drivers/net/bonding/bond

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
bond_mode_name
bond_dev_queue_xmit
bond_vlan_rx_add_vid
bond_vlan_rx_kill_vid
bond_set_carrier
bond_update_speed_duplex
bond_slave_link_status
bond_check_dev_link
bond_set_promiscuity
bond_set_allmulti
bond_resend_igmp_join_requests_delayed
bond_hw_addr_flush
bond_hw_addr_swap
bond_set_dev_addr
bond_get_old_active
bond_do_fail_over_mac
bond_choose_primary_or_current
bond_find_best_slave
bond_should_notify_peers
bond_change_active_slave
bond_select_active_slave
slave_enable_netpoll
slave_disable_netpoll
bond_poll_controller
bond_netpoll_cleanup
bond_netpoll_setup
slave_enable_netpoll
slave_disable_netpoll
bond_netpoll_cleanup
bond_fix_features
bond_compute_features
bond_setup_by_slave
bond_should_deliver_exact_match
bond_handle_frame
bond_lag_tx_type
bond_lag_hash_type
bond_master_upper_dev_link
bond_upper_dev_unlink
bond_alloc_slave
bond_free_slave
bond_fill_ifbond
bond_fill_ifslave
bond_netdev_notify_work
bond_queue_slave_event
bond_lower_state_changed
bond_enslave
__bond_release_one
bond_release
bond_release_and_destroy
bond_info_query
bond_slave_info_query
bond_miimon_inspect
bond_miimon_link_change
bond_miimon_commit
bond_mii_monitor
bond_upper_dev_walk
bond_has_this_ip
bond_arp_send
bond_verify_device_path
bond_arp_send_all
bond_validate_arp
bond_arp_rcv
bond_time_in_interval
bond_loadbalance_arp_mon
bond_ab_arp_inspect
bond_ab_arp_commit
bond_ab_arp_probe
bond_activebackup_arp_mon
bond_arp_monitor
bond_event_changename
bond_master_netdev_event
bond_slave_netdev_event
bond_netdev_event
bond_eth_hash
bond_flow_dissect
bond_xmit_hash
bond_work_init_all
bond_work_cancel_all
bond_open
bond_close
bond_fold_stats
bond_get_lowest_level_rcu
bond_get_stats
bond_do_ioctl
bond_change_rx_flags
bond_set_rx_mode
bond_neigh_init
bond_neigh_setup
bond_change_mtu
bond_set_mac_address
bond_xmit_slave_id
bond_rr_gen_slave_id
bond_xmit_roundrobin
bond_xmit_activebackup
bond_slave_arr_work_rearm
bond_slave_arr_handler
bond_update_slave_arr
bond_3ad_xor_xmit
bond_xmit_broadcast
bond_slave_override
bond_select_queue
__bond_start_xmit
bond_start_xmit
bond_ethtool_get_link_ksettings
bond_ethtool_get_drvinfo
bond_destructor
bond_setup
bond_uninit
bond_check_params
bond_init
bond_get_num_tx_queues
bond_create
bond_net_init
bond_net_exit
bonding_init
bonding_exit
   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33 
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/udp.h>
  46 #include <linux/slab.h>
  47 #include <linux/string.h>
  48 #include <linux/init.h>
  49 #include <linux/timer.h>
  50 #include <linux/socket.h>
  51 #include <linux/ctype.h>
  52 #include <linux/inet.h>
  53 #include <linux/bitops.h>
  54 #include <linux/io.h>
  55 #include <asm/dma.h>
  56 #include <linux/uaccess.h>
  57 #include <linux/errno.h>
  58 #include <linux/netdevice.h>
  59 #include <linux/inetdevice.h>
  60 #include <linux/igmp.h>
  61 #include <linux/etherdevice.h>
  62 #include <linux/skbuff.h>
  63 #include <net/sock.h>
  64 #include <linux/rtnetlink.h>
  65 #include <linux/smp.h>
  66 #include <linux/if_ether.h>
  67 #include <net/arp.h>
  68 #include <linux/mii.h>
  69 #include <linux/ethtool.h>
  70 #include <linux/if_vlan.h>
  71 #include <linux/if_bonding.h>
  72 #include <linux/jiffies.h>
  73 #include <linux/preempt.h>
  74 #include <net/route.h>
  75 #include <net/net_namespace.h>
  76 #include <net/netns/generic.h>
  77 #include <net/pkt_sched.h>
  78 #include <linux/rculist.h>
  79 #include <net/flow_dissector.h>
  80 #include <net/bonding.h>
  81 #include <net/bond_3ad.h>
  82 #include <net/bond_alb.h>
  83 
  84 #include "bonding_priv.h"
  85 
  86 /*---------------------------- Module parameters ----------------------------*/
  87 
  88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  89 
  90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  92 static int num_peer_notif = 1;
  93 static int miimon;
  94 static int updelay;
  95 static int downdelay;
  96 static int use_carrier  = 1;
  97 static char *mode;
  98 static char *primary;
  99 static char *primary_reselect;
 100 static char *lacp_rate;
 101 static int min_links;
 102 static char *ad_select;
 103 static char *xmit_hash_policy;
 104 static int arp_interval;
 105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 106 static char *arp_validate;
 107 static char *arp_all_targets;
 108 static char *fail_over_mac;
 109 static int all_slaves_active;
 110 static struct bond_params bonding_defaults;
 111 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 112 static int packets_per_slave = 1;
 113 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 114 
 115 module_param(max_bonds, int, 0);
 116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 117 module_param(tx_queues, int, 0);
 118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 119 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 120 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 121                                "failover event (alias of num_unsol_na)");
 122 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 123 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 124                                "failover event (alias of num_grat_arp)");
 125 module_param(miimon, int, 0);
 126 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 127 module_param(updelay, int, 0);
 128 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 129 module_param(downdelay, int, 0);
 130 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 131                             "in milliseconds");
 132 module_param(use_carrier, int, 0);
 133 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 134                               "0 for off, 1 for on (default)");
 135 module_param(mode, charp, 0);
 136 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 137                        "1 for active-backup, 2 for balance-xor, "
 138                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 139                        "6 for balance-alb");
 140 module_param(primary, charp, 0);
 141 MODULE_PARM_DESC(primary, "Primary network device to use");
 142 module_param(primary_reselect, charp, 0);
 143 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 144                                    "once it comes up; "
 145                                    "0 for always (default), "
 146                                    "1 for only if speed of primary is "
 147                                    "better, "
 148                                    "2 for only on active slave "
 149                                    "failure");
 150 module_param(lacp_rate, charp, 0);
 151 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 152                             "0 for slow, 1 for fast");
 153 module_param(ad_select, charp, 0);
 154 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
 155                             "0 for stable (default), 1 for bandwidth, "
 156                             "2 for count");
 157 module_param(min_links, int, 0);
 158 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 159 
 160 module_param(xmit_hash_policy, charp, 0);
 161 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
 162                                    "0 for layer 2 (default), 1 for layer 3+4, "
 163                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 164                                    "4 for encap layer 3+4");
 165 module_param(arp_interval, int, 0);
 166 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 167 module_param_array(arp_ip_target, charp, NULL, 0);
 168 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 169 module_param(arp_validate, charp, 0);
 170 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 171                                "0 for none (default), 1 for active, "
 172                                "2 for backup, 3 for all");
 173 module_param(arp_all_targets, charp, 0);
 174 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 175 module_param(fail_over_mac, charp, 0);
 176 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 177                                 "the same MAC; 0 for none (default), "
 178                                 "1 for active, 2 for follow");
 179 module_param(all_slaves_active, int, 0);
 180 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 181                                      "by setting active flag for all slaves; "
 182                                      "0 for never (default), 1 for always.");
 183 module_param(resend_igmp, int, 0);
 184 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 185                               "link failure");
 186 module_param(packets_per_slave, int, 0);
 187 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 188                                     "mode; 0 for a random slave, 1 packet per "
 189                                     "slave (default), >1 packets per slave.");
 190 module_param(lp_interval, uint, 0);
 191 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 192                               "the bonding driver sends learning packets to "
 193                               "each slaves peer switch. The default is 1.");
 194 
 195 /*----------------------------- Global variables ----------------------------*/
 196 
 197 #ifdef CONFIG_NET_POLL_CONTROLLER
 198 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 199 #endif
 200 
 201 unsigned int bond_net_id __read_mostly;
 202 
 203 /*-------------------------- Forward declarations ---------------------------*/
 204 
 205 static int bond_init(struct net_device *bond_dev);
 206 static void bond_uninit(struct net_device *bond_dev);
 207 static void bond_get_stats(struct net_device *bond_dev,
 208                            struct rtnl_link_stats64 *stats);
 209 static void bond_slave_arr_handler(struct work_struct *work);
 210 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 211                                   int mod);
 212 static void bond_netdev_notify_work(struct work_struct *work);
 213 
 214 /*---------------------------- General routines -----------------------------*/
 215 
 216 const char *bond_mode_name(int mode)
 217 {
 218         static const char *names[] = {
 219                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 220                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 221                 [BOND_MODE_XOR] = "load balancing (xor)",
 222                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 223                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 224                 [BOND_MODE_TLB] = "transmit load balancing",
 225                 [BOND_MODE_ALB] = "adaptive load balancing",
 226         };
 227 
 228         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 229                 return "unknown";
 230 
 231         return names[mode];
 232 }
 233 
 234 /*---------------------------------- VLAN -----------------------------------*/
 235 
 236 /**
 237  * bond_dev_queue_xmit - Prepare skb for xmit.
 238  *
 239  * @bond: bond device that got this skb for tx.
 240  * @skb: hw accel VLAN tagged skb to transmit
 241  * @slave_dev: slave that is supposed to xmit this skbuff
 242  */
 243 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 244                         struct net_device *slave_dev)
 245 {
 246         skb->dev = slave_dev;
 247 
 248         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 249                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 250         skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 251 
 252         if (unlikely(netpoll_tx_running(bond->dev)))
 253                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 254         else
 255                 dev_queue_xmit(skb);
 256 }
 257 
 258 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 259  * We don't protect the slave list iteration with a lock because:
 260  * a. This operation is performed in IOCTL context,
 261  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 262  * c. Holding a lock with BH disabled while directly calling a base driver
 263  *    entry point is generally a BAD idea.
 264  *
 265  * The design of synchronization/protection for this operation in the 8021q
 266  * module is good for one or more VLAN devices over a single physical device
 267  * and cannot be extended for a teaming solution like bonding, so there is a
 268  * potential race condition here where a net device from the vlan group might
 269  * be referenced (either by a base driver or the 8021q code) while it is being
 270  * removed from the system. However, it turns out we're not making matters
 271  * worse, and if it works for regular VLAN usage it will work here too.
 272 */
 273 
 274 /**
 275  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 276  * @bond_dev: bonding net device that got called
 277  * @vid: vlan id being added
 278  */
 279 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 280                                 __be16 proto, u16 vid)
 281 {
 282         struct bonding *bond = netdev_priv(bond_dev);
 283         struct slave *slave, *rollback_slave;
 284         struct list_head *iter;
 285         int res;
 286 
 287         bond_for_each_slave(bond, slave, iter) {
 288                 res = vlan_vid_add(slave->dev, proto, vid);
 289                 if (res)
 290                         goto unwind;
 291         }
 292 
 293         return 0;
 294 
 295 unwind:
 296         /* unwind to the slave that failed */
 297         bond_for_each_slave(bond, rollback_slave, iter) {
 298                 if (rollback_slave == slave)
 299                         break;
 300 
 301                 vlan_vid_del(rollback_slave->dev, proto, vid);
 302         }
 303 
 304         return res;
 305 }
 306 
 307 /**
 308  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 309  * @bond_dev: bonding net device that got called
 310  * @vid: vlan id being removed
 311  */
 312 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 313                                  __be16 proto, u16 vid)
 314 {
 315         struct bonding *bond = netdev_priv(bond_dev);
 316         struct list_head *iter;
 317         struct slave *slave;
 318 
 319         bond_for_each_slave(bond, slave, iter)
 320                 vlan_vid_del(slave->dev, proto, vid);
 321 
 322         if (bond_is_lb(bond))
 323                 bond_alb_clear_vlan(bond, vid);
 324 
 325         return 0;
 326 }
 327 
 328 /*------------------------------- Link status -------------------------------*/
 329 
 330 /* Set the carrier state for the master according to the state of its
 331  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 332  * do special 802.3ad magic.
 333  *
 334  * Returns zero if carrier state does not change, nonzero if it does.
 335  */
 336 int bond_set_carrier(struct bonding *bond)
 337 {
 338         struct list_head *iter;
 339         struct slave *slave;
 340 
 341         if (!bond_has_slaves(bond))
 342                 goto down;
 343 
 344         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 345                 return bond_3ad_set_carrier(bond);
 346 
 347         bond_for_each_slave(bond, slave, iter) {
 348                 if (slave->link == BOND_LINK_UP) {
 349                         if (!netif_carrier_ok(bond->dev)) {
 350                                 netif_carrier_on(bond->dev);
 351                                 return 1;
 352                         }
 353                         return 0;
 354                 }
 355         }
 356 
 357 down:
 358         if (netif_carrier_ok(bond->dev)) {
 359                 netif_carrier_off(bond->dev);
 360                 return 1;
 361         }
 362         return 0;
 363 }
 364 
 365 /* Get link speed and duplex from the slave's base driver
 366  * using ethtool. If for some reason the call fails or the
 367  * values are invalid, set speed and duplex to -1,
 368  * and return. Return 1 if speed or duplex settings are
 369  * UNKNOWN; 0 otherwise.
 370  */
 371 static int bond_update_speed_duplex(struct slave *slave)
 372 {
 373         struct net_device *slave_dev = slave->dev;
 374         struct ethtool_link_ksettings ecmd;
 375         int res;
 376 
 377         slave->speed = SPEED_UNKNOWN;
 378         slave->duplex = DUPLEX_UNKNOWN;
 379 
 380         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 381         if (res < 0)
 382                 return 1;
 383         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 384                 return 1;
 385         switch (ecmd.base.duplex) {
 386         case DUPLEX_FULL:
 387         case DUPLEX_HALF:
 388                 break;
 389         default:
 390                 return 1;
 391         }
 392 
 393         slave->speed = ecmd.base.speed;
 394         slave->duplex = ecmd.base.duplex;
 395 
 396         return 0;
 397 }
 398 
 399 const char *bond_slave_link_status(s8 link)
 400 {
 401         switch (link) {
 402         case BOND_LINK_UP:
 403                 return "up";
 404         case BOND_LINK_FAIL:
 405                 return "going down";
 406         case BOND_LINK_DOWN:
 407                 return "down";
 408         case BOND_LINK_BACK:
 409                 return "going back";
 410         default:
 411                 return "unknown";
 412         }
 413 }
 414 
 415 /* if <dev> supports MII link status reporting, check its link status.
 416  *
 417  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 418  * depending upon the setting of the use_carrier parameter.
 419  *
 420  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 421  * can't tell and just pretend it is), or 0, meaning that the link is
 422  * down.
 423  *
 424  * If reporting is non-zero, instead of faking link up, return -1 if
 425  * both ETHTOOL and MII ioctls fail (meaning the device does not
 426  * support them).  If use_carrier is set, return whatever it says.
 427  * It'd be nice if there was a good way to tell if a driver supports
 428  * netif_carrier, but there really isn't.
 429  */
 430 static int bond_check_dev_link(struct bonding *bond,
 431                                struct net_device *slave_dev, int reporting)
 432 {
 433         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 434         int (*ioctl)(struct net_device *, struct ifreq *, int);
 435         struct ifreq ifr;
 436         struct mii_ioctl_data *mii;
 437 
 438         if (!reporting && !netif_running(slave_dev))
 439                 return 0;
 440 
 441         if (bond->params.use_carrier)
 442                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 443 
 444         /* Try to get link status using Ethtool first. */
 445         if (slave_dev->ethtool_ops->get_link)
 446                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 447                         BMSR_LSTATUS : 0;
 448 
 449         /* Ethtool can't be used, fallback to MII ioctls. */
 450         ioctl = slave_ops->ndo_do_ioctl;
 451         if (ioctl) {
 452                 /* TODO: set pointer to correct ioctl on a per team member
 453                  *       bases to make this more efficient. that is, once
 454                  *       we determine the correct ioctl, we will always
 455                  *       call it and not the others for that team
 456                  *       member.
 457                  */
 458 
 459                 /* We cannot assume that SIOCGMIIPHY will also read a
 460                  * register; not all network drivers (e.g., e100)
 461                  * support that.
 462                  */
 463 
 464                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 465                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 466                 mii = if_mii(&ifr);
 467                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 468                         mii->reg_num = MII_BMSR;
 469                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 470                                 return mii->val_out & BMSR_LSTATUS;
 471                 }
 472         }
 473 
 474         /* If reporting, report that either there's no dev->do_ioctl,
 475          * or both SIOCGMIIREG and get_link failed (meaning that we
 476          * cannot report link status).  If not reporting, pretend
 477          * we're ok.
 478          */
 479         return reporting ? -1 : BMSR_LSTATUS;
 480 }
 481 
 482 /*----------------------------- Multicast list ------------------------------*/
 483 
 484 /* Push the promiscuity flag down to appropriate slaves */
 485 static int bond_set_promiscuity(struct bonding *bond, int inc)
 486 {
 487         struct list_head *iter;
 488         int err = 0;
 489 
 490         if (bond_uses_primary(bond)) {
 491                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 492 
 493                 if (curr_active)
 494                         err = dev_set_promiscuity(curr_active->dev, inc);
 495         } else {
 496                 struct slave *slave;
 497 
 498                 bond_for_each_slave(bond, slave, iter) {
 499                         err = dev_set_promiscuity(slave->dev, inc);
 500                         if (err)
 501                                 return err;
 502                 }
 503         }
 504         return err;
 505 }
 506 
 507 /* Push the allmulti flag down to all slaves */
 508 static int bond_set_allmulti(struct bonding *bond, int inc)
 509 {
 510         struct list_head *iter;
 511         int err = 0;
 512 
 513         if (bond_uses_primary(bond)) {
 514                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 515 
 516                 if (curr_active)
 517                         err = dev_set_allmulti(curr_active->dev, inc);
 518         } else {
 519                 struct slave *slave;
 520 
 521                 bond_for_each_slave(bond, slave, iter) {
 522                         err = dev_set_allmulti(slave->dev, inc);
 523                         if (err)
 524                                 return err;
 525                 }
 526         }
 527         return err;
 528 }
 529 
 530 /* Retrieve the list of registered multicast addresses for the bonding
 531  * device and retransmit an IGMP JOIN request to the current active
 532  * slave.
 533  */
 534 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 535 {
 536         struct bonding *bond = container_of(work, struct bonding,
 537                                             mcast_work.work);
 538 
 539         if (!rtnl_trylock()) {
 540                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 541                 return;
 542         }
 543         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 544 
 545         if (bond->igmp_retrans > 1) {
 546                 bond->igmp_retrans--;
 547                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 548         }
 549         rtnl_unlock();
 550 }
 551 
 552 /* Flush bond's hardware addresses from slave */
 553 static void bond_hw_addr_flush(struct net_device *bond_dev,
 554                                struct net_device *slave_dev)
 555 {
 556         struct bonding *bond = netdev_priv(bond_dev);
 557 
 558         dev_uc_unsync(slave_dev, bond_dev);
 559         dev_mc_unsync(slave_dev, bond_dev);
 560 
 561         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 562                 /* del lacpdu mc addr from mc list */
 563                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 564 
 565                 dev_mc_del(slave_dev, lacpdu_multicast);
 566         }
 567 }
 568 
 569 /*--------------------------- Active slave change ---------------------------*/
 570 
 571 /* Update the hardware address list and promisc/allmulti for the new and
 572  * old active slaves (if any).  Modes that are not using primary keep all
 573  * slaves up date at all times; only the modes that use primary need to call
 574  * this function to swap these settings during a failover.
 575  */
 576 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 577                               struct slave *old_active)
 578 {
 579         if (old_active) {
 580                 if (bond->dev->flags & IFF_PROMISC)
 581                         dev_set_promiscuity(old_active->dev, -1);
 582 
 583                 if (bond->dev->flags & IFF_ALLMULTI)
 584                         dev_set_allmulti(old_active->dev, -1);
 585 
 586                 bond_hw_addr_flush(bond->dev, old_active->dev);
 587         }
 588 
 589         if (new_active) {
 590                 /* FIXME: Signal errors upstream. */
 591                 if (bond->dev->flags & IFF_PROMISC)
 592                         dev_set_promiscuity(new_active->dev, 1);
 593 
 594                 if (bond->dev->flags & IFF_ALLMULTI)
 595                         dev_set_allmulti(new_active->dev, 1);
 596 
 597                 netif_addr_lock_bh(bond->dev);
 598                 dev_uc_sync(new_active->dev, bond->dev);
 599                 dev_mc_sync(new_active->dev, bond->dev);
 600                 netif_addr_unlock_bh(bond->dev);
 601         }
 602 }
 603 
 604 /**
 605  * bond_set_dev_addr - clone slave's address to bond
 606  * @bond_dev: bond net device
 607  * @slave_dev: slave net device
 608  *
 609  * Should be called with RTNL held.
 610  */
 611 static int bond_set_dev_addr(struct net_device *bond_dev,
 612                              struct net_device *slave_dev)
 613 {
 614         int err;
 615 
 616         slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 617                   bond_dev, slave_dev, slave_dev->addr_len);
 618         err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
 619         if (err)
 620                 return err;
 621 
 622         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 623         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 624         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 625         return 0;
 626 }
 627 
 628 static struct slave *bond_get_old_active(struct bonding *bond,
 629                                          struct slave *new_active)
 630 {
 631         struct slave *slave;
 632         struct list_head *iter;
 633 
 634         bond_for_each_slave(bond, slave, iter) {
 635                 if (slave == new_active)
 636                         continue;
 637 
 638                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 639                         return slave;
 640         }
 641 
 642         return NULL;
 643 }
 644 
 645 /* bond_do_fail_over_mac
 646  *
 647  * Perform special MAC address swapping for fail_over_mac settings
 648  *
 649  * Called with RTNL
 650  */
 651 static void bond_do_fail_over_mac(struct bonding *bond,
 652                                   struct slave *new_active,
 653                                   struct slave *old_active)
 654 {
 655         u8 tmp_mac[MAX_ADDR_LEN];
 656         struct sockaddr_storage ss;
 657         int rv;
 658 
 659         switch (bond->params.fail_over_mac) {
 660         case BOND_FOM_ACTIVE:
 661                 if (new_active) {
 662                         rv = bond_set_dev_addr(bond->dev, new_active->dev);
 663                         if (rv)
 664                                 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
 665                                           -rv);
 666                 }
 667                 break;
 668         case BOND_FOM_FOLLOW:
 669                 /* if new_active && old_active, swap them
 670                  * if just old_active, do nothing (going to no active slave)
 671                  * if just new_active, set new_active to bond's MAC
 672                  */
 673                 if (!new_active)
 674                         return;
 675 
 676                 if (!old_active)
 677                         old_active = bond_get_old_active(bond, new_active);
 678 
 679                 if (old_active) {
 680                         bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 681                                           new_active->dev->addr_len);
 682                         bond_hw_addr_copy(ss.__data,
 683                                           old_active->dev->dev_addr,
 684                                           old_active->dev->addr_len);
 685                         ss.ss_family = new_active->dev->type;
 686                 } else {
 687                         bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 688                                           bond->dev->addr_len);
 689                         ss.ss_family = bond->dev->type;
 690                 }
 691 
 692                 rv = dev_set_mac_address(new_active->dev,
 693                                          (struct sockaddr *)&ss, NULL);
 694                 if (rv) {
 695                         slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
 696                                   -rv);
 697                         goto out;
 698                 }
 699 
 700                 if (!old_active)
 701                         goto out;
 702 
 703                 bond_hw_addr_copy(ss.__data, tmp_mac,
 704                                   new_active->dev->addr_len);
 705                 ss.ss_family = old_active->dev->type;
 706 
 707                 rv = dev_set_mac_address(old_active->dev,
 708                                          (struct sockaddr *)&ss, NULL);
 709                 if (rv)
 710                         slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
 711                                   -rv);
 712 out:
 713                 break;
 714         default:
 715                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
 716                            bond->params.fail_over_mac);
 717                 break;
 718         }
 719 
 720 }
 721 
 722 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
 723 {
 724         struct slave *prim = rtnl_dereference(bond->primary_slave);
 725         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
 726 
 727         if (!prim || prim->link != BOND_LINK_UP) {
 728                 if (!curr || curr->link != BOND_LINK_UP)
 729                         return NULL;
 730                 return curr;
 731         }
 732 
 733         if (bond->force_primary) {
 734                 bond->force_primary = false;
 735                 return prim;
 736         }
 737 
 738         if (!curr || curr->link != BOND_LINK_UP)
 739                 return prim;
 740 
 741         /* At this point, prim and curr are both up */
 742         switch (bond->params.primary_reselect) {
 743         case BOND_PRI_RESELECT_ALWAYS:
 744                 return prim;
 745         case BOND_PRI_RESELECT_BETTER:
 746                 if (prim->speed < curr->speed)
 747                         return curr;
 748                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
 749                         return curr;
 750                 return prim;
 751         case BOND_PRI_RESELECT_FAILURE:
 752                 return curr;
 753         default:
 754                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
 755                            bond->params.primary_reselect);
 756                 return curr;
 757         }
 758 }
 759 
 760 /**
 761  * bond_find_best_slave - select the best available slave to be the active one
 762  * @bond: our bonding struct
 763  */
 764 static struct slave *bond_find_best_slave(struct bonding *bond)
 765 {
 766         struct slave *slave, *bestslave = NULL;
 767         struct list_head *iter;
 768         int mintime = bond->params.updelay;
 769 
 770         slave = bond_choose_primary_or_current(bond);
 771         if (slave)
 772                 return slave;
 773 
 774         bond_for_each_slave(bond, slave, iter) {
 775                 if (slave->link == BOND_LINK_UP)
 776                         return slave;
 777                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
 778                     slave->delay < mintime) {
 779                         mintime = slave->delay;
 780                         bestslave = slave;
 781                 }
 782         }
 783 
 784         return bestslave;
 785 }
 786 
 787 static bool bond_should_notify_peers(struct bonding *bond)
 788 {
 789         struct slave *slave;
 790 
 791         rcu_read_lock();
 792         slave = rcu_dereference(bond->curr_active_slave);
 793         rcu_read_unlock();
 794 
 795         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
 796                    slave ? slave->dev->name : "NULL");
 797 
 798         if (!slave || !bond->send_peer_notif ||
 799             bond->send_peer_notif %
 800             max(1, bond->params.peer_notif_delay) != 0 ||
 801             !netif_carrier_ok(bond->dev) ||
 802             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
 803                 return false;
 804 
 805         return true;
 806 }
 807 
 808 /**
 809  * change_active_interface - change the active slave into the specified one
 810  * @bond: our bonding struct
 811  * @new: the new slave to make the active one
 812  *
 813  * Set the new slave to the bond's settings and unset them on the old
 814  * curr_active_slave.
 815  * Setting include flags, mc-list, promiscuity, allmulti, etc.
 816  *
 817  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
 818  * because it is apparently the best available slave we have, even though its
 819  * updelay hasn't timed out yet.
 820  *
 821  * Caller must hold RTNL.
 822  */
 823 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
 824 {
 825         struct slave *old_active;
 826 
 827         ASSERT_RTNL();
 828 
 829         old_active = rtnl_dereference(bond->curr_active_slave);
 830 
 831         if (old_active == new_active)
 832                 return;
 833 
 834         if (new_active) {
 835                 new_active->last_link_up = jiffies;
 836 
 837                 if (new_active->link == BOND_LINK_BACK) {
 838                         if (bond_uses_primary(bond)) {
 839                                 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
 840                                            (bond->params.updelay - new_active->delay) * bond->params.miimon);
 841                         }
 842 
 843                         new_active->delay = 0;
 844                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
 845                                                   BOND_SLAVE_NOTIFY_NOW);
 846 
 847                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 848                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
 849 
 850                         if (bond_is_lb(bond))
 851                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
 852                 } else {
 853                         if (bond_uses_primary(bond)) {
 854                                 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
 855                         }
 856                 }
 857         }
 858 
 859         if (bond_uses_primary(bond))
 860                 bond_hw_addr_swap(bond, new_active, old_active);
 861 
 862         if (bond_is_lb(bond)) {
 863                 bond_alb_handle_active_change(bond, new_active);
 864                 if (old_active)
 865                         bond_set_slave_inactive_flags(old_active,
 866                                                       BOND_SLAVE_NOTIFY_NOW);
 867                 if (new_active)
 868                         bond_set_slave_active_flags(new_active,
 869                                                     BOND_SLAVE_NOTIFY_NOW);
 870         } else {
 871                 rcu_assign_pointer(bond->curr_active_slave, new_active);
 872         }
 873 
 874         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
 875                 if (old_active)
 876                         bond_set_slave_inactive_flags(old_active,
 877                                                       BOND_SLAVE_NOTIFY_NOW);
 878 
 879                 if (new_active) {
 880                         bool should_notify_peers = false;
 881 
 882                         bond_set_slave_active_flags(new_active,
 883                                                     BOND_SLAVE_NOTIFY_NOW);
 884 
 885                         if (bond->params.fail_over_mac)
 886                                 bond_do_fail_over_mac(bond, new_active,
 887                                                       old_active);
 888 
 889                         if (netif_running(bond->dev)) {
 890                                 bond->send_peer_notif =
 891                                         bond->params.num_peer_notif *
 892                                         max(1, bond->params.peer_notif_delay);
 893                                 should_notify_peers =
 894                                         bond_should_notify_peers(bond);
 895                         }
 896 
 897                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
 898                         if (should_notify_peers) {
 899                                 bond->send_peer_notif--;
 900                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
 901                                                          bond->dev);
 902                         }
 903                 }
 904         }
 905 
 906         /* resend IGMP joins since active slave has changed or
 907          * all were sent on curr_active_slave.
 908          * resend only if bond is brought up with the affected
 909          * bonding modes and the retransmission is enabled
 910          */
 911         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
 912             ((bond_uses_primary(bond) && new_active) ||
 913              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
 914                 bond->igmp_retrans = bond->params.resend_igmp;
 915                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 916         }
 917 }
 918 
 919 /**
 920  * bond_select_active_slave - select a new active slave, if needed
 921  * @bond: our bonding struct
 922  *
 923  * This functions should be called when one of the following occurs:
 924  * - The old curr_active_slave has been released or lost its link.
 925  * - The primary_slave has got its link back.
 926  * - A slave has got its link back and there's no old curr_active_slave.
 927  *
 928  * Caller must hold RTNL.
 929  */
 930 void bond_select_active_slave(struct bonding *bond)
 931 {
 932         struct slave *best_slave;
 933         int rv;
 934 
 935         ASSERT_RTNL();
 936 
 937         best_slave = bond_find_best_slave(bond);
 938         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
 939                 bond_change_active_slave(bond, best_slave);
 940                 rv = bond_set_carrier(bond);
 941                 if (!rv)
 942                         return;
 943 
 944                 if (netif_carrier_ok(bond->dev))
 945                         netdev_info(bond->dev, "active interface up!\n");
 946                 else
 947                         netdev_info(bond->dev, "now running without any active interface!\n");
 948         }
 949 }
 950 
 951 #ifdef CONFIG_NET_POLL_CONTROLLER
 952 static inline int slave_enable_netpoll(struct slave *slave)
 953 {
 954         struct netpoll *np;
 955         int err = 0;
 956 
 957         np = kzalloc(sizeof(*np), GFP_KERNEL);
 958         err = -ENOMEM;
 959         if (!np)
 960                 goto out;
 961 
 962         err = __netpoll_setup(np, slave->dev);
 963         if (err) {
 964                 kfree(np);
 965                 goto out;
 966         }
 967         slave->np = np;
 968 out:
 969         return err;
 970 }
 971 static inline void slave_disable_netpoll(struct slave *slave)
 972 {
 973         struct netpoll *np = slave->np;
 974 
 975         if (!np)
 976                 return;
 977 
 978         slave->np = NULL;
 979 
 980         __netpoll_free(np);
 981 }
 982 
 983 static void bond_poll_controller(struct net_device *bond_dev)
 984 {
 985         struct bonding *bond = netdev_priv(bond_dev);
 986         struct slave *slave = NULL;
 987         struct list_head *iter;
 988         struct ad_info ad_info;
 989 
 990         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 991                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
 992                         return;
 993 
 994         bond_for_each_slave_rcu(bond, slave, iter) {
 995                 if (!bond_slave_is_up(slave))
 996                         continue;
 997 
 998                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 999                         struct aggregator *agg =
1000                             SLAVE_AD_INFO(slave)->port.aggregator;
1001 
1002                         if (agg &&
1003                             agg->aggregator_identifier != ad_info.aggregator_id)
1004                                 continue;
1005                 }
1006 
1007                 netpoll_poll_dev(slave->dev);
1008         }
1009 }
1010 
1011 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1012 {
1013         struct bonding *bond = netdev_priv(bond_dev);
1014         struct list_head *iter;
1015         struct slave *slave;
1016 
1017         bond_for_each_slave(bond, slave, iter)
1018                 if (bond_slave_is_up(slave))
1019                         slave_disable_netpoll(slave);
1020 }
1021 
1022 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1023 {
1024         struct bonding *bond = netdev_priv(dev);
1025         struct list_head *iter;
1026         struct slave *slave;
1027         int err = 0;
1028 
1029         bond_for_each_slave(bond, slave, iter) {
1030                 err = slave_enable_netpoll(slave);
1031                 if (err) {
1032                         bond_netpoll_cleanup(dev);
1033                         break;
1034                 }
1035         }
1036         return err;
1037 }
1038 #else
1039 static inline int slave_enable_netpoll(struct slave *slave)
1040 {
1041         return 0;
1042 }
1043 static inline void slave_disable_netpoll(struct slave *slave)
1044 {
1045 }
1046 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1047 {
1048 }
1049 #endif
1050 
1051 /*---------------------------------- IOCTL ----------------------------------*/
1052 
1053 static netdev_features_t bond_fix_features(struct net_device *dev,
1054                                            netdev_features_t features)
1055 {
1056         struct bonding *bond = netdev_priv(dev);
1057         struct list_head *iter;
1058         netdev_features_t mask;
1059         struct slave *slave;
1060 
1061         mask = features;
1062 
1063         features &= ~NETIF_F_ONE_FOR_ALL;
1064         features |= NETIF_F_ALL_FOR_ALL;
1065 
1066         bond_for_each_slave(bond, slave, iter) {
1067                 features = netdev_increment_features(features,
1068                                                      slave->dev->features,
1069                                                      mask);
1070         }
1071         features = netdev_add_tso_features(features, mask);
1072 
1073         return features;
1074 }
1075 
1076 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1077                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1078                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1079 
1080 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1081                                  NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1082 
1083 #define BOND_MPLS_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1084                                  NETIF_F_ALL_TSO)
1085 
1086 static void bond_compute_features(struct bonding *bond)
1087 {
1088         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1089                                         IFF_XMIT_DST_RELEASE_PERM;
1090         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1091         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1092         netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
1093         struct net_device *bond_dev = bond->dev;
1094         struct list_head *iter;
1095         struct slave *slave;
1096         unsigned short max_hard_header_len = ETH_HLEN;
1097         unsigned int gso_max_size = GSO_MAX_SIZE;
1098         u16 gso_max_segs = GSO_MAX_SEGS;
1099 
1100         if (!bond_has_slaves(bond))
1101                 goto done;
1102         vlan_features &= NETIF_F_ALL_FOR_ALL;
1103         mpls_features &= NETIF_F_ALL_FOR_ALL;
1104 
1105         bond_for_each_slave(bond, slave, iter) {
1106                 vlan_features = netdev_increment_features(vlan_features,
1107                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1108 
1109                 enc_features = netdev_increment_features(enc_features,
1110                                                          slave->dev->hw_enc_features,
1111                                                          BOND_ENC_FEATURES);
1112 
1113                 mpls_features = netdev_increment_features(mpls_features,
1114                                                           slave->dev->mpls_features,
1115                                                           BOND_MPLS_FEATURES);
1116 
1117                 dst_release_flag &= slave->dev->priv_flags;
1118                 if (slave->dev->hard_header_len > max_hard_header_len)
1119                         max_hard_header_len = slave->dev->hard_header_len;
1120 
1121                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1122                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1123         }
1124         bond_dev->hard_header_len = max_hard_header_len;
1125 
1126 done:
1127         bond_dev->vlan_features = vlan_features;
1128         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1129                                     NETIF_F_HW_VLAN_CTAG_TX |
1130                                     NETIF_F_HW_VLAN_STAG_TX |
1131                                     NETIF_F_GSO_UDP_L4;
1132         bond_dev->mpls_features = mpls_features;
1133         bond_dev->gso_max_segs = gso_max_segs;
1134         netif_set_gso_max_size(bond_dev, gso_max_size);
1135 
1136         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1137         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1138             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1139                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1140 
1141         netdev_change_features(bond_dev);
1142 }
1143 
1144 static void bond_setup_by_slave(struct net_device *bond_dev,
1145                                 struct net_device *slave_dev)
1146 {
1147         bond_dev->header_ops        = slave_dev->header_ops;
1148 
1149         bond_dev->type              = slave_dev->type;
1150         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1151         bond_dev->addr_len          = slave_dev->addr_len;
1152 
1153         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1154                 slave_dev->addr_len);
1155 }
1156 
1157 /* On bonding slaves other than the currently active slave, suppress
1158  * duplicates except for alb non-mcast/bcast.
1159  */
1160 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1161                                             struct slave *slave,
1162                                             struct bonding *bond)
1163 {
1164         if (bond_is_slave_inactive(slave)) {
1165                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1166                     skb->pkt_type != PACKET_BROADCAST &&
1167                     skb->pkt_type != PACKET_MULTICAST)
1168                         return false;
1169                 return true;
1170         }
1171         return false;
1172 }
1173 
1174 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1175 {
1176         struct sk_buff *skb = *pskb;
1177         struct slave *slave;
1178         struct bonding *bond;
1179         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1180                           struct slave *);
1181         int ret = RX_HANDLER_ANOTHER;
1182 
1183         skb = skb_share_check(skb, GFP_ATOMIC);
1184         if (unlikely(!skb))
1185                 return RX_HANDLER_CONSUMED;
1186 
1187         *pskb = skb;
1188 
1189         slave = bond_slave_get_rcu(skb->dev);
1190         bond = slave->bond;
1191 
1192         recv_probe = READ_ONCE(bond->recv_probe);
1193         if (recv_probe) {
1194                 ret = recv_probe(skb, bond, slave);
1195                 if (ret == RX_HANDLER_CONSUMED) {
1196                         consume_skb(skb);
1197                         return ret;
1198                 }
1199         }
1200 
1201         /*
1202          * For packets determined by bond_should_deliver_exact_match() call to
1203          * be suppressed we want to make an exception for link-local packets.
1204          * This is necessary for e.g. LLDP daemons to be able to monitor
1205          * inactive slave links without being forced to bind to them
1206          * explicitly.
1207          *
1208          * At the same time, packets that are passed to the bonding master
1209          * (including link-local ones) can have their originating interface
1210          * determined via PACKET_ORIGDEV socket option.
1211          */
1212         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1213                 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1214                         return RX_HANDLER_PASS;
1215                 return RX_HANDLER_EXACT;
1216         }
1217 
1218         skb->dev = bond->dev;
1219 
1220         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1221             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1222             skb->pkt_type == PACKET_HOST) {
1223 
1224                 if (unlikely(skb_cow_head(skb,
1225                                           skb->data - skb_mac_header(skb)))) {
1226                         kfree_skb(skb);
1227                         return RX_HANDLER_CONSUMED;
1228                 }
1229                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1230                                   bond->dev->addr_len);
1231         }
1232 
1233         return ret;
1234 }
1235 
1236 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1237 {
1238         switch (BOND_MODE(bond)) {
1239         case BOND_MODE_ROUNDROBIN:
1240                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1241         case BOND_MODE_ACTIVEBACKUP:
1242                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1243         case BOND_MODE_BROADCAST:
1244                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1245         case BOND_MODE_XOR:
1246         case BOND_MODE_8023AD:
1247                 return NETDEV_LAG_TX_TYPE_HASH;
1248         default:
1249                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1250         }
1251 }
1252 
1253 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1254                                                enum netdev_lag_tx_type type)
1255 {
1256         if (type != NETDEV_LAG_TX_TYPE_HASH)
1257                 return NETDEV_LAG_HASH_NONE;
1258 
1259         switch (bond->params.xmit_policy) {
1260         case BOND_XMIT_POLICY_LAYER2:
1261                 return NETDEV_LAG_HASH_L2;
1262         case BOND_XMIT_POLICY_LAYER34:
1263                 return NETDEV_LAG_HASH_L34;
1264         case BOND_XMIT_POLICY_LAYER23:
1265                 return NETDEV_LAG_HASH_L23;
1266         case BOND_XMIT_POLICY_ENCAP23:
1267                 return NETDEV_LAG_HASH_E23;
1268         case BOND_XMIT_POLICY_ENCAP34:
1269                 return NETDEV_LAG_HASH_E34;
1270         default:
1271                 return NETDEV_LAG_HASH_UNKNOWN;
1272         }
1273 }
1274 
1275 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1276                                       struct netlink_ext_ack *extack)
1277 {
1278         struct netdev_lag_upper_info lag_upper_info;
1279         enum netdev_lag_tx_type type;
1280 
1281         type = bond_lag_tx_type(bond);
1282         lag_upper_info.tx_type = type;
1283         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1284 
1285         return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1286                                             &lag_upper_info, extack);
1287 }
1288 
1289 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1290 {
1291         netdev_upper_dev_unlink(slave->dev, bond->dev);
1292         slave->dev->flags &= ~IFF_SLAVE;
1293 }
1294 
1295 static struct slave *bond_alloc_slave(struct bonding *bond)
1296 {
1297         struct slave *slave = NULL;
1298 
1299         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1300         if (!slave)
1301                 return NULL;
1302 
1303         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1304                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1305                                                GFP_KERNEL);
1306                 if (!SLAVE_AD_INFO(slave)) {
1307                         kfree(slave);
1308                         return NULL;
1309                 }
1310         }
1311         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1312 
1313         return slave;
1314 }
1315 
1316 static void bond_free_slave(struct slave *slave)
1317 {
1318         struct bonding *bond = bond_get_bond_by_slave(slave);
1319 
1320         cancel_delayed_work_sync(&slave->notify_work);
1321         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1322                 kfree(SLAVE_AD_INFO(slave));
1323 
1324         kfree(slave);
1325 }
1326 
1327 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1328 {
1329         info->bond_mode = BOND_MODE(bond);
1330         info->miimon = bond->params.miimon;
1331         info->num_slaves = bond->slave_cnt;
1332 }
1333 
1334 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1335 {
1336         strcpy(info->slave_name, slave->dev->name);
1337         info->link = slave->link;
1338         info->state = bond_slave_state(slave);
1339         info->link_failure_count = slave->link_failure_count;
1340 }
1341 
1342 static void bond_netdev_notify_work(struct work_struct *_work)
1343 {
1344         struct slave *slave = container_of(_work, struct slave,
1345                                            notify_work.work);
1346 
1347         if (rtnl_trylock()) {
1348                 struct netdev_bonding_info binfo;
1349 
1350                 bond_fill_ifslave(slave, &binfo.slave);
1351                 bond_fill_ifbond(slave->bond, &binfo.master);
1352                 netdev_bonding_info_change(slave->dev, &binfo);
1353                 rtnl_unlock();
1354         } else {
1355                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1356         }
1357 }
1358 
1359 void bond_queue_slave_event(struct slave *slave)
1360 {
1361         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1362 }
1363 
1364 void bond_lower_state_changed(struct slave *slave)
1365 {
1366         struct netdev_lag_lower_state_info info;
1367 
1368         info.link_up = slave->link == BOND_LINK_UP ||
1369                        slave->link == BOND_LINK_FAIL;
1370         info.tx_enabled = bond_is_active_slave(slave);
1371         netdev_lower_state_changed(slave->dev, &info);
1372 }
1373 
1374 /* enslave device <slave> to bond device <master> */
1375 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1376                  struct netlink_ext_ack *extack)
1377 {
1378         struct bonding *bond = netdev_priv(bond_dev);
1379         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1380         struct slave *new_slave = NULL, *prev_slave;
1381         struct sockaddr_storage ss;
1382         int link_reporting;
1383         int res = 0, i;
1384 
1385         if (!bond->params.use_carrier &&
1386             slave_dev->ethtool_ops->get_link == NULL &&
1387             slave_ops->ndo_do_ioctl == NULL) {
1388                 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1389         }
1390 
1391         /* already in-use? */
1392         if (netdev_is_rx_handler_busy(slave_dev)) {
1393                 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1394                 slave_err(bond_dev, slave_dev,
1395                           "Error: Device is in use and cannot be enslaved\n");
1396                 return -EBUSY;
1397         }
1398 
1399         if (bond_dev == slave_dev) {
1400                 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1401                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1402                 return -EPERM;
1403         }
1404 
1405         /* vlan challenged mutual exclusion */
1406         /* no need to lock since we're protected by rtnl_lock */
1407         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1408                 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1409                 if (vlan_uses_dev(bond_dev)) {
1410                         NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1411                         slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1412                         return -EPERM;
1413                 } else {
1414                         slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1415                 }
1416         } else {
1417                 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1418         }
1419 
1420         /* Old ifenslave binaries are no longer supported.  These can
1421          * be identified with moderate accuracy by the state of the slave:
1422          * the current ifenslave will set the interface down prior to
1423          * enslaving it; the old ifenslave will not.
1424          */
1425         if (slave_dev->flags & IFF_UP) {
1426                 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1427                 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1428                 return -EPERM;
1429         }
1430 
1431         /* set bonding device ether type by slave - bonding netdevices are
1432          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1433          * there is a need to override some of the type dependent attribs/funcs.
1434          *
1435          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1436          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1437          */
1438         if (!bond_has_slaves(bond)) {
1439                 if (bond_dev->type != slave_dev->type) {
1440                         slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1441                                   bond_dev->type, slave_dev->type);
1442 
1443                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1444                                                        bond_dev);
1445                         res = notifier_to_errno(res);
1446                         if (res) {
1447                                 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1448                                 return -EBUSY;
1449                         }
1450 
1451                         /* Flush unicast and multicast addresses */
1452                         dev_uc_flush(bond_dev);
1453                         dev_mc_flush(bond_dev);
1454 
1455                         if (slave_dev->type != ARPHRD_ETHER)
1456                                 bond_setup_by_slave(bond_dev, slave_dev);
1457                         else {
1458                                 ether_setup(bond_dev);
1459                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1460                         }
1461 
1462                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1463                                                  bond_dev);
1464                 }
1465         } else if (bond_dev->type != slave_dev->type) {
1466                 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1467                 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1468                           slave_dev->type, bond_dev->type);
1469                 return -EINVAL;
1470         }
1471 
1472         if (slave_dev->type == ARPHRD_INFINIBAND &&
1473             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1474                 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1475                 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1476                            slave_dev->type);
1477                 res = -EOPNOTSUPP;
1478                 goto err_undo_flags;
1479         }
1480 
1481         if (!slave_ops->ndo_set_mac_address ||
1482             slave_dev->type == ARPHRD_INFINIBAND) {
1483                 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1484                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1485                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1486                         if (!bond_has_slaves(bond)) {
1487                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1488                                 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1489                         } else {
1490                                 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1491                                 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");
1492                                 res = -EOPNOTSUPP;
1493                                 goto err_undo_flags;
1494                         }
1495                 }
1496         }
1497 
1498         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1499 
1500         /* If this is the first slave, then we need to set the master's hardware
1501          * address to be the same as the slave's.
1502          */
1503         if (!bond_has_slaves(bond) &&
1504             bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1505                 res = bond_set_dev_addr(bond->dev, slave_dev);
1506                 if (res)
1507                         goto err_undo_flags;
1508         }
1509 
1510         new_slave = bond_alloc_slave(bond);
1511         if (!new_slave) {
1512                 res = -ENOMEM;
1513                 goto err_undo_flags;
1514         }
1515 
1516         new_slave->bond = bond;
1517         new_slave->dev = slave_dev;
1518         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1519          * is set via sysfs or module option if desired.
1520          */
1521         new_slave->queue_id = 0;
1522 
1523         /* Save slave's original mtu and then set it to match the bond */
1524         new_slave->original_mtu = slave_dev->mtu;
1525         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1526         if (res) {
1527                 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1528                 goto err_free;
1529         }
1530 
1531         /* Save slave's original ("permanent") mac address for modes
1532          * that need it, and for restoring it upon release, and then
1533          * set it to the master's address
1534          */
1535         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1536                           slave_dev->addr_len);
1537 
1538         if (!bond->params.fail_over_mac ||
1539             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1540                 /* Set slave to master's mac address.  The application already
1541                  * set the master's mac address to that of the first slave
1542                  */
1543                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1544                 ss.ss_family = slave_dev->type;
1545                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1546                                           extack);
1547                 if (res) {
1548                         slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1549                         goto err_restore_mtu;
1550                 }
1551         }
1552 
1553         /* set slave flag before open to prevent IPv6 addrconf */
1554         slave_dev->flags |= IFF_SLAVE;
1555 
1556         /* open the slave since the application closed it */
1557         res = dev_open(slave_dev, extack);
1558         if (res) {
1559                 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1560                 goto err_restore_mac;
1561         }
1562 
1563         slave_dev->priv_flags |= IFF_BONDING;
1564         /* initialize slave stats */
1565         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1566 
1567         if (bond_is_lb(bond)) {
1568                 /* bond_alb_init_slave() must be called before all other stages since
1569                  * it might fail and we do not want to have to undo everything
1570                  */
1571                 res = bond_alb_init_slave(bond, new_slave);
1572                 if (res)
1573                         goto err_close;
1574         }
1575 
1576         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1577         if (res) {
1578                 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1579                 goto err_close;
1580         }
1581 
1582         prev_slave = bond_last_slave(bond);
1583 
1584         new_slave->delay = 0;
1585         new_slave->link_failure_count = 0;
1586 
1587         if (bond_update_speed_duplex(new_slave) &&
1588             bond_needs_speed_duplex(bond))
1589                 new_slave->link = BOND_LINK_DOWN;
1590 
1591         new_slave->last_rx = jiffies -
1592                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1593         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1594                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1595 
1596         if (bond->params.miimon && !bond->params.use_carrier) {
1597                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1598 
1599                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1600                         /* miimon is set but a bonded network driver
1601                          * does not support ETHTOOL/MII and
1602                          * arp_interval is not set.  Note: if
1603                          * use_carrier is enabled, we will never go
1604                          * here (because netif_carrier is always
1605                          * supported); thus, we don't need to change
1606                          * the messages for netif_carrier.
1607                          */
1608                         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");
1609                 } else if (link_reporting == -1) {
1610                         /* unable get link status using mii/ethtool */
1611                         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");
1612                 }
1613         }
1614 
1615         /* check for initial state */
1616         new_slave->link = BOND_LINK_NOCHANGE;
1617         if (bond->params.miimon) {
1618                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1619                         if (bond->params.updelay) {
1620                                 bond_set_slave_link_state(new_slave,
1621                                                           BOND_LINK_BACK,
1622                                                           BOND_SLAVE_NOTIFY_NOW);
1623                                 new_slave->delay = bond->params.updelay;
1624                         } else {
1625                                 bond_set_slave_link_state(new_slave,
1626                                                           BOND_LINK_UP,
1627                                                           BOND_SLAVE_NOTIFY_NOW);
1628                         }
1629                 } else {
1630                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1631                                                   BOND_SLAVE_NOTIFY_NOW);
1632                 }
1633         } else if (bond->params.arp_interval) {
1634                 bond_set_slave_link_state(new_slave,
1635                                           (netif_carrier_ok(slave_dev) ?
1636                                           BOND_LINK_UP : BOND_LINK_DOWN),
1637                                           BOND_SLAVE_NOTIFY_NOW);
1638         } else {
1639                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1640                                           BOND_SLAVE_NOTIFY_NOW);
1641         }
1642 
1643         if (new_slave->link != BOND_LINK_DOWN)
1644                 new_slave->last_link_up = jiffies;
1645         slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1646                   new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1647                   (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1648 
1649         if (bond_uses_primary(bond) && bond->params.primary[0]) {
1650                 /* if there is a primary slave, remember it */
1651                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1652                         rcu_assign_pointer(bond->primary_slave, new_slave);
1653                         bond->force_primary = true;
1654                 }
1655         }
1656 
1657         switch (BOND_MODE(bond)) {
1658         case BOND_MODE_ACTIVEBACKUP:
1659                 bond_set_slave_inactive_flags(new_slave,
1660                                               BOND_SLAVE_NOTIFY_NOW);
1661                 break;
1662         case BOND_MODE_8023AD:
1663                 /* in 802.3ad mode, the internal mechanism
1664                  * will activate the slaves in the selected
1665                  * aggregator
1666                  */
1667                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1668                 /* if this is the first slave */
1669                 if (!prev_slave) {
1670                         SLAVE_AD_INFO(new_slave)->id = 1;
1671                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1672                          * can be called only after the mac address of the bond is set
1673                          */
1674                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1675                 } else {
1676                         SLAVE_AD_INFO(new_slave)->id =
1677                                 SLAVE_AD_INFO(prev_slave)->id + 1;
1678                 }
1679 
1680                 bond_3ad_bind_slave(new_slave);
1681                 break;
1682         case BOND_MODE_TLB:
1683         case BOND_MODE_ALB:
1684                 bond_set_active_slave(new_slave);
1685                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1686                 break;
1687         default:
1688                 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
1689 
1690                 /* always active in trunk mode */
1691                 bond_set_active_slave(new_slave);
1692 
1693                 /* In trunking mode there is little meaning to curr_active_slave
1694                  * anyway (it holds no special properties of the bond device),
1695                  * so we can change it without calling change_active_interface()
1696                  */
1697                 if (!rcu_access_pointer(bond->curr_active_slave) &&
1698                     new_slave->link == BOND_LINK_UP)
1699                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
1700 
1701                 break;
1702         } /* switch(bond_mode) */
1703 
1704 #ifdef CONFIG_NET_POLL_CONTROLLER
1705         if (bond->dev->npinfo) {
1706                 if (slave_enable_netpoll(new_slave)) {
1707                         slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1708                         res = -EBUSY;
1709                         goto err_detach;
1710                 }
1711         }
1712 #endif
1713 
1714         if (!(bond_dev->features & NETIF_F_LRO))
1715                 dev_disable_lro(slave_dev);
1716 
1717         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1718                                          new_slave);
1719         if (res) {
1720                 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
1721                 goto err_detach;
1722         }
1723 
1724         res = bond_master_upper_dev_link(bond, new_slave, extack);
1725         if (res) {
1726                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1727                 goto err_unregister;
1728         }
1729 
1730         res = bond_sysfs_slave_add(new_slave);
1731         if (res) {
1732                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1733                 goto err_upper_unlink;
1734         }
1735 
1736         /* If the mode uses primary, then the following is handled by
1737          * bond_change_active_slave().
1738          */
1739         if (!bond_uses_primary(bond)) {
1740                 /* set promiscuity level to new slave */
1741                 if (bond_dev->flags & IFF_PROMISC) {
1742                         res = dev_set_promiscuity(slave_dev, 1);
1743                         if (res)
1744                                 goto err_sysfs_del;
1745                 }
1746 
1747                 /* set allmulti level to new slave */
1748                 if (bond_dev->flags & IFF_ALLMULTI) {
1749                         res = dev_set_allmulti(slave_dev, 1);
1750                         if (res) {
1751                                 if (bond_dev->flags & IFF_PROMISC)
1752                                         dev_set_promiscuity(slave_dev, -1);
1753                                 goto err_sysfs_del;
1754                         }
1755                 }
1756 
1757                 netif_addr_lock_bh(bond_dev);
1758                 dev_mc_sync_multiple(slave_dev, bond_dev);
1759                 dev_uc_sync_multiple(slave_dev, bond_dev);
1760                 netif_addr_unlock_bh(bond_dev);
1761 
1762                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1763                         /* add lacpdu mc addr to mc list */
1764                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1765 
1766                         dev_mc_add(slave_dev, lacpdu_multicast);
1767                 }
1768         }
1769 
1770         bond->slave_cnt++;
1771         bond_compute_features(bond);
1772         bond_set_carrier(bond);
1773 
1774         if (bond_uses_primary(bond)) {
1775                 block_netpoll_tx();
1776                 bond_select_active_slave(bond);
1777                 unblock_netpoll_tx();
1778         }
1779 
1780         if (bond_mode_can_use_xmit_hash(bond))
1781                 bond_update_slave_arr(bond, NULL);
1782 
1783 
1784         slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
1785                    bond_is_active_slave(new_slave) ? "an active" : "a backup",
1786                    new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1787 
1788         /* enslave is successful */
1789         bond_queue_slave_event(new_slave);
1790         return 0;
1791 
1792 /* Undo stages on error */
1793 err_sysfs_del:
1794         bond_sysfs_slave_del(new_slave);
1795 
1796 err_upper_unlink:
1797         bond_upper_dev_unlink(bond, new_slave);
1798 
1799 err_unregister:
1800         netdev_rx_handler_unregister(slave_dev);
1801 
1802 err_detach:
1803         vlan_vids_del_by_dev(slave_dev, bond_dev);
1804         if (rcu_access_pointer(bond->primary_slave) == new_slave)
1805                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1806         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1807                 block_netpoll_tx();
1808                 bond_change_active_slave(bond, NULL);
1809                 bond_select_active_slave(bond);
1810                 unblock_netpoll_tx();
1811         }
1812         /* either primary_slave or curr_active_slave might've changed */
1813         synchronize_rcu();
1814         slave_disable_netpoll(new_slave);
1815 
1816 err_close:
1817         if (!netif_is_bond_master(slave_dev))
1818                 slave_dev->priv_flags &= ~IFF_BONDING;
1819         dev_close(slave_dev);
1820 
1821 err_restore_mac:
1822         slave_dev->flags &= ~IFF_SLAVE;
1823         if (!bond->params.fail_over_mac ||
1824             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1825                 /* XXX TODO - fom follow mode needs to change master's
1826                  * MAC if this slave's MAC is in use by the bond, or at
1827                  * least print a warning.
1828                  */
1829                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1830                                   new_slave->dev->addr_len);
1831                 ss.ss_family = slave_dev->type;
1832                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
1833         }
1834 
1835 err_restore_mtu:
1836         dev_set_mtu(slave_dev, new_slave->original_mtu);
1837 
1838 err_free:
1839         bond_free_slave(new_slave);
1840 
1841 err_undo_flags:
1842         /* Enslave of first slave has failed and we need to fix master's mac */
1843         if (!bond_has_slaves(bond)) {
1844                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1845                                             slave_dev->dev_addr))
1846                         eth_hw_addr_random(bond_dev);
1847                 if (bond_dev->type != ARPHRD_ETHER) {
1848                         dev_close(bond_dev);
1849                         ether_setup(bond_dev);
1850                         bond_dev->flags |= IFF_MASTER;
1851                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1852                 }
1853         }
1854 
1855         return res;
1856 }
1857 
1858 /* Try to release the slave device <slave> from the bond device <master>
1859  * It is legal to access curr_active_slave without a lock because all the function
1860  * is RTNL-locked. If "all" is true it means that the function is being called
1861  * while destroying a bond interface and all slaves are being released.
1862  *
1863  * The rules for slave state should be:
1864  *   for Active/Backup:
1865  *     Active stays on all backups go down
1866  *   for Bonded connections:
1867  *     The first up interface should be left on and all others downed.
1868  */
1869 static int __bond_release_one(struct net_device *bond_dev,
1870                               struct net_device *slave_dev,
1871                               bool all, bool unregister)
1872 {
1873         struct bonding *bond = netdev_priv(bond_dev);
1874         struct slave *slave, *oldcurrent;
1875         struct sockaddr_storage ss;
1876         int old_flags = bond_dev->flags;
1877         netdev_features_t old_features = bond_dev->features;
1878 
1879         /* slave is not a slave or master is not master of this slave */
1880         if (!(slave_dev->flags & IFF_SLAVE) ||
1881             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1882                 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
1883                 return -EINVAL;
1884         }
1885 
1886         block_netpoll_tx();
1887 
1888         slave = bond_get_slave_by_dev(bond, slave_dev);
1889         if (!slave) {
1890                 /* not a slave of this bond */
1891                 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
1892                 unblock_netpoll_tx();
1893                 return -EINVAL;
1894         }
1895 
1896         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1897 
1898         bond_sysfs_slave_del(slave);
1899 
1900         /* recompute stats just before removing the slave */
1901         bond_get_stats(bond->dev, &bond->bond_stats);
1902 
1903         bond_upper_dev_unlink(bond, slave);
1904         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1905          * for this slave anymore.
1906          */
1907         netdev_rx_handler_unregister(slave_dev);
1908 
1909         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1910                 bond_3ad_unbind_slave(slave);
1911 
1912         if (bond_mode_can_use_xmit_hash(bond))
1913                 bond_update_slave_arr(bond, slave);
1914 
1915         slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
1916                     bond_is_active_slave(slave) ? "active" : "backup");
1917 
1918         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1919 
1920         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1921 
1922         if (!all && (!bond->params.fail_over_mac ||
1923                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1924                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1925                     bond_has_slaves(bond))
1926                         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",
1927                                    slave->perm_hwaddr);
1928         }
1929 
1930         if (rtnl_dereference(bond->primary_slave) == slave)
1931                 RCU_INIT_POINTER(bond->primary_slave, NULL);
1932 
1933         if (oldcurrent == slave)
1934                 bond_change_active_slave(bond, NULL);
1935 
1936         if (bond_is_lb(bond)) {
1937                 /* Must be called only after the slave has been
1938                  * detached from the list and the curr_active_slave
1939                  * has been cleared (if our_slave == old_current),
1940                  * but before a new active slave is selected.
1941                  */
1942                 bond_alb_deinit_slave(bond, slave);
1943         }
1944 
1945         if (all) {
1946                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1947         } else if (oldcurrent == slave) {
1948                 /* Note that we hold RTNL over this sequence, so there
1949                  * is no concern that another slave add/remove event
1950                  * will interfere.
1951                  */
1952                 bond_select_active_slave(bond);
1953         }
1954 
1955         if (!bond_has_slaves(bond)) {
1956                 bond_set_carrier(bond);
1957                 eth_hw_addr_random(bond_dev);
1958         }
1959 
1960         unblock_netpoll_tx();
1961         synchronize_rcu();
1962         bond->slave_cnt--;
1963 
1964         if (!bond_has_slaves(bond)) {
1965                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1966                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1967         }
1968 
1969         bond_compute_features(bond);
1970         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1971             (old_features & NETIF_F_VLAN_CHALLENGED))
1972                 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
1973 
1974         vlan_vids_del_by_dev(slave_dev, bond_dev);
1975 
1976         /* If the mode uses primary, then this case was handled above by
1977          * bond_change_active_slave(..., NULL)
1978          */
1979         if (!bond_uses_primary(bond)) {
1980                 /* unset promiscuity level from slave
1981                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
1982                  * of the IFF_PROMISC flag in the bond_dev, but we need the
1983                  * value of that flag before that change, as that was the value
1984                  * when this slave was attached, so we cache at the start of the
1985                  * function and use it here. Same goes for ALLMULTI below
1986                  */
1987                 if (old_flags & IFF_PROMISC)
1988                         dev_set_promiscuity(slave_dev, -1);
1989 
1990                 /* unset allmulti level from slave */
1991                 if (old_flags & IFF_ALLMULTI)
1992                         dev_set_allmulti(slave_dev, -1);
1993 
1994                 bond_hw_addr_flush(bond_dev, slave_dev);
1995         }
1996 
1997         slave_disable_netpoll(slave);
1998 
1999         /* close slave before restoring its mac address */
2000         dev_close(slave_dev);
2001 
2002         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2003             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2004                 /* restore original ("permanent") mac address */
2005                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2006                                   slave->dev->addr_len);
2007                 ss.ss_family = slave_dev->type;
2008                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2009         }
2010 
2011         if (unregister)
2012                 __dev_set_mtu(slave_dev, slave->original_mtu);
2013         else
2014                 dev_set_mtu(slave_dev, slave->original_mtu);
2015 
2016         if (!netif_is_bond_master(slave_dev))
2017                 slave_dev->priv_flags &= ~IFF_BONDING;
2018 
2019         bond_free_slave(slave);
2020 
2021         return 0;
2022 }
2023 
2024 /* A wrapper used because of ndo_del_link */
2025 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2026 {
2027         return __bond_release_one(bond_dev, slave_dev, false, false);
2028 }
2029 
2030 /* First release a slave and then destroy the bond if no more slaves are left.
2031  * Must be under rtnl_lock when this function is called.
2032  */
2033 static int bond_release_and_destroy(struct net_device *bond_dev,
2034                                     struct net_device *slave_dev)
2035 {
2036         struct bonding *bond = netdev_priv(bond_dev);
2037         int ret;
2038 
2039         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2040         if (ret == 0 && !bond_has_slaves(bond)) {
2041                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2042                 netdev_info(bond_dev, "Destroying bond\n");
2043                 bond_remove_proc_entry(bond);
2044                 unregister_netdevice(bond_dev);
2045         }
2046         return ret;
2047 }
2048 
2049 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2050 {
2051         struct bonding *bond = netdev_priv(bond_dev);
2052         bond_fill_ifbond(bond, info);
2053 }
2054 
2055 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2056 {
2057         struct bonding *bond = netdev_priv(bond_dev);
2058         struct list_head *iter;
2059         int i = 0, res = -ENODEV;
2060         struct slave *slave;
2061 
2062         bond_for_each_slave(bond, slave, iter) {
2063                 if (i++ == (int)info->slave_id) {
2064                         res = 0;
2065                         bond_fill_ifslave(slave, info);
2066                         break;
2067                 }
2068         }
2069 
2070         return res;
2071 }
2072 
2073 /*-------------------------------- Monitoring -------------------------------*/
2074 
2075 /* called with rcu_read_lock() */
2076 static int bond_miimon_inspect(struct bonding *bond)
2077 {
2078         int link_state, commit = 0;
2079         struct list_head *iter;
2080         struct slave *slave;
2081         bool ignore_updelay;
2082 
2083         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2084 
2085         bond_for_each_slave_rcu(bond, slave, iter) {
2086                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2087 
2088                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2089 
2090                 switch (slave->link) {
2091                 case BOND_LINK_UP:
2092                         if (link_state)
2093                                 continue;
2094 
2095                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2096                         commit++;
2097                         slave->delay = bond->params.downdelay;
2098                         if (slave->delay) {
2099                                 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2100                                            (BOND_MODE(bond) ==
2101                                             BOND_MODE_ACTIVEBACKUP) ?
2102                                             (bond_is_active_slave(slave) ?
2103                                              "active " : "backup ") : "",
2104                                            bond->params.downdelay * bond->params.miimon);
2105                         }
2106                         /*FALLTHRU*/
2107                 case BOND_LINK_FAIL:
2108                         if (link_state) {
2109                                 /* recovered before downdelay expired */
2110                                 bond_propose_link_state(slave, BOND_LINK_UP);
2111                                 slave->last_link_up = jiffies;
2112                                 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2113                                            (bond->params.downdelay - slave->delay) *
2114                                            bond->params.miimon);
2115                                 commit++;
2116                                 continue;
2117                         }
2118 
2119                         if (slave->delay <= 0) {
2120                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2121                                 commit++;
2122                                 continue;
2123                         }
2124 
2125                         slave->delay--;
2126                         break;
2127 
2128                 case BOND_LINK_DOWN:
2129                         if (!link_state)
2130                                 continue;
2131 
2132                         bond_propose_link_state(slave, BOND_LINK_BACK);
2133                         commit++;
2134                         slave->delay = bond->params.updelay;
2135 
2136                         if (slave->delay) {
2137                                 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2138                                            ignore_updelay ? 0 :
2139                                            bond->params.updelay *
2140                                            bond->params.miimon);
2141                         }
2142                         /*FALLTHRU*/
2143                 case BOND_LINK_BACK:
2144                         if (!link_state) {
2145                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2146                                 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2147                                            (bond->params.updelay - slave->delay) *
2148                                            bond->params.miimon);
2149                                 commit++;
2150                                 continue;
2151                         }
2152 
2153                         if (ignore_updelay)
2154                                 slave->delay = 0;
2155 
2156                         if (slave->delay <= 0) {
2157                                 bond_propose_link_state(slave, BOND_LINK_UP);
2158                                 commit++;
2159                                 ignore_updelay = false;
2160                                 continue;
2161                         }
2162 
2163                         slave->delay--;
2164                         break;
2165                 }
2166         }
2167 
2168         return commit;
2169 }
2170 
2171 static void bond_miimon_link_change(struct bonding *bond,
2172                                     struct slave *slave,
2173                                     char link)
2174 {
2175         switch (BOND_MODE(bond)) {
2176         case BOND_MODE_8023AD:
2177                 bond_3ad_handle_link_change(slave, link);
2178                 break;
2179         case BOND_MODE_TLB:
2180         case BOND_MODE_ALB:
2181                 bond_alb_handle_link_change(bond, slave, link);
2182                 break;
2183         case BOND_MODE_XOR:
2184                 bond_update_slave_arr(bond, NULL);
2185                 break;
2186         }
2187 }
2188 
2189 static void bond_miimon_commit(struct bonding *bond)
2190 {
2191         struct list_head *iter;
2192         struct slave *slave, *primary;
2193 
2194         bond_for_each_slave(bond, slave, iter) {
2195                 switch (slave->link_new_state) {
2196                 case BOND_LINK_NOCHANGE:
2197                         /* For 802.3ad mode, check current slave speed and
2198                          * duplex again in case its port was disabled after
2199                          * invalid speed/duplex reporting but recovered before
2200                          * link monitoring could make a decision on the actual
2201                          * link status
2202                          */
2203                         if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2204                             slave->link == BOND_LINK_UP)
2205                                 bond_3ad_adapter_speed_duplex_changed(slave);
2206                         continue;
2207 
2208                 case BOND_LINK_UP:
2209                         if (bond_update_speed_duplex(slave) &&
2210                             bond_needs_speed_duplex(bond)) {
2211                                 slave->link = BOND_LINK_DOWN;
2212                                 if (net_ratelimit())
2213                                         slave_warn(bond->dev, slave->dev,
2214                                                    "failed to get link speed/duplex\n");
2215                                 continue;
2216                         }
2217                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2218                                                   BOND_SLAVE_NOTIFY_NOW);
2219                         slave->last_link_up = jiffies;
2220 
2221                         primary = rtnl_dereference(bond->primary_slave);
2222                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2223                                 /* prevent it from being the active one */
2224                                 bond_set_backup_slave(slave);
2225                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2226                                 /* make it immediately active */
2227                                 bond_set_active_slave(slave);
2228                         }
2229 
2230                         slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2231                                    slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2232                                    slave->duplex ? "full" : "half");
2233 
2234                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2235 
2236                         if (!bond->curr_active_slave || slave == primary)
2237                                 goto do_failover;
2238 
2239                         continue;
2240 
2241                 case BOND_LINK_DOWN:
2242                         if (slave->link_failure_count < UINT_MAX)
2243                                 slave->link_failure_count++;
2244 
2245                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2246                                                   BOND_SLAVE_NOTIFY_NOW);
2247 
2248                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2249                             BOND_MODE(bond) == BOND_MODE_8023AD)
2250                                 bond_set_slave_inactive_flags(slave,
2251                                                               BOND_SLAVE_NOTIFY_NOW);
2252 
2253                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2254 
2255                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2256 
2257                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2258                                 goto do_failover;
2259 
2260                         continue;
2261 
2262                 default:
2263                         slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2264                                   slave->link_new_state);
2265                         bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2266 
2267                         continue;
2268                 }
2269 
2270 do_failover:
2271                 block_netpoll_tx();
2272                 bond_select_active_slave(bond);
2273                 unblock_netpoll_tx();
2274         }
2275 
2276         bond_set_carrier(bond);
2277 }
2278 
2279 /* bond_mii_monitor
2280  *
2281  * Really a wrapper that splits the mii monitor into two phases: an
2282  * inspection, then (if inspection indicates something needs to be done)
2283  * an acquisition of appropriate locks followed by a commit phase to
2284  * implement whatever link state changes are indicated.
2285  */
2286 static void bond_mii_monitor(struct work_struct *work)
2287 {
2288         struct bonding *bond = container_of(work, struct bonding,
2289                                             mii_work.work);
2290         bool should_notify_peers = false;
2291         bool commit;
2292         unsigned long delay;
2293         struct slave *slave;
2294         struct list_head *iter;
2295 
2296         delay = msecs_to_jiffies(bond->params.miimon);
2297 
2298         if (!bond_has_slaves(bond))
2299                 goto re_arm;
2300 
2301         rcu_read_lock();
2302         should_notify_peers = bond_should_notify_peers(bond);
2303         commit = !!bond_miimon_inspect(bond);
2304         if (bond->send_peer_notif) {
2305                 rcu_read_unlock();
2306                 if (rtnl_trylock()) {
2307                         bond->send_peer_notif--;
2308                         rtnl_unlock();
2309                 }
2310         } else {
2311                 rcu_read_unlock();
2312         }
2313 
2314         if (commit) {
2315                 /* Race avoidance with bond_close cancel of workqueue */
2316                 if (!rtnl_trylock()) {
2317                         delay = 1;
2318                         should_notify_peers = false;
2319                         goto re_arm;
2320                 }
2321 
2322                 bond_for_each_slave(bond, slave, iter) {
2323                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2324                 }
2325                 bond_miimon_commit(bond);
2326 
2327                 rtnl_unlock();  /* might sleep, hold no other locks */
2328         }
2329 
2330 re_arm:
2331         if (bond->params.miimon)
2332                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2333 
2334         if (should_notify_peers) {
2335                 if (!rtnl_trylock())
2336                         return;
2337                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2338                 rtnl_unlock();
2339         }
2340 }
2341 
2342 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2343 {
2344         __be32 ip = *((__be32 *)data);
2345 
2346         return ip == bond_confirm_addr(upper, 0, ip);
2347 }
2348 
2349 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2350 {
2351         bool ret = false;
2352 
2353         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2354                 return true;
2355 
2356         rcu_read_lock();
2357         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2358                 ret = true;
2359         rcu_read_unlock();
2360 
2361         return ret;
2362 }
2363 
2364 /* We go to the (large) trouble of VLAN tagging ARP frames because
2365  * switches in VLAN mode (especially if ports are configured as
2366  * "native" to a VLAN) might not pass non-tagged frames.
2367  */
2368 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2369                           __be32 src_ip, struct bond_vlan_tag *tags)
2370 {
2371         struct sk_buff *skb;
2372         struct bond_vlan_tag *outer_tag = tags;
2373         struct net_device *slave_dev = slave->dev;
2374         struct net_device *bond_dev = slave->bond->dev;
2375 
2376         slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2377                   arp_op, &dest_ip, &src_ip);
2378 
2379         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2380                          NULL, slave_dev->dev_addr, NULL);
2381 
2382         if (!skb) {
2383                 net_err_ratelimited("ARP packet allocation failed\n");
2384                 return;
2385         }
2386 
2387         if (!tags || tags->vlan_proto == VLAN_N_VID)
2388                 goto xmit;
2389 
2390         tags++;
2391 
2392         /* Go through all the tags backwards and add them to the packet */
2393         while (tags->vlan_proto != VLAN_N_VID) {
2394                 if (!tags->vlan_id) {
2395                         tags++;
2396                         continue;
2397                 }
2398 
2399                 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2400                           ntohs(outer_tag->vlan_proto), tags->vlan_id);
2401                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2402                                                 tags->vlan_id);
2403                 if (!skb) {
2404                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2405                         return;
2406                 }
2407 
2408                 tags++;
2409         }
2410         /* Set the outer tag */
2411         if (outer_tag->vlan_id) {
2412                 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2413                           ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2414                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2415                                        outer_tag->vlan_id);
2416         }
2417 
2418 xmit:
2419         arp_xmit(skb);
2420 }
2421 
2422 /* Validate the device path between the @start_dev and the @end_dev.
2423  * The path is valid if the @end_dev is reachable through device
2424  * stacking.
2425  * When the path is validated, collect any vlan information in the
2426  * path.
2427  */
2428 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2429                                               struct net_device *end_dev,
2430                                               int level)
2431 {
2432         struct bond_vlan_tag *tags;
2433         struct net_device *upper;
2434         struct list_head  *iter;
2435 
2436         if (start_dev == end_dev) {
2437                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2438                 if (!tags)
2439                         return ERR_PTR(-ENOMEM);
2440                 tags[level].vlan_proto = VLAN_N_VID;
2441                 return tags;
2442         }
2443 
2444         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2445                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2446                 if (IS_ERR_OR_NULL(tags)) {
2447                         if (IS_ERR(tags))
2448                                 return tags;
2449                         continue;
2450                 }
2451                 if (is_vlan_dev(upper)) {
2452                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2453                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2454                 }
2455 
2456                 return tags;
2457         }
2458 
2459         return NULL;
2460 }
2461 
2462 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2463 {
2464         struct rtable *rt;
2465         struct bond_vlan_tag *tags;
2466         __be32 *targets = bond->params.arp_targets, addr;
2467         int i;
2468 
2469         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2470                 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2471                           __func__, &targets[i]);
2472                 tags = NULL;
2473 
2474                 /* Find out through which dev should the packet go */
2475                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2476                                      RTO_ONLINK, 0);
2477                 if (IS_ERR(rt)) {
2478                         /* there's no route to target - try to send arp
2479                          * probe to generate any traffic (arp_validate=0)
2480                          */
2481                         if (bond->params.arp_validate)
2482                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2483                                                      bond->dev->name,
2484                                                      &targets[i]);
2485                         bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2486                                       0, tags);
2487                         continue;
2488                 }
2489 
2490                 /* bond device itself */
2491                 if (rt->dst.dev == bond->dev)
2492                         goto found;
2493 
2494                 rcu_read_lock();
2495                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2496                 rcu_read_unlock();
2497 
2498                 if (!IS_ERR_OR_NULL(tags))
2499                         goto found;
2500 
2501                 /* Not our device - skip */
2502                 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2503                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2504 
2505                 ip_rt_put(rt);
2506                 continue;
2507 
2508 found:
2509                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2510                 ip_rt_put(rt);
2511                 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2512                 kfree(tags);
2513         }
2514 }
2515 
2516 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2517 {
2518         int i;
2519 
2520         if (!sip || !bond_has_this_ip(bond, tip)) {
2521                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2522                            __func__, &sip, &tip);
2523                 return;
2524         }
2525 
2526         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2527         if (i == -1) {
2528                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2529                            __func__, &sip);
2530                 return;
2531         }
2532         slave->last_rx = jiffies;
2533         slave->target_last_arp_rx[i] = jiffies;
2534 }
2535 
2536 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2537                  struct slave *slave)
2538 {
2539         struct arphdr *arp = (struct arphdr *)skb->data;
2540         struct slave *curr_active_slave, *curr_arp_slave;
2541         unsigned char *arp_ptr;
2542         __be32 sip, tip;
2543         int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2544         unsigned int alen;
2545 
2546         if (!slave_do_arp_validate(bond, slave)) {
2547                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2548                     !slave_do_arp_validate_only(bond))
2549                         slave->last_rx = jiffies;
2550                 return RX_HANDLER_ANOTHER;
2551         } else if (!is_arp) {
2552                 return RX_HANDLER_ANOTHER;
2553         }
2554 
2555         alen = arp_hdr_len(bond->dev);
2556 
2557         slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2558                    __func__, skb->dev->name);
2559 
2560         if (alen > skb_headlen(skb)) {
2561                 arp = kmalloc(alen, GFP_ATOMIC);
2562                 if (!arp)
2563                         goto out_unlock;
2564                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2565                         goto out_unlock;
2566         }
2567 
2568         if (arp->ar_hln != bond->dev->addr_len ||
2569             skb->pkt_type == PACKET_OTHERHOST ||
2570             skb->pkt_type == PACKET_LOOPBACK ||
2571             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2572             arp->ar_pro != htons(ETH_P_IP) ||
2573             arp->ar_pln != 4)
2574                 goto out_unlock;
2575 
2576         arp_ptr = (unsigned char *)(arp + 1);
2577         arp_ptr += bond->dev->addr_len;
2578         memcpy(&sip, arp_ptr, 4);
2579         arp_ptr += 4 + bond->dev->addr_len;
2580         memcpy(&tip, arp_ptr, 4);
2581 
2582         slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2583                   __func__, slave->dev->name, bond_slave_state(slave),
2584                   bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2585                   &sip, &tip);
2586 
2587         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2588         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2589 
2590         /* We 'trust' the received ARP enough to validate it if:
2591          *
2592          * (a) the slave receiving the ARP is active (which includes the
2593          * current ARP slave, if any), or
2594          *
2595          * (b) the receiving slave isn't active, but there is a currently
2596          * active slave and it received valid arp reply(s) after it became
2597          * the currently active slave, or
2598          *
2599          * (c) there is an ARP slave that sent an ARP during the prior ARP
2600          * interval, and we receive an ARP reply on any slave.  We accept
2601          * these because switch FDB update delays may deliver the ARP
2602          * reply to a slave other than the sender of the ARP request.
2603          *
2604          * Note: for (b), backup slaves are receiving the broadcast ARP
2605          * request, not a reply.  This request passes from the sending
2606          * slave through the L2 switch(es) to the receiving slave.  Since
2607          * this is checking the request, sip/tip are swapped for
2608          * validation.
2609          *
2610          * This is done to avoid endless looping when we can't reach the
2611          * arp_ip_target and fool ourselves with our own arp requests.
2612          */
2613         if (bond_is_active_slave(slave))
2614                 bond_validate_arp(bond, slave, sip, tip);
2615         else if (curr_active_slave &&
2616                  time_after(slave_last_rx(bond, curr_active_slave),
2617                             curr_active_slave->last_link_up))
2618                 bond_validate_arp(bond, slave, tip, sip);
2619         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2620                  bond_time_in_interval(bond,
2621                                        dev_trans_start(curr_arp_slave->dev), 1))
2622                 bond_validate_arp(bond, slave, sip, tip);
2623 
2624 out_unlock:
2625         if (arp != (struct arphdr *)skb->data)
2626                 kfree(arp);
2627         return RX_HANDLER_ANOTHER;
2628 }
2629 
2630 /* function to verify if we're in the arp_interval timeslice, returns true if
2631  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2632  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2633  */
2634 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2635                                   int mod)
2636 {
2637         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2638 
2639         return time_in_range(jiffies,
2640                              last_act - delta_in_ticks,
2641                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
2642 }
2643 
2644 /* This function is called regularly to monitor each slave's link
2645  * ensuring that traffic is being sent and received when arp monitoring
2646  * is used in load-balancing mode. if the adapter has been dormant, then an
2647  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2648  * arp monitoring in active backup mode.
2649  */
2650 static void bond_loadbalance_arp_mon(struct bonding *bond)
2651 {
2652         struct slave *slave, *oldcurrent;
2653         struct list_head *iter;
2654         int do_failover = 0, slave_state_changed = 0;
2655 
2656         if (!bond_has_slaves(bond))
2657                 goto re_arm;
2658 
2659         rcu_read_lock();
2660 
2661         oldcurrent = rcu_dereference(bond->curr_active_slave);
2662         /* see if any of the previous devices are up now (i.e. they have
2663          * xmt and rcv traffic). the curr_active_slave does not come into
2664          * the picture unless it is null. also, slave->last_link_up is not
2665          * needed here because we send an arp on each slave and give a slave
2666          * as long as it needs to get the tx/rx within the delta.
2667          * TODO: what about up/down delay in arp mode? it wasn't here before
2668          *       so it can wait
2669          */
2670         bond_for_each_slave_rcu(bond, slave, iter) {
2671                 unsigned long trans_start = dev_trans_start(slave->dev);
2672 
2673                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2674 
2675                 if (slave->link != BOND_LINK_UP) {
2676                         if (bond_time_in_interval(bond, trans_start, 1) &&
2677                             bond_time_in_interval(bond, slave->last_rx, 1)) {
2678 
2679                                 bond_propose_link_state(slave, BOND_LINK_UP);
2680                                 slave_state_changed = 1;
2681 
2682                                 /* primary_slave has no meaning in round-robin
2683                                  * mode. the window of a slave being up and
2684                                  * curr_active_slave being null after enslaving
2685                                  * is closed.
2686                                  */
2687                                 if (!oldcurrent) {
2688                                         slave_info(bond->dev, slave->dev, "link status definitely up\n");
2689                                         do_failover = 1;
2690                                 } else {
2691                                         slave_info(bond->dev, slave->dev, "interface is now up\n");
2692                                 }
2693                         }
2694                 } else {
2695                         /* slave->link == BOND_LINK_UP */
2696 
2697                         /* not all switches will respond to an arp request
2698                          * when the source ip is 0, so don't take the link down
2699                          * if we don't know our ip yet
2700                          */
2701                         if (!bond_time_in_interval(bond, trans_start, 2) ||
2702                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
2703 
2704                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2705                                 slave_state_changed = 1;
2706 
2707                                 if (slave->link_failure_count < UINT_MAX)
2708                                         slave->link_failure_count++;
2709 
2710                                 slave_info(bond->dev, slave->dev, "interface is now down\n");
2711 
2712                                 if (slave == oldcurrent)
2713                                         do_failover = 1;
2714                         }
2715                 }
2716 
2717                 /* note: if switch is in round-robin mode, all links
2718                  * must tx arp to ensure all links rx an arp - otherwise
2719                  * links may oscillate or not come up at all; if switch is
2720                  * in something like xor mode, there is nothing we can
2721                  * do - all replies will be rx'ed on same link causing slaves
2722                  * to be unstable during low/no traffic periods
2723                  */
2724                 if (bond_slave_is_up(slave))
2725                         bond_arp_send_all(bond, slave);
2726         }
2727 
2728         rcu_read_unlock();
2729 
2730         if (do_failover || slave_state_changed) {
2731                 if (!rtnl_trylock())
2732                         goto re_arm;
2733 
2734                 bond_for_each_slave(bond, slave, iter) {
2735                         if (slave->link_new_state != BOND_LINK_NOCHANGE)
2736                                 slave->link = slave->link_new_state;
2737                 }
2738 
2739                 if (slave_state_changed) {
2740                         bond_slave_state_change(bond);
2741                         if (BOND_MODE(bond) == BOND_MODE_XOR)
2742                                 bond_update_slave_arr(bond, NULL);
2743                 }
2744                 if (do_failover) {
2745                         block_netpoll_tx();
2746                         bond_select_active_slave(bond);
2747                         unblock_netpoll_tx();
2748                 }
2749                 rtnl_unlock();
2750         }
2751 
2752 re_arm:
2753         if (bond->params.arp_interval)
2754                 queue_delayed_work(bond->wq, &bond->arp_work,
2755                                    msecs_to_jiffies(bond->params.arp_interval));
2756 }
2757 
2758 /* Called to inspect slaves for active-backup mode ARP monitor link state
2759  * changes.  Sets proposed link state in slaves to specify what action
2760  * should take place for the slave.  Returns 0 if no changes are found, >0
2761  * if changes to link states must be committed.
2762  *
2763  * Called with rcu_read_lock held.
2764  */
2765 static int bond_ab_arp_inspect(struct bonding *bond)
2766 {
2767         unsigned long trans_start, last_rx;
2768         struct list_head *iter;
2769         struct slave *slave;
2770         int commit = 0;
2771 
2772         bond_for_each_slave_rcu(bond, slave, iter) {
2773                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2774                 last_rx = slave_last_rx(bond, slave);
2775 
2776                 if (slave->link != BOND_LINK_UP) {
2777                         if (bond_time_in_interval(bond, last_rx, 1)) {
2778                                 bond_propose_link_state(slave, BOND_LINK_UP);
2779                                 commit++;
2780                         }
2781                         continue;
2782                 }
2783 
2784                 /* Give slaves 2*delta after being enslaved or made
2785                  * active.  This avoids bouncing, as the last receive
2786                  * times need a full ARP monitor cycle to be updated.
2787                  */
2788                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2789                         continue;
2790 
2791                 /* Backup slave is down if:
2792                  * - No current_arp_slave AND
2793                  * - more than 3*delta since last receive AND
2794                  * - the bond has an IP address
2795                  *
2796                  * Note: a non-null current_arp_slave indicates
2797                  * the curr_active_slave went down and we are
2798                  * searching for a new one; under this condition
2799                  * we only take the curr_active_slave down - this
2800                  * gives each slave a chance to tx/rx traffic
2801                  * before being taken out
2802                  */
2803                 if (!bond_is_active_slave(slave) &&
2804                     !rcu_access_pointer(bond->current_arp_slave) &&
2805                     !bond_time_in_interval(bond, last_rx, 3)) {
2806                         bond_propose_link_state(slave, BOND_LINK_DOWN);
2807                         commit++;
2808                 }
2809 
2810                 /* Active slave is down if:
2811                  * - more than 2*delta since transmitting OR
2812                  * - (more than 2*delta since receive AND
2813                  *    the bond has an IP address)
2814                  */
2815                 trans_start = dev_trans_start(slave->dev);
2816                 if (bond_is_active_slave(slave) &&
2817                     (!bond_time_in_interval(bond, trans_start, 2) ||
2818                      !bond_time_in_interval(bond, last_rx, 2))) {
2819                         bond_propose_link_state(slave, BOND_LINK_DOWN);
2820                         commit++;
2821                 }
2822         }
2823 
2824         return commit;
2825 }
2826 
2827 /* Called to commit link state changes noted by inspection step of
2828  * active-backup mode ARP monitor.
2829  *
2830  * Called with RTNL hold.
2831  */
2832 static void bond_ab_arp_commit(struct bonding *bond)
2833 {
2834         unsigned long trans_start;
2835         struct list_head *iter;
2836         struct slave *slave;
2837 
2838         bond_for_each_slave(bond, slave, iter) {
2839                 switch (slave->link_new_state) {
2840                 case BOND_LINK_NOCHANGE:
2841                         continue;
2842 
2843                 case BOND_LINK_UP:
2844                         trans_start = dev_trans_start(slave->dev);
2845                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
2846                             (!rtnl_dereference(bond->curr_active_slave) &&
2847                              bond_time_in_interval(bond, trans_start, 1))) {
2848                                 struct slave *current_arp_slave;
2849 
2850                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2851                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
2852                                                           BOND_SLAVE_NOTIFY_NOW);
2853                                 if (current_arp_slave) {
2854                                         bond_set_slave_inactive_flags(
2855                                                 current_arp_slave,
2856                                                 BOND_SLAVE_NOTIFY_NOW);
2857                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2858                                 }
2859 
2860                                 slave_info(bond->dev, slave->dev, "link status definitely up\n");
2861 
2862                                 if (!rtnl_dereference(bond->curr_active_slave) ||
2863                                     slave == rtnl_dereference(bond->primary_slave))
2864                                         goto do_failover;
2865 
2866                         }
2867 
2868                         continue;
2869 
2870                 case BOND_LINK_DOWN:
2871                         if (slave->link_failure_count < UINT_MAX)
2872                                 slave->link_failure_count++;
2873 
2874                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2875                                                   BOND_SLAVE_NOTIFY_NOW);
2876                         bond_set_slave_inactive_flags(slave,
2877                                                       BOND_SLAVE_NOTIFY_NOW);
2878 
2879                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2880 
2881                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
2882                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2883                                 goto do_failover;
2884                         }
2885 
2886                         continue;
2887 
2888                 default:
2889                         slave_err(bond->dev, slave->dev,
2890                                   "impossible: link_new_state %d on slave\n",
2891                                   slave->link_new_state);
2892                         continue;
2893                 }
2894 
2895 do_failover:
2896                 block_netpoll_tx();
2897                 bond_select_active_slave(bond);
2898                 unblock_netpoll_tx();
2899         }
2900 
2901         bond_set_carrier(bond);
2902 }
2903 
2904 /* Send ARP probes for active-backup mode ARP monitor.
2905  *
2906  * Called with rcu_read_lock held.
2907  */
2908 static bool bond_ab_arp_probe(struct bonding *bond)
2909 {
2910         struct slave *slave, *before = NULL, *new_slave = NULL,
2911                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2912                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2913         struct list_head *iter;
2914         bool found = false;
2915         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2916 
2917         if (curr_arp_slave && curr_active_slave)
2918                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2919                             curr_arp_slave->dev->name,
2920                             curr_active_slave->dev->name);
2921 
2922         if (curr_active_slave) {
2923                 bond_arp_send_all(bond, curr_active_slave);
2924                 return should_notify_rtnl;
2925         }
2926 
2927         /* if we don't have a curr_active_slave, search for the next available
2928          * backup slave from the current_arp_slave and make it the candidate
2929          * for becoming the curr_active_slave
2930          */
2931 
2932         if (!curr_arp_slave) {
2933                 curr_arp_slave = bond_first_slave_rcu(bond);
2934                 if (!curr_arp_slave)
2935                         return should_notify_rtnl;
2936         }
2937 
2938         bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2939 
2940         bond_for_each_slave_rcu(bond, slave, iter) {
2941                 if (!found && !before && bond_slave_is_up(slave))
2942                         before = slave;
2943 
2944                 if (found && !new_slave && bond_slave_is_up(slave))
2945                         new_slave = slave;
2946                 /* if the link state is up at this point, we
2947                  * mark it down - this can happen if we have
2948                  * simultaneous link failures and
2949                  * reselect_active_interface doesn't make this
2950                  * one the current slave so it is still marked
2951                  * up when it is actually down
2952                  */
2953                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2954                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2955                                                   BOND_SLAVE_NOTIFY_LATER);
2956                         if (slave->link_failure_count < UINT_MAX)
2957                                 slave->link_failure_count++;
2958 
2959                         bond_set_slave_inactive_flags(slave,
2960                                                       BOND_SLAVE_NOTIFY_LATER);
2961 
2962                         slave_info(bond->dev, slave->dev, "backup interface is now down\n");
2963                 }
2964                 if (slave == curr_arp_slave)
2965                         found = true;
2966         }
2967 
2968         if (!new_slave && before)
2969                 new_slave = before;
2970 
2971         if (!new_slave)
2972                 goto check_state;
2973 
2974         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2975                                   BOND_SLAVE_NOTIFY_LATER);
2976         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2977         bond_arp_send_all(bond, new_slave);
2978         new_slave->last_link_up = jiffies;
2979         rcu_assign_pointer(bond->current_arp_slave, new_slave);
2980 
2981 check_state:
2982         bond_for_each_slave_rcu(bond, slave, iter) {
2983                 if (slave->should_notify || slave->should_notify_link) {
2984                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2985                         break;
2986                 }
2987         }
2988         return should_notify_rtnl;
2989 }
2990 
2991 static void bond_activebackup_arp_mon(struct bonding *bond)
2992 {
2993         bool should_notify_peers = false;
2994         bool should_notify_rtnl = false;
2995         int delta_in_ticks;
2996 
2997         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2998 
2999         if (!bond_has_slaves(bond))
3000                 goto re_arm;
3001 
3002         rcu_read_lock();
3003 
3004         should_notify_peers = bond_should_notify_peers(bond);
3005 
3006         if (bond_ab_arp_inspect(bond)) {
3007                 rcu_read_unlock();
3008 
3009                 /* Race avoidance with bond_close flush of workqueue */
3010                 if (!rtnl_trylock()) {
3011                         delta_in_ticks = 1;
3012                         should_notify_peers = false;
3013                         goto re_arm;
3014                 }
3015 
3016                 bond_ab_arp_commit(bond);
3017 
3018                 rtnl_unlock();
3019                 rcu_read_lock();
3020         }
3021 
3022         should_notify_rtnl = bond_ab_arp_probe(bond);
3023         rcu_read_unlock();
3024 
3025 re_arm:
3026         if (bond->params.arp_interval)
3027                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3028 
3029         if (should_notify_peers || should_notify_rtnl) {
3030                 if (!rtnl_trylock())
3031                         return;
3032 
3033                 if (should_notify_peers)
3034                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3035                                                  bond->dev);
3036                 if (should_notify_rtnl) {
3037                         bond_slave_state_notify(bond);
3038                         bond_slave_link_notify(bond);
3039                 }
3040 
3041                 rtnl_unlock();
3042         }
3043 }
3044 
3045 static void bond_arp_monitor(struct work_struct *work)
3046 {
3047         struct bonding *bond = container_of(work, struct bonding,
3048                                             arp_work.work);
3049 
3050         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3051                 bond_activebackup_arp_mon(bond);
3052         else
3053                 bond_loadbalance_arp_mon(bond);
3054 }
3055 
3056 /*-------------------------- netdev event handling --------------------------*/
3057 
3058 /* Change device name */
3059 static int bond_event_changename(struct bonding *bond)
3060 {
3061         bond_remove_proc_entry(bond);
3062         bond_create_proc_entry(bond);
3063 
3064         bond_debug_reregister(bond);
3065 
3066         return NOTIFY_DONE;
3067 }
3068 
3069 static int bond_master_netdev_event(unsigned long event,
3070                                     struct net_device *bond_dev)
3071 {
3072         struct bonding *event_bond = netdev_priv(bond_dev);
3073 
3074         netdev_dbg(bond_dev, "%s called\n", __func__);
3075 
3076         switch (event) {
3077         case NETDEV_CHANGENAME:
3078                 return bond_event_changename(event_bond);
3079         case NETDEV_UNREGISTER:
3080                 bond_remove_proc_entry(event_bond);
3081                 break;
3082         case NETDEV_REGISTER:
3083                 bond_create_proc_entry(event_bond);
3084                 break;
3085         default:
3086                 break;
3087         }
3088 
3089         return NOTIFY_DONE;
3090 }
3091 
3092 static int bond_slave_netdev_event(unsigned long event,
3093                                    struct net_device *slave_dev)
3094 {
3095         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3096         struct bonding *bond;
3097         struct net_device *bond_dev;
3098 
3099         /* A netdev event can be generated while enslaving a device
3100          * before netdev_rx_handler_register is called in which case
3101          * slave will be NULL
3102          */
3103         if (!slave) {
3104                 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3105                 return NOTIFY_DONE;
3106         }
3107 
3108         bond_dev = slave->bond->dev;
3109         bond = slave->bond;
3110         primary = rtnl_dereference(bond->primary_slave);
3111 
3112         slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3113 
3114         switch (event) {
3115         case NETDEV_UNREGISTER:
3116                 if (bond_dev->type != ARPHRD_ETHER)
3117                         bond_release_and_destroy(bond_dev, slave_dev);
3118                 else
3119                         __bond_release_one(bond_dev, slave_dev, false, true);
3120                 break;
3121         case NETDEV_UP:
3122         case NETDEV_CHANGE:
3123                 /* For 802.3ad mode only:
3124                  * Getting invalid Speed/Duplex values here will put slave
3125                  * in weird state. Mark it as link-fail if the link was
3126                  * previously up or link-down if it hasn't yet come up, and
3127                  * let link-monitoring (miimon) set it right when correct
3128                  * speeds/duplex are available.
3129                  */
3130                 if (bond_update_speed_duplex(slave) &&
3131                     BOND_MODE(bond) == BOND_MODE_8023AD) {
3132                         if (slave->last_link_up)
3133                                 slave->link = BOND_LINK_FAIL;
3134                         else
3135                                 slave->link = BOND_LINK_DOWN;
3136                 }
3137 
3138                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3139                         bond_3ad_adapter_speed_duplex_changed(slave);
3140                 /* Fallthrough */
3141         case NETDEV_DOWN:
3142                 /* Refresh slave-array if applicable!
3143                  * If the setup does not use miimon or arpmon (mode-specific!),
3144                  * then these events will not cause the slave-array to be
3145                  * refreshed. This will cause xmit to use a slave that is not
3146                  * usable. Avoid such situation by refeshing the array at these
3147                  * events. If these (miimon/arpmon) parameters are configured
3148                  * then array gets refreshed twice and that should be fine!
3149                  */
3150                 if (bond_mode_can_use_xmit_hash(bond))
3151                         bond_update_slave_arr(bond, NULL);
3152                 break;
3153         case NETDEV_CHANGEMTU:
3154                 /* TODO: Should slaves be allowed to
3155                  * independently alter their MTU?  For
3156                  * an active-backup bond, slaves need
3157                  * not be the same type of device, so
3158                  * MTUs may vary.  For other modes,
3159                  * slaves arguably should have the
3160                  * same MTUs. To do this, we'd need to
3161                  * take over the slave's change_mtu
3162                  * function for the duration of their
3163                  * servitude.
3164                  */
3165                 break;
3166         case NETDEV_CHANGENAME:
3167                 /* we don't care if we don't have primary set */
3168                 if (!bond_uses_primary(bond) ||
3169                     !bond->params.primary[0])
3170                         break;
3171 
3172                 if (slave == primary) {
3173                         /* slave's name changed - he's no longer primary */
3174                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3175                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3176                         /* we have a new primary slave */
3177                         rcu_assign_pointer(bond->primary_slave, slave);
3178                 } else { /* we didn't change primary - exit */
3179                         break;
3180                 }
3181 
3182                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3183                             primary ? slave_dev->name : "none");
3184 
3185                 block_netpoll_tx();
3186                 bond_select_active_slave(bond);
3187                 unblock_netpoll_tx();
3188                 break;
3189         case NETDEV_FEAT_CHANGE:
3190                 bond_compute_features(bond);
3191                 break;
3192         case NETDEV_RESEND_IGMP:
3193                 /* Propagate to master device */
3194                 call_netdevice_notifiers(event, slave->bond->dev);
3195                 break;
3196         default:
3197                 break;
3198         }
3199 
3200         return NOTIFY_DONE;
3201 }
3202 
3203 /* bond_netdev_event: handle netdev notifier chain events.
3204  *
3205  * This function receives events for the netdev chain.  The caller (an
3206  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3207  * locks for us to safely manipulate the slave devices (RTNL lock,
3208  * dev_probe_lock).
3209  */
3210 static int bond_netdev_event(struct notifier_block *this,
3211                              unsigned long event, void *ptr)
3212 {
3213         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3214 
3215         netdev_dbg(event_dev, "%s received %s\n",
3216                    __func__, netdev_cmd_to_name(event));
3217 
3218         if (!(event_dev->priv_flags & IFF_BONDING))
3219                 return NOTIFY_DONE;
3220 
3221         if (event_dev->flags & IFF_MASTER) {
3222                 int ret;
3223 
3224                 ret = bond_master_netdev_event(event, event_dev);
3225                 if (ret != NOTIFY_DONE)
3226                         return ret;
3227         }
3228 
3229         if (event_dev->flags & IFF_SLAVE)
3230                 return bond_slave_netdev_event(event, event_dev);
3231 
3232         return NOTIFY_DONE;
3233 }
3234 
3235 static struct notifier_block bond_netdev_notifier = {
3236         .notifier_call = bond_netdev_event,
3237 };
3238 
3239 /*---------------------------- Hashing Policies -----------------------------*/
3240 
3241 /* L2 hash helper */
3242 static inline u32 bond_eth_hash(struct sk_buff *skb)
3243 {
3244         struct ethhdr *ep, hdr_tmp;
3245 
3246         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3247         if (ep)
3248                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3249         return 0;
3250 }
3251 
3252 /* Extract the appropriate headers based on bond's xmit policy */
3253 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3254                               struct flow_keys *fk)
3255 {
3256         const struct ipv6hdr *iph6;
3257         const struct iphdr *iph;
3258         int noff, proto = -1;
3259 
3260         if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3261                 return skb_flow_dissect_flow_keys(skb, fk, 0);
3262 
3263         fk->ports.ports = 0;
3264         noff = skb_network_offset(skb);
3265         if (skb->protocol == htons(ETH_P_IP)) {
3266                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3267                         return false;
3268                 iph = ip_hdr(skb);
3269                 iph_to_flow_copy_v4addrs(fk, iph);
3270                 noff += iph->ihl << 2;
3271                 if (!ip_is_fragment(iph))
3272                         proto = iph->protocol;
3273         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3274                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3275                         return false;
3276                 iph6 = ipv6_hdr(skb);
3277                 iph_to_flow_copy_v6addrs(fk, iph6);
3278                 noff += sizeof(*iph6);
3279                 proto = iph6->nexthdr;
3280         } else {
3281                 return false;
3282         }
3283         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3284                 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3285 
3286         return true;
3287 }
3288 
3289 /**
3290  * bond_xmit_hash - generate a hash value based on the xmit policy
3291  * @bond: bonding device
3292  * @skb: buffer to use for headers
3293  *
3294  * This function will extract the necessary headers from the skb buffer and use
3295  * them to generate a hash based on the xmit_policy set in the bonding device
3296  */
3297 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3298 {
3299         struct flow_keys flow;
3300         u32 hash;
3301 
3302         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3303             skb->l4_hash)
3304                 return skb->hash;
3305 
3306         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3307             !bond_flow_dissect(bond, skb, &flow))
3308                 return bond_eth_hash(skb);
3309 
3310         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3311             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3312                 hash = bond_eth_hash(skb);
3313         else
3314                 hash = (__force u32)flow.ports.ports;
3315         hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3316                 (__force u32)flow_get_u32_src(&flow);
3317         hash ^= (hash >> 16);
3318         hash ^= (hash >> 8);
3319 
3320         return hash >> 1;
3321 }
3322 
3323 /*-------------------------- Device entry points ----------------------------*/
3324 
3325 void bond_work_init_all(struct bonding *bond)
3326 {
3327         INIT_DELAYED_WORK(&bond->mcast_work,
3328                           bond_resend_igmp_join_requests_delayed);
3329         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3330         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3331         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3332         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3333         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3334 }
3335 
3336 static void bond_work_cancel_all(struct bonding *bond)
3337 {
3338         cancel_delayed_work_sync(&bond->mii_work);
3339         cancel_delayed_work_sync(&bond->arp_work);
3340         cancel_delayed_work_sync(&bond->alb_work);
3341         cancel_delayed_work_sync(&bond->ad_work);
3342         cancel_delayed_work_sync(&bond->mcast_work);
3343         cancel_delayed_work_sync(&bond->slave_arr_work);
3344 }
3345 
3346 static int bond_open(struct net_device *bond_dev)
3347 {
3348         struct bonding *bond = netdev_priv(bond_dev);
3349         struct list_head *iter;
3350         struct slave *slave;
3351 
3352         /* reset slave->backup and slave->inactive */
3353         if (bond_has_slaves(bond)) {
3354                 bond_for_each_slave(bond, slave, iter) {
3355                         if (bond_uses_primary(bond) &&
3356                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3357                                 bond_set_slave_inactive_flags(slave,
3358                                                               BOND_SLAVE_NOTIFY_NOW);
3359                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3360                                 bond_set_slave_active_flags(slave,
3361                                                             BOND_SLAVE_NOTIFY_NOW);
3362                         }
3363                 }
3364         }
3365 
3366         if (bond_is_lb(bond)) {
3367                 /* bond_alb_initialize must be called before the timer
3368                  * is started.
3369                  */
3370                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3371                         return -ENOMEM;
3372                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3373                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3374         }
3375 
3376         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3377                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3378 
3379         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3380                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3381                 bond->recv_probe = bond_arp_rcv;
3382         }
3383 
3384         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3385                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3386                 /* register to receive LACPDUs */
3387                 bond->recv_probe = bond_3ad_lacpdu_recv;
3388                 bond_3ad_initiate_agg_selection(bond, 1);
3389         }
3390 
3391         if (bond_mode_can_use_xmit_hash(bond))
3392                 bond_update_slave_arr(bond, NULL);
3393 
3394         return 0;
3395 }
3396 
3397 static int bond_close(struct net_device *bond_dev)
3398 {
3399         struct bonding *bond = netdev_priv(bond_dev);
3400 
3401         bond_work_cancel_all(bond);
3402         bond->send_peer_notif = 0;
3403         if (bond_is_lb(bond))
3404                 bond_alb_deinitialize(bond);
3405         bond->recv_probe = NULL;
3406 
3407         return 0;
3408 }
3409 
3410 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3411  * that some drivers can provide 32bit values only.
3412  */
3413 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3414                             const struct rtnl_link_stats64 *_new,
3415                             const struct rtnl_link_stats64 *_old)
3416 {
3417         const u64 *new = (const u64 *)_new;
3418         const u64 *old = (const u64 *)_old;
3419         u64 *res = (u64 *)_res;
3420         int i;
3421 
3422         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3423                 u64 nv = new[i];
3424                 u64 ov = old[i];
3425                 s64 delta = nv - ov;
3426 
3427                 /* detects if this particular field is 32bit only */
3428                 if (((nv | ov) >> 32) == 0)
3429                         delta = (s64)(s32)((u32)nv - (u32)ov);
3430 
3431                 /* filter anomalies, some drivers reset their stats
3432                  * at down/up events.
3433                  */
3434                 if (delta > 0)
3435                         res[i] += delta;
3436         }
3437 }
3438 
3439 #ifdef CONFIG_LOCKDEP
3440 static int bond_get_lowest_level_rcu(struct net_device *dev)
3441 {
3442         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
3443         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
3444         int cur = 0, max = 0;
3445 
3446         now = dev;
3447         iter = &dev->adj_list.lower;
3448 
3449         while (1) {
3450                 next = NULL;
3451                 while (1) {
3452                         ldev = netdev_next_lower_dev_rcu(now, &iter);
3453                         if (!ldev)
3454                                 break;
3455 
3456                         next = ldev;
3457                         niter = &ldev->adj_list.lower;
3458                         dev_stack[cur] = now;
3459                         iter_stack[cur++] = iter;
3460                         if (max <= cur)
3461                                 max = cur;
3462                         break;
3463                 }
3464 
3465                 if (!next) {
3466                         if (!cur)
3467                                 return max;
3468                         next = dev_stack[--cur];
3469                         niter = iter_stack[cur];
3470                 }
3471 
3472                 now = next;
3473                 iter = niter;
3474         }
3475 
3476         return max;
3477 }
3478 #endif
3479 
3480 static void bond_get_stats(struct net_device *bond_dev,
3481                            struct rtnl_link_stats64 *stats)
3482 {
3483         struct bonding *bond = netdev_priv(bond_dev);
3484         struct rtnl_link_stats64 temp;
3485         struct list_head *iter;
3486         struct slave *slave;
3487         int nest_level = 0;
3488 
3489 
3490         rcu_read_lock();
3491 #ifdef CONFIG_LOCKDEP
3492         nest_level = bond_get_lowest_level_rcu(bond_dev);
3493 #endif
3494 
3495         spin_lock_nested(&bond->stats_lock, nest_level);
3496         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3497 
3498         bond_for_each_slave_rcu(bond, slave, iter) {
3499                 const struct rtnl_link_stats64 *new =
3500                         dev_get_stats(slave->dev, &temp);
3501 
3502                 bond_fold_stats(stats, new, &slave->slave_stats);
3503 
3504                 /* save off the slave stats for the next run */
3505                 memcpy(&slave->slave_stats, new, sizeof(*new));
3506         }
3507 
3508         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3509         spin_unlock(&bond->stats_lock);
3510         rcu_read_unlock();
3511 }
3512 
3513 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3514 {
3515         struct bonding *bond = netdev_priv(bond_dev);
3516         struct net_device *slave_dev = NULL;
3517         struct ifbond k_binfo;
3518         struct ifbond __user *u_binfo = NULL;
3519         struct ifslave k_sinfo;
3520         struct ifslave __user *u_sinfo = NULL;
3521         struct mii_ioctl_data *mii = NULL;
3522         struct bond_opt_value newval;
3523         struct net *net;
3524         int res = 0;
3525 
3526         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3527 
3528         switch (cmd) {
3529         case SIOCGMIIPHY:
3530                 mii = if_mii(ifr);
3531                 if (!mii)
3532                         return -EINVAL;
3533 
3534                 mii->phy_id = 0;
3535                 /* Fall Through */
3536         case SIOCGMIIREG:
3537                 /* We do this again just in case we were called by SIOCGMIIREG
3538                  * instead of SIOCGMIIPHY.
3539                  */
3540                 mii = if_mii(ifr);
3541                 if (!mii)
3542                         return -EINVAL;
3543 
3544                 if (mii->reg_num == 1) {
3545                         mii->val_out = 0;
3546                         if (netif_carrier_ok(bond->dev))
3547                                 mii->val_out = BMSR_LSTATUS;
3548                 }
3549 
3550                 return 0;
3551         case BOND_INFO_QUERY_OLD:
3552         case SIOCBONDINFOQUERY:
3553                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3554 
3555                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3556                         return -EFAULT;
3557 
3558                 bond_info_query(bond_dev, &k_binfo);
3559                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3560                         return -EFAULT;
3561 
3562                 return 0;
3563         case BOND_SLAVE_INFO_QUERY_OLD:
3564         case SIOCBONDSLAVEINFOQUERY:
3565                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3566 
3567                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3568                         return -EFAULT;
3569 
3570                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3571                 if (res == 0 &&
3572                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3573                         return -EFAULT;
3574 
3575                 return res;
3576         default:
3577                 break;
3578         }
3579 
3580         net = dev_net(bond_dev);
3581 
3582         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3583                 return -EPERM;
3584 
3585         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3586 
3587         slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
3588 
3589         if (!slave_dev)
3590                 return -ENODEV;
3591 
3592         switch (cmd) {
3593         case BOND_ENSLAVE_OLD:
3594         case SIOCBONDENSLAVE:
3595                 res = bond_enslave(bond_dev, slave_dev, NULL);
3596                 break;
3597         case BOND_RELEASE_OLD:
3598         case SIOCBONDRELEASE:
3599                 res = bond_release(bond_dev, slave_dev);
3600                 if (!res)
3601                         netdev_update_lockdep_key(slave_dev);
3602                 break;
3603         case BOND_SETHWADDR_OLD:
3604         case SIOCBONDSETHWADDR:
3605                 res = bond_set_dev_addr(bond_dev, slave_dev);
3606                 break;
3607         case BOND_CHANGE_ACTIVE_OLD:
3608         case SIOCBONDCHANGEACTIVE:
3609                 bond_opt_initstr(&newval, slave_dev->name);
3610                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3611                                             &newval);
3612                 break;
3613         default:
3614                 res = -EOPNOTSUPP;
3615         }
3616 
3617         return res;
3618 }
3619 
3620 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3621 {
3622         struct bonding *bond = netdev_priv(bond_dev);
3623 
3624         if (change & IFF_PROMISC)
3625                 bond_set_promiscuity(bond,
3626                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
3627 
3628         if (change & IFF_ALLMULTI)
3629                 bond_set_allmulti(bond,
3630                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3631 }
3632 
3633 static void bond_set_rx_mode(struct net_device *bond_dev)
3634 {
3635         struct bonding *bond = netdev_priv(bond_dev);
3636         struct list_head *iter;
3637         struct slave *slave;
3638 
3639         rcu_read_lock();
3640         if (bond_uses_primary(bond)) {
3641                 slave = rcu_dereference(bond->curr_active_slave);
3642                 if (slave) {
3643                         dev_uc_sync(slave->dev, bond_dev);
3644                         dev_mc_sync(slave->dev, bond_dev);
3645                 }
3646         } else {
3647                 bond_for_each_slave_rcu(bond, slave, iter) {
3648                         dev_uc_sync_multiple(slave->dev, bond_dev);
3649                         dev_mc_sync_multiple(slave->dev, bond_dev);
3650                 }
3651         }
3652         rcu_read_unlock();
3653 }
3654 
3655 static int bond_neigh_init(struct neighbour *n)
3656 {
3657         struct bonding *bond = netdev_priv(n->dev);
3658         const struct net_device_ops *slave_ops;
3659         struct neigh_parms parms;
3660         struct slave *slave;
3661         int ret = 0;
3662 
3663         rcu_read_lock();
3664         slave = bond_first_slave_rcu(bond);
3665         if (!slave)
3666                 goto out;
3667         slave_ops = slave->dev->netdev_ops;
3668         if (!slave_ops->ndo_neigh_setup)
3669                 goto out;
3670 
3671         /* TODO: find another way [1] to implement this.
3672          * Passing a zeroed structure is fragile,
3673          * but at least we do not pass garbage.
3674          *
3675          * [1] One way would be that ndo_neigh_setup() never touch
3676          *     struct neigh_parms, but propagate the new neigh_setup()
3677          *     back to ___neigh_create() / neigh_parms_alloc()
3678          */
3679         memset(&parms, 0, sizeof(parms));
3680         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3681 
3682         if (ret)
3683                 goto out;
3684 
3685         if (parms.neigh_setup)
3686                 ret = parms.neigh_setup(n);
3687 out:
3688         rcu_read_unlock();
3689         return ret;
3690 }
3691 
3692 /* The bonding ndo_neigh_setup is called at init time beofre any
3693  * slave exists. So we must declare proxy setup function which will
3694  * be used at run time to resolve the actual slave neigh param setup.
3695  *
3696  * It's also called by master devices (such as vlans) to setup their
3697  * underlying devices. In that case - do nothing, we're already set up from
3698  * our init.
3699  */
3700 static int bond_neigh_setup(struct net_device *dev,
3701                             struct neigh_parms *parms)
3702 {
3703         /* modify only our neigh_parms */
3704         if (parms->dev == dev)
3705                 parms->neigh_setup = bond_neigh_init;
3706 
3707         return 0;
3708 }
3709 
3710 /* Change the MTU of all of a master's slaves to match the master */
3711 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3712 {
3713         struct bonding *bond = netdev_priv(bond_dev);
3714         struct slave *slave, *rollback_slave;
3715         struct list_head *iter;
3716         int res = 0;
3717 
3718         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3719 
3720         bond_for_each_slave(bond, slave, iter) {
3721                 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
3722                            slave, slave->dev->netdev_ops->ndo_change_mtu);
3723 
3724                 res = dev_set_mtu(slave->dev, new_mtu);
3725 
3726                 if (res) {
3727                         /* If we failed to set the slave's mtu to the new value
3728                          * we must abort the operation even in ACTIVE_BACKUP
3729                          * mode, because if we allow the backup slaves to have
3730                          * different mtu values than the active slave we'll
3731                          * need to change their mtu when doing a failover. That
3732                          * means changing their mtu from timer context, which
3733                          * is probably not a good idea.
3734                          */
3735                         slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
3736                                   res, new_mtu);
3737                         goto unwind;
3738                 }
3739         }
3740 
3741         bond_dev->mtu = new_mtu;
3742 
3743         return 0;
3744 
3745 unwind:
3746         /* unwind from head to the slave that failed */
3747         bond_for_each_slave(bond, rollback_slave, iter) {
3748                 int tmp_res;
3749 
3750                 if (rollback_slave == slave)
3751                         break;
3752 
3753                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3754                 if (tmp_res)
3755                         slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
3756                                   tmp_res);
3757         }
3758 
3759         return res;
3760 }
3761 
3762 /* Change HW address
3763  *
3764  * Note that many devices must be down to change the HW address, and
3765  * downing the master releases all slaves.  We can make bonds full of
3766  * bonding devices to test this, however.
3767  */
3768 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3769 {
3770         struct bonding *bond = netdev_priv(bond_dev);
3771         struct slave *slave, *rollback_slave;
3772         struct sockaddr_storage *ss = addr, tmp_ss;
3773         struct list_head *iter;
3774         int res = 0;
3775 
3776         if (BOND_MODE(bond) == BOND_MODE_ALB)
3777                 return bond_alb_set_mac_address(bond_dev, addr);
3778 
3779 
3780         netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
3781 
3782         /* If fail_over_mac is enabled, do nothing and return success.
3783          * Returning an error causes ifenslave to fail.
3784          */
3785         if (bond->params.fail_over_mac &&
3786             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3787                 return 0;
3788 
3789         if (!is_valid_ether_addr(ss->__data))
3790                 return -EADDRNOTAVAIL;
3791 
3792         bond_for_each_slave(bond, slave, iter) {
3793                 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
3794                           __func__, slave);
3795                 res = dev_set_mac_address(slave->dev, addr, NULL);
3796                 if (res) {
3797                         /* TODO: consider downing the slave
3798                          * and retry ?
3799                          * User should expect communications
3800                          * breakage anyway until ARP finish
3801                          * updating, so...
3802                          */
3803                         slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
3804                                   __func__, res);
3805                         goto unwind;
3806                 }
3807         }
3808 
3809         /* success */
3810         memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3811         return 0;
3812 
3813 unwind:
3814         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3815         tmp_ss.ss_family = bond_dev->type;
3816 
3817         /* unwind from head to the slave that failed */
3818         bond_for_each_slave(bond, rollback_slave, iter) {
3819                 int tmp_res;
3820 
3821                 if (rollback_slave == slave)
3822                         break;
3823 
3824                 tmp_res = dev_set_mac_address(rollback_slave->dev,
3825                                               (struct sockaddr *)&tmp_ss, NULL);
3826                 if (tmp_res) {
3827                         slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
3828                                    __func__, tmp_res);
3829                 }
3830         }
3831 
3832         return res;
3833 }
3834 
3835 /**
3836  * bond_xmit_slave_id - transmit skb through slave with slave_id
3837  * @bond: bonding device that is transmitting
3838  * @skb: buffer to transmit
3839  * @slave_id: slave id up to slave_cnt-1 through which to transmit
3840  *
3841  * This function tries to transmit through slave with slave_id but in case
3842  * it fails, it tries to find the first available slave for transmission.
3843  * The skb is consumed in all cases, thus the function is void.
3844  */
3845 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3846 {
3847         struct list_head *iter;
3848         struct slave *slave;
3849         int i = slave_id;
3850 
3851         /* Here we start from the slave with slave_id */
3852         bond_for_each_slave_rcu(bond, slave, iter) {
3853                 if (--i < 0) {
3854                         if (bond_slave_can_tx(slave)) {
3855                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3856                                 return;
3857                         }
3858                 }
3859         }
3860 
3861         /* Here we start from the first slave up to slave_id */
3862         i = slave_id;
3863         bond_for_each_slave_rcu(bond, slave, iter) {
3864                 if (--i < 0)
3865                         break;
3866                 if (bond_slave_can_tx(slave)) {
3867                         bond_dev_queue_xmit(bond, skb, slave->dev);
3868                         return;
3869                 }
3870         }
3871         /* no slave that can tx has been found */
3872         bond_tx_drop(bond->dev, skb);
3873 }
3874 
3875 /**
3876  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3877  * @bond: bonding device to use
3878  *
3879  * Based on the value of the bonding device's packets_per_slave parameter
3880  * this function generates a slave id, which is usually used as the next
3881  * slave to transmit through.
3882  */
3883 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3884 {
3885         u32 slave_id;
3886         struct reciprocal_value reciprocal_packets_per_slave;
3887         int packets_per_slave = bond->params.packets_per_slave;
3888 
3889         switch (packets_per_slave) {
3890         case 0:
3891                 slave_id = prandom_u32();
3892                 break;
3893         case 1:
3894                 slave_id = bond->rr_tx_counter;
3895                 break;
3896         default:
3897                 reciprocal_packets_per_slave =
3898                         bond->params.reciprocal_packets_per_slave;
3899                 slave_id = reciprocal_divide(bond->rr_tx_counter,
3900                                              reciprocal_packets_per_slave);
3901                 break;
3902         }
3903         bond->rr_tx_counter++;
3904 
3905         return slave_id;
3906 }
3907 
3908 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3909                                         struct net_device *bond_dev)
3910 {
3911         struct bonding *bond = netdev_priv(bond_dev);
3912         struct slave *slave;
3913         int slave_cnt;
3914         u32 slave_id;
3915 
3916         /* Start with the curr_active_slave that joined the bond as the
3917          * default for sending IGMP traffic.  For failover purposes one
3918          * needs to maintain some consistency for the interface that will
3919          * send the join/membership reports.  The curr_active_slave found
3920          * will send all of this type of traffic.
3921          */
3922         if (skb->protocol == htons(ETH_P_IP)) {
3923                 int noff = skb_network_offset(skb);
3924                 struct iphdr *iph;
3925 
3926                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3927                         goto non_igmp;
3928 
3929                 iph = ip_hdr(skb);
3930                 if (iph->protocol == IPPROTO_IGMP) {
3931                         slave = rcu_dereference(bond->curr_active_slave);
3932                         if (slave)
3933                                 bond_dev_queue_xmit(bond, skb, slave->dev);
3934                         else
3935                                 bond_xmit_slave_id(bond, skb, 0);
3936                         return NETDEV_TX_OK;
3937                 }
3938         }
3939 
3940 non_igmp:
3941         slave_cnt = READ_ONCE(bond->slave_cnt);
3942         if (likely(slave_cnt)) {
3943                 slave_id = bond_rr_gen_slave_id(bond);
3944                 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3945         } else {
3946                 bond_tx_drop(bond_dev, skb);
3947         }
3948         return NETDEV_TX_OK;
3949 }
3950 
3951 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3952  * the bond has a usable interface.
3953  */
3954 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3955                                           struct net_device *bond_dev)
3956 {
3957         struct bonding *bond = netdev_priv(bond_dev);
3958         struct slave *slave;
3959 
3960         slave = rcu_dereference(bond->curr_active_slave);
3961         if (slave)
3962                 bond_dev_queue_xmit(bond, skb, slave->dev);
3963         else
3964                 bond_tx_drop(bond_dev, skb);
3965 
3966         return NETDEV_TX_OK;
3967 }
3968 
3969 /* Use this to update slave_array when (a) it's not appropriate to update
3970  * slave_array right away (note that update_slave_array() may sleep)
3971  * and / or (b) RTNL is not held.
3972  */
3973 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3974 {
3975         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3976 }
3977 
3978 /* Slave array work handler. Holds only RTNL */
3979 static void bond_slave_arr_handler(struct work_struct *work)
3980 {
3981         struct bonding *bond = container_of(work, struct bonding,
3982                                             slave_arr_work.work);
3983         int ret;
3984 
3985         if (!rtnl_trylock())
3986                 goto err;
3987 
3988         ret = bond_update_slave_arr(bond, NULL);
3989         rtnl_unlock();
3990         if (ret) {
3991                 pr_warn_ratelimited("Failed to update slave array from WT\n");
3992                 goto err;
3993         }
3994         return;
3995 
3996 err:
3997         bond_slave_arr_work_rearm(bond, 1);
3998 }
3999 
4000 /* Build the usable slaves array in control path for modes that use xmit-hash
4001  * to determine the slave interface -
4002  * (a) BOND_MODE_8023AD
4003  * (b) BOND_MODE_XOR
4004  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4005  *
4006  * The caller is expected to hold RTNL only and NO other lock!
4007  */
4008 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4009 {
4010         struct slave *slave;
4011         struct list_head *iter;
4012         struct bond_up_slave *new_arr, *old_arr;
4013         int agg_id = 0;
4014         int ret = 0;
4015 
4016 #ifdef CONFIG_LOCKDEP
4017         WARN_ON(lockdep_is_held(&bond->mode_lock));
4018 #endif
4019 
4020         new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
4021                           GFP_KERNEL);
4022         if (!new_arr) {
4023                 ret = -ENOMEM;
4024                 pr_err("Failed to build slave-array.\n");
4025                 goto out;
4026         }
4027         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4028                 struct ad_info ad_info;
4029 
4030                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4031                         pr_debug("bond_3ad_get_active_agg_info failed\n");
4032                         kfree_rcu(new_arr, rcu);
4033                         /* No active aggragator means it's not safe to use
4034                          * the previous array.
4035                          */
4036                         old_arr = rtnl_dereference(bond->slave_arr);
4037                         if (old_arr) {
4038                                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4039                                 kfree_rcu(old_arr, rcu);
4040                         }
4041                         goto out;
4042                 }
4043                 agg_id = ad_info.aggregator_id;
4044         }
4045         bond_for_each_slave(bond, slave, iter) {
4046                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4047                         struct aggregator *agg;
4048 
4049                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
4050                         if (!agg || agg->aggregator_identifier != agg_id)
4051                                 continue;
4052                 }
4053                 if (!bond_slave_can_tx(slave))
4054                         continue;
4055                 if (skipslave == slave)
4056                         continue;
4057 
4058                 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4059                           new_arr->count);
4060 
4061                 new_arr->arr[new_arr->count++] = slave;
4062         }
4063 
4064         old_arr = rtnl_dereference(bond->slave_arr);
4065         rcu_assign_pointer(bond->slave_arr, new_arr);
4066         if (old_arr)
4067                 kfree_rcu(old_arr, rcu);
4068 out:
4069         if (ret != 0 && skipslave) {
4070                 int idx;
4071 
4072                 /* Rare situation where caller has asked to skip a specific
4073                  * slave but allocation failed (most likely!). BTW this is
4074                  * only possible when the call is initiated from
4075                  * __bond_release_one(). In this situation; overwrite the
4076                  * skipslave entry in the array with the last entry from the
4077                  * array to avoid a situation where the xmit path may choose
4078                  * this to-be-skipped slave to send a packet out.
4079                  */
4080                 old_arr = rtnl_dereference(bond->slave_arr);
4081                 for (idx = 0; old_arr != NULL && idx < old_arr->count; idx++) {
4082                         if (skipslave == old_arr->arr[idx]) {
4083                                 old_arr->arr[idx] =
4084                                     old_arr->arr[old_arr->count-1];
4085                                 old_arr->count--;
4086                                 break;
4087                         }
4088                 }
4089         }
4090         return ret;
4091 }
4092 
4093 /* Use this Xmit function for 3AD as well as XOR modes. The current
4094  * usable slave array is formed in the control path. The xmit function
4095  * just calculates hash and sends the packet out.
4096  */
4097 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4098                                      struct net_device *dev)
4099 {
4100         struct bonding *bond = netdev_priv(dev);
4101         struct slave *slave;
4102         struct bond_up_slave *slaves;
4103         unsigned int count;
4104 
4105         slaves = rcu_dereference(bond->slave_arr);
4106         count = slaves ? READ_ONCE(slaves->count) : 0;
4107         if (likely(count)) {
4108                 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4109                 bond_dev_queue_xmit(bond, skb, slave->dev);
4110         } else {
4111                 bond_tx_drop(dev, skb);
4112         }
4113 
4114         return NETDEV_TX_OK;
4115 }
4116 
4117 /* in broadcast mode, we send everything to all usable interfaces. */
4118 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4119                                        struct net_device *bond_dev)
4120 {
4121         struct bonding *bond = netdev_priv(bond_dev);
4122         struct slave *slave = NULL;
4123         struct list_head *iter;
4124 
4125         bond_for_each_slave_rcu(bond, slave, iter) {
4126                 if (bond_is_last_slave(bond, slave))
4127                         break;
4128                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4129                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4130 
4131                         if (!skb2) {
4132                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4133                                                     bond_dev->name, __func__);
4134                                 continue;
4135                         }
4136                         bond_dev_queue_xmit(bond, skb2, slave->dev);
4137                 }
4138         }
4139         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4140                 bond_dev_queue_xmit(bond, skb, slave->dev);
4141         else
4142                 bond_tx_drop(bond_dev, skb);
4143 
4144         return NETDEV_TX_OK;
4145 }
4146 
4147 /*------------------------- Device initialization ---------------------------*/
4148 
4149 /* Lookup the slave that corresponds to a qid */
4150 static inline int bond_slave_override(struct bonding *bond,
4151                                       struct sk_buff *skb)
4152 {
4153         struct slave *slave = NULL;
4154         struct list_head *iter;
4155 
4156         if (!skb_rx_queue_recorded(skb))
4157                 return 1;
4158 
4159         /* Find out if any slaves have the same mapping as this skb. */
4160         bond_for_each_slave_rcu(bond, slave, iter) {
4161                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4162                         if (bond_slave_is_up(slave) &&
4163                             slave->link == BOND_LINK_UP) {
4164                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4165                                 return 0;
4166                         }
4167                         /* If the slave isn't UP, use default transmit policy. */
4168                         break;
4169                 }
4170         }
4171 
4172         return 1;
4173 }
4174 
4175 
4176 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4177                              struct net_device *sb_dev)
4178 {
4179         /* This helper function exists to help dev_pick_tx get the correct
4180          * destination queue.  Using a helper function skips a call to
4181          * skb_tx_hash and will put the skbs in the queue we expect on their
4182          * way down to the bonding driver.
4183          */
4184         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4185 
4186         /* Save the original txq to restore before passing to the driver */
4187         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4188 
4189         if (unlikely(txq >= dev->real_num_tx_queues)) {
4190                 do {
4191                         txq -= dev->real_num_tx_queues;
4192                 } while (txq >= dev->real_num_tx_queues);
4193         }
4194         return txq;
4195 }
4196 
4197 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4198 {
4199         struct bonding *bond = netdev_priv(dev);
4200 
4201         if (bond_should_override_tx_queue(bond) &&
4202             !bond_slave_override(bond, skb))
4203                 return NETDEV_TX_OK;
4204 
4205         switch (BOND_MODE(bond)) {
4206         case BOND_MODE_ROUNDROBIN:
4207                 return bond_xmit_roundrobin(skb, dev);
4208         case BOND_MODE_ACTIVEBACKUP:
4209                 return bond_xmit_activebackup(skb, dev);
4210         case BOND_MODE_8023AD:
4211         case BOND_MODE_XOR:
4212                 return bond_3ad_xor_xmit(skb, dev);
4213         case BOND_MODE_BROADCAST:
4214                 return bond_xmit_broadcast(skb, dev);
4215         case BOND_MODE_ALB:
4216                 return bond_alb_xmit(skb, dev);
4217         case BOND_MODE_TLB:
4218                 return bond_tlb_xmit(skb, dev);
4219         default:
4220                 /* Should never happen, mode already checked */
4221                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4222                 WARN_ON_ONCE(1);
4223                 bond_tx_drop(dev, skb);
4224                 return NETDEV_TX_OK;
4225         }
4226 }
4227 
4228 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4229 {
4230         struct bonding *bond = netdev_priv(dev);
4231         netdev_tx_t ret = NETDEV_TX_OK;
4232 
4233         /* If we risk deadlock from transmitting this in the
4234          * netpoll path, tell netpoll to queue the frame for later tx
4235          */
4236         if (unlikely(is_netpoll_tx_blocked(dev)))
4237                 return NETDEV_TX_BUSY;
4238 
4239         rcu_read_lock();
4240         if (bond_has_slaves(bond))
4241                 ret = __bond_start_xmit(skb, dev);
4242         else
4243                 bond_tx_drop(dev, skb);
4244         rcu_read_unlock();
4245 
4246         return ret;
4247 }
4248 
4249 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4250                                            struct ethtool_link_ksettings *cmd)
4251 {
4252         struct bonding *bond = netdev_priv(bond_dev);
4253         unsigned long speed = 0;
4254         struct list_head *iter;
4255         struct slave *slave;
4256 
4257         cmd->base.duplex = DUPLEX_UNKNOWN;
4258         cmd->base.port = PORT_OTHER;
4259 
4260         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4261          * do not need to check mode.  Though link speed might not represent
4262          * the true receive or transmit bandwidth (not all modes are symmetric)
4263          * this is an accurate maximum.
4264          */
4265         bond_for_each_slave(bond, slave, iter) {
4266                 if (bond_slave_can_tx(slave)) {
4267                         if (slave->speed != SPEED_UNKNOWN)
4268                                 speed += slave->speed;
4269                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4270                             slave->duplex != DUPLEX_UNKNOWN)
4271                                 cmd->base.duplex = slave->duplex;
4272                 }
4273         }
4274         cmd->base.speed = speed ? : SPEED_UNKNOWN;
4275 
4276         return 0;
4277 }
4278 
4279 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4280                                      struct ethtool_drvinfo *drvinfo)
4281 {
4282         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4283         strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4284         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4285                  BOND_ABI_VERSION);
4286 }
4287 
4288 static const struct ethtool_ops bond_ethtool_ops = {
4289         .get_drvinfo            = bond_ethtool_get_drvinfo,
4290         .get_link               = ethtool_op_get_link,
4291         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
4292 };
4293 
4294 static const struct net_device_ops bond_netdev_ops = {
4295         .ndo_init               = bond_init,
4296         .ndo_uninit             = bond_uninit,
4297         .ndo_open               = bond_open,
4298         .ndo_stop               = bond_close,
4299         .ndo_start_xmit         = bond_start_xmit,
4300         .ndo_select_queue       = bond_select_queue,
4301         .ndo_get_stats64        = bond_get_stats,
4302         .ndo_do_ioctl           = bond_do_ioctl,
4303         .ndo_change_rx_flags    = bond_change_rx_flags,
4304         .ndo_set_rx_mode        = bond_set_rx_mode,
4305         .ndo_change_mtu         = bond_change_mtu,
4306         .ndo_set_mac_address    = bond_set_mac_address,
4307         .ndo_neigh_setup        = bond_neigh_setup,
4308         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4309         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4310 #ifdef CONFIG_NET_POLL_CONTROLLER
4311         .ndo_netpoll_setup      = bond_netpoll_setup,
4312         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4313         .ndo_poll_controller    = bond_poll_controller,
4314 #endif
4315         .ndo_add_slave          = bond_enslave,
4316         .ndo_del_slave          = bond_release,
4317         .ndo_fix_features       = bond_fix_features,
4318         .ndo_features_check     = passthru_features_check,
4319 };
4320 
4321 static const struct device_type bond_type = {
4322         .name = "bond",
4323 };
4324 
4325 static void bond_destructor(struct net_device *bond_dev)
4326 {
4327         struct bonding *bond = netdev_priv(bond_dev);
4328         if (bond->wq)
4329                 destroy_workqueue(bond->wq);
4330 }
4331 
4332 void bond_setup(struct net_device *bond_dev)
4333 {
4334         struct bonding *bond = netdev_priv(bond_dev);
4335 
4336         spin_lock_init(&bond->mode_lock);
4337         bond->params = bonding_defaults;
4338 
4339         /* Initialize pointers */
4340         bond->dev = bond_dev;
4341 
4342         /* Initialize the device entry points */
4343         ether_setup(bond_dev);
4344         bond_dev->max_mtu = ETH_MAX_MTU;
4345         bond_dev->netdev_ops = &bond_netdev_ops;
4346         bond_dev->ethtool_ops = &bond_ethtool_ops;
4347 
4348         bond_dev->needs_free_netdev = true;
4349         bond_dev->priv_destructor = bond_destructor;
4350 
4351         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4352 
4353         /* Initialize the device options */
4354         bond_dev->flags |= IFF_MASTER;
4355         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4356         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4357 
4358         /* don't acquire bond device's netif_tx_lock when transmitting */
4359         bond_dev->features |= NETIF_F_LLTX;
4360 
4361         /* By default, we declare the bond to be fully
4362          * VLAN hardware accelerated capable. Special
4363          * care is taken in the various xmit functions
4364          * when there are slaves that are not hw accel
4365          * capable
4366          */
4367 
4368         /* Don't allow bond devices to change network namespaces. */
4369         bond_dev->features |= NETIF_F_NETNS_LOCAL;
4370 
4371         bond_dev->hw_features = BOND_VLAN_FEATURES |
4372                                 NETIF_F_HW_VLAN_CTAG_RX |
4373                                 NETIF_F_HW_VLAN_CTAG_FILTER;
4374 
4375         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4376         bond_dev->features |= bond_dev->hw_features;
4377         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
4378 }
4379 
4380 /* Destroy a bonding device.
4381  * Must be under rtnl_lock when this function is called.
4382  */
4383 static void bond_uninit(struct net_device *bond_dev)
4384 {
4385         struct bonding *bond = netdev_priv(bond_dev);
4386         struct list_head *iter;
4387         struct slave *slave;
4388         struct bond_up_slave *arr;
4389 
4390         bond_netpoll_cleanup(bond_dev);
4391 
4392         /* Release the bonded slaves */
4393         bond_for_each_slave(bond, slave, iter)
4394                 __bond_release_one(bond_dev, slave->dev, true, true);
4395         netdev_info(bond_dev, "Released all slaves\n");
4396 
4397         arr = rtnl_dereference(bond->slave_arr);
4398         if (arr) {
4399                 RCU_INIT_POINTER(bond->slave_arr, NULL);
4400                 kfree_rcu(arr, rcu);
4401         }
4402 
4403         list_del(&bond->bond_list);
4404 
4405         lockdep_unregister_key(&bond->stats_lock_key);
4406         bond_debug_unregister(bond);
4407 }
4408 
4409 /*------------------------- Module initialization ---------------------------*/
4410 
4411 static int bond_check_params(struct bond_params *params)
4412 {
4413         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4414         struct bond_opt_value newval;
4415         const struct bond_opt_value *valptr;
4416         int arp_all_targets_value = 0;
4417         u16 ad_actor_sys_prio = 0;
4418         u16 ad_user_port_key = 0;
4419         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4420         int arp_ip_count;
4421         int bond_mode   = BOND_MODE_ROUNDROBIN;
4422         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4423         int lacp_fast = 0;
4424         int tlb_dynamic_lb;
4425 
4426         /* Convert string parameters. */
4427         if (mode) {
4428                 bond_opt_initstr(&newval, mode);
4429                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4430                 if (!valptr) {
4431                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4432                         return -EINVAL;
4433                 }
4434                 bond_mode = valptr->value;
4435         }
4436 
4437         if (xmit_hash_policy) {
4438                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4439                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
4440                     bond_mode == BOND_MODE_BROADCAST) {
4441                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4442                                 bond_mode_name(bond_mode));
4443                 } else {
4444                         bond_opt_initstr(&newval, xmit_hash_policy);
4445                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4446                                                 &newval);
4447                         if (!valptr) {
4448                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4449                                        xmit_hash_policy);
4450                                 return -EINVAL;
4451                         }
4452                         xmit_hashtype = valptr->value;
4453                 }
4454         }
4455 
4456         if (lacp_rate) {
4457                 if (bond_mode != BOND_MODE_8023AD) {
4458                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4459                                 bond_mode_name(bond_mode));
4460                 } else {
4461                         bond_opt_initstr(&newval, lacp_rate);
4462                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4463                                                 &newval);
4464                         if (!valptr) {
4465                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4466                                        lacp_rate);
4467                                 return -EINVAL;
4468                         }
4469                         lacp_fast = valptr->value;
4470                 }
4471         }
4472 
4473         if (ad_select) {
4474                 bond_opt_initstr(&newval, ad_select);
4475                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4476                                         &newval);
4477                 if (!valptr) {
4478                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4479                         return -EINVAL;
4480                 }
4481                 params->ad_select = valptr->value;
4482                 if (bond_mode != BOND_MODE_8023AD)
4483                         pr_warn("ad_select param only affects 802.3ad mode\n");
4484         } else {
4485                 params->ad_select = BOND_AD_STABLE;
4486         }
4487 
4488         if (max_bonds < 0) {
4489                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4490                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4491                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4492         }
4493 
4494         if (miimon < 0) {
4495                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4496                         miimon, INT_MAX);
4497                 miimon = 0;
4498         }
4499 
4500         if (updelay < 0) {
4501                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4502                         updelay, INT_MAX);
4503                 updelay = 0;
4504         }
4505 
4506         if (downdelay < 0) {
4507                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4508                         downdelay, INT_MAX);
4509                 downdelay = 0;
4510         }
4511 
4512         if ((use_carrier != 0) && (use_carrier != 1)) {
4513                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4514                         use_carrier);
4515                 use_carrier = 1;
4516         }
4517 
4518         if (num_peer_notif < 0 || num_peer_notif > 255) {
4519                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4520                         num_peer_notif);
4521                 num_peer_notif = 1;
4522         }
4523 
4524         /* reset values for 802.3ad/TLB/ALB */
4525         if (!bond_mode_uses_arp(bond_mode)) {
4526                 if (!miimon) {
4527                         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");
4528                         pr_warn("Forcing miimon to 100msec\n");
4529                         miimon = BOND_DEFAULT_MIIMON;
4530                 }
4531         }
4532 
4533         if (tx_queues < 1 || tx_queues > 255) {
4534                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4535                         tx_queues, BOND_DEFAULT_TX_QUEUES);
4536                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4537         }
4538 
4539         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4540                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4541                         all_slaves_active);
4542                 all_slaves_active = 0;
4543         }
4544 
4545         if (resend_igmp < 0 || resend_igmp > 255) {
4546                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4547                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4548                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4549         }
4550 
4551         bond_opt_initval(&newval, packets_per_slave);
4552         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4553                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4554                         packets_per_slave, USHRT_MAX);
4555                 packets_per_slave = 1;
4556         }
4557 
4558         if (bond_mode == BOND_MODE_ALB) {
4559                 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",
4560                           updelay);
4561         }
4562 
4563         if (!miimon) {
4564                 if (updelay || downdelay) {
4565                         /* just warn the user the up/down delay will have
4566                          * no effect since miimon is zero...
4567                          */
4568                         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",
4569                                 updelay, downdelay);
4570                 }
4571         } else {
4572                 /* don't allow arp monitoring */
4573                 if (arp_interval) {
4574                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4575                                 miimon, arp_interval);
4576                         arp_interval = 0;
4577                 }
4578 
4579                 if ((updelay % miimon) != 0) {
4580                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4581                                 updelay, miimon, (updelay / miimon) * miimon);
4582                 }
4583 
4584                 updelay /= miimon;
4585 
4586                 if ((downdelay % miimon) != 0) {
4587                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4588                                 downdelay, miimon,
4589                                 (downdelay / miimon) * miimon);
4590                 }
4591 
4592                 downdelay /= miimon;
4593         }
4594 
4595         if (arp_interval < 0) {
4596                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4597                         arp_interval, INT_MAX);
4598                 arp_interval = 0;
4599         }
4600 
4601         for (arp_ip_count = 0, i = 0;
4602              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4603                 __be32 ip;
4604 
4605                 /* not a complete check, but good enough to catch mistakes */
4606                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4607                     !bond_is_ip_target_ok(ip)) {
4608                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4609                                 arp_ip_target[i]);
4610                         arp_interval = 0;
4611                 } else {
4612                         if (bond_get_targets_ip(arp_target, ip) == -1)
4613                                 arp_target[arp_ip_count++] = ip;
4614                         else
4615                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4616                                         &ip);
4617                 }
4618         }
4619 
4620         if (arp_interval && !arp_ip_count) {
4621                 /* don't allow arping if no arp_ip_target given... */
4622                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4623                         arp_interval);
4624                 arp_interval = 0;
4625         }
4626 
4627         if (arp_validate) {
4628                 if (!arp_interval) {
4629                         pr_err("arp_validate requires arp_interval\n");
4630                         return -EINVAL;
4631                 }
4632 
4633                 bond_opt_initstr(&newval, arp_validate);
4634                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4635                                         &newval);
4636                 if (!valptr) {
4637                         pr_err("Error: invalid arp_validate \"%s\"\n",
4638                                arp_validate);
4639                         return -EINVAL;
4640                 }
4641                 arp_validate_value = valptr->value;
4642         } else {
4643                 arp_validate_value = 0;
4644         }
4645 
4646         if (arp_all_targets) {
4647                 bond_opt_initstr(&newval, arp_all_targets);
4648                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4649                                         &newval);
4650                 if (!valptr) {
4651                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4652                                arp_all_targets);
4653                         arp_all_targets_value = 0;
4654                 } else {
4655                         arp_all_targets_value = valptr->value;
4656                 }
4657         }
4658 
4659         if (miimon) {
4660                 pr_info("MII link monitoring set to %d ms\n", miimon);
4661         } else if (arp_interval) {
4662                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4663                                           arp_validate_value);
4664                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4665                         arp_interval, valptr->string, arp_ip_count);
4666 
4667                 for (i = 0; i < arp_ip_count; i++)
4668                         pr_cont(" %s", arp_ip_target[i]);
4669 
4670                 pr_cont("\n");
4671 
4672         } else if (max_bonds) {
4673                 /* miimon and arp_interval not set, we need one so things
4674                  * work as expected, see bonding.txt for details
4675                  */
4676                 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");
4677         }
4678 
4679         if (primary && !bond_mode_uses_primary(bond_mode)) {
4680                 /* currently, using a primary only makes sense
4681                  * in active backup, TLB or ALB modes
4682                  */
4683                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4684                         primary, bond_mode_name(bond_mode));
4685                 primary = NULL;
4686         }
4687 
4688         if (primary && primary_reselect) {
4689                 bond_opt_initstr(&newval, primary_reselect);
4690                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4691                                         &newval);
4692                 if (!valptr) {
4693                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4694                                primary_reselect);
4695                         return -EINVAL;
4696                 }
4697                 primary_reselect_value = valptr->value;
4698         } else {
4699                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4700         }
4701 
4702         if (fail_over_mac) {
4703                 bond_opt_initstr(&newval, fail_over_mac);
4704                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4705                                         &newval);
4706                 if (!valptr) {
4707                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4708                                fail_over_mac);
4709                         return -EINVAL;
4710                 }
4711                 fail_over_mac_value = valptr->value;
4712                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4713                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4714         } else {
4715                 fail_over_mac_value = BOND_FOM_NONE;
4716         }
4717 
4718         bond_opt_initstr(&newval, "default");
4719         valptr = bond_opt_parse(
4720                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4721                                      &newval);
4722         if (!valptr) {
4723                 pr_err("Error: No ad_actor_sys_prio default value");
4724                 return -EINVAL;
4725         }
4726         ad_actor_sys_prio = valptr->value;
4727 
4728         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4729                                 &newval);
4730         if (!valptr) {
4731                 pr_err("Error: No ad_user_port_key default value");
4732                 return -EINVAL;
4733         }
4734         ad_user_port_key = valptr->value;
4735 
4736         bond_opt_initstr(&newval, "default");
4737         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4738         if (!valptr) {
4739                 pr_err("Error: No tlb_dynamic_lb default value");
4740                 return -EINVAL;
4741         }
4742         tlb_dynamic_lb = valptr->value;
4743 
4744         if (lp_interval == 0) {
4745                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4746                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4747                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4748         }
4749 
4750         /* fill params struct with the proper values */
4751         params->mode = bond_mode;
4752         params->xmit_policy = xmit_hashtype;
4753         params->miimon = miimon;
4754         params->num_peer_notif = num_peer_notif;
4755         params->arp_interval = arp_interval;
4756         params->arp_validate = arp_validate_value;
4757         params->arp_all_targets = arp_all_targets_value;
4758         params->updelay = updelay;
4759         params->downdelay = downdelay;
4760         params->peer_notif_delay = 0;
4761         params->use_carrier = use_carrier;
4762         params->lacp_fast = lacp_fast;
4763         params->primary[0] = 0;
4764         params->primary_reselect = primary_reselect_value;
4765         params->fail_over_mac = fail_over_mac_value;
4766         params->tx_queues = tx_queues;
4767         params->all_slaves_active = all_slaves_active;
4768         params->resend_igmp = resend_igmp;
4769         params->min_links = min_links;
4770         params->lp_interval = lp_interval;
4771         params->packets_per_slave = packets_per_slave;
4772         params->tlb_dynamic_lb = tlb_dynamic_lb;
4773         params->ad_actor_sys_prio = ad_actor_sys_prio;
4774         eth_zero_addr(params->ad_actor_system);
4775         params->ad_user_port_key = ad_user_port_key;
4776         if (packets_per_slave > 0) {
4777                 params->reciprocal_packets_per_slave =
4778                         reciprocal_value(packets_per_slave);
4779         } else {
4780                 /* reciprocal_packets_per_slave is unused if
4781                  * packets_per_slave is 0 or 1, just initialize it
4782                  */
4783                 params->reciprocal_packets_per_slave =
4784                         (struct reciprocal_value) { 0 };
4785         }
4786 
4787         if (primary) {
4788                 strncpy(params->primary, primary, IFNAMSIZ);
4789                 params->primary[IFNAMSIZ - 1] = 0;
4790         }
4791 
4792         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4793 
4794         return 0;
4795 }
4796 
4797 /* Called from registration process */
4798 static int bond_init(struct net_device *bond_dev)
4799 {
4800         struct bonding *bond = netdev_priv(bond_dev);
4801         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4802 
4803         netdev_dbg(bond_dev, "Begin bond_init\n");
4804 
4805         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4806         if (!bond->wq)
4807                 return -ENOMEM;
4808 
4809         spin_lock_init(&bond->stats_lock);
4810         lockdep_register_key(&bond->stats_lock_key);
4811         lockdep_set_class(&bond->stats_lock, &bond->stats_lock_key);
4812 
4813         list_add_tail(&bond->bond_list, &bn->dev_list);
4814 
4815         bond_prepare_sysfs_group(bond);
4816 
4817         bond_debug_register(bond);
4818 
4819         /* Ensure valid dev_addr */
4820         if (is_zero_ether_addr(bond_dev->dev_addr) &&
4821             bond_dev->addr_assign_type == NET_ADDR_PERM)
4822                 eth_hw_addr_random(bond_dev);
4823 
4824         return 0;
4825 }
4826 
4827 unsigned int bond_get_num_tx_queues(void)
4828 {
4829         return tx_queues;
4830 }
4831 
4832 /* Create a new bond based on the specified name and bonding parameters.
4833  * If name is NULL, obtain a suitable "bond%d" name for us.
4834  * Caller must NOT hold rtnl_lock; we need to release it here before we
4835  * set up our sysfs entries.
4836  */
4837 int bond_create(struct net *net, const char *name)
4838 {
4839         struct net_device *bond_dev;
4840         struct bonding *bond;
4841         struct alb_bond_info *bond_info;
4842         int res;
4843 
4844         rtnl_lock();
4845 
4846         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4847                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
4848                                    bond_setup, tx_queues);
4849         if (!bond_dev) {
4850                 pr_err("%s: eek! can't alloc netdev!\n", name);
4851                 rtnl_unlock();
4852                 return -ENOMEM;
4853         }
4854 
4855         /*
4856          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4857          * It is set to 0 by default which is wrong.
4858          */
4859         bond = netdev_priv(bond_dev);
4860         bond_info = &(BOND_ALB_INFO(bond));
4861         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4862 
4863         dev_net_set(bond_dev, net);
4864         bond_dev->rtnl_link_ops = &bond_link_ops;
4865 
4866         res = register_netdevice(bond_dev);
4867 
4868         netif_carrier_off(bond_dev);
4869 
4870         bond_work_init_all(bond);
4871 
4872         rtnl_unlock();
4873         if (res < 0)
4874                 free_netdev(bond_dev);
4875         return res;
4876 }
4877 
4878 static int __net_init bond_net_init(struct net *net)
4879 {
4880         struct bond_net *bn = net_generic(net, bond_net_id);
4881 
4882         bn->net = net;
4883         INIT_LIST_HEAD(&bn->dev_list);
4884 
4885         bond_create_proc_dir(bn);
4886         bond_create_sysfs(bn);
4887 
4888         return 0;
4889 }
4890 
4891 static void __net_exit bond_net_exit(struct net *net)
4892 {
4893         struct bond_net *bn = net_generic(net, bond_net_id);
4894         struct bonding *bond, *tmp_bond;
4895         LIST_HEAD(list);
4896 
4897         bond_destroy_sysfs(bn);
4898 
4899         /* Kill off any bonds created after unregistering bond rtnl ops */
4900         rtnl_lock();
4901         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4902                 unregister_netdevice_queue(bond->dev, &list);
4903         unregister_netdevice_many(&list);
4904         rtnl_unlock();
4905 
4906         bond_destroy_proc_dir(bn);
4907 }
4908 
4909 static struct pernet_operations bond_net_ops = {
4910         .init = bond_net_init,
4911         .exit = bond_net_exit,
4912         .id   = &bond_net_id,
4913         .size = sizeof(struct bond_net),
4914 };
4915 
4916 static int __init bonding_init(void)
4917 {
4918         int i;
4919         int res;
4920 
4921         pr_info("%s", bond_version);
4922 
4923         res = bond_check_params(&bonding_defaults);
4924         if (res)
4925                 goto out;
4926 
4927         res = register_pernet_subsys(&bond_net_ops);
4928         if (res)
4929                 goto out;
4930 
4931         res = bond_netlink_init();
4932         if (res)
4933                 goto err_link;
4934 
4935         bond_create_debugfs();
4936 
4937         for (i = 0; i < max_bonds; i++) {
4938                 res = bond_create(&init_net, NULL);
4939                 if (res)
4940                         goto err;
4941         }
4942 
4943         register_netdevice_notifier(&bond_netdev_notifier);
4944 out:
4945         return res;
4946 err:
4947         bond_destroy_debugfs();
4948         bond_netlink_fini();
4949 err_link:
4950         unregister_pernet_subsys(&bond_net_ops);
4951         goto out;
4952 
4953 }
4954 
4955 static void __exit bonding_exit(void)
4956 {
4957         unregister_netdevice_notifier(&bond_netdev_notifier);
4958 
4959         bond_destroy_debugfs();
4960 
4961         bond_netlink_fini();
4962         unregister_pernet_subsys(&bond_net_ops);
4963 
4964 #ifdef CONFIG_NET_POLL_CONTROLLER
4965         /* Make sure we don't have an imbalance on our netpoll blocking */
4966         WARN_ON(atomic_read(&netpoll_block_tx));
4967 #endif
4968 }
4969 
4970 module_init(bonding_init);
4971 module_exit(bonding_exit);
4972 MODULE_LICENSE("GPL");
4973 MODULE_VERSION(DRV_VERSION);
4974 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4975 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
/* [<][>][^][v][top][bottom][index][help] */
root/drivers/net/bonding/bond_main.c

DEFINITIONS
