_VLAN_STAG_TX |
1295                                     NETIF_F_GSO_UDP_L4;
1296 #ifdef CONFIG_XFRM_OFFLOAD
1297         bond_dev->hw_enc_features |= xfrm_features;
1298 #endif /* CONFIG_XFRM_OFFLOAD */
1299         bond_dev->mpls_features = mpls_features;
1300         bond_dev->gso_max_segs = gso_max_segs;
1301         netif_set_gso_max_size(bond_dev, gso_max_size);
1302
1303         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1304         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1305             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1306                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1307
1308         netdev_change_features(bond_dev);
1309 }
1310
1311 static void bond_setup_by_slave(struct net_device *bond_dev,
1312                                 struct net_device *slave_dev)
1313 {
1314         bond_dev->header_ops        = slave_dev->header_ops;
1315
1316         bond_dev->type              = slave_dev->type;
1317         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1318         bond_dev->needed_headroom   = slave_dev->needed_headroom;
1319         bond_dev->addr_len          = slave_dev->addr_len;
1320
1321         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1322                 slave_dev->addr_len);
1323 }
1324
1325 /* On bonding slaves other than the currently active slave, suppress
1326  * duplicates except for alb non-mcast/bcast.
1327  */
1328 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1329                                             struct slave *slave,
1330                                             struct bonding *bond)
1331 {
1332         if (bond_is_slave_inactive(slave)) {
1333                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1334                     skb->pkt_type != PACKET_BROADCAST &&
1335                     skb->pkt_type != PACKET_MULTICAST)
1336                         return false;
1337                 return true;
1338         }
1339         return false;
1340 }
1341
1342 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1343 {
1344         struct sk_buff *skb = *pskb;
1345         struct slave *slave;
1346         struct bonding *bond;
1347         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1348                           struct slave *);
1349         int ret = RX_HANDLER_ANOTHER;
1350
1351         skb = skb_share_check(skb, GFP_ATOMIC);
1352         if (unlikely(!skb))
1353                 return RX_HANDLER_CONSUMED;
1354
1355         *pskb = skb;
1356
1357         slave = bond_slave_get_rcu(skb->dev);
1358         bond = slave->bond;
1359
1360         recv_probe = READ_ONCE(bond->recv_probe);
1361         if (recv_probe) {
1362                 ret = recv_probe(skb, bond, slave);
1363                 if (ret == RX_HANDLER_CONSUMED) {
1364                         consume_skb(skb);
1365                         return ret;
1366                 }
1367         }
1368
1369         /*
1370          * For packets determined by bond_should_deliver_exact_match() call to
1371          * be suppressed we want to make an exception for link-local packets.
1372          * This is necessary for e.g. LLDP daemons to be able to monitor
1373          * inactive slave links without being forced to bind to them
1374          * explicitly.
1375          *
1376          * At the same time, packets that are passed to the bonding master
1377          * (including link-local ones) can have their originating interface
1378          * determined via PACKET_ORIGDEV socket option.
1379          */
1380         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1381                 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1382                         return RX_HANDLER_PASS;
1383                 return RX_HANDLER_EXACT;
1384         }
1385
1386         skb->dev = bond->dev;
1387
1388         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1389             netif_is_bridge_port(bond->dev) &&
1390             skb->pkt_type == PACKET_HOST) {
1391
1392                 if (unlikely(skb_cow_head(skb,
1393                                           skb->data - skb_mac_header(skb)))) {
1394                         kfree_skb(skb);
1395                         return RX_HANDLER_CONSUMED;
1396                 }
1397                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1398                                   bond->dev->addr_len);
1399         }
1400
1401         return ret;
1402 }
1403
1404 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1405 {
1406         switch (BOND_MODE(bond)) {
1407         case BOND_MODE_ROUNDROBIN:
1408                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1409         case BOND_MODE_ACTIVEBACKUP:
1410                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1411         case BOND_MODE_BROADCAST:
1412                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1413         case BOND_MODE_XOR:
1414         case BOND_MODE_8023AD:
1415                 return NETDEV_LAG_TX_TYPE_HASH;
1416         default:
1417                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1418         }
1419 }
1420
1421 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1422                                                enum netdev_lag_tx_type type)
1423 {
1424         if (type != NETDEV_LAG_TX_TYPE_HASH)
1425                 return NETDEV_LAG_HASH_NONE;
1426
1427         switch (bond->params.xmit_policy) {
1428         case BOND_XMIT_POLICY_LAYER2:
1429                 return NETDEV_LAG_HASH_L2;
1430         case BOND_XMIT_POLICY_LAYER34:
1431                 return NETDEV_LAG_HASH_L34;
1432         case BOND_XMIT_POLICY_LAYER23:
1433                 return NETDEV_LAG_HASH_L23;
1434         case BOND_XMIT_POLICY_ENCAP23:
1435                 return NETDEV_LAG_HASH_E23;
1436         case BOND_XMIT_POLICY_ENCAP34:
1437                 return NETDEV_LAG_HASH_E34;
1438         default:
1439                 return NETDEV_LAG_HASH_UNKNOWN;
1440         }
1441 }
1442
1443 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1444                                       struct netlink_ext_ack *extack)
1445 {
1446         struct netdev_lag_upper_info lag_upper_info;
1447         enum netdev_lag_tx_type type;
1448
1449         type = bond_lag_tx_type(bond);
1450         lag_upper_info.tx_type = type;
1451         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1452
1453         return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1454                                             &lag_upper_info, extack);
1455 }
1456
1457 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1458 {
1459         netdev_upper_dev_unlink(slave->dev, bond->dev);
1460         slave->dev->flags &= ~IFF_SLAVE;
1461 }
1462
1463 static struct slave *bond_alloc_slave(struct bonding *bond)
1464 {
1465         struct slave *slave = NULL;
1466
1467         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1468         if (!slave)
1469                 return NULL;
1470
1471         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1472                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1473                                                GFP_KERNEL);
1474                 if (!SLAVE_AD_INFO(slave)) {
1475                         kfree(slave);
1476                         return NULL;
1477                 }
1478         }
1479         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1480
1481         return slave;
1482 }
1483
1484 static void bond_free_slave(struct slave *slave)
1485 {
1486         struct bonding *bond = bond_get_bond_by_slave(slave);
1487
1488         cancel_delayed_work_sync(&slave->notify_work);
1489         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1490                 kfree(SLAVE_AD_INFO(slave));
1491
1492         kfree(slave);
1493 }
1494
1495 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1496 {
1497         info->bond_mode = BOND_MODE(bond);
1498         info->miimon = bond->params.miimon;
1499         info->num_slaves = bond->slave_cnt;
1500 }
1501
1502 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1503 {
1504         strcpy(info->slave_name, slave->dev->name);
1505         info->link = slave->link;
1506         info->state = bond_slave_state(slave);
1507         info->link_failure_count = slave->link_failure_count;
1508 }
1509
1510 static void bond_netdev_notify_work(struct work_struct *_work)
1511 {
1512         struct slave *slave = container_of(_work, struct slave,
1513                                            notify_work.work);
1514
1515         if (rtnl_trylock()) {
1516                 struct netdev_bonding_info binfo;
1517
1518                 bond_fill_ifslave(slave, &binfo.slave);
1519                 bond_fill_ifbond(slave->bond, &binfo.master);
1520                 netdev_bonding_info_change(slave->dev, &binfo);
1521                 rtnl_unlock();
1522         } else {
1523                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1524         }
1525 }
1526
1527 void bond_queue_slave_event(struct slave *slave)
1528 {
1529         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1530 }
1531
1532 void bond_lower_state_changed(struct slave *slave)
1533 {
1534         struct netdev_lag_lower_state_info info;
1535
1536         info.link_up = slave->link == BOND_LINK_UP ||
1537                        slave->link == BOND_LINK_FAIL;
1538         info.tx_enabled = bond_is_active_slave(slave);
1539         netdev_lower_state_changed(slave->dev, &info);
1540 }
1541
1542 /* enslave device <slave> to bond device <master> */
1543 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1544                  struct netlink_ext_ack *extack)
1545 {
1546         struct bonding *bond = netdev_priv(bond_dev);
1547         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1548         struct slave *new_slave = NULL, *prev_slave;
1549         struct sockaddr_storage ss;
1550         int link_reporting;
1551         int res = 0, i;
1552
1553         if (!bond->params.use_carrier &&
1554             slave_dev->ethtool_ops->get_link == NULL &&
1555             slave_ops->ndo_do_ioctl == NULL) {
1556                 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1557         }
1558
1559         /* already in-use? */
1560         if (netdev_is_rx_handler_busy(slave_dev)) {
1561                 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1562                 slave_err(bond_dev, slave_dev,
1563                           "Error: Device is in use and cannot be enslaved\n");
1564                 return -EBUSY;
1565         }
1566
1567         if (bond_dev == slave_dev) {
1568                 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1569                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1570                 return -EPERM;
1571         }
1572
1573         /* vlan challenged mutual exclusion */
1574         /* no need to lock since we're protected by rtnl_lock */
1575         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1576                 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1577                 if (vlan_uses_dev(bond_dev)) {
1578                         NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1579                         slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1580                         return -EPERM;
1581                 } else {
1582                         slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1583                 }
1584         } else {
1585                 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1586         }
1587
1588         if (slave_dev->features & NETIF_F_HW_ESP)
1589                 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1590
1591         /* Old ifenslave binaries are no longer supported.  These can
1592          * be identified with moderate accuracy by the state of the slave:
1593          * the current ifenslave will set the interface down prior to
1594          * enslaving it; the old ifenslave will not.
1595          */
1596         if (slave_dev->flags & IFF_UP) {
1597                 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1598                 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1599                 return -EPERM;
1600         }
1601
1602         /* set bonding device ether type by slave - bonding netdevices are
1603          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1604          * there is a need to override some of the type dependent attribs/funcs.
1605          *
1606          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1607          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1608          */
1609         if (!bond_has_slaves(bond)) {
1610                 if (bond_dev->type != slave_dev->type) {
1611                         slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1612                                   bond_dev->type, slave_dev->type);
1613
1614                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1615                                                        bond_dev);
1616                         res = notifier_to_errno(res);
1617                         if (res) {
1618                                 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1619                                 return -EBUSY;
1620                         }
1621
1622                         /* Flush unicast and multicast addresses */
1623                         dev_uc_flush(bond_dev);
1624                         dev_mc_flush(bond_dev);
1625
1626                         if (slave_dev->type != ARPHRD_ETHER)
1627                                 bond_setup_by_slave(bond_dev, slave_dev);
1628                         else {
1629                                 ether_setup(bond_dev);
1630                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1631                         }
1632
1633                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1634                                                  bond_dev);
1635                 }
1636         } else if (bond_dev->type != slave_dev->type) {
1637                 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1638                 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1639                           slave_dev->type, bond_dev->type);
1640                 return -EINVAL;
1641         }
1642
1643         if (slave_dev->type == ARPHRD_INFINIBAND &&
1644             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1645                 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1646                 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1647                            slave_dev->type);
1648                 res = -EOPNOTSUPP;
1649                 goto err_undo_flags;
1650         }
1651
1652         if (!slave_ops->ndo_set_mac_address ||
1653             slave_dev->type == ARPHRD_INFINIBAND) {
1654                 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1655                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1656                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1657                         if (!bond_has_slaves(bond)) {
1658                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1659                                 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1660                         } else {
1661                                 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1662                                 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");
1663                                 res = -EOPNOTSUPP;
1664                                 goto err_undo_flags;
1665                         }
1666                 }
1667         }
1668
1669         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1670
1671         /* If this is the first slave, then we need to set the master's hardware
1672          * address to be the same as the slave's.
1673          */
1674         if (!bond_has_slaves(bond) &&
1675             bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1676                 res = bond_set_dev_addr(bond->dev, slave_dev);
1677                 if (res)
1678                         goto err_undo_flags;
1679         }
1680
1681         new_slave = bond_alloc_slave(bond);
1682         if (!new_slave) {
1683                 res = -ENOMEM;
1684                 goto err_undo_flags;
1685         }
1686
1687         new_slave->bond = bond;
1688         new_slave->dev = slave_dev;
1689         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1690          * is set via sysfs or module option if desired.
1691          */
1692         new_slave->queue_id = 0;
1693
1694         /* Save slave's original mtu and then set it to match the bond */
1695         new_slave->original_mtu = slave_dev->mtu;
1696         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1697         if (res) {
1698                 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1699                 goto err_free;
1700         }
1701
1702         /* Save slave's original ("permanent") mac address for modes
1703          * that need it, and for restoring it upon release, and then
1704          * set it to the master's address
1705          */
1706         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1707                           slave_dev->addr_len);
1708
1709         if (!bond->params.fail_over_mac ||
1710             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1711                 /* Set slave to master's mac address.  The application already
1712                  * set the master's mac address to that of the first slave
1713                  */
1714                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1715                 ss.ss_family = slave_dev->type;
1716                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1717                                           extack);
1718                 if (res) {
1719                         slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1720                         goto err_restore_mtu;
1721                 }
1722         }
1723
1724         /* set slave flag before open to prevent IPv6 addrconf */
1725         slave_dev->flags |= IFF_SLAVE;
1726
1727         /* open the slave since the application closed it */
1728         res = dev_open(slave_dev, extack);
1729         if (res) {
1730                 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1731                 goto err_restore_mac;
1732         }
1733
1734         slave_dev->priv_flags |= IFF_BONDING;
1735         /* initialize slave stats */
1736         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1737
1738         if (bond_is_lb(bond)) {
1739                 /* bond_alb_init_slave() must be called before all other stages since
1740                  * it might fail and we do not want to have to undo everything
1741                  */
1742                 res = bond_alb_init_slave(bond, new_slave);
1743                 if (res)
1744                         goto err_close;
1745         }
1746
1747         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1748         if (res) {
1749                 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1750                 goto err_close;
1751         }
1752
1753         prev_slave = bond_last_slave(bond);
1754
1755         new_slave->delay = 0;
1756         new_slave->link_failure_count = 0;
1757
1758         if (bond_update_speed_duplex(new_slave) &&
1759             bond_needs_speed_duplex(bond))
1760                 new_slave->link = BOND_LINK_DOWN;
1761
1762         new_slave->last_rx = jiffies -
1763                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1764         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1765                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1766
1767         if (bond->params.miimon && !bond->params.use_carrier) {
1768                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1769
1770                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1771                         /* miimon is set but a bonded network driver
1772                          * does not support ETHTOOL/MII and
1773                          * arp_interval is not set.  Note: if
1774                          * use_carrier is enabled, we will never go
1775                          * here (because netif_carrier is always
1776                          * supported); thus, we don't need to change
1777                          * the messages for netif_carrier.
1778                          */
1779                         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");
1780                 } else if (link_reporting == -1) {
1781                         /* unable get link status using mii/ethtool */
1782                         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");
1783                 }
1784         }
1785
1786         /* check for initial state */
1787         new_slave->link = BOND_LINK_NOCHANGE;
1788         if (bond->params.miimon) {
1789                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1790                         if (bond->params.updelay) {
1791                                 bond_set_slave_link_state(new_slave,
1792                                                           BOND_LINK_BACK,
1793                                                           BOND_SLAVE_NOTIFY_NOW);
1794                                 new_slave->delay = bond->params.updelay;
1795                         } else {
1796                                 bond_set_slave_link_state(new_slave,
1797                                                           BOND_LINK_UP,
1798                                                           BOND_SLAVE_NOTIFY_NOW);
1799                         }
1800                 } else {
1801                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1802                                                   BOND_SLAVE_NOTIFY_NOW);
1803                 }
1804         } else if (bond->params.arp_interval) {
1805                 bond_set_slave_link_state(new_slave,
1806                                           (netif_carrier_ok(slave_dev) ?
1807                                           BOND_LINK_UP : BOND_LINK_DOWN),
1808                                           BOND_SLAVE_NOTIFY_NOW);
1809         } else {
1810                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1811                                           BOND_SLAVE_NOTIFY_NOW);
1812         }
1813
1814         if (new_slave->link != BOND_LINK_DOWN)
1815                 new_slave->last_link_up = jiffies;
1816         slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1817                   new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1818                   (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1819
1820         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1821                 /* if there is a primary slave, remember it */
1822                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1823                         rcu_assign_pointer(bond->primary_slave, new_slave);
1824                         bond->force_primary = true;
1825                 }
1826         }
1827
1828         switch (BOND_MODE(bond)) {
1829         case BOND_MODE_ACTIVEBACKUP:
1830                 bond_set_slave_inactive_flags(new_slave,
1831                                               BOND_SLAVE_NOTIFY_NOW);
1832                 break;
1833         case BOND_MODE_8023AD:
1834                 /* in 802.3ad mode, the internal mechanism
1835                  * will activate the slaves in the selected
1836                  * aggregator
1837                  */
1838                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1839                 /* if this is the first slave */
1840                 if (!prev_slave) {
1841                         SLAVE_AD_INFO(new_slave)->id = 1;
1842                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1843                          * can be called only after the mac address of the bond is set
1844                          */
1845                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1846                 } else {
1847                         SLAVE_AD_INFO(new_slave)->id =
1848                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1849                 }
1850
1851                 bond_3ad_bind_slave(new_slave);
1852                 break;
1853         case BOND_MODE_TLB:
1854         case BOND_MODE_ALB:
1855                 bond_set_active_slave(new_slave);
1856                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1857                 break;
1858         default:
1859                 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
1860
1861                 /* always active in trunk mode */
1862                 bond_set_active_slave(new_slave);
1863
1864                 /* In trunking mode there is little meaning to curr_active_slave
1865                  * anyway (it holds no special properties of the bond device),
1866                  * so we can change it without calling change_active_interface()
1867                  */
1868                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1869                     new_slave->link == BOND_LINK_UP)
1870                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1871
1872                 break;
1873         } /* switch(bond_mode) */
1874
1875 #ifdef CONFIG_NET_POLL_CONTROLLER
1876         if (bond->dev->npinfo) {
1877                 if (slave_enable_netpoll(new_slave)) {
1878                         slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1879                         res = -EBUSY;
1880                         goto err_detach;
1881                 }
1882         }
1883 #endif
1884
1885         if (!(bond_dev->features & NETIF_F_LRO))
1886                 dev_disable_lro(slave_dev);
1887
1888         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1889                                          new_slave);
1890         if (res) {
1891                 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
1892                 goto err_detach;
1893         }
1894
1895         res = bond_master_upper_dev_link(bond, new_slave, extack);
1896         if (res) {
1897                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1898                 goto err_unregister;
1899         }
1900
1901         res = bond_sysfs_slave_add(new_slave);
1902         if (res) {
1903                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1904                 goto err_upper_unlink;
1905         }
1906
1907         /* If the mode uses primary, then the following is handled by
1908          * bond_change_active_slave().
1909          */
1910         if (!bond_uses_primary(bond)) {
1911                 /* set promiscuity level to new slave */
1912                 if (bond_dev->flags & IFF_PROMISC) {
1913                         res = dev_set_promiscuity(slave_dev, 1);
1914                         if (res)
1915                                 goto err_sysfs_del;
1916                 }
1917
1918                 /* set allmulti level to new slave */
1919                 if (bond_dev->flags & IFF_ALLMULTI) {
1920                         res = dev_set_allmulti(slave_dev, 1);
1921                         if (res) {
1922                                 if (bond_dev->flags & IFF_PROMISC)
1923                                         dev_set_promiscuity(slave_dev, -1);
1924                                 goto err_sysfs_del;
1925                         }
1926                 }
1927
1928                 netif_addr_lock_bh(bond_dev);
1929                 dev_mc_sync_multiple(slave_dev, bond_dev);
1930                 dev_uc_sync_multiple(slave_dev, bond_dev);
1931                 netif_addr_unlock_bh(bond_dev);
1932
1933                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1934                         /* add lacpdu mc addr to mc list */
1935                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1936
1937                         dev_mc_add(slave_dev, lacpdu_multicast);
1938                 }
1939         }
1940
1941         bond->slave_cnt++;
1942         bond_compute_features(bond);
1943         bond_set_carrier(bond);
1944
1945         if (bond_uses_primary(bond)) {
1946                 block_netpoll_tx();
1947                 bond_select_active_slave(bond);
1948                 unblock_netpoll_tx();
1949         }
1950
1951         if (bond_mode_can_use_xmit_hash(bond))
1952                 bond_update_slave_arr(bond, NULL);
1953
1954
1955         slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
1956                    bond_is_active_slave(new_slave) ? "an active" : "a backup",
1957                    new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1958
1959         /* enslave is successful */
1960         bond_queue_slave_event(new_slave);
1961         return 0;
1962
1963 /* Undo stages on error */
1964 err_sysfs_del:
1965         bond_sysfs_slave_del(new_slave);
1966
1967 err_upper_unlink:
1968         bond_upper_dev_unlink(bond, new_slave);
1969
1970 err_unregister:
1971         netdev_rx_handler_unregister(slave_dev);
1972
1973 err_detach:
1974         vlan_vids_del_by_dev(slave_dev, bond_dev);
1975         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1976                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1977         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1978                 block_netpoll_tx();
1979                 bond_change_active_slave(bond, NULL);
1980                 bond_select_active_slave(bond);
1981                 unblock_netpoll_tx();
1982         }
1983         /* either primary_slave or curr_active_slave might've changed */
1984         synchronize_rcu();
1985         slave_disable_netpoll(new_slave);
1986
1987 err_close:
1988         if (!netif_is_bond_master(slave_dev))
1989                 slave_dev->priv_flags &= ~IFF_BONDING;
1990         dev_close(slave_dev);
1991
1992 err_restore_mac:
1993         slave_dev->flags &= ~IFF_SLAVE;
1994         if (!bond->params.fail_over_mac ||
1995             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1996                 /* XXX TODO - fom follow mode needs to change master's
1997                  * MAC if this slave's MAC is in use by the bond, or at
1998                  * least print a warning.
1999                  */
2000                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2001                                   new_slave->dev->addr_len);
2002                 ss.ss_family = slave_dev->type;
2003                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2004         }
2005
2006 err_restore_mtu:
2007         dev_set_mtu(slave_dev, new_slave->original_mtu);
2008
2009 err_free:
2010         bond_free_slave(new_slave);
2011
2012 err_undo_flags:
2013         /* Enslave of first slave has failed and we need to fix master's mac */
2014         if (!bond_has_slaves(bond)) {
2015                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2016                                             slave_dev->dev_addr))
2017                         eth_hw_addr_random(bond_dev);
2018                 if (bond_dev->type != ARPHRD_ETHER) {
2019                         dev_close(bond_dev);
2020                         ether_setup(bond_dev);
2021                         bond_dev->flags |= IFF_MASTER;
2022                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2023                 }
2024         }
2025
2026         return res;
2027 }
2028
2029 /* Try to release the slave device <slave> from the bond device <master>
2030  * It is legal to access curr_active_slave without a lock because all the function
2031  * is RTNL-locked. If "all" is true it means that the function is being called
2032  * while destroying a bond interface and all slaves are being released.
2033  *
2034  * The rules for slave state should be:
2035  *   for Active/Backup:
2036  *     Active stays on all backups go down
2037  *   for Bonded connections:
2038  *     The first up interface should be left on and all others downed.
2039  */
2040 static int __bond_release_one(struct net_device *bond_dev,
2041                               struct net_device *slave_dev,
2042                               bool all, bool unregister)
2043 {
2044         struct bonding *bond = netdev_priv(bond_dev);
2045         struct slave *slave, *oldcurrent;
2046         struct sockaddr_storage ss;
2047         int old_flags = bond_dev->flags;
2048         netdev_features_t old_features = bond_dev->features;
2049
2050         /* slave is not a slave or master is not master of this slave */
2051         if (!(slave_dev->flags & IFF_SLAVE) ||
2052             !netdev_has_upper_dev(slave_dev, bond_dev)) {
2053                 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2054                 return -EINVAL;
2055         }
2056
2057         block_netpoll_tx();
2058
2059         slave = bond_get_slave_by_dev(bond, slave_dev);
2060         if (!slave) {
2061                 /* not a slave of this bond */
2062                 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2063                 unblock_netpoll_tx();
2064                 return -EINVAL;
2065         }
2066
2067         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2068
2069         bond_sysfs_slave_del(slave);
2070
2071         /* recompute stats just before removing the slave */
2072         bond_get_stats(bond->dev, &bond->bond_stats);
2073
2074         bond_upper_dev_unlink(bond, slave);
2075         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2076          * for this slave anymore.
2077          */
2078         netdev_rx_handler_unregister(slave_dev);
2079
2080         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2081                 bond_3ad_unbind_slave(slave);
2082
2083         if (bond_mode_can_use_xmit_hash(bond))
2084                 bond_update_slave_arr(bond, slave);
2085
2086         slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2087                     bond_is_active_slave(slave) ? "active" : "backup");
2088
2089         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2090
2091         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2092
2093         if (!all && (!bond->params.fail_over_mac ||
2094                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2095                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2096                     bond_has_slaves(bond))
2097                         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",
2098                                    slave->perm_hwaddr);
2099         }
2100
2101         if (rtnl_dereference(bond->primary_slave) == slave)
2102                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2103
2104         if (oldcurrent == slave)
2105                 bond_change_active_slave(bond, NULL);
2106
2107         if (bond_is_lb(bond)) {
2108                 /* Must be called only after the slave has been
2109                  * detached from the list and the curr_active_slave
2110                  * has been cleared (if our_slave == old_current),
2111                  * but before a new active slave is selected.
2112                  */
2113                 bond_alb_deinit_slave(bond, slave);
2114         }
2115
2116         if (all) {
2117                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2118         } else if (oldcurrent == slave) {
2119                 /* Note that we hold RTNL over this sequence, so there
2120                  * is no concern that another slave add/remove event
2121                  * will interfere.
2122                  */
2123                 bond_select_active_slave(bond);
2124         }
2125
2126         if (!bond_has_slaves(bond)) {
2127                 bond_set_carrier(bond);
2128                 eth_hw_addr_random(bond_dev);
2129         }
2130
2131         unblock_netpoll_tx();
2132         synchronize_rcu();
2133         bond->slave_cnt--;
2134
2135         if (!bond_has_slaves(bond)) {
2136                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2137                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2138         }
2139
2140         bond_compute_features(bond);
2141         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2142             (old_features & NETIF_F_VLAN_CHALLENGED))
2143                 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2144
2145         vlan_vids_del_by_dev(slave_dev, bond_dev);
2146
2147         /* If the mode uses primary, then this case was handled above by
2148          * bond_change_active_slave(..., NULL)
2149          */
2150         if (!bond_uses_primary(bond)) {
2151                 /* unset promiscuity level from slave
2152                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
2153                  * of the IFF_PROMISC flag in the bond_dev, but we need the
2154                  * value of that flag before that change, as that was the value
2155                  * when this slave was attached, so we cache at the start of the
2156                  * function and use it here. Same goes for ALLMULTI below
2157                  */
2158                 if (old_flags & IFF_PROMISC)
2159                         dev_set_promiscuity(slave_dev, -1);
2160
2161                 /* unset allmulti level from slave */
2162                 if (old_flags & IFF_ALLMULTI)
2163                         dev_set_allmulti(slave_dev, -1);
2164
2165                 bond_hw_addr_flush(bond_dev, slave_dev);
2166         }
2167
2168         slave_disable_netpoll(slave);
2169
2170         /* close slave before restoring its mac address */
2171         dev_close(slave_dev);
2172
2173         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2174             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2175                 /* restore original ("permanent") mac address */
2176                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2177                                   slave->dev->addr_len);
2178                 ss.ss_family = slave_dev->type;
2179                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2180         }
2181
2182         if (unregister)
2183                 __dev_set_mtu(slave_dev, slave->original_mtu);
2184         else
2185                 dev_set_mtu(slave_dev, slave->original_mtu);
2186
2187         if (!netif_is_bond_master(slave_dev))
2188                 slave_dev->priv_flags &= ~IFF_BONDING;
2189
2190         bond_free_slave(slave);
2191
2192         return 0;
2193 }
2194
2195 /* A wrapper used because of ndo_del_link */
2196 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2197 {
2198         return __bond_release_one(bond_dev, slave_dev, false, false);
2199 }
2200
2201 /* First release a slave and then destroy the bond if no more slaves are left.
2202  * Must be under rtnl_lock when this function is called.
2203  */
2204 static int bond_release_and_destroy(struct net_device *bond_dev,
2205                                     struct net_device *slave_dev)
2206 {
2207         struct bonding *bond = netdev_priv(bond_dev);
2208         int ret;
2209
2210         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2211         if (ret == 0 && !bond_has_slaves(bond) &&
2212             bond_dev->reg_state != NETREG_UNREGISTERING) {
2213                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2214                 netdev_info(bond_dev, "Destroying bond\n");
2215                 bond_remove_proc_entry(bond);
2216                 unregister_netdevice(bond_dev);
2217         }
2218         return ret;
2219 }
2220
2221 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2222 {
2223         struct bonding *bond = netdev_priv(bond_dev);
2224         bond_fill_ifbond(bond, info);
2225 }
2226
2227 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2228 {
2229         struct bonding *bond = netdev_priv(bond_dev);
2230         struct list_head *iter;
2231         int i = 0, res = -ENODEV;
2232         struct slave *slave;
2233
2234         bond_for_each_slave(bond, slave, iter) {
2235                 if (i++ == (int)info->slave_id) {
2236                         res = 0;
2237                         bond_fill_ifslave(slave, info);
2238                         break;
2239                 }
2240         }
2241
2242         return res;
2243 }
2244
2245 /*-------------------------------- Monitoring -------------------------------*/
2246
2247 /* called with rcu_read_lock() */
2248 static int bond_miimon_inspect(struct bonding *bond)
2249 {
2250         int link_state, commit = 0;
2251         struct list_head *iter;
2252         struct slave *slave;
2253         bool ignore_updelay;
2254
2255         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2256
2257         bond_for_each_slave_rcu(bond, slave, iter) {
2258                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2259
2260                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2261
2262                 switch (slave->link) {
2263                 case BOND_LINK_UP:
2264                         if (link_state)
2265                                 continue;
2266
2267                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2268                         commit++;
2269                         slave->delay = bond->params.downdelay;
2270                         if (slave->delay) {
2271                                 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2272                                            (BOND_MODE(bond) ==
2273                                             BOND_MODE_ACTIVEBACKUP) ?
2274                                             (bond_is_active_slave(slave) ?
2275                                              "active " : "backup ") : "",
2276                                            bond->params.downdelay * bond->params.miimon);
2277                         }
2278                         fallthrough;
2279                 case BOND_LINK_FAIL:
2280                         if (link_state) {
2281                                 /* recovered before downdelay expired */
2282                                 bond_propose_link_state(slave, BOND_LINK_UP);
2283                                 slave->last_link_up = jiffies;
2284                                 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2285                                            (bond->params.downdelay - slave->delay) *
2286                                            bond->params.miimon);
2287                                 commit++;
2288                                 continue;
2289                         }
2290
2291                         if (slave->delay <= 0) {
2292                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2293                                 commit++;
2294                                 continue;
2295                         }
2296
2297                         slave->delay--;
2298                         break;
2299
2300                 case BOND_LINK_DOWN:
2301                         if (!link_state)
2302                                 continue;
2303
2304                         bond_propose_link_state(slave, BOND_LINK_BACK);
2305                         commit++;
2306                         slave->delay = bond->params.updelay;
2307
2308                         if (slave->delay) {
2309                                 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2310                                            ignore_updelay ? 0 :
2311                                            bond->params.updelay *
2312                                            bond->params.miimon);
2313                         }
2314                         fallthrough;
2315                 case BOND_LINK_BACK:
2316                         if (!link_state) {
2317                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2318                                 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2319                                            (bond->params.updelay - slave->delay) *
2320                                            bond->params.miimon);
2321                                 commit++;
2322                                 continue;
2323                         }
2324
2325                         if (ignore_updelay)
2326                                 slave->delay = 0;
2327
2328                         if (slave->delay <= 0) {
2329                                 bond_propose_link_state(slave, BOND_LINK_UP);
2330                                 commit++;
2331                                 ignore_updelay = false;
2332                                 continue;
2333                         }
2334
2335                         slave->delay--;
2336                         break;
2337                 }
2338         }
2339
2340         return commit;
2341 }
2342
2343 static void bond_miimon_link_change(struct bonding *bond,
2344                                     struct slave *slave,
2345                                     char link)
2346 {
2347         switch (BOND_MODE(bond)) {
2348         case BOND_MODE_8023AD:
2349                 bond_3ad_handle_link_change(slave, link);
2350                 break;
2351         case BOND_MODE_TLB:
2352         case BOND_MODE_ALB:
2353                 bond_alb_handle_link_change(bond, slave, link);
2354                 break;
2355         case BOND_MODE_XOR:
2356                 bond_update_slave_arr(bond, NULL);
2357                 break;
2358         }
2359 }
2360
2361 static void bond_miimon_commit(struct bonding *bond)
2362 {
2363         struct list_head *iter;
2364         struct slave *slave, *primary;
2365
2366         bond_for_each_slave(bond, slave, iter) {
2367                 switch (slave->link_new_state) {
2368                 case BOND_LINK_NOCHANGE:
2369                         /* For 802.3ad mode, check current slave speed and
2370                          * duplex again in case its port was disabled after
2371                          * invalid speed/duplex reporting but recovered before
2372                          * link monitoring could make a decision on the actual
2373                          * link status
2374                          */
2375                         if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2376                             slave->link == BOND_LINK_UP)
2377                                 bond_3ad_adapter_speed_duplex_changed(slave);
2378                         continue;
2379
2380                 case BOND_LINK_UP:
2381                         if (bond_update_speed_duplex(slave) &&
2382                             bond_needs_speed_duplex(bond)) {
2383                                 slave->link = BOND_LINK_DOWN;
2384                                 if (net_ratelimit())
2385                                         slave_warn(bond->dev, slave->dev,
2386                                                    "failed to get link speed/duplex\n");
2387                                 continue;
2388                         }
2389                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2390                                                   BOND_SLAVE_NOTIFY_NOW);
2391                         slave->last_link_up = jiffies;
2392
2393                         primary = rtnl_dereference(bond->primary_slave);
2394                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2395                                 /* prevent it from being the active one */
2396                                 bond_set_backup_slave(slave);
2397                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2398                                 /* make it immediately active */
2399                                 bond_set_active_slave(slave);
2400                         }
2401
2402                         slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2403                                    slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2404                                    slave->duplex ? "full" : "half");
2405
2406                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2407
2408                         if (!bond->curr_active_slave || slave == primary)
2409                                 goto do_failover;
2410
2411                         continue;
2412
2413                 case BOND_LINK_DOWN:
2414                         if (slave->link_failure_count < UINT_MAX)
2415                                 slave->link_failure_count++;
2416
2417                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2418                                                   BOND_SLAVE_NOTIFY_NOW);
2419
2420                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2421                             BOND_MODE(bond) == BOND_MODE_8023AD)
2422                                 bond_set_slave_inactive_flags(slave,
2423                                                               BOND_SLAVE_NOTIFY_NOW);
2424
2425                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2426
2427                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2428
2429                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2430                                 goto do_failover;
2431
2432                         continue;
2433
2434                 default:
2435                         slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2436                                   slave->link_new_state);
2437                         bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2438
2439                         continue;
2440                 }
2441
2442 do_failover:
2443                 block_netpoll_tx();
2444                 bond_select_active_slave(bond);
2445                 unblock_netpoll_tx();
2446         }
2447
2448         bond_set_carrier(bond);
2449 }
2450
2451 /* bond_mii_monitor
2452  *
2453  * Really a wrapper that splits the mii monitor into two phases: an
2454  * inspection, then (if inspection indicates something needs to be done)
2455  * an acquisition of appropriate locks followed by a commit phase to
2456  * implement whatever link state changes are indicated.
2457  */
2458 static void bond_mii_monitor(struct work_struct *work)
2459 {
2460         struct bonding *bond = container_of(work, struct bonding,
2461                                             mii_work.work);
2462         bool should_notify_peers = false;
2463         bool commit;
2464         unsigned long delay;
2465         struct slave *slave;
2466         struct list_head *iter;
2467
2468         delay = msecs_to_jiffies(bond->params.miimon);
2469
2470         if (!bond_has_slaves(bond))
2471                 goto re_arm;
2472
2473         rcu_read_lock();
2474         should_notify_peers = bond_should_notify_peers(bond);
2475         commit = !!bond_miimon_inspect(bond);
2476         if (bond->send_peer_notif) {
2477                 rcu_read_unlock();
2478                 if (rtnl_trylock()) {
2479                         bond->send_peer_notif--;
2480                         rtnl_unlock();
2481                 }
2482         } else {
2483                 rcu_read_unlock();
2484         }
2485
2486         if (commit) {
2487                 /* Race avoidance with bond_close cancel of workqueue */
2488                 if (!rtnl_trylock()) {
2489                         delay = 1;
2490                         should_notify_peers = false;
2491                         goto re_arm;
2492                 }
2493
2494                 bond_for_each_slave(bond, slave, iter) {
2495                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2496                 }
2497                 bond_miimon_commit(bond);
2498
2499                 rtnl_unlock();  /* might sleep, hold no other locks */
2500         }
2501
2502 re_arm:
2503         if (bond->params.miimon)
2504                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2505
2506         if (should_notify_peers) {
2507                 if (!rtnl_trylock())
2508                         return;
2509                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2510                 rtnl_unlock();
2511         }
2512 }
2513
2514 static int bond_upper_dev_walk(struct net_device *upper,
2515                                struct netdev_nested_priv *priv)
2516 {
2517         __be32 ip = *(__be32 *)priv->data;
2518
2519         return ip == bond_confirm_addr(upper, 0, ip);
2520 }
2521
2522 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2523 {
2524         struct netdev_nested_priv priv = {
2525                 .data = (void *)&ip,
2526         };
2527         bool ret = false;
2528
2529         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2530                 return true;
2531
2532         rcu_read_lock();
2533         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2534                 ret = true;
2535         rcu_read_unlock();
2536
2537         return ret;
2538 }
2539
2540 /* We go to the (large) trouble of VLAN tagging ARP frames because
2541  * switches in VLAN mode (especially if ports are configured as
2542  * "native" to a VLAN) might not pass non-tagged frames.
2543  */
2544 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2545                           __be32 src_ip, struct bond_vlan_tag *tags)
2546 {
2547         struct sk_buff *skb;
2548         struct bond_vlan_tag *outer_tag = tags;
2549         struct net_device *slave_dev = slave->dev;
2550         struct net_device *bond_dev = slave->bond->dev;
2551
2552         slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2553                   arp_op, &dest_ip, &src_ip);
2554
2555         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2556                          NULL, slave_dev->dev_addr, NULL);
2557
2558         if (!skb) {
2559                 net_err_ratelimited("ARP packet allocation failed\n");
2560                 return;
2561         }
2562
2563         if (!tags || tags->vlan_proto == VLAN_N_VID)
2564                 goto xmit;
2565
2566         tags++;
2567
2568         /* Go through all the tags backwards and add them to the packet */
2569         while (tags->vlan_proto != VLAN_N_VID) {
2570                 if (!tags->vlan_id) {
2571                         tags++;
2572                         continue;
2573                 }
2574
2575                 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2576                           ntohs(outer_tag->vlan_proto), tags->vlan_id);
2577                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2578                                                 tags->vlan_id);
2579                 if (!skb) {
2580                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2581                         return;
2582                 }
2583
2584                 tags++;
2585         }
2586         /* Set the outer tag */
2587         if (outer_tag->vlan_id) {
2588                 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2589                           ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2590                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2591                                        outer_tag->vlan_id);
2592         }
2593
2594 xmit:
2595         arp_xmit(skb);
2596 }
2597
2598 /* Validate the device path between the @start_dev and the @end_dev.
2599  * The path is valid if the @end_dev is reachable through device
2600  * stacking.
2601  * When the path is validated, collect any vlan information in the
2602  * path.
2603  */
2604 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2605                                               struct net_device *end_dev,
2606                                               int level)
2607 {
2608         struct bond_vlan_tag *tags;
2609         struct net_device *upper;
2610         struct list_head  *iter;
2611
2612         if (start_dev == end_dev) {
2613                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2614                 if (!tags)
2615                         return ERR_PTR(-ENOMEM);
2616                 tags[level].vlan_proto = VLAN_N_VID;
2617                 return tags;
2618         }
2619
2620         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2621                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2622                 if (IS_ERR_OR_NULL(tags)) {
2623                         if (IS_ERR(tags))
2624                                 return tags;
2625                         continue;
2626                 }
2627                 if (is_vlan_dev(upper)) {
2628                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2629                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2630                 }
2631
2632                 return tags;
2633         }
2634
2635         return NULL;
2636 }
2637
2638 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2639 {
2640         struct rtable *rt;
2641         struct bond_vlan_tag *tags;
2642         __be32 *targets = bond->params.arp_targets, addr;
2643         int i;
2644
2645         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2646                 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2647                           __func__, &targets[i]);
2648                 tags = NULL;
2649
2650                 /* Find out through which dev should the packet go */
2651                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2652                                      RTO_ONLINK, 0);
2653                 if (IS_ERR(rt)) {
2654                         /* there's no route to target - try to send arp
2655                          * probe to generate any traffic (arp_validate=0)
2656                          */
2657                         if (bond->params.arp_validate)
2658                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2659                                                      bond->dev->name,
2660                                                      &targets[i]);
2661                         bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2662                                       0, tags);
2663                         continue;
2664                 }
2665
2666                 /* bond device itself */
2667                 if (rt->dst.dev == bond->dev)
2668                         goto found;
2669
2670                 rcu_read_lock();
2671                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2672                 rcu_read_unlock();
2673
2674                 if (!IS_ERR_OR_NULL(tags))
2675                         goto found;
2676
2677                 /* Not our device - skip */
2678                 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2679                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2680
2681                 ip_rt_put(rt);
2682                 continue;
2683
2684 found:
2685                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2686                 ip_rt_put(rt);
2687                 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2688                 kfree(tags);
2689         }
2690 }
2691
2692 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2693 {
2694         int i;
2695
2696         if (!sip || !bond_has_this_ip(bond, tip)) {
2697                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2698                            __func__, &sip, &tip);
2699                 return;
2700         }
2701
2702         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2703         if (i == -1) {
2704                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2705                            __func__, &sip);
2706                 return;
2707         }
2708         slave->last_rx = jiffies;
2709         slave->target_last_arp_rx[i] = jiffies;
2710 }
2711
2712 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2713                  struct slave *slave)
2714 {
2715         struct arphdr *arp = (struct arphdr *)skb->data;
2716         struct slave *curr_active_slave, *curr_arp_slave;
2717         unsigned char *arp_ptr;
2718         __be32 sip, tip;
2719         int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2720         unsigned int alen;
2721
2722         if (!slave_do_arp_validate(bond, slave)) {
2723                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2724                     !slave_do_arp_validate_only(bond))
2725                         slave->last_rx = jiffies;
2726                 return RX_HANDLER_ANOTHER;
2727         } else if (!is_arp) {
2728                 return RX_HANDLER_ANOTHER;
2729         }
2730
2731         alen = arp_hdr_len(bond->dev);
2732
2733         slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2734                    __func__, skb->dev->name);
2735
2736         if (alen > skb_headlen(skb)) {
2737                 arp = kmalloc(alen, GFP_ATOMIC);
2738                 if (!arp)
2739                         goto out_unlock;
2740                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2741                         goto out_unlock;
2742         }
2743
2744         if (arp->ar_hln != bond->dev->addr_len ||
2745             skb->pkt_type == PACKET_OTHERHOST ||
2746             skb->pkt_type == PACKET_LOOPBACK ||
2747             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2748             arp->ar_pro != htons(ETH_P_IP) ||
2749             arp->ar_pln != 4)
2750                 goto out_unlock;
2751
2752         arp_ptr = (unsigned char *)(arp + 1);
2753         arp_ptr += bond->dev->addr_len;
2754         memcpy(&sip, arp_ptr, 4);
2755         arp_ptr += 4 + bond->dev->addr_len;
2756         memcpy(&tip, arp_ptr, 4);
2757
2758         slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2759                   __func__, slave->dev->name, bond_slave_state(slave),
2760                   bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2761                   &sip, &tip);
2762
2763         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2764         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2765
2766         /* We 'trust' the received ARP enough to validate it if:
2767          *
2768          * (a) the slave receiving the ARP is active (which includes the
2769          * current ARP slave, if any), or
2770          *
2771          * (b) the receiving slave isn't active, but there is a currently
2772          * active slave and it received valid arp reply(s) after it became
2773          * the currently active slave, or
2774          *
2775          * (c) there is an ARP slave that sent an ARP during the prior ARP
2776          * interval, and we receive an ARP reply on any slave.  We accept
2777          * these because switch FDB update delays may deliver the ARP
2778          * reply to a slave other than the sender of the ARP request.
2779          *
2780          * Note: for (b), backup slaves are receiving the broadcast ARP
2781          * request, not a reply.  This request passes from the sending
2782          * slave through the L2 switch(es) to the receiving slave.  Since
2783          * this is checking the request, sip/tip are swapped for
2784          * validation.
2785          *
2786          * This is done to avoid endless looping when we can't reach the
2787          * arp_ip_target and fool ourselves with our own arp requests.
2788          */
2789         if (bond_is_active_slave(slave))
2790                 bond_validate_arp(bond, slave, sip, tip);
2791         else if (curr_active_slave &&
2792                  time_after(slave_last_rx(bond, curr_active_slave),
2793                             curr_active_slave->last_link_up))
2794                 bond_validate_arp(bond, slave, tip, sip);
2795         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2796                  bond_time_in_interval(bond,
2797                                        dev_trans_start(curr_arp_slave->dev), 1))
2798                 bond_validate_arp(bond, slave, sip, tip);
2799
2800 out_unlock:
2801         if (arp != (struct arphdr *)skb->data)
2802                 kfree(arp);
2803         return RX_HANDLER_ANOTHER;
2804 }
2805
2806 /* function to verify if we're in the arp_interval timeslice, returns true if
2807  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2808  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2809  */
2810 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2811                                   int mod)
2812 {
2813         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2814
2815         return time_in_range(jiffies,
2816                              last_act - delta_in_ticks,
2817                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2818 }
2819
2820 /* This function is called regularly to monitor each slave's link
2821  * ensuring that traffic is being sent and received when arp monitoring
2822  * is used in load-balancing mode. if the adapter has been dormant, then an
2823  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2824  * arp monitoring in active backup mode.
2825  */
2826 static void bond_loadbalance_arp_mon(struct bonding *bond)
2827 {
2828         struct slave *slave, *oldcurrent;
2829         struct list_head *iter;
2830         int do_failover = 0, slave_state_changed = 0;
2831
2832         if (!bond_has_slaves(bond))
2833                 goto re_arm;
2834
2835         rcu_read_lock();
2836
2837         oldcurrent = rcu_dereference(bond->curr_active_slave);
2838         /* see if any of the previous devices are up now (i.e. they have
2839          * xmt and rcv traffic). the curr_active_slave does not come into
2840          * the picture unless it is null. also, slave->last_link_up is not
2841          * needed here because we send an arp on each slave and give a slave
2842          * as long as it needs to get the tx/rx within the delta.
2843          * TODO: what about up/down delay in arp mode? it wasn't here before
2844          *       so it can wait
2845          */
2846         bond_for_each_slave_rcu(bond, slave, iter) {
2847                 unsigned long trans_start = dev_trans_start(slave->dev);
2848
2849                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2850
2851                 if (slave->link != BOND_LINK_UP) {
2852                         if (bond_time_in_interval(bond, trans_start, 1) &&
2853                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2854
2855                                 bond_propose_link_state(slave, BOND_LINK_UP);
2856                                 slave_state_changed = 1;
2857
2858                                 /* primary_slave has no meaning in round-robin
2859                                  * mode. the window of a slave being up and
2860                                  * curr_active_slave being null after enslaving
2861                                  * is closed.
2862                                  */
2863                                 if (!oldcurrent) {
2864                                         slave_info(bond->dev, slave->dev, "link status definitely up\n");
2865                                         do_failover = 1;
2866                                 } else {
2867                                         slave_info(bond->dev, slave->dev, "interface is now up\n");
2868                                 }
2869                         }
2870                 } else {
2871                         /* slave->link == BOND_LINK_UP */
2872
2873                         /* not all switches will respond to an arp request
2874                          * when the source ip is 0, so don't take the link down
2875                          * if we don't know our ip yet
2876                          */
2877                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2878                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2879
2880                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2881                                 slave_state_changed = 1;
2882
2883                                 if (slave->link_failure_count < UINT_MAX)
2884                                         slave->link_failure_count++;
2885
2886                                 slave_info(bond->dev, slave->dev, "interface is now down\n");
2887
2888                                 if (slave == oldcurrent)
2889                                         do_failover = 1;
2890                         }
2891                 }
2892
2893                 /* note: if switch is in round-robin mode, all links
2894                  * must tx arp to ensure all links rx an arp - otherwise
2895                  * links may oscillate or not come up at all; if switch is
2896                  * in something like xor mode, there is nothing we can
2897                  * do - all replies will be rx'ed on same link causing slaves
2898                  * to be unstable during low/no traffic periods
2899                  */
2900                 if (bond_slave_is_up(slave))
2901                         bond_arp_send_all(bond, slave);
2902         }
2903
2904         rcu_read_unlock();
2905
2906         if (do_failover || slave_state_changed) {
2907                 if (!rtnl_trylock())
2908                         goto re_arm;
2909
2910                 bond_for_each_slave(bond, slave, iter) {
2911                         if (slave->link_new_state != BOND_LINK_NOCHANGE)
2912                                 slave->link = slave->link_new_state;
2913                 }
2914
2915                 if (slave_state_changed) {
2916                         bond_slave_state_change(bond);
2917                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2918                                 bond_update_slave_arr(bond, NULL);
2919                 }
2920                 if (do_failover) {
2921                         block_netpoll_tx();
2922                         bond_select_active_slave(bond);
2923                         unblock_netpoll_tx();
2924                 }
2925                 rtnl_unlock();
2926         }
2927
2928 re_arm:
2929         if (bond->params.arp_interval)
2930                 queue_delayed_work(bond->wq, &bond->arp_work,
2931                                    msecs_to_jiffies(bond->params.arp_interval));
2932 }
2933
2934 /* Called to inspect slaves for active-backup mode ARP monitor link state
2935  * changes.  Sets proposed link state in slaves to specify what action
2936  * should take place for the slave.  Returns 0 if no changes are found, >0
2937  * if changes to link states must be committed.
2938  *
2939  * Called with rcu_read_lock held.
2940  */
2941 static int bond_ab_arp_inspect(struct bonding *bond)
2942 {
2943         unsigned long trans_start, last_rx;
2944         struct list_head *iter;
2945         struct slave *slave;
2946         int commit = 0;
2947
2948         bond_for_each_slave_rcu(bond, slave, iter) {
2949                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2950                 last_rx = slave_last_rx(bond, slave);
2951
2952                 if (slave->link != BOND_LINK_UP) {
2953                         if (bond_time_in_interval(bond, last_rx, 1)) {
2954                                 bond_propose_link_state(slave, BOND_LINK_UP);
2955                                 commit++;
2956                         } else if (slave->link == BOND_LINK_BACK) {
2957                                 bond_propose_link_state(slave, BOND_LINK_FAIL);
2958                                 commit++;
2959                         }
2960                         continue;
2961                 }
2962
2963                 /* Give slaves 2*delta after being enslaved or made
2964                  * active.  This avoids bouncing, as the last receive
2965                  * times need a full ARP monitor cycle to be updated.
2966                  */
2967                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2968                         continue;
2969
2970                 /* Backup slave is down if:
2971                  * - No current_arp_slave AND
2972                  * - more than 3*delta since last receive AND
2973                  * - the bond has an IP address
2974                  *
2975                  * Note: a non-null current_arp_slave indicates
2976                  * the curr_active_slave went down and we are
2977                  * searching for a new one; under this condition
2978                  * we only take the curr_active_slave down - this
2979                  * gives each slave a chance to tx/rx traffic
2980                  * before being taken out
2981                  */
2982                 if (!bond_is_active_slave(slave) &&
2983                     !rcu_access_pointer(bond->current_arp_slave) &&
2984                     !bond_time_in_interval(bond, last_rx, 3)) {
2985                         bond_propose_link_state(slave, BOND_LINK_DOWN);
2986                         commit++;
2987                 }
2988
2989                 /* Active slave is down if:
2990                  * - more than 2*delta since transmitting OR
2991                  * - (more than 2*delta since receive AND
2992                  *    the bond has an IP address)
2993                  */
2994                 trans_start = dev_trans_start(slave->dev);
2995                 if (bond_is_active_slave(slave) &&
2996                     (!bond_time_in_interval(bond, trans_start, 2) ||
2997                      !bond_time_in_interval(bond, last_rx, 2))) {
2998                         bond_propose_link_state(slave, BOND_LINK_DOWN);
2999                         commit++;
3000                 }
3001         }
3002
3003         return commit;
3004 }
3005
3006 /* Called to commit link state changes noted by inspection step of
3007  * active-backup mode ARP monitor.
3008  *
3009  * Called with RTNL hold.
3010  */
3011 static void bond_ab_arp_commit(struct bonding *bond)
3012 {
3013         unsigned long trans_start;
3014         struct list_head *iter;
3015         struct slave *slave;
3016
3017         bond_for_each_slave(bond, slave, iter) {
3018                 switch (slave->link_new_state) {
3019                 case BOND_LINK_NOCHANGE:
3020                         continue;
3021
3022                 case BOND_LINK_UP:
3023                         trans_start = dev_trans_start(slave->dev);
3024                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
3025                             (!rtnl_dereference(bond->curr_active_slave) &&
3026                              bond_time_in_interval(bond, trans_start, 1))) {
3027                                 struct slave *current_arp_slave;
3028
3029                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3030                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
3031                                                           BOND_SLAVE_NOTIFY_NOW);
3032                                 if (current_arp_slave) {
3033                                         bond_set_slave_inactive_flags(
3034                                                 current_arp_slave,
3035                                                 BOND_SLAVE_NOTIFY_NOW);
3036                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3037                                 }
3038
3039                                 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3040
3041                                 if (!rtnl_dereference(bond->curr_active_slave) ||
3042                                     slave == rtnl_dereference(bond->primary_slave))
3043                                         goto do_failover;
3044
3045                         }
3046
3047                         continue;
3048
3049                 case BOND_LINK_DOWN:
3050                         if (slave->link_failure_count < UINT_MAX)
3051                                 slave->link_failure_count++;
3052
3053                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3054                                                   BOND_SLAVE_NOTIFY_NOW);
3055                         bond_set_slave_inactive_flags(slave,
3056                                                       BOND_SLAVE_NOTIFY_NOW);
3057
3058                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3059
3060                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
3061                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3062                                 goto do_failover;
3063                         }
3064
3065                         continue;
3066
3067                 case BOND_LINK_FAIL:
3068                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3069                                                   BOND_SLAVE_NOTIFY_NOW);
3070                         bond_set_slave_inactive_flags(slave,
3071                                                       BOND_SLAVE_NOTIFY_NOW);
3072
3073                         /* A slave has just been enslaved and has become
3074                          * the current active slave.
3075                          */
3076                         if (rtnl_dereference(bond->curr_active_slave))
3077                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3078                         continue;
3079
3080                 default:
3081                         slave_err(bond->dev, slave->dev,
3082                                   "impossible: link_new_state %d on slave\n",
3083                                   slave->link_new_state);
3084                         continue;
3085                 }
3086
3087 do_failover:
3088                 block_netpoll_tx();
3089                 bond_select_active_slave(bond);
3090                 unblock_netpoll_tx();
3091         }
3092
3093         bond_set_carrier(bond);
3094 }
3095
3096 /* Send ARP probes for active-backup mode ARP monitor.
3097  *
3098  * Called with rcu_read_lock held.
3099  */
3100 static bool bond_ab_arp_probe(struct bonding *bond)
3101 {
3102         struct slave *slave, *before = NULL, *new_slave = NULL,
3103                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3104                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3105         struct list_head *iter;
3106         bool found = false;
3107         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3108
3109         if (curr_arp_slave && curr_active_slave)
3110                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3111                             curr_arp_slave->dev->name,
3112                             curr_active_slave->dev->name);
3113
3114         if (curr_active_slave) {
3115                 bond_arp_send_all(bond, curr_active_slave);
3116                 return should_notify_rtnl;
3117         }
3118
3119         /* if we don't have a curr_active_slave, search for the next available
3120          * backup slave from the current_arp_slave and make it the candidate
3121          * for becoming the curr_active_slave
3122          */
3123
3124         if (!curr_arp_slave) {
3125                 curr_arp_slave = bond_first_slave_rcu(bond);
3126                 if (!curr_arp_slave)
3127                         return should_notify_rtnl;
3128         }
3129
3130         bond_for_each_slave_rcu(bond, slave, iter) {
3131                 if (!found && !before && bond_slave_is_up(slave))
3132                         before = slave;
3133
3134                 if (found && !new_slave && bond_slave_is_up(slave))
3135                         new_slave = slave;
3136                 /* if the link state is up at this point, we
3137                  * mark it down - this can happen if we have
3138                  * simultaneous link failures and
3139                  * reselect_active_interface doesn't make this
3140                  * one the current slave so it is still marked
3141                  * up when it is actually down
3142                  */
3143                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3144                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3145                                                   BOND_SLAVE_NOTIFY_LATER);
3146                         if (slave->link_failure_count < UINT_MAX)
3147                                 slave->link_failure_count++;
3148
3149                         bond_set_slave_inactive_flags(slave,
3150                                                       BOND_SLAVE_NOTIFY_LATER);
3151
3152                         slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3153                 }
3154                 if (slave == curr_arp_slave)
3155                         found = true;
3156         }
3157
3158         if (!new_slave && before)
3159                 new_slave = before;
3160
3161         if (!new_slave)
3162                 goto check_state;
3163
3164         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3165                                   BOND_SLAVE_NOTIFY_LATER);
3166         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3167         bond_arp_send_all(bond, new_slave);
3168         new_slave->last_link_up = jiffies;
3169         rcu_assign_pointer(bond->current_arp_slave, new_slave);
3170
3171 check_state:
3172         bond_for_each_slave_rcu(bond, slave, iter) {
3173                 if (slave->should_notify || slave->should_notify_link) {
3174                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3175                         break;
3176                 }
3177         }
3178         return should_notify_rtnl;
3179 }
3180
3181 static void bond_activebackup_arp_mon(struct bonding *bond)
3182 {
3183         bool should_notify_peers = false;
3184         bool should_notify_rtnl = false;
3185         int delta_in_ticks;
3186
3187         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3188
3189         if (!bond_has_slaves(bond))
3190                 goto re_arm;
3191
3192         rcu_read_lock();
3193
3194         should_notify_peers = bond_should_notify_peers(bond);
3195
3196         if (bond_ab_arp_inspect(bond)) {
3197                 rcu_read_unlock();
3198
3199                 /* Race avoidance with bond_close flush of workqueue */
3200                 if (!rtnl_trylock()) {
3201                         delta_in_ticks = 1;
3202                         should_notify_peers = false;
3203                         goto re_arm;
3204                 }
3205
3206                 bond_ab_arp_commit(bond);
3207
3208                 rtnl_unlock();
3209                 rcu_read_lock();
3210         }
3211
3212         should_notify_rtnl = bond_ab_arp_probe(bond);
3213         rcu_read_unlock();
3214
3215 re_arm:
3216         if (bond->params.arp_interval)
3217                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3218
3219         if (should_notify_peers || should_notify_rtnl) {
3220                 if (!rtnl_trylock())
3221                         return;
3222
3223                 if (should_notify_peers)
3224                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3225                                                  bond->dev);
3226                 if (should_notify_rtnl) {
3227                         bond_slave_state_notify(bond);
3228                         bond_slave_link_notify(bond);
3229                 }
3230
3231                 rtnl_unlock();
3232         }
3233 }
3234
3235 static void bond_arp_monitor(struct work_struct *work)
3236 {
3237         struct bonding *bond = container_of(work, struct bonding,
3238                                             arp_work.work);
3239
3240         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3241                 bond_activebackup_arp_mon(bond);
3242         else
3243                 bond_loadbalance_arp_mon(bond);
3244 }
3245
3246 /*-------------------------- netdev event handling --------------------------*/
3247
3248 /* Change device name */
3249 static int bond_event_changename(struct bonding *bond)
3250 {
3251         bond_remove_proc_entry(bond);
3252         bond_create_proc_entry(bond);
3253
3254         bond_debug_reregister(bond);
3255
3256         return NOTIFY_DONE;
3257 }
3258
3259 static int bond_master_netdev_event(unsigned long event,
3260                                     struct net_device *bond_dev)
3261 {
3262         struct bonding *event_bond = netdev_priv(bond_dev);
3263
3264         netdev_dbg(bond_dev, "%s called\n", __func__);
3265
3266         switch (event) {
3267         case NETDEV_CHANGENAME:
3268                 return bond_event_changename(event_bond);
3269         case NETDEV_UNREGISTER:
3270                 bond_remove_proc_entry(event_bond);
3271                 break;
3272         case NETDEV_REGISTER:
3273                 bond_create_proc_entry(event_bond);
3274                 break;
3275         default:
3276                 break;
3277         }
3278
3279         return NOTIFY_DONE;
3280 }
3281
3282 static int bond_slave_netdev_event(unsigned long event,
3283                                    struct net_device *slave_dev)
3284 {
3285         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3286         struct bonding *bond;
3287         struct net_device *bond_dev;
3288
3289         /* A netdev event can be generated while enslaving a device
3290          * before netdev_rx_handler_register is called in which case
3291          * slave will be NULL
3292          */
3293         if (!slave) {
3294                 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3295                 return NOTIFY_DONE;
3296         }
3297
3298         bond_dev = slave->bond->dev;
3299         bond = slave->bond;
3300         primary = rtnl_dereference(bond->primary_slave);
3301
3302         slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3303
3304         switch (event) {
3305         case NETDEV_UNREGISTER:
3306                 if (bond_dev->type != ARPHRD_ETHER)
3307                         bond_release_and_destroy(bond_dev, slave_dev);
3308                 else
3309                         __bond_release_one(bond_dev, slave_dev, false, true);
3310                 break;
3311         case NETDEV_UP:
3312         case NETDEV_CHANGE:
3313                 /* For 802.3ad mode only:
3314                  * Getting invalid Speed/Duplex values here will put slave
3315                  * in weird state. Mark it as link-fail if the link was
3316                  * previously up or link-down if it hasn't yet come up, and
3317                  * let link-monitoring (miimon) set it right when correct
3318                  * speeds/duplex are available.
3319                  */
3320                 if (bond_update_speed_duplex(slave) &&
3321                     BOND_MODE(bond) == BOND_MODE_8023AD) {
3322                         if (slave->last_link_up)
3323                                 slave->link = BOND_LINK_FAIL;
3324                         else
3325                                 slave->link = BOND_LINK_DOWN;
3326                 }
3327
3328                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3329                         bond_3ad_adapter_speed_duplex_changed(slave);
3330                 fallthrough;
3331         case NETDEV_DOWN:
3332                 /* Refresh slave-array if applicable!
3333                  * If the setup does not use miimon or arpmon (mode-specific!),
3334                  * then these events will not cause the slave-array to be
3335                  * refreshed. This will cause xmit to use a slave that is not
3336                  * usable. Avoid such situation by refeshing the array at these
3337                  * events. If these (miimon/arpmon) parameters are configured
3338                  * then array gets refreshed twice and that should be fine!
3339                  */
3340                 if (bond_mode_can_use_xmit_hash(bond))
3341                         bond_update_slave_arr(bond, NULL);
3342                 break;
3343         case NETDEV_CHANGEMTU:
3344                 /* TODO: Should slaves be allowed to
3345                  * independently alter their MTU?  For
3346                  * an active-backup bond, slaves need
3347                  * not be the same type of device, so
3348                  * MTUs may vary.  For other modes,
3349                  * slaves arguably should have the
3350                  * same MTUs. To do this, we'd need to
3351                  * take over the slave's change_mtu
3352                  * function for the duration of their
3353                  * servitude.
3354                  */
3355                 break;
3356         case NETDEV_CHANGENAME:
3357                 /* we don't care if we don't have primary set */
3358                 if (!bond_uses_primary(bond) ||
3359                     !bond->params.primary[0])
3360                         break;
3361
3362                 if (slave == primary) {
3363                         /* slave's name changed - he's no longer primary */
3364                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3365                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3366                         /* we have a new primary slave */
3367                         rcu_assign_pointer(bond->primary_slave, slave);
3368                 } else { /* we didn't change primary - exit */
3369                         break;
3370                 }
3371
3372                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3373                             primary ? slave_dev->name : "none");
3374
3375                 block_netpoll_tx();
3376                 bond_select_active_slave(bond);
3377                 unblock_netpoll_tx();
3378                 break;
3379         case NETDEV_FEAT_CHANGE:
3380                 bond_compute_features(bond);
3381                 break;
3382         case NETDEV_RESEND_IGMP:
3383                 /* Propagate to master device */
3384                 call_netdevice_notifiers(event, slave->bond->dev);
3385                 break;
3386         default:
3387                 break;
3388         }
3389
3390         return NOTIFY_DONE;
3391 }
3392
3393 /* bond_netdev_event: handle netdev notifier chain events.
3394  *
3395  * This function receives events for the netdev chain.  The caller (an
3396  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3397  * locks for us to safely manipulate the slave devices (RTNL lock,
3398  * dev_probe_lock).
3399  */
3400 static int bond_netdev_event(struct notifier_block *this,
3401                              unsigned long event, void *ptr)
3402 {
3403         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3404
3405         netdev_dbg(event_dev, "%s received %s\n",
3406                    __func__, netdev_cmd_to_name(event));
3407
3408         if (!(event_dev->priv_flags & IFF_BONDING))
3409                 return NOTIFY_DONE;
3410
3411         if (event_dev->flags & IFF_MASTER) {
3412                 int ret;
3413
3414                 ret = bond_master_netdev_event(event, event_dev);
3415                 if (ret != NOTIFY_DONE)
3416                         return ret;
3417         }
3418
3419         if (event_dev->flags & IFF_SLAVE)
3420                 return bond_slave_netdev_event(event, event_dev);
3421
3422         return NOTIFY_DONE;
3423 }
3424
3425 static struct notifier_block bond_netdev_notifier = {
3426         .notifier_call = bond_netdev_event,
3427 };
3428
3429 /*---------------------------- Hashing Policies -----------------------------*/
3430
3431 /* L2 hash helper */
3432 static inline u32 bond_eth_hash(struct sk_buff *skb)
3433 {
3434         struct ethhdr *ep, hdr_tmp;
3435
3436         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3437         if (ep)
3438                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3439         return 0;
3440 }
3441
3442 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk,
3443                          int *noff, int *proto, bool l34)
3444 {
3445         const struct ipv6hdr *iph6;
3446         const struct iphdr *iph;
3447
3448         if (skb->protocol == htons(ETH_P_IP)) {
3449                 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph))))
3450                         return false;
3451                 iph = (const struct iphdr *)(skb->data + *noff);
3452                 iph_to_flow_copy_v4addrs(fk, iph);
3453                 *noff += iph->ihl << 2;
3454                 if (!ip_is_fragment(iph))
3455                         *proto = iph->protocol;
3456         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3457                 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6))))
3458                         return false;
3459                 iph6 = (const struct ipv6hdr *)(skb->data + *noff);
3460                 iph_to_flow_copy_v6addrs(fk, iph6);
3461                 *noff += sizeof(*iph6);
3462                 *proto = iph6->nexthdr;
3463         } else {
3464                 return false;
3465         }
3466
3467         if (l34 && *proto >= 0)
3468                 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto);
3469
3470         return true;
3471 }
3472
3473 /* Extract the appropriate headers based on bond's xmit policy */
3474 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3475                               struct flow_keys *fk)
3476 {
3477         bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
3478         int noff, proto = -1;
3479
3480         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23) {
3481                 memset(fk, 0, sizeof(*fk));
3482                 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3483                                           fk, NULL, 0, 0, 0, 0);
3484         }
3485
3486         fk->ports.ports = 0;
3487         memset(&fk->icmp, 0, sizeof(fk->icmp));
3488         noff = skb_network_offset(skb);
3489         if (!bond_flow_ip(skb, fk, &noff, &proto, l34))
3490                 return false;
3491
3492         /* ICMP error packets contains at least 8 bytes of the header
3493          * of the packet which generated the error. Use this information
3494          * to correlate ICMP error packets within the same flow which
3495          * generated the error.
3496          */
3497         if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
3498                 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data,
3499                                       skb_transport_offset(skb),
3500                                       skb_headlen(skb));
3501                 if (proto == IPPROTO_ICMP) {
3502                         if (!icmp_is_err(fk->icmp.type))
3503                                 return true;
3504
3505                         noff += sizeof(struct icmphdr);
3506                 } else if (proto == IPPROTO_ICMPV6) {
3507                         if (!icmpv6_is_err(fk->icmp.type))
3508                                 return true;
3509
3510                         noff += sizeof(struct icmp6hdr);
3511                 }
3512                 return bond_flow_ip(skb, fk, &noff, &proto, l34);
3513         }
3514
3515         return true;
3516 }
3517
3518 /**
3519  * bond_xmit_hash - generate a hash value based on the xmit policy
3520  * @bond: bonding device
3521  * @skb: buffer to use for headers
3522  *
3523  * This function will extract the necessary headers from the skb buffer and use
3524  * them to generate a hash based on the xmit_policy set in the bonding device
3525  */
3526 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3527 {
3528         struct flow_keys flow;
3529         u32 hash;
3530
3531         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3532             skb->l4_hash)
3533                 return skb->hash;
3534
3535         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3536             !bond_flow_dissect(bond, skb, &flow))
3537                 return bond_eth_hash(skb);
3538
3539         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3540             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
3541                 hash = bond_eth_hash(skb);
3542         } else {
3543                 if (flow.icmp.id)
3544                         memcpy(&hash, &flow.icmp, sizeof(hash));
3545                 else
3546                         memcpy(&hash, &flow.ports.ports, sizeof(hash));
3547         }
3548         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3549                 (__force u32)flow_get_u32_src(&flow);
3550         hash ^= (hash >> 16);
3551         hash ^= (hash >> 8);
3552
3553         return hash >> 1;
3554 }
3555
3556 /*-------------------------- Device entry points ----------------------------*/
3557
3558 void bond_work_init_all(struct bonding *bond)
3559 {
3560         INIT_DELAYED_WORK(&bond->mcast_work,
3561                           bond_resend_igmp_join_requests_delayed);
3562         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3563         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3564         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3565         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3566         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3567 }
3568
3569 static void bond_work_cancel_all(struct bonding *bond)
3570 {
3571         cancel_delayed_work_sync(&bond->mii_work);
3572         cancel_delayed_work_sync(&bond->arp_work);
3573         cancel_delayed_work_sync(&bond->alb_work);
3574         cancel_delayed_work_sync(&bond->ad_work);
3575         cancel_delayed_work_sync(&bond->mcast_work);
3576         cancel_delayed_work_sync(&bond->slave_arr_work);
3577 }
3578
3579 static int bond_open(struct net_device *bond_dev)
3580 {
3581         struct bonding *bond = netdev_priv(bond_dev);
3582         struct list_head *iter;
3583         struct slave *slave;
3584
3585         /* reset slave->backup and slave->inactive */
3586         if (bond_has_slaves(bond)) {
3587                 bond_for_each_slave(bond, slave, iter) {
3588                         if (bond_uses_primary(bond) &&
3589                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3590                                 bond_set_slave_inactive_flags(slave,
3591                                                               BOND_SLAVE_NOTIFY_NOW);
3592                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3593                                 bond_set_slave_active_flags(slave,
3594                                                             BOND_SLAVE_NOTIFY_NOW);
3595                         }
3596                 }
3597         }
3598
3599         if (bond_is_lb(bond)) {
3600                 /* bond_alb_initialize must be called before the timer
3601                  * is started.
3602                  */
3603                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3604                         return -ENOMEM;
3605                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3606                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3607         }
3608
3609         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3610                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3611
3612         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3613                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3614                 bond->recv_probe = bond_arp_rcv;
3615         }
3616
3617         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3618                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3619                 /* register to receive LACPDUs */
3620                 bond->recv_probe = bond_3ad_lacpdu_recv;
3621                 bond_3ad_initiate_agg_selection(bond, 1);
3622         }
3623
3624         if (bond_mode_can_use_xmit_hash(bond))
3625                 bond_update_slave_arr(bond, NULL);
3626
3627         return 0;
3628 }
3629
3630 static int bond_close(struct net_device *bond_dev)
3631 {
3632         struct bonding *bond = netdev_priv(bond_dev);
3633
3634         bond_work_cancel_all(bond);
3635         bond->send_peer_notif = 0;
3636         if (bond_is_lb(bond))
3637                 bond_alb_deinitialize(bond);
3638         bond->recv_probe = NULL;
3639
3640         return 0;
3641 }
3642
3643 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3644  * that some drivers can provide 32bit values only.
3645  */
3646 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3647                             const struct rtnl_link_stats64 *_new,
3648                             const struct rtnl_link_stats64 *_old)
3649 {
3650         const u64 *new = (const u64 *)_new;
3651         const u64 *old = (const u64 *)_old;
3652         u64 *res = (u64 *)_res;
3653         int i;
3654
3655         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3656                 u64 nv = new[i];
3657                 u64 ov = old[i];
3658                 s64 delta = nv - ov;
3659
3660                 /* detects if this particular field is 32bit only */
3661                 if (((nv | ov) >> 32) == 0)
3662                         delta = (s64)(s32)((u32)nv - (u32)ov);
3663
3664                 /* filter anomalies, some drivers reset their stats
3665                  * at down/up events.
3666                  */
3667                 if (delta > 0)
3668                         res[i] += delta;
3669         }
3670 }
3671
3672 #ifdef CONFIG_LOCKDEP
3673 static int bond_get_lowest_level_rcu(struct net_device *dev)
3674 {
3675         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
3676         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
3677         int cur = 0, max = 0;
3678
3679         now = dev;
3680         iter = &dev->adj_list.lower;
3681
3682         while (1) {
3683                 next = NULL;
3684                 while (1) {
3685                         ldev = netdev_next_lower_dev_rcu(now, &iter);
3686                         if (!ldev)
3687                                 break;
3688
3689                         next = ldev;
3690                         niter = &ldev->adj_list.lower;
3691                         dev_stack[cur] = now;
3692                         iter_stack[cur++] = iter;
3693                         if (max <= cur)
3694                                 max = cur;
3695                         break;
3696                 }
3697
3698                 if (!next) {
3699                         if (!cur)
3700                                 return max;
3701                         next = dev_stack[--cur];
3702                         niter = iter_stack[cur];
3703                 }
3704
3705                 now = next;
3706                 iter = niter;
3707         }
3708
3709         return max;
3710 }
3711 #endif
3712
3713 static void bond_get_stats(struct net_device *bond_dev,
3714                            struct rtnl_link_stats64 *stats)
3715 {
3716         struct bonding *bond = netdev_priv(bond_dev);
3717         struct rtnl_link_stats64 temp;
3718         struct list_head *iter;
3719         struct slave *slave;
3720         int nest_level = 0;
3721
3722
3723         rcu_read_lock();
3724 #ifdef CONFIG_LOCKDEP
3725         nest_level = bond_get_lowest_level_rcu(bond_dev);
3726 #endif
3727
3728         spin_lock_nested(&bond->stats_lock, nest_level);
3729         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3730
3731         bond_for_each_slave_rcu(bond, slave, iter) {
3732                 const struct rtnl_link_stats64 *new =
3733                         dev_get_stats(slave->dev, &temp);
3734
3735                 bond_fold_stats(stats, new, &slave->slave_stats);
3736
3737                 /* save off the slave stats for the next run */
3738                 memcpy(&slave->slave_stats, new, sizeof(*new));
3739         }
3740
3741         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3742         spin_unlock(&bond->stats_lock);
3743         rcu_read_unlock();
3744 }
3745
3746 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3747 {
3748         struct bonding *bond = netdev_priv(bond_dev);
3749         struct net_device *slave_dev = NULL;
3750         struct ifbond k_binfo;
3751         struct ifbond __user *u_binfo = NULL;
3752         struct ifslave k_sinfo;
3753         struct ifslave __user *u_sinfo = NULL;
3754         struct mii_ioctl_data *mii = NULL;
3755         struct bond_opt_value newval;
3756         struct net *net;
3757         int res = 0;
3758
3759         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3760
3761         switch (cmd) {
3762         case SIOCGMIIPHY:
3763                 mii = if_mii(ifr);
3764                 if (!mii)
3765                         return -EINVAL;
3766
3767                 mii->phy_id = 0;
3768                 fallthrough;
3769         case SIOCGMIIREG:
3770                 /* We do this again just in case we were called by SIOCGMIIREG
3771                  * instead of SIOCGMIIPHY.
3772                  */
3773                 mii = if_mii(ifr);
3774                 if (!mii)
3775                         return -EINVAL;
3776
3777                 if (mii->reg_num == 1) {
3778                         mii->val_out = 0;
3779                         if (netif_carrier_ok(bond->dev))
3780                                 mii->val_out = BMSR_LSTATUS;
3781                 }
3782
3783                 return 0;
3784         case BOND_INFO_QUERY_OLD:
3785         case SIOCBONDINFOQUERY:
3786                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3787
3788                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3789                         return -EFAULT;
3790
3791                 bond_info_query(bond_dev, &k_binfo);
3792                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3793                         return -EFAULT;
3794
3795                 return 0;
3796         case BOND_SLAVE_INFO_QUERY_OLD:
3797         case SIOCBONDSLAVEINFOQUERY:
3798                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3799
3800                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3801                         return -EFAULT;
3802
3803                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3804                 if (res == 0 &&
3805                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3806                         return -EFAULT;
3807
3808                 return res;
3809         default:
3810                 break;
3811         }
3812
3813         net = dev_net(bond_dev);
3814
3815         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3816                 return -EPERM;
3817
3818         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3819
3820         slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
3821
3822         if (!slave_dev)
3823                 return -ENODEV;
3824
3825         switch (cmd) {
3826         case BOND_ENSLAVE_OLD:
3827         case SIOCBONDENSLAVE:
3828                 res = bond_enslave(bond_dev, slave_dev, NULL);
3829                 break;
3830         case BOND_RELEASE_OLD:
3831         case SIOCBONDRELEASE:
3832                 res = bond_release(bond_dev, slave_dev);
3833                 break;
3834         case BOND_SETHWADDR_OLD:
3835         case SIOCBONDSETHWADDR:
3836                 res = bond_set_dev_addr(bond_dev, slave_dev);
3837                 break;
3838         case BOND_CHANGE_ACTIVE_OLD:
3839         case SIOCBONDCHANGEACTIVE:
3840                 bond_opt_initstr(&newval, slave_dev->name);
3841                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3842                                             &newval);
3843                 break;
3844         default:
3845                 res = -EOPNOTSUPP;
3846         }
3847
3848         return res;
3849 }
3850
3851 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3852 {
3853         struct bonding *bond = netdev_priv(bond_dev);
3854
3855         if (change & IFF_PROMISC)
3856                 bond_set_promiscuity(bond,
3857                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3858
3859         if (change & IFF_ALLMULTI)
3860                 bond_set_allmulti(bond,
3861                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3862 }
3863
3864 static void bond_set_rx_mode(struct net_device *bond_dev)
3865 {
3866         struct bonding *bond = netdev_priv(bond_dev);
3867         struct list_head *iter;
3868         struct slave *slave;
3869
3870         rcu_read_lock();
3871         if (bond_uses_primary(bond)) {
3872                 slave = rcu_dereference(bond->curr_active_slave);
3873                 if (slave) {
3874                         dev_uc_sync(slave->dev, bond_dev);
3875                         dev_mc_sync(slave->dev, bond_dev);
3876                 }
3877         } else {
3878                 bond_for_each_slave_rcu(bond, slave, iter) {
3879                         dev_uc_sync_multiple(slave->dev, bond_dev);
3880                         dev_mc_sync_multiple(slave->dev, bond_dev);
3881                 }
3882         }
3883         rcu_read_unlock();
3884 }
3885
3886 static int bond_neigh_init(struct neighbour *n)
3887 {
3888         struct bonding *bond = netdev_priv(n->dev);
3889         const struct net_device_ops *slave_ops;
3890         struct neigh_parms parms;
3891         struct slave *slave;
3892         int ret = 0;
3893
3894         rcu_read_lock();
3895         slave = bond_first_slave_rcu(bond);
3896         if (!slave)
3897                 goto out;
3898         slave_ops = slave->dev->netdev_ops;
3899         if (!slave_ops->ndo_neigh_setup)
3900                 goto out;
3901
3902         /* TODO: find another way [1] to implement this.
3903          * Passing a zeroed structure is fragile,
3904          * but at least we do not pass garbage.
3905          *
3906          * [1] One way would be that ndo_neigh_setup() never touch
3907          *     struct neigh_parms, but propagate the new neigh_setup()
3908          *     back to ___neigh_create() / neigh_parms_alloc()
3909          */
3910         memset(&parms, 0, sizeof(parms));
3911         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3912
3913         if (ret)
3914                 goto out;
3915
3916         if (parms.neigh_setup)
3917                 ret = parms.neigh_setup(n);
3918 out:
3919         rcu_read_unlock();
3920         return ret;
3921 }
3922
3923 /* The bonding ndo_neigh_setup is called at init time beofre any
3924  * slave exists. So we must declare proxy setup function which will
3925  * be used at run time to resolve the actual slave neigh param setup.
3926  *
3927  * It's also called by master devices (such as vlans) to setup their
3928  * underlying devices. In that case - do nothing, we're already set up from
3929  * our init.
3930  */
3931 static int bond_neigh_setup(struct net_device *dev,
3932                             struct neigh_parms *parms)
3933 {
3934         /* modify only our neigh_parms */
3935         if (parms->dev == dev)
3936                 parms->neigh_setup = bond_neigh_init;
3937
3938         return 0;
3939 }
3940
3941 /* Change the MTU of all of a master's slaves to match the master */
3942 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3943 {
3944         struct bonding *bond = netdev_priv(bond_dev);
3945         struct slave *slave, *rollback_slave;
3946         struct list_head *iter;
3947         int res = 0;
3948
3949         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3950
3951         bond_for_each_slave(bond, slave, iter) {
3952                 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
3953                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3954
3955                 res = dev_set_mtu(slave->dev, new_mtu);
3956
3957                 if (res) {
3958                         /* If we failed to set the slave's mtu to the new value
3959                          * we must abort the operation even in ACTIVE_BACKUP
3960                          * mode, because if we allow the backup slaves to have
3961                          * different mtu values than the active slave we'll
3962                          * need to change their mtu when doing a failover. That
3963                          * means changing their mtu from timer context, which
3964                          * is probably not a good idea.
3965                          */
3966                         slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
3967                                   res, new_mtu);
3968                         goto unwind;
3969                 }
3970         }
3971
3972         bond_dev->mtu = new_mtu;
3973
3974         return 0;
3975
3976 unwind:
3977         /* unwind from head to the slave that failed */
3978         bond_for_each_slave(bond, rollback_slave, iter) {
3979                 int tmp_res;
3980
3981                 if (rollback_slave == slave)
3982                         break;
3983
3984                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3985                 if (tmp_res)
3986                         slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
3987                                   tmp_res);
3988         }
3989
3990         return res;
3991 }
3992
3993 /* Change HW address
3994  *
3995  * Note that many devices must be down to change the HW address, and
3996  * downing the master releases all slaves.  We can make bonds full of
3997  * bonding devices to test this, however.
3998  */
3999 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4000 {
4001         struct bonding *bond = netdev_priv(bond_dev);
4002         struct slave *slave, *rollback_slave;
4003         struct sockaddr_storage *ss = addr, tmp_ss;
4004         struct list_head *iter;
4005         int res = 0;
4006
4007         if (BOND_MODE(bond) == BOND_MODE_ALB)
4008                 return bond_alb_set_mac_address(bond_dev, addr);
4009
4010
4011         netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4012
4013         /* If fail_over_mac is enabled, do nothing and return success.
4014          * Returning an error causes ifenslave to fail.
4015          */
4016         if (bond->params.fail_over_mac &&
4017             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4018                 return 0;
4019
4020         if (!is_valid_ether_addr(ss->__data))
4021                 return -EADDRNOTAVAIL;
4022
4023         bond_for_each_slave(bond, slave, iter) {
4024                 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4025                           __func__, slave);
4026                 res = dev_set_mac_address(slave->dev, addr, NULL);
4027                 if (res) {
4028                         /* TODO: consider downing the slave
4029                          * and retry ?
4030                          * User should expect communications
4031                          * breakage anyway until ARP finish
4032                          * updating, so...
4033                          */
4034                         slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4035                                   __func__, res);
4036                         goto unwind;
4037                 }
4038         }
4039
4040         /* success */
4041         memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
4042         return 0;
4043
4044 unwind:
4045         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4046         tmp_ss.ss_family = bond_dev->type;
4047
4048         /* unwind from head to the slave that failed */
4049         bond_for_each_slave(bond, rollback_slave, iter) {
4050                 int tmp_res;
4051
4052                 if (rollback_slave == slave)
4053                         break;
4054
4055                 tmp_res = dev_set_mac_address(rollback_slave->dev,
4056                                               (struct sockaddr *)&tmp_ss, NULL);
4057                 if (tmp_res) {
4058                         slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4059                                    __func__, tmp_res);
4060                 }
4061         }
4062
4063         return res;
4064 }
4065
4066 /**
4067  * bond_get_slave_by_id - get xmit slave with slave_id
4068  * @bond: bonding device that is transmitting
4069  * @slave_id: slave id up to slave_cnt-1 through which to transmit
4070  *
4071  * This function tries to get slave with slave_id but in case
4072  * it fails, it tries to find the first available slave for transmission.
4073  */
4074 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4075                                           int slave_id)
4076 {
4077         struct list_head *iter;
4078         struct slave *slave;
4079         int i = slave_id;
4080
4081         /* Here we start from the slave with slave_id */
4082         bond_for_each_slave_rcu(bond, slave, iter) {
4083                 if (--i < 0) {
4084                         if (bond_slave_can_tx(slave))
4085                                 return slave;
4086                 }
4087         }
4088
4089         /* Here we start from the first slave up to slave_id */
4090         i = slave_id;
4091         bond_for_each_slave_rcu(bond, slave, iter) {
4092                 if (--i < 0)
4093                         break;
4094                 if (bond_slave_can_tx(slave))
4095                         return slave;
4096         }
4097         /* no slave that can tx has been found */
4098         return NULL;
4099 }
4100
4101 /**
4102  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4103  * @bond: bonding device to use
4104  *
4105  * Based on the value of the bonding device's packets_per_slave parameter
4106  * this function generates a slave id, which is usually used as the next
4107  * slave to transmit through.
4108  */
4109 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4110 {
4111         u32 slave_id;
4112         struct reciprocal_value reciprocal_packets_per_slave;
4113         int packets_per_slave = bond->params.packets_per_slave;
4114
4115         switch (packets_per_slave) {
4116         case 0:
4117                 slave_id = prandom_u32();
4118                 break;
4119         case 1:
4120                 slave_id = bond->rr_tx_counter;
4121                 break;
4122         default:
4123                 reciprocal_packets_per_slave =
4124                         bond->params.reciprocal_packets_per_slave;
4125                 slave_id = reciprocal_divide(bond->rr_tx_counter,
4126                                              reciprocal_packets_per_slave);
4127                 break;
4128         }
4129         bond->rr_tx_counter++;
4130
4131         return slave_id;
4132 }
4133
4134 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4135                                                     struct sk_buff *skb)
4136 {
4137         struct slave *slave;
4138         int slave_cnt;
4139         u32 slave_id;
4140
4141         /* Start with the curr_active_slave that joined the bond as the
4142          * default for sending IGMP traffic.  For failover purposes one
4143          * needs to maintain some consistency for the interface that will
4144          * send the join/membership reports.  The curr_active_slave found
4145          * will send all of this type of traffic.
4146          */
4147         if (skb->protocol == htons(ETH_P_IP)) {
4148                 int noff = skb_network_offset(skb);
4149                 struct iphdr *iph;
4150
4151                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4152                         goto non_igmp;
4153
4154                 iph = ip_hdr(skb);
4155                 if (iph->protocol == IPPROTO_IGMP) {
4156                         slave = rcu_dereference(bond->curr_active_slave);
4157                         if (slave)
4158                                 return slave;
4159                         return bond_get_slave_by_id(bond, 0);
4160                 }
4161         }
4162
4163 non_igmp:
4164         slave_cnt = READ_ONCE(bond->slave_cnt);
4165         if (likely(slave_cnt)) {
4166                 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4167                 return bond_get_slave_by_id(bond, slave_id);
4168         }
4169         return NULL;
4170 }
4171
4172 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4173                                         struct net_device *bond_dev)
4174 {
4175         struct bonding *bond = netdev_priv(bond_dev);
4176         struct slave *slave;
4177
4178         slave = bond_xmit_roundrobin_slave_get(bond, skb);
4179         if (likely(slave))
4180                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4181
4182         return bond_tx_drop(bond_dev, skb);
4183 }
4184
4185 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond,
4186                                                       struct sk_buff *skb)
4187 {
4188         return rcu_dereference(bond->curr_active_slave);
4189 }
4190
4191 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4192  * the bond has a usable interface.
4193  */
4194 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4195                                           struct net_device *bond_dev)
4196 {
4197         struct bonding *bond = netdev_priv(bond_dev);
4198         struct slave *slave;
4199
4200         slave = bond_xmit_activebackup_slave_get(bond, skb);
4201         if (slave)
4202                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4203
4204         return bond_tx_drop(bond_dev, skb);
4205 }
4206
4207 /* Use this to update slave_array when (a) it's not appropriate to update
4208  * slave_array right away (note that update_slave_array() may sleep)
4209  * and / or (b) RTNL is not held.
4210  */
4211 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4212 {
4213         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4214 }
4215
4216 /* Slave array work handler. Holds only RTNL */
4217 static void bond_slave_arr_handler(struct work_struct *work)
4218 {
4219         struct bonding *bond = container_of(work, struct bonding,
4220                                             slave_arr_work.work);
4221         int ret;
4222
4223         if (!rtnl_trylock())
4224                 goto err;
4225
4226         ret = bond_update_slave_arr(bond, NULL);
4227         rtnl_unlock();
4228         if (ret) {
4229                 pr_warn_ratelimited("Failed to update slave array from WT\n");
4230                 goto err;
4231         }
4232         return;
4233
4234 err:
4235         bond_slave_arr_work_rearm(bond, 1);
4236 }
4237
4238 static void bond_skip_slave(struct bond_up_slave *slaves,
4239                             struct slave *skipslave)
4240 {
4241         int idx;
4242
4243         /* Rare situation where caller has asked to skip a specific
4244          * slave but allocation failed (most likely!). BTW this is
4245          * only possible when the call is initiated from
4246          * __bond_release_one(). In this situation; overwrite the
4247          * skipslave entry in the array with the last entry from the
4248          * array to avoid a situation where the xmit path may choose
4249          * this to-be-skipped slave to send a packet out.
4250          */
4251         for (idx = 0; slaves && idx < slaves->count; idx++) {
4252                 if (skipslave == slaves->arr[idx]) {
4253                         slaves->arr[idx] =
4254                                 slaves->arr[slaves->count - 1];
4255                         slaves->count--;
4256                         break;
4257                 }
4258         }
4259 }
4260
4261 static void bond_set_slave_arr(struct bonding *bond,
4262                                struct bond_up_slave *usable_slaves,
4263                                struct bond_up_slave *all_slaves)
4264 {
4265         struct bond_up_slave *usable, *all;
4266
4267         usable = rtnl_dereference(bond->usable_slaves);
4268         rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4269         kfree_rcu(usable, rcu);
4270
4271         all = rtnl_dereference(bond->all_slaves);
4272         rcu_assign_pointer(bond->all_slaves, all_slaves);
4273         kfree_rcu(all, rcu);
4274 }
4275
4276 static void bond_reset_slave_arr(struct bonding *bond)
4277 {
4278         struct bond_up_slave *usable, *all;
4279
4280         usable = rtnl_dereference(bond->usable_slaves);
4281         if (usable) {
4282                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4283                 kfree_rcu(usable, rcu);
4284         }
4285
4286         all = rtnl_dereference(bond->all_slaves);
4287         if (all) {
4288                 RCU_INIT_POINTER(bond->all_slaves, NULL);
4289                 kfree_rcu(all, rcu);
4290         }
4291 }
4292
4293 /* Build the usable slaves array in control path for modes that use xmit-hash
4294  * to determine the slave interface -
4295  * (a) BOND_MODE_8023AD
4296  * (b) BOND_MODE_XOR
4297  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4298  *
4299  * The caller is expected to hold RTNL only and NO other lock!
4300  */
4301 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4302 {
4303         struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
4304         struct slave *slave;
4305         struct list_head *iter;
4306         int agg_id = 0;
4307         int ret = 0;
4308
4309 #ifdef CONFIG_LOCKDEP
4310         WARN_ON(lockdep_is_held(&bond->mode_lock));
4311 #endif
4312
4313         usable_slaves = kzalloc(struct_size(usable_slaves, arr,
4314                                             bond->slave_cnt), GFP_KERNEL);
4315         all_slaves = kzalloc(struct_size(all_slaves, arr,
4316                                          bond->slave_cnt), GFP_KERNEL);
4317         if (!usable_slaves || !all_slaves) {
4318                 ret = -ENOMEM;
4319                 goto out;
4320         }
4321         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4322                 struct ad_info ad_info;
4323
4324                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4325                         pr_debug("bond_3ad_get_active_agg_info failed\n");
4326                         /* No active aggragator means it's not safe to use
4327                          * the previous array.
4328                          */
4329                         bond_reset_slave_arr(bond);
4330                         goto out;
4331                 }
4332                 agg_id = ad_info.aggregator_id;
4333         }
4334         bond_for_each_slave(bond, slave, iter) {
4335                 if (skipslave == slave)
4336                         continue;
4337
4338                 all_slaves->arr[all_slaves->count++] = slave;
4339                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4340                         struct aggregator *agg;
4341
4342                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
4343                         if (!agg || agg->aggregator_identifier != agg_id)
4344                                 continue;
4345                 }
4346                 if (!bond_slave_can_tx(slave))
4347                         continue;
4348
4349                 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4350                           usable_slaves->count);
4351
4352                 usable_slaves->arr[usable_slaves->count++] = slave;
4353         }
4354
4355         bond_set_slave_arr(bond, usable_slaves, all_slaves);
4356         return ret;
4357 out:
4358         if (ret != 0 && skipslave) {
4359                 bond_skip_slave(rtnl_dereference(bond->all_slaves),
4360                                 skipslave);
4361                 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
4362                                 skipslave);
4363         }
4364         kfree_rcu(all_slaves, rcu);
4365         kfree_rcu(usable_slaves, rcu);
4366
4367         return ret;
4368 }
4369
4370 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
4371                                                  struct sk_buff *skb,
4372                                                  struct bond_up_slave *slaves)
4373 {
4374         struct slave *slave;
4375         unsigned int count;
4376         u32 hash;
4377
4378         hash = bond_xmit_hash(bond, skb);
4379         count = slaves ? READ_ONCE(slaves->count) : 0;
4380         if (unlikely(!count))
4381                 return NULL;
4382
4383         slave = slaves->arr[hash % count];
4384         return slave;
4385 }
4386
4387 /* Use this Xmit function for 3AD as well as XOR modes. The current
4388  * usable slave array is formed in the control path. The xmit function
4389  * just calculates hash and sends the packet out.
4390  */
4391 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4392                                      struct net_device *dev)
4393 {
4394         struct bonding *bond = netdev_priv(dev);
4395         struct bond_up_slave *slaves;
4396         struct slave *slave;
4397
4398         slaves = rcu_dereference(bond->usable_slaves);
4399         slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4400         if (likely(slave))
4401                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4402
4403         return bond_tx_drop(dev, skb);
4404 }
4405
4406 /* in broadcast mode, we send everything to all usable interfaces. */
4407 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4408                                        struct net_device *bond_dev)
4409 {
4410         struct bonding *bond = netdev_priv(bond_dev);
4411         struct slave *slave = NULL;
4412         struct list_head *iter;
4413
4414         bond_for_each_slave_rcu(bond, slave, iter) {
4415                 if (bond_is_last_slave(bond, slave))
4416                         break;
4417                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4418                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4419
4420                         if (!skb2) {
4421                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4422                                                     bond_dev->name, __func__);
4423                                 continue;
4424                         }
4425                         bond_dev_queue_xmit(bond, skb2, slave->dev);
4426                 }
4427         }
4428         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4429                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4430
4431         return bond_tx_drop(bond_dev, skb);
4432 }
4433
4434 /*------------------------- Device initialization ---------------------------*/
4435
4436 /* Lookup the slave that corresponds to a qid */
4437 static inline int bond_slave_override(struct bonding *bond,
4438                                       struct sk_buff *skb)
4439 {
4440         struct slave *slave = NULL;
4441         struct list_head *iter;
4442
4443         if (!skb_rx_queue_recorded(skb))
4444                 return 1;
4445
4446         /* Find out if any slaves have the same mapping as this skb. */
4447         bond_for_each_slave_rcu(bond, slave, iter) {
4448                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4449                         if (bond_slave_is_up(slave) &&
4450                             slave->link == BOND_LINK_UP) {
4451                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4452                                 return 0;
4453                         }
4454                         /* If the slave isn't UP, use default transmit policy. */
4455                         break;
4456                 }
4457         }
4458
4459         return 1;
4460 }
4461
4462
4463 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4464                              struct net_device *sb_dev)
4465 {
4466         /* This helper function exists to help dev_pick_tx get the correct
4467          * destination queue.  Using a helper function skips a call to
4468          * skb_tx_hash and will put the skbs in the queue we expect on their
4469          * way down to the bonding driver.
4470          */
4471         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4472
4473         /* Save the original txq to restore before passing to the driver */
4474         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4475
4476         if (unlikely(txq >= dev->real_num_tx_queues)) {
4477                 do {
4478                         txq -= dev->real_num_tx_queues;
4479                 } while (txq >= dev->real_num_tx_queues);
4480         }
4481         return txq;
4482 }
4483
4484 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
4485                                               struct sk_buff *skb,
4486                                               bool all_slaves)
4487 {
4488         struct bonding *bond = netdev_priv(master_dev);
4489         struct bond_up_slave *slaves;
4490         struct slave *slave = NULL;
4491
4492         switch (BOND_MODE(bond)) {
4493         case BOND_MODE_ROUNDROBIN:
4494                 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4495                 break;
4496         case BOND_MODE_ACTIVEBACKUP:
4497                 slave = bond_xmit_activebackup_slave_get(bond, skb);
4498                 break;
4499         case BOND_MODE_8023AD:
4500         case BOND_MODE_XOR:
4501                 if (all_slaves)
4502                         slaves = rcu_dereference(bond->all_slaves);
4503                 else
4504                         slaves = rcu_dereference(bond->usable_slaves);
4505                 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4506                 break;
4507         case BOND_MODE_BROADCAST:
4508                 break;
4509         case BOND_MODE_ALB:
4510                 slave = bond_xmit_alb_slave_get(bond, skb);
4511                 break;
4512         case BOND_MODE_TLB:
4513                 slave = bond_xmit_tlb_slave_get(bond, skb);
4514                 break;
4515         default:
4516                 /* Should never happen, mode already checked */
4517                 WARN_ONCE(true, "Unknown bonding mode");
4518                 break;
4519         }
4520
4521         if (slave)
4522                 return slave->dev;
4523         return NULL;
4524 }
4525
4526 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4527 {
4528         struct bonding *bond = netdev_priv(dev);
4529
4530         if (bond_should_override_tx_queue(bond) &&
4531             !bond_slave_override(bond, skb))
4532                 return NETDEV_TX_OK;
4533
4534         switch (BOND_MODE(bond)) {
4535         case BOND_MODE_ROUNDROBIN:
4536                 return bond_xmit_roundrobin(skb, dev);
4537         case BOND_MODE_ACTIVEBACKUP:
4538                 return bond_xmit_activebackup(skb, dev);
4539         case BOND_MODE_8023AD:
4540         case BOND_MODE_XOR:
4541                 return bond_3ad_xor_xmit(skb, dev);
4542         case BOND_MODE_BROADCAST:
4543                 return bond_xmit_broadcast(skb, dev);
4544         case BOND_MODE_ALB:
4545                 return bond_alb_xmit(skb, dev);
4546         case BOND_MODE_TLB:
4547                 return bond_tlb_xmit(skb, dev);
4548         default:
4549                 /* Should never happen, mode already checked */
4550                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4551                 WARN_ON_ONCE(1);
4552                 return bond_tx_drop(dev, skb);
4553         }
4554 }
4555
4556 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4557 {
4558         struct bonding *bond = netdev_priv(dev);
4559         netdev_tx_t ret = NETDEV_TX_OK;
4560
4561         /* If we risk deadlock from transmitting this in the
4562          * netpoll path, tell netpoll to queue the frame for later tx
4563          */
4564         if (unlikely(is_netpoll_tx_blocked(dev)))
4565                 return NETDEV_TX_BUSY;
4566
4567         rcu_read_lock();
4568         if (bond_has_slaves(bond))
4569                 ret = __bond_start_xmit(skb, dev);
4570         else
4571                 ret = bond_tx_drop(dev, skb);
4572         rcu_read_unlock();
4573
4574         return ret;
4575 }
4576
4577 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
4578 {
4579         if (speed == 0 || speed == SPEED_UNKNOWN)
4580                 speed = slave->speed;
4581         else
4582                 speed = min(speed, slave->speed);
4583
4584         return speed;
4585 }
4586
4587 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4588                                            struct ethtool_link_ksettings *cmd)
4589 {
4590         struct bonding *bond = netdev_priv(bond_dev);
4591         struct list_head *iter;
4592         struct slave *slave;
4593         u32 speed = 0;
4594
4595         cmd->base.duplex = DUPLEX_UNKNOWN;
4596         cmd->base.port = PORT_OTHER;
4597
4598         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4599          * do not need to check mode.  Though link speed might not represent
4600          * the true receive or transmit bandwidth (not all modes are symmetric)
4601          * this is an accurate maximum.
4602          */
4603         bond_for_each_slave(bond, slave, iter) {
4604                 if (bond_slave_can_tx(slave)) {
4605                         if (slave->speed != SPEED_UNKNOWN) {
4606                                 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
4607                                         speed = bond_mode_bcast_speed(slave,
4608                                                                       speed);
4609                                 else
4610                                         speed += slave->speed;
4611                         }
4612                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4613                             slave->duplex != DUPLEX_UNKNOWN)
4614                                 cmd->base.duplex = slave->duplex;
4615                 }
4616         }
4617         cmd->base.speed = speed ? : SPEED_UNKNOWN;
4618
4619         return 0;
4620 }
4621
4622 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4623                                      struct ethtool_drvinfo *drvinfo)
4624 {
4625         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4626         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4627                  BOND_ABI_VERSION);
4628 }
4629
4630 static const struct ethtool_ops bond_ethtool_ops = {
4631         .get_drvinfo            = bond_ethtool_get_drvinfo,
4632         .get_link               = ethtool_op_get_link,
4633         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
4634 };
4635
4636 static const struct net_device_ops bond_netdev_ops = {
4637         .ndo_init               = bond_init,
4638         .ndo_uninit             = bond_uninit,
4639         .ndo_open               = bond_open,
4640         .ndo_stop               = bond_close,
4641         .ndo_start_xmit         = bond_start_xmit,
4642         .ndo_select_queue       = bond_select_queue,
4643         .ndo_get_stats64        = bond_get_stats,
4644         .ndo_do_ioctl           = bond_do_ioctl,
4645         .ndo_change_rx_flags    = bond_change_rx_flags,
4646         .ndo_set_rx_mode        = bond_set_rx_mode,
4647         .ndo_change_mtu         = bond_change_mtu,
4648         .ndo_set_mac_address    = bond_set_mac_address,
4649         .ndo_neigh_setup        = bond_neigh_setup,
4650         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4651         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4652 #ifdef CONFIG_NET_POLL_CONTROLLER
4653         .ndo_netpoll_setup      = bond_netpoll_setup,
4654         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4655         .ndo_poll_controller    = bond_poll_controller,
4656 #endif
4657         .ndo_add_slave          = bond_enslave,
4658         .ndo_del_slave          = bond_release,
4659         .ndo_fix_features       = bond_fix_features,
4660         .ndo_features_check     = passthru_features_check,
4661         .ndo_get_xmit_slave     = bond_xmit_get_slave,
4662 };
4663
4664 static const struct device_type bond_type = {
4665         .name = "bond",
4666 };
4667
4668 static void bond_destructor(struct net_device *bond_dev)
4669 {
4670         struct bonding *bond = netdev_priv(bond_dev);
4671         if (bond->wq)
4672                 destroy_workqueue(bond->wq);
4673 }
4674
4675 void bond_setup(struct net_device *bond_dev)
4676 {
4677         struct bonding *bond = netdev_priv(bond_dev);
4678
4679         spin_lock_init(&bond->mode_lock);
4680         bond->params = bonding_defaults;
4681
4682         /* Initialize pointers */
4683         bond->dev = bond_dev;
4684
4685         /* Initialize the device entry points */
4686         ether_setup(bond_dev);
4687         bond_dev->max_mtu = ETH_MAX_MTU;
4688         bond_dev->netdev_ops = &bond_netdev_ops;
4689         bond_dev->ethtool_ops = &bond_ethtool_ops;
4690
4691         bond_dev->needs_free_netdev = true;
4692         bond_dev->priv_destructor = bond_destructor;
4693
4694         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4695
4696         /* Initialize the device options */
4697         bond_dev->flags |= IFF_MASTER;
4698         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4699         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4700
4701 #ifdef CONFIG_XFRM_OFFLOAD
4702         /* set up xfrm device ops (only supported in active-backup right now) */
4703         bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
4704         bond->xs = NULL;
4705 #endif /* CONFIG_XFRM_OFFLOAD */
4706
4707         /* don't acquire bond device's netif_tx_lock when transmitting */
4708         bond_dev->features |= NETIF_F_LLTX;
4709
4710         /* By default, we declare the bond to be fully
4711          * VLAN hardware accelerated capable. Special
4712          * care is taken in the various xmit functions
4713          * when there are slaves that are not hw accel
4714          * capable
4715          */
4716
4717         /* Don't allow bond devices to change network namespaces. */
4718         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4719
4720         bond_dev->hw_features = BOND_VLAN_FEATURES |
4721                                 NETIF_F_HW_VLAN_CTAG_RX |
4722                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4723
4724         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4725 #ifdef CONFIG_XFRM_OFFLOAD
4726         bond_dev->hw_features |= BOND_XFRM_FEATURES;
4727 #endif /* CONFIG_XFRM_OFFLOAD */
4728         bond_dev->features |= bond_dev->hw_features;
4729         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4730 #ifdef CONFIG_XFRM_OFFLOAD
4731         /* Disable XFRM features if this isn't an active-backup config */
4732         if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)
4733                 bond_dev->features &= ~BOND_XFRM_FEATURES;
4734 #endif /* CONFIG_XFRM_OFFLOAD */
4735 }
4736
4737 /* Destroy a bonding device.
4738  * Must be under rtnl_lock when this function is called.
4739  */
4740 static void bond_uninit(struct net_device *bond_dev)
4741 {
4742         struct bonding *bond = netdev_priv(bond_dev);
4743         struct bond_up_slave *usable, *all;
4744         struct list_head *iter;
4745         struct slave *slave;
4746
4747         bond_netpoll_cleanup(bond_dev);
4748
4749         /* Release the bonded slaves */
4750         bond_for_each_slave(bond, slave, iter)
4751                 __bond_release_one(bond_dev, slave->dev, true, true);
4752         netdev_info(bond_dev, "Released all slaves\n");
4753
4754         usable = rtnl_dereference(bond->usable_slaves);
4755         if (usable) {
4756                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4757                 kfree_rcu(usable, rcu);
4758         }
4759
4760         all = rtnl_dereference(bond->all_slaves);
4761         if (all) {
4762                 RCU_INIT_POINTER(bond->all_slaves, NULL);
4763                 kfree_rcu(all, rcu);
4764         }
4765
4766         list_del(&bond->bond_list);
4767
4768         bond_debug_unregister(bond);
4769 }
4770
4771 /*------------------------- Module initialization ---------------------------*/
4772
4773 static int bond_check_params(struct bond_params *params)
4774 {
4775         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4776         struct bond_opt_value newval;
4777         const struct bond_opt_value *valptr;
4778         int arp_all_targets_value = 0;
4779         u16 ad_actor_sys_prio = 0;
4780         u16 ad_user_port_key = 0;
4781         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4782         int arp_ip_count;
4783         int bond_mode   = BOND_MODE_ROUNDROBIN;
4784         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4785         int lacp_fast = 0;
4786         int tlb_dynamic_lb;
4787
4788         /* Convert string parameters. */
4789         if (mode) {
4790                 bond_opt_initstr(&newval, mode);
4791                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4792                 if (!valptr) {
4793                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4794                         return -EINVAL;
4795                 }
4796                 bond_mode = valptr->value;
4797         }
4798
4799         if (xmit_hash_policy) {
4800                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4801                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
4802                     bond_mode == BOND_MODE_BROADCAST) {
4803                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4804                                 bond_mode_name(bond_mode));
4805                 } else {
4806                         bond_opt_initstr(&newval, xmit_hash_policy);
4807                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4808                                                 &newval);
4809                         if (!valptr) {
4810                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4811                                        xmit_hash_policy);
4812                                 return -EINVAL;
4813                         }
4814                         xmit_hashtype = valptr->value;
4815                 }
4816         }
4817
4818         if (lacp_rate) {
4819                 if (bond_mode != BOND_MODE_8023AD) {
4820                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4821                                 bond_mode_name(bond_mode));
4822                 } else {
4823                         bond_opt_initstr(&newval, lacp_rate);
4824                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4825                                                 &newval);
4826                         if (!valptr) {
4827                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4828                                        lacp_rate);
4829                                 return -EINVAL;
4830                         }
4831                         lacp_fast = valptr->value;
4832                 }
4833         }
4834
4835         if (ad_select) {
4836                 bond_opt_initstr(&newval, ad_select);
4837                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4838                                         &newval);
4839                 if (!valptr) {
4840                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4841                         return -EINVAL;
4842                 }
4843                 params->ad_select = valptr->value;
4844                 if (bond_mode != BOND_MODE_8023AD)
4845                         pr_warn("ad_select param only affects 802.3ad mode\n");
4846         } else {
4847                 params->ad_select = BOND_AD_STABLE;
4848         }
4849
4850         if (max_bonds < 0) {
4851                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4852                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4853                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4854         }
4855
4856         if (miimon < 0) {
4857                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4858                         miimon, INT_MAX);
4859                 miimon = 0;
4860         }
4861
4862         if (updelay < 0) {
4863                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4864                         updelay, INT_MAX);
4865                 updelay = 0;
4866         }
4867
4868         if (downdelay < 0) {
4869                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4870                         downdelay, INT_MAX);
4871                 downdelay = 0;
4872         }
4873
4874         if ((use_carrier != 0) && (use_carrier != 1)) {
4875                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4876                         use_carrier);
4877                 use_carrier = 1;
4878         }
4879
4880         if (num_peer_notif < 0 || num_peer_notif > 255) {
4881                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4882                         num_peer_notif);
4883                 num_peer_notif = 1;
4884         }
4885
4886         /* reset values for 802.3ad/TLB/ALB */
4887         if (!bond_mode_uses_arp(bond_mode)) {
4888                 if (!miimon) {
4889                         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");
4890                         pr_warn("Forcing miimon to 100msec\n");
4891                         miimon = BOND_DEFAULT_MIIMON;
4892                 }
4893         }
4894
4895         if (tx_queues < 1 || tx_queues > 255) {
4896                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4897                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4898                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4899         }
4900
4901         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4902                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4903                         all_slaves_active);
4904                 all_slaves_active = 0;
4905         }
4906
4907         if (resend_igmp < 0 || resend_igmp > 255) {
4908                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4909                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4910                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4911         }
4912
4913         bond_opt_initval(&newval, packets_per_slave);
4914         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4915                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4916                         packets_per_slave, USHRT_MAX);
4917                 packets_per_slave = 1;
4918         }
4919
4920         if (bond_mode == BOND_MODE_ALB) {
4921                 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",
4922                           updelay);
4923         }
4924
4925         if (!miimon) {
4926                 if (updelay || downdelay) {
4927                         /* just warn the user the up/down delay will have
4928                          * no effect since miimon is zero...
4929                          */
4930                         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",
4931                                 updelay, downdelay);
4932                 }
4933         } else {
4934                 /* don't allow arp monitoring */
4935                 if (arp_interval) {
4936                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4937                                 miimon, arp_interval);
4938                         arp_interval = 0;
4939                 }
4940
4941                 if ((updelay % miimon) != 0) {
4942                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4943                                 updelay, miimon, (updelay / miimon) * miimon);
4944                 }
4945
4946                 updelay /= miimon;
4947
4948                 if ((downdelay % miimon) != 0) {
4949                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4950                                 downdelay, miimon,
4951                                 (downdelay / miimon) * miimon);
4952                 }
4953
4954                 downdelay /= miimon;
4955         }
4956
4957         if (arp_interval < 0) {
4958                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4959                         arp_interval, INT_MAX);
4960                 arp_interval = 0;
4961         }
4962
4963         for (arp_ip_count = 0, i = 0;
4964              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4965                 __be32 ip;
4966
4967                 /* not a complete check, but good enough to catch mistakes */
4968                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4969                     !bond_is_ip_target_ok(ip)) {
4970                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4971                                 arp_ip_target[i]);
4972                         arp_interval = 0;
4973                 } else {
4974                         if (bond_get_targets_ip(arp_target, ip) == -1)
4975                                 arp_target[arp_ip_count++] = ip;
4976                         else
4977                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4978                                         &ip);
4979                 }
4980         }
4981
4982         if (arp_interval && !arp_ip_count) {
4983                 /* don't allow arping if no arp_ip_target given... */
4984                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4985                         arp_interval);
4986                 arp_interval = 0;
4987         }
4988
4989         if (arp_validate) {
4990                 if (!arp_interval) {
4991                         pr_err("arp_validate requires arp_interval\n");
4992                         return -EINVAL;
4993                 }
4994
4995                 bond_opt_initstr(&newval, arp_validate);
4996                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4997                                         &newval);
4998                 if (!valptr) {
4999                         pr_err("Error: invalid arp_validate \"%s\"\n",
5000                                arp_validate);
5001                         return -EINVAL;
5002                 }
5003                 arp_validate_value = valptr->value;
5004         } else {
5005                 arp_validate_value = 0;
5006         }
5007
5008         if (arp_all_targets) {
5009                 bond_opt_initstr(&newval, arp_all_targets);
5010                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
5011                                         &newval);
5012                 if (!valptr) {
5013                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
5014                                arp_all_targets);
5015                         arp_all_targets_value = 0;
5016                 } else {
5017                         arp_all_targets_value = valptr->value;
5018                 }
5019         }
5020
5021         if (miimon) {
5022                 pr_info("MII link monitoring set to %d ms\n", miimon);
5023         } else if (arp_interval) {
5024                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
5025                                           arp_validate_value);
5026                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5027                         arp_interval, valptr->string, arp_ip_count);
5028
5029                 for (i = 0; i < arp_ip_count; i++)
5030                         pr_cont(" %s", arp_ip_target[i]);
5031
5032                 pr_cont("\n");
5033
5034         } else if (max_bonds) {
5035                 /* miimon and arp_interval not set, we need one so things
5036                  * work as expected, see bonding.txt for details
5037                  */
5038                 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");
5039         }
5040
5041         if (primary && !bond_mode_uses_primary(bond_mode)) {
5042                 /* currently, using a primary only makes sense
5043                  * in active backup, TLB or ALB modes
5044                  */
5045                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
5046                         primary, bond_mode_name(bond_mode));
5047                 primary = NULL;
5048         }
5049
5050         if (primary && primary_reselect) {
5051                 bond_opt_initstr(&newval, primary_reselect);
5052                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
5053                                         &newval);
5054                 if (!valptr) {
5055                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5056                                primary_reselect);
5057                         return -EINVAL;
5058                 }
5059                 primary_reselect_value = valptr->value;
5060         } else {
5061                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5062         }
5063
5064         if (fail_over_mac) {
5065                 bond_opt_initstr(&newval, fail_over_mac);
5066                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
5067                                         &newval);
5068                 if (!valptr) {
5069                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5070                                fail_over_mac);
5071                         return -EINVAL;
5072                 }
5073                 fail_over_mac_value = valptr->value;
5074                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5075                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
5076         } else {
5077                 fail_over_mac_value = BOND_FOM_NONE;
5078         }
5079
5080         bond_opt_initstr(&newval, "default");
5081         valptr = bond_opt_parse(
5082                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
5083                                      &newval);
5084         if (!valptr) {
5085                 pr_err("Error: No ad_actor_sys_prio default value");
5086                 return -EINVAL;
5087         }
5088         ad_actor_sys_prio = valptr->value;
5089
5090         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
5091                                 &newval);
5092         if (!valptr) {
5093                 pr_err("Error: No ad_user_port_key default value");
5094                 return -EINVAL;
5095         }
5096         ad_user_port_key = valptr->value;
5097
5098         bond_opt_initstr(&newval, "default");
5099         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
5100         if (!valptr) {
5101                 pr_err("Error: No tlb_dynamic_lb default value");
5102                 return -EINVAL;
5103         }
5104         tlb_dynamic_lb = valptr->value;
5105
5106         if (lp_interval == 0) {
5107                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
5108                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
5109                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
5110         }
5111
5112         /* fill params struct with the proper values */
5113         params->mode = bond_mode;
5114         params->xmit_policy = xmit_hashtype;
5115         params->miimon = miimon;
5116         params->num_peer_notif = num_peer_notif;
5117         params->arp_interval = arp_interval;
5118         params->arp_validate = arp_validate_value;
5119         params->arp_all_targets = arp_all_targets_value;
5120         params->updelay = updelay;
5121         params->downdelay = downdelay;
5122         params->peer_notif_delay = 0;
5123         params->use_carrier = use_carrier;
5124         params->lacp_fast = lacp_fast;
5125         params->primary[0] = 0;
5126         params->primary_reselect = primary_reselect_value;
5127         params->fail_over_mac = fail_over_mac_value;
5128         params->tx_queues = tx_queues;
5129         params->all_slaves_active = all_slaves_active;
5130         params->resend_igmp = resend_igmp;
5131         params->min_links = min_links;
5132         params->lp_interval = lp_interval;
5133         params->packets_per_slave = packets_per_slave;
5134         params->tlb_dynamic_lb = tlb_dynamic_lb;
5135         params->ad_actor_sys_prio = ad_actor_sys_prio;
5136         eth_zero_addr(params->ad_actor_system);
5137         params->ad_user_port_key = ad_user_port_key;
5138         if (packets_per_slave > 0) {
5139                 params->reciprocal_packets_per_slave =
5140                         reciprocal_value(packets_per_slave);
5141         } else {
5142                 /* reciprocal_packets_per_slave is unused if
5143                  * packets_per_slave is 0 or 1, just initialize it
5144                  */
5145                 params->reciprocal_packets_per_slave =
5146                         (struct reciprocal_value) { 0 };
5147         }
5148
5149         if (primary) {
5150                 strncpy(params->primary, primary, IFNAMSIZ);
5151                 params->primary[IFNAMSIZ - 1] = 0;
5152         }
5153
5154         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5155
5156         return 0;
5157 }
5158
5159 /* Called from registration process */
5160 static int bond_init(struct net_device *bond_dev)
5161 {
5162         struct bonding *bond = netdev_priv(bond_dev);
5163         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5164
5165         netdev_dbg(bond_dev, "Begin bond_init\n");
5166
5167         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
5168         if (!bond->wq)
5169                 return -ENOMEM;
5170
5171         spin_lock_init(&bond->stats_lock);
5172         netdev_lockdep_set_classes(bond_dev);
5173
5174         list_add_tail(&bond->bond_list, &bn->dev_list);
5175
5176         bond_prepare_sysfs_group(bond);
5177
5178         bond_debug_register(bond);
5179
5180         /* Ensure valid dev_addr */
5181         if (is_zero_ether_addr(bond_dev->dev_addr) &&
5182             bond_dev->addr_assign_type == NET_ADDR_PERM)
5183                 eth_hw_addr_random(bond_dev);
5184
5185         return 0;
5186 }
5187
5188 unsigned int bond_get_num_tx_queues(void)
5189 {
5190         return tx_queues;
5191 }
5192
5193 /* Create a new bond based on the specified name and bonding parameters.
5194  * If name is NULL, obtain a suitable "bond%d" name for us.
5195  * Caller must NOT hold rtnl_lock; we need to release it here before we
5196  * set up our sysfs entries.
5197  */
5198 int bond_create(struct net *net, const char *name)
5199 {
5200         struct net_device *bond_dev;
5201         struct bonding *bond;
5202         struct alb_bond_info *bond_info;
5203         int res;
5204
5205         rtnl_lock();
5206
5207         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
5208                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
5209                                    bond_setup, tx_queues);
5210         if (!bond_dev) {
5211                 pr_err("%s: eek! can't alloc netdev!\n", name);
5212                 rtnl_unlock();
5213                 return -ENOMEM;
5214         }
5215
5216         /*
5217          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
5218          * It is set to 0 by default which is wrong.
5219          */
5220         bond = netdev_priv(bond_dev);
5221         bond_info = &(BOND_ALB_INFO(bond));
5222         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
5223
5224         dev_net_set(bond_dev, net);
5225         bond_dev->rtnl_link_ops = &bond_link_ops;
5226
5227         res = register_netdevice(bond_dev);
5228         if (res < 0) {
5229                 free_netdev(bond_dev);
5230                 rtnl_unlock();
5231
5232                 return res;
5233         }
5234
5235         netif_carrier_off(bond_dev);
5236
5237         bond_work_init_all(bond);
5238
5239         rtnl_unlock();
5240         return 0;
5241 }
5242
5243 static int __net_init bond_net_init(struct net *net)
5244 {
5245         struct bond_net *bn = net_generic(net, bond_net_id);
5246
5247         bn->net = net;
5248         INIT_LIST_HEAD(&bn->dev_list);
5249
5250         bond_create_proc_dir(bn);
5251         bond_create_sysfs(bn);
5252
5253         return 0;
5254 }
5255
5256 static void __net_exit bond_net_exit(struct net *net)
5257 {
5258         struct bond_net *bn = net_generic(net, bond_net_id);
5259         struct bonding *bond, *tmp_bond;
5260         LIST_HEAD(list);
5261
5262         bond_destroy_sysfs(bn);
5263
5264         /* Kill off any bonds created after unregistering bond rtnl ops */
5265         rtnl_lock();
5266         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
5267                 unregister_netdevice_queue(bond->dev, &list);
5268         unregister_netdevice_many(&list);
5269         rtnl_unlock();
5270
5271         bond_destroy_proc_dir(bn);
5272 }
5273
5274 static struct pernet_operations bond_net_ops = {
5275         .init = bond_net_init,
5276         .exit = bond_net_exit,
5277         .id   = &bond_net_id,
5278         .size = sizeof(struct bond_net),
5279 };
5280
5281 static int __init bonding_init(void)
5282 {
5283         int i;
5284         int res;
5285
5286         res = bond_check_params(&bonding_defaults);
5287         if (res)
5288                 goto out;
5289
5290         res = register_pernet_subsys(&bond_net_ops);
5291         if (res)
5292                 goto out;
5293
5294         res = bond_netlink_init();
5295         if (res)
5296                 goto err_link;
5297
5298         bond_create_debugfs();
5299
5300         for (i = 0; i < max_bonds; i++) {
5301                 res = bond_create(&init_net, NULL);
5302                 if (res)
5303                         goto err;
5304         }
5305
5306         skb_flow_dissector_init(&flow_keys_bonding,
5307                                 flow_keys_bonding_keys,
5308                                 ARRAY_SIZE(flow_keys_bonding_keys));
5309
5310         register_netdevice_notifier(&bond_netdev_notifier);
5311 out:
5312         return res;
5313 err:
5314         bond_destroy_debugfs();
5315         bond_netlink_fini();
5316 err_link:
5317         unregister_pernet_subsys(&bond_net_ops);
5318         goto out;
5319
5320 }
5321
5322 static void __exit bonding_exit(void)
5323 {
5324         unregister_netdevice_notifier(&bond_netdev_notifier);
5325
5326         bond_destroy_debugfs();
5327
5328         bond_netlink_fini();
5329         unregister_pernet_subsys(&bond_net_ops);
5330
5331 #ifdef CONFIG_NET_POLL_CONTROLLER
5332         /* Make sure we don't have an imbalance on our netpoll blocking */
5333         WARN_ON(atomic_read(&netpoll_block_tx));
5334 #endif
5335 }
5336
5337 module_init(bonding_init);
5338 module_exit(bonding_exit);
5339 MODULE_LICENSE("GPL");
5340 MODULE_DESCRIPTION(DRV_DESCRIPTION);
5341 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");