drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/interrupt.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/ioport.h>
  41 #include <linux/in.h>
  42 #include <net/ip.h>
  43 #include <linux/ip.h>
  44 #include <linux/icmp.h>
  45 #include <linux/icmpv6.h>
  46 #include <linux/tcp.h>
  47 #include <linux/udp.h>
  48 #include <linux/slab.h>
  49 #include <linux/string.h>
  50 #include <linux/init.h>
  51 #include <linux/timer.h>
  52 #include <linux/socket.h>
  53 #include <linux/ctype.h>
  54 #include <linux/inet.h>
  55 #include <linux/bitops.h>
  56 #include <linux/io.h>
  57 #include <asm/dma.h>
  58 #include <linux/uaccess.h>
  59 #include <linux/errno.h>
  60 #include <linux/netdevice.h>
  61 #include <linux/inetdevice.h>
  62 #include <linux/igmp.h>
  63 #include <linux/etherdevice.h>
  64 #include <linux/skbuff.h>
  65 #include <net/sock.h>
  66 #include <linux/rtnetlink.h>
  67 #include <linux/smp.h>
  68 #include <linux/if_ether.h>
  69 #include <net/arp.h>
  70 #include <linux/mii.h>
  71 #include <linux/ethtool.h>
  72 #include <linux/if_vlan.h>
  73 #include <linux/if_bonding.h>
  74 #include <linux/jiffies.h>
  75 #include <linux/preempt.h>
  76 #include <net/route.h>
  77 #include <net/net_namespace.h>
  78 #include <net/netns/generic.h>
  79 #include <net/pkt_sched.h>
  80 #include <linux/rculist.h>
  81 #include <net/flow_dissector.h>
  82 #include <net/xfrm.h>
  83 #include <net/bonding.h>
  84 #include <net/bond_3ad.h>
  85 #include <net/bond_alb.h>
  86 #if IS_ENABLED(CONFIG_TLS_DEVICE)
  87 #include <net/tls.h>
  88 #endif
  89
  90 #include "bonding_priv.h"
  91
  92 /*---------------------------- Module parameters ----------------------------*/
  93
  94 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  95
  96 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  97 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  98 static int num_peer_notif = 1;
  99 static int miimon;
 100 static int updelay;
 101 static int downdelay;
 102 static int use_carrier  = 1;
 103 static char *mode;
 104 static char *primary;
 105 static char *primary_reselect;
 106 static char *lacp_rate;
 107 static int min_links;
 108 static char *ad_select;
 109 static char *xmit_hash_policy;
 110 static int arp_interval;
 111 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 112 static char *arp_validate;
 113 static char *arp_all_targets;
 114 static char *fail_over_mac;
 115 static int all_slaves_active;
 116 static struct bond_params bonding_defaults;
 117 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 118 static int packets_per_slave = 1;
 119 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 120
 121 module_param(max_bonds, int, 0);
 122 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 123 module_param(tx_queues, int, 0);
 124 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 125 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 126 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 127                                "failover event (alias of num_unsol_na)");
 128 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 129 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 130                                "failover event (alias of num_grat_arp)");
 131 module_param(miimon, int, 0);
 132 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 133 module_param(updelay, int, 0);
 134 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 135 module_param(downdelay, int, 0);
 136 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 137                             "in milliseconds");
 138 module_param(use_carrier, int, 0);
 139 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 140                               "0 for off, 1 for on (default)");
 141 module_param(mode, charp, 0);
 142 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 143                        "1 for active-backup, 2 for balance-xor, "
 144                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 145                        "6 for balance-alb");
 146 module_param(primary, charp, 0);
 147 MODULE_PARM_DESC(primary, "Primary network device to use");
 148 module_param(primary_reselect, charp, 0);
 149 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 150                                    "once it comes up; "
 151                                    "0 for always (default), "
 152                                    "1 for only if speed of primary is "
 153                                    "better, "
 154                                    "2 for only on active slave "
 155                                    "failure");
 156 module_param(lacp_rate, charp, 0);
 157 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 158                             "0 for slow, 1 for fast");
 159 module_param(ad_select, charp, 0);
 160 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
 161                             "0 for stable (default), 1 for bandwidth, "
 162                             "2 for count");
 163 module_param(min_links, int, 0);
 164 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 165
 166 module_param(xmit_hash_policy, charp, 0);
 167 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
 168                                    "0 for layer 2 (default), 1 for layer 3+4, "
 169                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 170                                    "4 for encap layer 3+4, 5 for vlan+srcmac");
 171 module_param(arp_interval, int, 0);
 172 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 173 module_param_array(arp_ip_target, charp, NULL, 0);
 174 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 175 module_param(arp_validate, charp, 0);
 176 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 177                                "0 for none (default), 1 for active, "
 178                                "2 for backup, 3 for all");
 179 module_param(arp_all_targets, charp, 0);
 180 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 181 module_param(fail_over_mac, charp, 0);
 182 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 183                                 "the same MAC; 0 for none (default), "
 184                                 "1 for active, 2 for follow");
 185 module_param(all_slaves_active, int, 0);
 186 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 187                                      "by setting active flag for all slaves; "
 188                                      "0 for never (default), 1 for always.");
 189 module_param(resend_igmp, int, 0);
 190 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 191                               "link failure");
 192 module_param(packets_per_slave, int, 0);
 193 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 194                                     "mode; 0 for a random slave, 1 packet per "
 195                                     "slave (default), >1 packets per slave.");
 196 module_param(lp_interval, uint, 0);
 197 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 198                               "the bonding driver sends learning packets to "
 199                               "each slaves peer switch. The default is 1.");
 200
 201 /*----------------------------- Global variables ----------------------------*/
 202
 203 #ifdef CONFIG_NET_POLL_CONTROLLER
 204 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 205 #endif
 206
 207 unsigned int bond_net_id __read_mostly;
 208
 209 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
 210         {
 211                 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
 212                 .offset = offsetof(struct flow_keys, control),
 213         },
 214         {
 215                 .key_id = FLOW_DISSECTOR_KEY_BASIC,
 216                 .offset = offsetof(struct flow_keys, basic),
 217         },
 218         {
 219                 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 220                 .offset = offsetof(struct flow_keys, addrs.v4addrs),
 221         },
 222         {
 223                 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 224                 .offset = offsetof(struct flow_keys, addrs.v6addrs),
 225         },
 226         {
 227                 .key_id = FLOW_DISSECTOR_KEY_TIPC,
 228                 .offset = offsetof(struct flow_keys, addrs.tipckey),
 229         },
 230         {
 231                 .key_id = FLOW_DISSECTOR_KEY_PORTS,
 232                 .offset = offsetof(struct flow_keys, ports),
 233         },
 234         {
 235                 .key_id = FLOW_DISSECTOR_KEY_ICMP,
 236                 .offset = offsetof(struct flow_keys, icmp),
 237         },
 238         {
 239                 .key_id = FLOW_DISSECTOR_KEY_VLAN,
 240                 .offset = offsetof(struct flow_keys, vlan),
 241         },
 242         {
 243                 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
 244                 .offset = offsetof(struct flow_keys, tags),
 245         },
 246         {
 247                 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
 248                 .offset = offsetof(struct flow_keys, keyid),
 249         },
 250 };
 251
 252 static struct flow_dissector flow_keys_bonding __read_mostly;
 253
 254 /*-------------------------- Forward declarations ---------------------------*/
 255
 256 static int bond_init(struct net_device *bond_dev);
 257 static void bond_uninit(struct net_device *bond_dev);
 258 static void bond_get_stats(struct net_device *bond_dev,
 259                            struct rtnl_link_stats64 *stats);
 260 static void bond_slave_arr_handler(struct work_struct *work);
 261 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 262                                   int mod);
 263 static void bond_netdev_notify_work(struct work_struct *work);
 264
 265 /*---------------------------- General routines -----------------------------*/
 266
 267 const char *bond_mode_name(int mode)
 268 {
 269         static const char *names[] = {
 270                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 271                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 272                 [BOND_MODE_XOR] = "load balancing (xor)",
 273                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 274                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 275                 [BOND_MODE_TLB] = "transmit load balancing",
 276                 [BOND_MODE_ALB] = "adaptive load balancing",
 277         };
 278
 279         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 280                 return "unknown";
 281
 282         return names[mode];
 283 }
 284
 285 /**
 286  * bond_dev_queue_xmit - Prepare skb for xmit.
 287  *
 288  * @bond: bond device that got this skb for tx.
 289  * @skb: hw accel VLAN tagged skb to transmit
 290  * @slave_dev: slave that is supposed to xmit this skbuff
 291  */
 292 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 293                         struct net_device *slave_dev)
 294 {
 295         skb->dev = slave_dev;
 296
 297         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 298                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 299         skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 300
 301         if (unlikely(netpoll_tx_running(bond->dev)))
 302                 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 303
 304         return dev_queue_xmit(skb);
 305 }
 306
 307 bool bond_sk_check(struct bonding *bond)
 308 {
 309         switch (BOND_MODE(bond)) {
 310         case BOND_MODE_8023AD:
 311         case BOND_MODE_XOR:
 312                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
 313                         return true;
 314                 fallthrough;
 315         default:
 316                 return false;
 317         }
 318 }
 319
 320 /*---------------------------------- VLAN -----------------------------------*/
 321
 322 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 323  * We don't protect the slave list iteration with a lock because:
 324  * a. This operation is performed in IOCTL context,
 325  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 326  * c. Holding a lock with BH disabled while directly calling a base driver
 327  *    entry point is generally a BAD idea.
 328  *
 329  * The design of synchronization/protection for this operation in the 8021q
 330  * module is good for one or more VLAN devices over a single physical device
 331  * and cannot be extended for a teaming solution like bonding, so there is a
 332  * potential race condition here where a net device from the vlan group might
 333  * be referenced (either by a base driver or the 8021q code) while it is being
 334  * removed from the system. However, it turns out we're not making matters
 335  * worse, and if it works for regular VLAN usage it will work here too.
 336 */
 337
 338 /**
 339  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 340  * @bond_dev: bonding net device that got called
 341  * @proto: network protocol ID
 342  * @vid: vlan id being added
 343  */
 344 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 345                                 __be16 proto, u16 vid)
 346 {
 347         struct bonding *bond = netdev_priv(bond_dev);
 348         struct slave *slave, *rollback_slave;
 349         struct list_head *iter;
 350         int res;
 351
 352         bond_for_each_slave(bond, slave, iter) {
 353                 res = vlan_vid_add(slave->dev, proto, vid);
 354                 if (res)
 355                         goto unwind;
 356         }
 357
 358         return 0;
 359
 360 unwind:
 361         /* unwind to the slave that failed */
 362         bond_for_each_slave(bond, rollback_slave, iter) {
 363                 if (rollback_slave == slave)
 364                         break;
 365
 366                 vlan_vid_del(rollback_slave->dev, proto, vid);
 367         }
 368
 369         return res;
 370 }
 371
 372 /**
 373  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 374  * @bond_dev: bonding net device that got called
 375  * @proto: network protocol ID
 376  * @vid: vlan id being removed
 377  */
 378 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 379                                  __be16 proto, u16 vid)
 380 {
 381         struct bonding *bond = netdev_priv(bond_dev);
 382         struct list_head *iter;
 383         struct slave *slave;
 384
 385         bond_for_each_slave(bond, slave, iter)
 386                 vlan_vid_del(slave->dev, proto, vid);
 387
 388         if (bond_is_lb(bond))
 389                 bond_alb_clear_vlan(bond, vid);
 390
 391         return 0;
 392 }
 393
 394 /*---------------------------------- XFRM -----------------------------------*/
 395
 396 #ifdef CONFIG_XFRM_OFFLOAD
 397 /**
 398  * bond_ipsec_add_sa - program device with a security association
 399  * @xs: pointer to transformer state struct
 400  **/
 401 static int bond_ipsec_add_sa(struct xfrm_state *xs)
 402 {
 403         struct net_device *bond_dev = xs->xso.dev;
 404         struct bond_ipsec *ipsec;
 405         struct bonding *bond;
 406         struct slave *slave;
 407         int err;
 408
 409         if (!bond_dev)
 410                 return -EINVAL;
 411
 412         rcu_read_lock();
 413         bond = netdev_priv(bond_dev);
 414         slave = rcu_dereference(bond->curr_active_slave);
 415         if (!slave) {
 416                 rcu_read_unlock();
 417                 return -ENODEV;
 418         }
 419
 420         if (!slave->dev->xfrmdev_ops ||
 421             !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
 422             netif_is_bond_master(slave->dev)) {
 423                 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
 424                 rcu_read_unlock();
 425                 return -EINVAL;
 426         }
 427
 428         ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
 429         if (!ipsec) {
 430                 rcu_read_unlock();
 431                 return -ENOMEM;
 432         }
 433         xs->xso.real_dev = slave->dev;
 434
 435         err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
 436         if (!err) {
 437                 ipsec->xs = xs;
 438                 INIT_LIST_HEAD(&ipsec->list);
 439                 spin_lock_bh(&bond->ipsec_lock);
 440                 list_add(&ipsec->list, &bond->ipsec_list);
 441                 spin_unlock_bh(&bond->ipsec_lock);
 442         } else {
 443                 kfree(ipsec);
 444         }
 445         rcu_read_unlock();
 446         return err;
 447 }
 448
 449 static void bond_ipsec_add_sa_all(struct bonding *bond)
 450 {
 451         struct net_device *bond_dev = bond->dev;
 452         struct bond_ipsec *ipsec;
 453         struct slave *slave;
 454
 455         rcu_read_lock();
 456         slave = rcu_dereference(bond->curr_active_slave);
 457         if (!slave)
 458                 goto out;
 459
 460         if (!slave->dev->xfrmdev_ops ||
 461             !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
 462             netif_is_bond_master(slave->dev)) {
 463                 spin_lock_bh(&bond->ipsec_lock);
 464                 if (!list_empty(&bond->ipsec_list))
 465                         slave_warn(bond_dev, slave->dev,
 466                                    "%s: no slave xdo_dev_state_add\n",
 467                                    __func__);
 468                 spin_unlock_bh(&bond->ipsec_lock);
 469                 goto out;
 470         }
 471
 472         spin_lock_bh(&bond->ipsec_lock);
 473         list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 474                 ipsec->xs->xso.real_dev = slave->dev;
 475                 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) {
 476                         slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
 477                         ipsec->xs->xso.real_dev = NULL;
 478                 }
 479         }
 480         spin_unlock_bh(&bond->ipsec_lock);
 481 out:
 482         rcu_read_unlock();
 483 }
 484
 485 /**
 486  * bond_ipsec_del_sa - clear out this specific SA
 487  * @xs: pointer to transformer state struct
 488  **/
 489 static void bond_ipsec_del_sa(struct xfrm_state *xs)
 490 {
 491         struct net_device *bond_dev = xs->xso.dev;
 492         struct bond_ipsec *ipsec;
 493         struct bonding *bond;
 494         struct slave *slave;
 495
 496         if (!bond_dev)
 497                 return;
 498
 499         rcu_read_lock();
 500         bond = netdev_priv(bond_dev);
 501         slave = rcu_dereference(bond->curr_active_slave);
 502
 503         if (!slave)
 504                 goto out;
 505
 506         if (!xs->xso.real_dev)
 507                 goto out;
 508
 509         WARN_ON(xs->xso.real_dev != slave->dev);
 510
 511         if (!slave->dev->xfrmdev_ops ||
 512             !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
 513             netif_is_bond_master(slave->dev)) {
 514                 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
 515                 goto out;
 516         }
 517
 518         slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
 519 out:
 520         spin_lock_bh(&bond->ipsec_lock);
 521         list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 522                 if (ipsec->xs == xs) {
 523                         list_del(&ipsec->list);
 524                         kfree(ipsec);
 525                         break;
 526                 }
 527         }
 528         spin_unlock_bh(&bond->ipsec_lock);
 529         rcu_read_unlock();
 530 }
 531
 532 static void bond_ipsec_del_sa_all(struct bonding *bond)
 533 {
 534         struct net_device *bond_dev = bond->dev;
 535         struct bond_ipsec *ipsec;
 536         struct slave *slave;
 537
 538         rcu_read_lock();
 539         slave = rcu_dereference(bond->curr_active_slave);
 540         if (!slave) {
 541                 rcu_read_unlock();
 542                 return;
 543         }
 544
 545         spin_lock_bh(&bond->ipsec_lock);
 546         list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 547                 if (!ipsec->xs->xso.real_dev)
 548                         continue;
 549
 550                 if (!slave->dev->xfrmdev_ops ||
 551                     !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
 552                     netif_is_bond_master(slave->dev)) {
 553                         slave_warn(bond_dev, slave->dev,
 554                                    "%s: no slave xdo_dev_state_delete\n",
 555                                    __func__);
 556                 } else {
 557                         slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
 558                 }
 559                 ipsec->xs->xso.real_dev = NULL;
 560         }
 561         spin_unlock_bh(&bond->ipsec_lock);
 562         rcu_read_unlock();
 563 }
 564
 565 /**
 566  * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
 567  * @skb: current data packet
 568  * @xs: pointer to transformer state struct
 569  **/
 570 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
 571 {
 572         struct net_device *bond_dev = xs->xso.dev;
 573         struct net_device *real_dev;
 574         struct slave *curr_active;
 575         struct bonding *bond;
 576         int err;
 577
 578         bond = netdev_priv(bond_dev);
 579         rcu_read_lock();
 580         curr_active = rcu_dereference(bond->curr_active_slave);
 581         real_dev = curr_active->dev;
 582
 583         if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
 584                 err = false;
 585                 goto out;
 586         }
 587
 588         if (!xs->xso.real_dev) {
 589                 err = false;
 590                 goto out;
 591         }
 592
 593         if (!real_dev->xfrmdev_ops ||
 594             !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
 595             netif_is_bond_master(real_dev)) {
 596                 err = false;
 597                 goto out;
 598         }
 599
 600         err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
 601 out:
 602         rcu_read_unlock();
 603         return err;
 604 }
 605
 606 static const struct xfrmdev_ops bond_xfrmdev_ops = {
 607         .xdo_dev_state_add = bond_ipsec_add_sa,
 608         .xdo_dev_state_delete = bond_ipsec_del_sa,
 609         .xdo_dev_offload_ok = bond_ipsec_offload_ok,
 610 };
 611 #endif /* CONFIG_XFRM_OFFLOAD */
 612
 613 /*------------------------------- Link status -------------------------------*/
 614
 615 /* Set the carrier state for the master according to the state of its
 616  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 617  * do special 802.3ad magic.
 618  *
 619  * Returns zero if carrier state does not change, nonzero if it does.
 620  */
 621 int bond_set_carrier(struct bonding *bond)
 622 {
 623         struct list_head *iter;
 624         struct slave *slave;
 625
 626         if (!bond_has_slaves(bond))
 627                 goto down;
 628
 629         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 630                 return bond_3ad_set_carrier(bond);
 631
 632         bond_for_each_slave(bond, slave, iter) {
 633                 if (slave->link == BOND_LINK_UP) {
 634                         if (!netif_carrier_ok(bond->dev)) {
 635                                 netif_carrier_on(bond->dev);
 636                                 return 1;
 637                         }
 638                         return 0;
 639                 }
 640         }
 641
 642 down:
 643         if (netif_carrier_ok(bond->dev)) {
 644                 netif_carrier_off(bond->dev);
 645                 return 1;
 646         }
 647         return 0;
 648 }
 649
 650 /* Get link speed and duplex from the slave's base driver
 651  * using ethtool. If for some reason the call fails or the
 652  * values are invalid, set speed and duplex to -1,
 653  * and return. Return 1 if speed or duplex settings are
 654  * UNKNOWN; 0 otherwise.
 655  */
 656 static int bond_update_speed_duplex(struct slave *slave)
 657 {
 658         struct net_device *slave_dev = slave->dev;
 659         struct ethtool_link_ksettings ecmd;
 660         int res;
 661
 662         slave->speed = SPEED_UNKNOWN;
 663         slave->duplex = DUPLEX_UNKNOWN;
 664
 665         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 666         if (res < 0)
 667                 return 1;
 668         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 669                 return 1;
 670         switch (ecmd.base.duplex) {
 671         case DUPLEX_FULL:
 672         case DUPLEX_HALF:
 673                 break;
 674         default:
 675                 return 1;
 676         }
 677
 678         slave->speed = ecmd.base.speed;
 679         slave->duplex = ecmd.base.duplex;
 680
 681         return 0;
 682 }
 683
 684 const char *bond_slave_link_status(s8 link)
 685 {
 686         switch (link) {
 687         case BOND_LINK_UP:
 688                 return "up";
 689         case BOND_LINK_FAIL:
 690                 return "going down";
 691         case BOND_LINK_DOWN:
 692                 return "down";
 693         case BOND_LINK_BACK:
 694                 return "going back";
 695         default:
 696                 return "unknown";
 697         }
 698 }
 699
 700 /* if <dev> supports MII link status reporting, check its link status.
 701  *
 702  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 703  * depending upon the setting of the use_carrier parameter.
 704  *
 705  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 706  * can't tell and just pretend it is), or 0, meaning that the link is
 707  * down.
 708  *
 709  * If reporting is non-zero, instead of faking link up, return -1 if
 710  * both ETHTOOL and MII ioctls fail (meaning the device does not
 711  * support them).  If use_carrier is set, return whatever it says.
 712  * It'd be nice if there was a good way to tell if a driver supports
 713  * netif_carrier, but there really isn't.
 714  */
 715 static int bond_check_dev_link(struct bonding *bond,
 716                                struct net_device *slave_dev, int reporting)
 717 {
 718         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 719         int (*ioctl)(struct net_device *, struct ifreq *, int);
 720         struct ifreq ifr;
 721         struct mii_ioctl_data *mii;
 722
 723         if (!reporting && !netif_running(slave_dev))
 724                 return 0;
 725
 726         if (bond->params.use_carrier)
 727                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 728
 729         /* Try to get link status using Ethtool first. */
 730         if (slave_dev->ethtool_ops->get_link)
 731                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 732                         BMSR_LSTATUS : 0;
 733
 734         /* Ethtool can't be used, fallback to MII ioctls. */
 735         ioctl = slave_ops->ndo_do_ioctl;
 736         if (ioctl) {
 737                 /* TODO: set pointer to correct ioctl on a per team member
 738                  *       bases to make this more efficient. that is, once
 739                  *       we determine the correct ioctl, we will always
 740                  *       call it and not the others for that team
 741                  *       member.
 742                  */
 743
 744                 /* We cannot assume that SIOCGMIIPHY will also read a
 745                  * register; not all network drivers (e.g., e100)
 746                  * support that.
 747                  */
 748
 749                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 750                 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 751                 mii = if_mii(&ifr);
 752                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 753                         mii->reg_num = MII_BMSR;
 754                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 755                                 return mii->val_out & BMSR_LSTATUS;
 756                 }
 757         }
 758
 759         /* If reporting, report that either there's no dev->do_ioctl,
 760          * or both SIOCGMIIREG and get_link failed (meaning that we
 761          * cannot report link status).  If not reporting, pretend
 762          * we're ok.
 763          */
 764         return reporting ? -1 : BMSR_LSTATUS;
 765 }
 766
 767 /*----------------------------- Multicast list ------------------------------*/
 768
 769 /* Push the promiscuity flag down to appropriate slaves */
 770 static int bond_set_promiscuity(struct bonding *bond, int inc)
 771 {
 772         struct list_head *iter;
 773         int err = 0;
 774
 775         if (bond_uses_primary(bond)) {
 776                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 777
 778                 if (curr_active)
 779                         err = dev_set_promiscuity(curr_active->dev, inc);
 780         } else {
 781                 struct slave *slave;
 782
 783                 bond_for_each_slave(bond, slave, iter) {
 784                         err = dev_set_promiscuity(slave->dev, inc);
 785                         if (err)
 786                                 return err;
 787                 }
 788         }
 789         return err;
 790 }
 791
 792 /* Push the allmulti flag down to all slaves */
 793 static int bond_set_allmulti(struct bonding *bond, int inc)
 794 {
 795         struct list_head *iter;
 796         int err = 0;
 797
 798         if (bond_uses_primary(bond)) {
 799                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 800
 801                 if (curr_active)
 802                         err = dev_set_allmulti(curr_active->dev, inc);
 803         } else {
 804                 struct slave *slave;
 805
 806                 bond_for_each_slave(bond, slave, iter) {
 807                         err = dev_set_allmulti(slave->dev, inc);
 808                         if (err)
 809                                 return err;
 810                 }
 811         }
 812         return err;
 813 }
 814
 815 /* Retrieve the list of registered multicast addresses for the bonding
 816  * device and retransmit an IGMP JOIN request to the current active
 817  * slave.
 818  */
 819 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 820 {
 821         struct bonding *bond = container_of(work, struct bonding,
 822                                             mcast_work.work);
 823
 824         if (!rtnl_trylock()) {
 825                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 826                 return;
 827         }
 828         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 829
 830         if (bond->igmp_retrans > 1) {
 831                 bond->igmp_retrans--;
 832                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 833         }
 834         rtnl_unlock();
 835 }
 836
 837 /* Flush bond's hardware addresses from slave */
 838 static void bond_hw_addr_flush(struct net_device *bond_dev,
 839                                struct net_device *slave_dev)
 840 {
 841         struct bonding *bond = netdev_priv(bond_dev);
 842
 843         dev_uc_unsync(slave_dev, bond_dev);
 844         dev_mc_unsync(slave_dev, bond_dev);
 845
 846         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 847                 /* del lacpdu mc addr from mc list */
 848                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 849
 850                 dev_mc_del(slave_dev, lacpdu_multicast);
 851         }
 852 }
 853
 854 /*--------------------------- Active slave change ---------------------------*/
 855
 856 /* Update the hardware address list and promisc/allmulti for the new and
 857  * old active slaves (if any).  Modes that are not using primary keep all
 858  * slaves up date at all times; only the modes that use primary need to call
 859  * this function to swap these settings during a failover.
 860  */
 861 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 862                               struct slave *old_active)
 863 {
 864         if (old_active) {
 865                 if (bond->dev->flags & IFF_PROMISC)
 866                         dev_set_promiscuity(old_active->dev, -1);
 867
 868                 if (bond->dev->flags & IFF_ALLMULTI)
 869                         dev_set_allmulti(old_active->dev, -1);
 870
 871                 bond_hw_addr_flush(bond->dev, old_active->dev);
 872         }
 873
 874         if (new_active) {
 875                 /* FIXME: Signal errors upstream. */
 876                 if (bond->dev->flags & IFF_PROMISC)
 877                         dev_set_promiscuity(new_active->dev, 1);
 878
 879                 if (bond->dev->flags & IFF_ALLMULTI)
 880                         dev_set_allmulti(new_active->dev, 1);
 881
 882                 netif_addr_lock_bh(bond->dev);
 883                 dev_uc_sync(new_active->dev, bond->dev);
 884                 dev_mc_sync(new_active->dev, bond->dev);
 885                 netif_addr_unlock_bh(bond->dev);
 886         }
 887 }
 888
 889 /**
 890  * bond_set_dev_addr - clone slave's address to bond
 891  * @bond_dev: bond net device
 892  * @slave_dev: slave net device
 893  *
 894  * Should be called with RTNL held.
 895  */
 896 static int bond_set_dev_addr(struct net_device *bond_dev,
 897                              struct net_device *slave_dev)
 898 {
 899         int err;
 900
 901         slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 902                   bond_dev, slave_dev, slave_dev->addr_len);
 903         err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
 904         if (err)
 905                 return err;
 906
 907         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 908         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 909         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 910         return 0;
 911 }
 912
 913 static struct slave *bond_get_old_active(struct bonding *bond,
 914                                          struct slave *new_active)
 915 {
 916         struct slave *slave;
 917         struct list_head *iter;
 918
 919         bond_for_each_slave(bond, slave, iter) {
 920                 if (slave == new_active)
 921                         continue;
 922
 923                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 924                         return slave;
 925         }
 926
 927         return NULL;
 928 }
 929
 930 /* bond_do_fail_over_mac
 931  *
 932  * Perform special MAC address swapping for fail_over_mac settings
 933  *
 934  * Called with RTNL
 935  */
 936 static void bond_do_fail_over_mac(struct bonding *bond,
 937                                   struct slave *new_active,
 938                                   struct slave *old_active)
 939 {
 940         u8 tmp_mac[MAX_ADDR_LEN];
 941         struct sockaddr_storage ss;
 942         int rv;
 943
 944         switch (bond->params.fail_over_mac) {
 945         case BOND_FOM_ACTIVE:
 946                 if (new_active) {
 947                         rv = bond_set_dev_addr(bond->dev, new_active->dev);
 948                         if (rv)
 949                                 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
 950                                           -rv);
 951                 }
 952                 break;
 953         case BOND_FOM_FOLLOW:
 954                 /* if new_active && old_active, swap them
 955                  * if just old_active, do nothing (going to no active slave)
 956                  * if just new_active, set new_active to bond's MAC
 957                  */
 958                 if (!new_active)
 959                         return;
 960
 961                 if (!old_active)
 962                         old_active = bond_get_old_active(bond, new_active);
 963
 964                 if (old_active) {
 965                         bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 966                                           new_active->dev->addr_len);
 967                         bond_hw_addr_copy(ss.__data,
 968                                           old_active->dev->dev_addr,
 969                                           old_active->dev->addr_len);
 970                         ss.ss_family = new_active->dev->type;
 971                 } else {
 972                         bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 973                                           bond->dev->addr_len);
 974                         ss.ss_family = bond->dev->type;
 975                 }
 976
 977                 rv = dev_set_mac_address(new_active->dev,
 978                                          (struct sockaddr *)&ss, NULL);
 979                 if (rv) {
 980                         slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
 981                                   -rv);
 982                         goto out;
 983                 }
 984
 985                 if (!old_active)
 986                         goto out;
 987
 988                 bond_hw_addr_copy(ss.__data, tmp_mac,
 989                                   new_active->dev->addr_len);
 990                 ss.ss_family = old_active->dev->type;
 991
 992                 rv = dev_set_mac_address(old_active->dev,
 993                                          (struct sockaddr *)&ss, NULL);
 994                 if (rv)
 995                         slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
 996                                   -rv);
 997 out:
 998                 break;
 999         default:
1000                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1001                            bond->params.fail_over_mac);
1002                 break;
1003         }
1004
1005 }
1006
1007 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1008 {
1009         struct slave *prim = rtnl_dereference(bond->primary_slave);
1010         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1011
1012         if (!prim || prim->link != BOND_LINK_UP) {
1013                 if (!curr || curr->link != BOND_LINK_UP)
1014                         return NULL;
1015                 return curr;
1016         }
1017
1018         if (bond->force_primary) {
1019                 bond->force_primary = false;
1020                 return prim;
1021         }
1022
1023         if (!curr || curr->link != BOND_LINK_UP)
1024                 return prim;
1025
1026         /* At this point, prim and curr are both up */
1027         switch (bond->params.primary_reselect) {
1028         case BOND_PRI_RESELECT_ALWAYS:
1029                 return prim;
1030         case BOND_PRI_RESELECT_BETTER:
1031                 if (prim->speed < curr->speed)
1032                         return curr;
1033                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1034                         return curr;
1035                 return prim;
1036         case BOND_PRI_RESELECT_FAILURE:
1037                 return curr;
1038         default:
1039                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1040                            bond->params.primary_reselect);
1041                 return curr;
1042         }
1043 }
1044
1045 /**
1046  * bond_find_best_slave - select the best available slave to be the active one
1047  * @bond: our bonding struct
1048  */
1049 static struct slave *bond_find_best_slave(struct bonding *bond)
1050 {
1051         struct slave *slave, *bestslave = NULL;
1052         struct list_head *iter;
1053         int mintime = bond->params.updelay;
1054
1055         slave = bond_choose_primary_or_current(bond);
1056         if (slave)
1057                 return slave;
1058
1059         bond_for_each_slave(bond, slave, iter) {
1060                 if (slave->link == BOND_LINK_UP)
1061                         return slave;
1062                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1063                     slave->delay < mintime) {
1064                         mintime = slave->delay;
1065                         bestslave = slave;
1066                 }
1067         }
1068
1069         return bestslave;
1070 }
1071
1072 static bool bond_should_notify_peers(struct bonding *bond)
1073 {
1074         struct slave *slave;
1075
1076         rcu_read_lock();
1077         slave = rcu_dereference(bond->curr_active_slave);
1078         rcu_read_unlock();
1079
1080         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1081                    slave ? slave->dev->name : "NULL");
1082
1083         if (!slave || !bond->send_peer_notif ||
1084             bond->send_peer_notif %
1085             max(1, bond->params.peer_notif_delay) != 0 ||
1086             !netif_carrier_ok(bond->dev) ||
1087             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1088                 return false;
1089
1090         return true;
1091 }
1092
1093 /**
1094  * bond_change_active_slave - change the active slave into the specified one
1095  * @bond: our bonding struct
1096  * @new_active: the new slave to make the active one
1097  *
1098  * Set the new slave to the bond's settings and unset them on the old
1099  * curr_active_slave.
1100  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1101  *
1102  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1103  * because it is apparently the best available slave we have, even though its
1104  * updelay hasn't timed out yet.
1105  *
1106  * Caller must hold RTNL.
1107  */
1108 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1109 {
1110         struct slave *old_active;
1111
1112         ASSERT_RTNL();
1113
1114         old_active = rtnl_dereference(bond->curr_active_slave);
1115
1116         if (old_active == new_active)
1117                 return;
1118
1119 #ifdef CONFIG_XFRM_OFFLOAD
1120         bond_ipsec_del_sa_all(bond);
1121 #endif /* CONFIG_XFRM_OFFLOAD */
1122
1123         if (new_active) {
1124                 new_active->last_link_up = jiffies;
1125
1126                 if (new_active->link == BOND_LINK_BACK) {
1127                         if (bond_uses_primary(bond)) {
1128                                 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1129                                            (bond->params.updelay - new_active->delay) * bond->params.miimon);
1130                         }
1131
1132                         new_active->delay = 0;
1133                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
1134                                                   BOND_SLAVE_NOTIFY_NOW);
1135
1136                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1137                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1138
1139                         if (bond_is_lb(bond))
1140                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1141                 } else {
1142                         if (bond_uses_primary(bond))
1143                                 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1144                 }
1145         }
1146
1147         if (bond_uses_primary(bond))
1148                 bond_hw_addr_swap(bond, new_active, old_active);
1149
1150         if (bond_is_lb(bond)) {
1151                 bond_alb_handle_active_change(bond, new_active);
1152                 if (old_active)
1153                         bond_set_slave_inactive_flags(old_active,
1154                                                       BOND_SLAVE_NOTIFY_NOW);
1155                 if (new_active)
1156                         bond_set_slave_active_flags(new_active,
1157                                                     BOND_SLAVE_NOTIFY_NOW);
1158         } else {
1159                 rcu_assign_pointer(bond->curr_active_slave, new_active);
1160         }
1161
1162         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1163                 if (old_active)
1164                         bond_set_slave_inactive_flags(old_active,
1165                                                       BOND_SLAVE_NOTIFY_NOW);
1166
1167                 if (new_active) {
1168                         bool should_notify_peers = false;
1169
1170                         bond_set_slave_active_flags(new_active,
1171                                                     BOND_SLAVE_NOTIFY_NOW);
1172
1173                         if (bond->params.fail_over_mac)
1174                                 bond_do_fail_over_mac(bond, new_active,
1175                                                       old_active);
1176
1177                         if (netif_running(bond->dev)) {
1178                                 bond->send_peer_notif =
1179                                         bond->params.num_peer_notif *
1180                                         max(1, bond->params.peer_notif_delay);
1181                                 should_notify_peers =
1182                                         bond_should_notify_peers(bond);
1183                         }
1184
1185                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1186                         if (should_notify_peers) {
1187                                 bond->send_peer_notif--;
1188                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1189                                                          bond->dev);
1190                         }
1191                 }
1192         }
1193
1194 #ifdef CONFIG_XFRM_OFFLOAD
1195         bond_ipsec_add_sa_all(bond);
1196 #endif /* CONFIG_XFRM_OFFLOAD */
1197
1198         /* resend IGMP joins since active slave has changed or
1199          * all were sent on curr_active_slave.
1200          * resend only if bond is brought up with the affected
1201          * bonding modes and the retransmission is enabled
1202          */
1203         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1204             ((bond_uses_primary(bond) && new_active) ||
1205              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1206                 bond->igmp_retrans = bond->params.resend_igmp;
1207                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1208         }
1209 }
1210
1211 /**
1212  * bond_select_active_slave - select a new active slave, if needed
1213  * @bond: our bonding struct
1214  *
1215  * This functions should be called when one of the following occurs:
1216  * - The old curr_active_slave has been released or lost its link.
1217  * - The primary_slave has got its link back.
1218  * - A slave has got its link back and there's no old curr_active_slave.
1219  *
1220  * Caller must hold RTNL.
1221  */
1222 void bond_select_active_slave(struct bonding *bond)
1223 {
1224         struct slave *best_slave;
1225         int rv;
1226
1227         ASSERT_RTNL();
1228
1229         best_slave = bond_find_best_slave(bond);
1230         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1231                 bond_change_active_slave(bond, best_slave);
1232                 rv = bond_set_carrier(bond);
1233                 if (!rv)
1234                         return;
1235
1236                 if (netif_carrier_ok(bond->dev))
1237                         netdev_info(bond->dev, "active interface up!\n");
1238                 else
1239                         netdev_info(bond->dev, "now running without any active interface!\n");
1240         }
1241 }
1242
1243 #ifdef CONFIG_NET_POLL_CONTROLLER
1244 static inline int slave_enable_netpoll(struct slave *slave)
1245 {
1246         struct netpoll *np;
1247         int err = 0;
1248
1249         np = kzalloc(sizeof(*np), GFP_KERNEL);
1250         err = -ENOMEM;
1251         if (!np)
1252                 goto out;
1253
1254         err = __netpoll_setup(np, slave->dev);
1255         if (err) {
1256                 kfree(np);
1257                 goto out;
1258         }
1259         slave->np = np;
1260 out:
1261         return err;
1262 }
1263 static inline void slave_disable_netpoll(struct slave *slave)
1264 {
1265         struct netpoll *np = slave->np;
1266
1267         if (!np)
1268                 return;
1269
1270         slave->np = NULL;
1271
1272         __netpoll_free(np);
1273 }
1274
1275 static void bond_poll_controller(struct net_device *bond_dev)
1276 {
1277         struct bonding *bond = netdev_priv(bond_dev);
1278         struct slave *slave = NULL;
1279         struct list_head *iter;
1280         struct ad_info ad_info;
1281
1282         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1283                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1284                         return;
1285
1286         bond_for_each_slave_rcu(bond, slave, iter) {
1287                 if (!bond_slave_is_up(slave))
1288                         continue;
1289
1290                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1291                         struct aggregator *agg =
1292                             SLAVE_AD_INFO(slave)->port.aggregator;
1293
1294                         if (agg &&
1295                             agg->aggregator_identifier != ad_info.aggregator_id)
1296                                 continue;
1297                 }
1298
1299                 netpoll_poll_dev(slave->dev);
1300         }
1301 }
1302
1303 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1304 {
1305         struct bonding *bond = netdev_priv(bond_dev);
1306         struct list_head *iter;
1307         struct slave *slave;
1308
1309         bond_for_each_slave(bond, slave, iter)
1310                 if (bond_slave_is_up(slave))
1311                         slave_disable_netpoll(slave);
1312 }
1313
1314 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1315 {
1316         struct bonding *bond = netdev_priv(dev);
1317         struct list_head *iter;
1318         struct slave *slave;
1319         int err = 0;
1320
1321         bond_for_each_slave(bond, slave, iter) {
1322                 err = slave_enable_netpoll(slave);
1323                 if (err) {
1324                         bond_netpoll_cleanup(dev);
1325                         break;
1326                 }
1327         }
1328         return err;
1329 }
1330 #else
1331 static inline int slave_enable_netpoll(struct slave *slave)
1332 {
1333         return 0;
1334 }
1335 static inline void slave_disable_netpoll(struct slave *slave)
1336 {
1337 }
1338 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1339 {
1340 }
1341 #endif
1342
1343 /*---------------------------------- IOCTL ----------------------------------*/
1344
1345 static netdev_features_t bond_fix_features(struct net_device *dev,
1346                                            netdev_features_t features)
1347 {
1348         struct bonding *bond = netdev_priv(dev);
1349         struct list_head *iter;
1350         netdev_features_t mask;
1351         struct slave *slave;
1352
1353 #if IS_ENABLED(CONFIG_TLS_DEVICE)
1354         if (bond_sk_check(bond))
1355                 features |= BOND_TLS_FEATURES;
1356         else
1357                 features &= ~BOND_TLS_FEATURES;
1358 #endif
1359
1360         mask = features;
1361
1362         features &= ~NETIF_F_ONE_FOR_ALL;
1363         features |= NETIF_F_ALL_FOR_ALL;
1364
1365         bond_for_each_slave(bond, slave, iter) {
1366                 features = netdev_increment_features(features,
1367                                                      slave->dev->features,
1368                                                      mask);
1369         }
1370         features = netdev_add_tso_features(features, mask);
1371
1372         return features;
1373 }
1374
1375 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1376                                  NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1377                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1378
1379 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1380                                  NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1381
1382 #define BOND_MPLS_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1383                                  NETIF_F_GSO_SOFTWARE)
1384
1385
1386 static void bond_compute_features(struct bonding *bond)
1387 {
1388         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1389                                         IFF_XMIT_DST_RELEASE_PERM;
1390         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1391         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1392 #ifdef CONFIG_XFRM_OFFLOAD
1393         netdev_features_t xfrm_features  = BOND_XFRM_FEATURES;
1394 #endif /* CONFIG_XFRM_OFFLOAD */
1395         netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
1396         struct net_device *bond_dev = bond->dev;
1397         struct list_head *iter;
1398         struct slave *slave;
1399         unsigned short max_hard_header_len = ETH_HLEN;
1400         unsigned int gso_max_size = GSO_MAX_SIZE;
1401         u16 gso_max_segs = GSO_MAX_SEGS;
1402
1403         if (!bond_has_slaves(bond))
1404                 goto done;
1405         vlan_features &= NETIF_F_ALL_FOR_ALL;
1406         mpls_features &= NETIF_F_ALL_FOR_ALL;
1407
1408         bond_for_each_slave(bond, slave, iter) {
1409                 vlan_features = netdev_increment_features(vlan_features,
1410                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1411
1412                 enc_features = netdev_increment_features(enc_features,
1413                                                          slave->dev->hw_enc_features,
1414                                                          BOND_ENC_FEATURES);
1415
1416 #ifdef CONFIG_XFRM_OFFLOAD
1417                 xfrm_features = netdev_increment_features(xfrm_features,
1418                                                           slave->dev->hw_enc_features,
1419                                                           BOND_XFRM_FEATURES);
1420 #endif /* CONFIG_XFRM_OFFLOAD */
1421
1422                 mpls_features = netdev_increment_features(mpls_features,
1423                                                           slave->dev->mpls_features,
1424                                                           BOND_MPLS_FEATURES);
1425
1426                 dst_release_flag &= slave->dev->priv_flags;
1427                 if (slave->dev->hard_header_len > max_hard_header_len)
1428                         max_hard_header_len = slave->dev->hard_header_len;
1429
1430                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1431                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1432         }
1433         bond_dev->hard_header_len = max_hard_header_len;
1434
1435 done:
1436         bond_dev->vlan_features = vlan_features;
1437         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1438                                     NETIF_F_HW_VLAN_CTAG_TX |
1439                                     NETIF_F_HW_VLAN_STAG_TX;
1440 #ifdef CONFIG_XFRM_OFFLOAD
1441         bond_dev->hw_enc_features |= xfrm_features;
1442 #endif /* CONFIG_XFRM_OFFLOAD */
1443         bond_dev->mpls_features = mpls_features;
1444         bond_dev->gso_max_segs = gso_max_segs;
1445         netif_set_gso_max_size(bond_dev, gso_max_size);
1446
1447         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1448         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1449             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1450                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1451
1452         netdev_change_features(bond_dev);
1453 }
1454
1455 static void bond_setup_by_slave(struct net_device *bond_dev,
1456                                 struct net_device *slave_dev)
1457 {
1458         bond_dev->header_ops        = slave_dev->header_ops;
1459
1460         bond_dev->type              = slave_dev->type;
1461         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1462         bond_dev->needed_headroom   = slave_dev->needed_headroom;
1463         bond_dev->addr_len          = slave_dev->addr_len;
1464
1465         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1466                 slave_dev->addr_len);
1467 }
1468
1469 /* On bonding slaves other than the currently active slave, suppress
1470  * duplicates except for alb non-mcast/bcast.
1471  */
1472 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1473                                             struct slave *slave,
1474                                             struct bonding *bond)
1475 {
1476         if (bond_is_slave_inactive(slave)) {
1477                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1478                     skb->pkt_type != PACKET_BROADCAST &&
1479                     skb->pkt_type != PACKET_MULTICAST)
1480                         return false;
1481                 return true;
1482         }
1483         return false;
1484 }
1485
1486 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1487 {
1488         struct sk_buff *skb = *pskb;
1489         struct slave *slave;
1490         struct bonding *bond;
1491         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1492                           struct slave *);
1493         int ret = RX_HANDLER_ANOTHER;
1494
1495         skb = skb_share_check(skb, GFP_ATOMIC);
1496         if (unlikely(!skb))
1497                 return RX_HANDLER_CONSUMED;
1498
1499         *pskb = skb;
1500
1501         slave = bond_slave_get_rcu(skb->dev);
1502         bond = slave->bond;
1503
1504         recv_probe = READ_ONCE(bond->recv_probe);
1505         if (recv_probe) {
1506                 ret = recv_probe(skb, bond, slave);
1507                 if (ret == RX_HANDLER_CONSUMED) {
1508                         consume_skb(skb);
1509                         return ret;
1510                 }
1511         }
1512
1513         /*
1514          * For packets determined by bond_should_deliver_exact_match() call to
1515          * be suppressed we want to make an exception for link-local packets.
1516          * This is necessary for e.g. LLDP daemons to be able to monitor
1517          * inactive slave links without being forced to bind to them
1518          * explicitly.
1519          *
1520          * At the same time, packets that are passed to the bonding master
1521          * (including link-local ones) can have their originating interface
1522          * determined via PACKET_ORIGDEV socket option.
1523          */
1524         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1525                 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1526                         return RX_HANDLER_PASS;
1527                 return RX_HANDLER_EXACT;
1528         }
1529
1530         skb->dev = bond->dev;
1531
1532         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1533             netif_is_bridge_port(bond->dev) &&
1534             skb->pkt_type == PACKET_HOST) {
1535
1536                 if (unlikely(skb_cow_head(skb,
1537                                           skb->data - skb_mac_header(skb)))) {
1538                         kfree_skb(skb);
1539                         return RX_HANDLER_CONSUMED;
1540                 }
1541                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1542                                   bond->dev->addr_len);
1543         }
1544
1545         return ret;
1546 }
1547
1548 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1549 {
1550         switch (BOND_MODE(bond)) {
1551         case BOND_MODE_ROUNDROBIN:
1552                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1553         case BOND_MODE_ACTIVEBACKUP:
1554                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1555         case BOND_MODE_BROADCAST:
1556                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1557         case BOND_MODE_XOR:
1558         case BOND_MODE_8023AD:
1559                 return NETDEV_LAG_TX_TYPE_HASH;
1560         default:
1561                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1562         }
1563 }
1564
1565 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1566                                                enum netdev_lag_tx_type type)
1567 {
1568         if (type != NETDEV_LAG_TX_TYPE_HASH)
1569                 return NETDEV_LAG_HASH_NONE;
1570
1571         switch (bond->params.xmit_policy) {
1572         case BOND_XMIT_POLICY_LAYER2:
1573                 return NETDEV_LAG_HASH_L2;
1574         case BOND_XMIT_POLICY_LAYER34:
1575                 return NETDEV_LAG_HASH_L34;
1576         case BOND_XMIT_POLICY_LAYER23:
1577                 return NETDEV_LAG_HASH_L23;
1578         case BOND_XMIT_POLICY_ENCAP23:
1579                 return NETDEV_LAG_HASH_E23;
1580         case BOND_XMIT_POLICY_ENCAP34:
1581                 return NETDEV_LAG_HASH_E34;
1582         case BOND_XMIT_POLICY_VLAN_SRCMAC:
1583                 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1584         default:
1585                 return NETDEV_LAG_HASH_UNKNOWN;
1586         }
1587 }
1588
1589 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1590                                       struct netlink_ext_ack *extack)
1591 {
1592         struct netdev_lag_upper_info lag_upper_info;
1593         enum netdev_lag_tx_type type;
1594
1595         type = bond_lag_tx_type(bond);
1596         lag_upper_info.tx_type = type;
1597         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1598
1599         return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1600                                             &lag_upper_info, extack);
1601 }
1602
1603 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1604 {
1605         netdev_upper_dev_unlink(slave->dev, bond->dev);
1606         slave->dev->flags &= ~IFF_SLAVE;
1607 }
1608
1609 static void slave_kobj_release(struct kobject *kobj)
1610 {
1611         struct slave *slave = to_slave(kobj);
1612         struct bonding *bond = bond_get_bond_by_slave(slave);
1613
1614         cancel_delayed_work_sync(&slave->notify_work);
1615         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1616                 kfree(SLAVE_AD_INFO(slave));
1617
1618         kfree(slave);
1619 }
1620
1621 static struct kobj_type slave_ktype = {
1622         .release = slave_kobj_release,
1623 #ifdef CONFIG_SYSFS
1624         .sysfs_ops = &slave_sysfs_ops,
1625 #endif
1626 };
1627
1628 static int bond_kobj_init(struct slave *slave)
1629 {
1630         int err;
1631
1632         err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1633                                    &(slave->dev->dev.kobj), "bonding_slave");
1634         if (err)
1635                 kobject_put(&slave->kobj);
1636
1637         return err;
1638 }
1639
1640 static struct slave *bond_alloc_slave(struct bonding *bond,
1641                                       struct net_device *slave_dev)
1642 {
1643         struct slave *slave = NULL;
1644
1645         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1646         if (!slave)
1647                 return NULL;
1648
1649         slave->bond = bond;
1650         slave->dev = slave_dev;
1651         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1652
1653         if (bond_kobj_init(slave))
1654                 return NULL;
1655
1656         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1657                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1658                                                GFP_KERNEL);
1659                 if (!SLAVE_AD_INFO(slave)) {
1660                         kobject_put(&slave->kobj);
1661                         return NULL;
1662                 }
1663         }
1664
1665         return slave;
1666 }
1667
1668 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1669 {
1670         info->bond_mode = BOND_MODE(bond);
1671         info->miimon = bond->params.miimon;
1672         info->num_slaves = bond->slave_cnt;
1673 }
1674
1675 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1676 {
1677         strcpy(info->slave_name, slave->dev->name);
1678         info->link = slave->link;
1679         info->state = bond_slave_state(slave);
1680         info->link_failure_count = slave->link_failure_count;
1681 }
1682
1683 static void bond_netdev_notify_work(struct work_struct *_work)
1684 {
1685         struct slave *slave = container_of(_work, struct slave,
1686                                            notify_work.work);
1687
1688         if (rtnl_trylock()) {
1689                 struct netdev_bonding_info binfo;
1690
1691                 bond_fill_ifslave(slave, &binfo.slave);
1692                 bond_fill_ifbond(slave->bond, &binfo.master);
1693                 netdev_bonding_info_change(slave->dev, &binfo);
1694                 rtnl_unlock();
1695         } else {
1696                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1697         }
1698 }
1699
1700 void bond_queue_slave_event(struct slave *slave)
1701 {
1702         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1703 }
1704
1705 void bond_lower_state_changed(struct slave *slave)
1706 {
1707         struct netdev_lag_lower_state_info info;
1708
1709         info.link_up = slave->link == BOND_LINK_UP ||
1710                        slave->link == BOND_LINK_FAIL;
1711         info.tx_enabled = bond_is_active_slave(slave);
1712         netdev_lower_state_changed(slave->dev, &info);
1713 }
1714
1715 /* enslave device <slave> to bond device <master> */
1716 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1717                  struct netlink_ext_ack *extack)
1718 {
1719         struct bonding *bond = netdev_priv(bond_dev);
1720         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1721         struct slave *new_slave = NULL, *prev_slave;
1722         struct sockaddr_storage ss;
1723         int link_reporting;
1724         int res = 0, i;
1725
1726         if (slave_dev->flags & IFF_MASTER &&
1727             !netif_is_bond_master(slave_dev)) {
1728                 NL_SET_ERR_MSG(extack, "Device with IFF_MASTER cannot be enslaved");
1729                 netdev_err(bond_dev,
1730                            "Error: Device with IFF_MASTER cannot be enslaved\n");
1731                 return -EPERM;
1732         }
1733
1734         if (!bond->params.use_carrier &&
1735             slave_dev->ethtool_ops->get_link == NULL &&
1736             slave_ops->ndo_do_ioctl == NULL) {
1737                 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1738         }
1739
1740         /* already in-use? */
1741         if (netdev_is_rx_handler_busy(slave_dev)) {
1742                 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1743                 slave_err(bond_dev, slave_dev,
1744                           "Error: Device is in use and cannot be enslaved\n");
1745                 return -EBUSY;
1746         }
1747
1748         if (bond_dev == slave_dev) {
1749                 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1750                 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1751                 return -EPERM;
1752         }
1753
1754         /* vlan challenged mutual exclusion */
1755         /* no need to lock since we're protected by rtnl_lock */
1756         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1757                 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1758                 if (vlan_uses_dev(bond_dev)) {
1759                         NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1760                         slave_err(bond_dev, slave_dev, "Error: cannot enslave VLAN challenged slave on VLAN enabled bond\n");
1761                         return -EPERM;
1762                 } else {
1763                         slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1764                 }
1765         } else {
1766                 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1767         }
1768
1769         if (slave_dev->features & NETIF_F_HW_ESP)
1770                 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1771
1772         /* Old ifenslave binaries are no longer supported.  These can
1773          * be identified with moderate accuracy by the state of the slave:
1774          * the current ifenslave will set the interface down prior to
1775          * enslaving it; the old ifenslave will not.
1776          */
1777         if (slave_dev->flags & IFF_UP) {
1778                 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1779                 slave_err(bond_dev, slave_dev, "slave is up - this may be due to an out of date ifenslave\n");
1780                 return -EPERM;
1781         }
1782
1783         /* set bonding device ether type by slave - bonding netdevices are
1784          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1785          * there is a need to override some of the type dependent attribs/funcs.
1786          *
1787          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1788          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1789          */
1790         if (!bond_has_slaves(bond)) {
1791                 if (bond_dev->type != slave_dev->type) {
1792                         slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1793                                   bond_dev->type, slave_dev->type);
1794
1795                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1796                                                        bond_dev);
1797                         res = notifier_to_errno(res);
1798                         if (res) {
1799                                 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1800                                 return -EBUSY;
1801                         }
1802
1803                         /* Flush unicast and multicast addresses */
1804                         dev_uc_flush(bond_dev);
1805                         dev_mc_flush(bond_dev);
1806
1807                         if (slave_dev->type != ARPHRD_ETHER)
1808                                 bond_setup_by_slave(bond_dev, slave_dev);
1809                         else {
1810                                 ether_setup(bond_dev);
1811                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1812                         }
1813
1814                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1815                                                  bond_dev);
1816                 }
1817         } else if (bond_dev->type != slave_dev->type) {
1818                 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1819                 slave_err(bond_dev, slave_dev, "ether type (%d) is different from other slaves (%d), can not enslave it\n",
1820                           slave_dev->type, bond_dev->type);
1821                 return -EINVAL;
1822         }
1823
1824         if (slave_dev->type == ARPHRD_INFINIBAND &&
1825             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1826                 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1827                 slave_warn(bond_dev, slave_dev, "Type (%d) supports only active-backup mode\n",
1828                            slave_dev->type);
1829                 res = -EOPNOTSUPP;
1830                 goto err_undo_flags;
1831         }
1832
1833         if (!slave_ops->ndo_set_mac_address ||
1834             slave_dev->type == ARPHRD_INFINIBAND) {
1835                 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1836                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1837                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1838                         if (!bond_has_slaves(bond)) {
1839                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1840                                 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1841                         } else {
1842                                 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1843                                 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");
1844                                 res = -EOPNOTSUPP;
1845                                 goto err_undo_flags;
1846                         }
1847                 }
1848         }
1849
1850         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1851
1852         /* If this is the first slave, then we need to set the master's hardware
1853          * address to be the same as the slave's.
1854          */
1855         if (!bond_has_slaves(bond) &&
1856             bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1857                 res = bond_set_dev_addr(bond->dev, slave_dev);
1858                 if (res)
1859                         goto err_undo_flags;
1860         }
1861
1862         new_slave = bond_alloc_slave(bond, slave_dev);
1863         if (!new_slave) {
1864                 res = -ENOMEM;
1865                 goto err_undo_flags;
1866         }
1867
1868         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1869          * is set via sysfs or module option if desired.
1870          */
1871         new_slave->queue_id = 0;
1872
1873         /* Save slave's original mtu and then set it to match the bond */
1874         new_slave->original_mtu = slave_dev->mtu;
1875         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1876         if (res) {
1877                 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1878                 goto err_free;
1879         }
1880
1881         /* Save slave's original ("permanent") mac address for modes
1882          * that need it, and for restoring it upon release, and then
1883          * set it to the master's address
1884          */
1885         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1886                           slave_dev->addr_len);
1887
1888         if (!bond->params.fail_over_mac ||
1889             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1890                 /* Set slave to master's mac address.  The application already
1891                  * set the master's mac address to that of the first slave
1892                  */
1893                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1894                 ss.ss_family = slave_dev->type;
1895                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1896                                           extack);
1897                 if (res) {
1898                         slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1899                         goto err_restore_mtu;
1900                 }
1901         }
1902
1903         /* set slave flag before open to prevent IPv6 addrconf */
1904         slave_dev->flags |= IFF_SLAVE;
1905
1906         /* open the slave since the application closed it */
1907         res = dev_open(slave_dev, extack);
1908         if (res) {
1909                 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1910                 goto err_restore_mac;
1911         }
1912
1913         slave_dev->priv_flags |= IFF_BONDING;
1914         /* initialize slave stats */
1915         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1916
1917         if (bond_is_lb(bond)) {
1918                 /* bond_alb_init_slave() must be called before all other stages since
1919                  * it might fail and we do not want to have to undo everything
1920                  */
1921                 res = bond_alb_init_slave(bond, new_slave);
1922                 if (res)
1923                         goto err_close;
1924         }
1925
1926         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1927         if (res) {
1928                 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1929                 goto err_close;
1930         }
1931
1932         prev_slave = bond_last_slave(bond);
1933
1934         new_slave->delay = 0;
1935         new_slave->link_failure_count = 0;
1936
1937         if (bond_update_speed_duplex(new_slave) &&
1938             bond_needs_speed_duplex(bond))
1939                 new_slave->link = BOND_LINK_DOWN;
1940
1941         new_slave->last_rx = jiffies -
1942                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1943         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1944                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1945
1946         if (bond->params.miimon && !bond->params.use_carrier) {
1947                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1948
1949                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1950                         /* miimon is set but a bonded network driver
1951                          * does not support ETHTOOL/MII and
1952                          * arp_interval is not set.  Note: if
1953                          * use_carrier is enabled, we will never go
1954                          * here (because netif_carrier is always
1955                          * supported); thus, we don't need to change
1956                          * the messages for netif_carrier.
1957                          */
1958                         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");
1959                 } else if (link_reporting == -1) {
1960                         /* unable get link status using mii/ethtool */
1961                         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");
1962                 }
1963         }
1964
1965         /* check for initial state */
1966         new_slave->link = BOND_LINK_NOCHANGE;
1967         if (bond->params.miimon) {
1968                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1969                         if (bond->params.updelay) {
1970                                 bond_set_slave_link_state(new_slave,
1971                                                           BOND_LINK_BACK,
1972                                                           BOND_SLAVE_NOTIFY_NOW);
1973                                 new_slave->delay = bond->params.updelay;
1974                         } else {
1975                                 bond_set_slave_link_state(new_slave,
1976                                                           BOND_LINK_UP,
1977                                                           BOND_SLAVE_NOTIFY_NOW);
1978                         }
1979                 } else {
1980                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1981                                                   BOND_SLAVE_NOTIFY_NOW);
1982                 }
1983         } else if (bond->params.arp_interval) {
1984                 bond_set_slave_link_state(new_slave,
1985                                           (netif_carrier_ok(slave_dev) ?
1986                                           BOND_LINK_UP : BOND_LINK_DOWN),
1987                                           BOND_SLAVE_NOTIFY_NOW);
1988         } else {
1989                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1990                                           BOND_SLAVE_NOTIFY_NOW);
1991         }
1992
1993         if (new_slave->link != BOND_LINK_DOWN)
1994                 new_slave->last_link_up = jiffies;
1995         slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
1996                   new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1997                   (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1998
1999         if (bond_uses_primary(bond) && bond->params.primary[0]) {
2000                 /* if there is a primary slave, remember it */
2001                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2002                         rcu_assign_pointer(bond->primary_slave, new_slave);
2003                         bond->force_primary = true;
2004                 }
2005         }
2006
2007         switch (BOND_MODE(bond)) {
2008         case BOND_MODE_ACTIVEBACKUP:
2009                 bond_set_slave_inactive_flags(new_slave,
2010                                               BOND_SLAVE_NOTIFY_NOW);
2011                 break;
2012         case BOND_MODE_8023AD:
2013                 /* in 802.3ad mode, the internal mechanism
2014                  * will activate the slaves in the selected
2015                  * aggregator
2016                  */
2017                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2018                 /* if this is the first slave */
2019                 if (!prev_slave) {
2020                         SLAVE_AD_INFO(new_slave)->id = 1;
2021                         /* Initialize AD with the number of times that the AD timer is called in 1 second
2022                          * can be called only after the mac address of the bond is set
2023                          */
2024                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
2025                 } else {
2026                         SLAVE_AD_INFO(new_slave)->id =
2027                                 SLAVE_AD_INFO(prev_slave)->id + 1;
2028                 }
2029
2030                 bond_3ad_bind_slave(new_slave);
2031                 break;
2032         case BOND_MODE_TLB:
2033         case BOND_MODE_ALB:
2034                 bond_set_active_slave(new_slave);
2035                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2036                 break;
2037         default:
2038                 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2039
2040                 /* always active in trunk mode */
2041                 bond_set_active_slave(new_slave);
2042
2043                 /* In trunking mode there is little meaning to curr_active_slave
2044                  * anyway (it holds no special properties of the bond device),
2045                  * so we can change it without calling change_active_interface()
2046                  */
2047                 if (!rcu_access_pointer(bond->curr_active_slave) &&
2048                     new_slave->link == BOND_LINK_UP)
2049                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
2050
2051                 break;
2052         } /* switch(bond_mode) */
2053
2054 #ifdef CONFIG_NET_POLL_CONTROLLER
2055         if (bond->dev->npinfo) {
2056                 if (slave_enable_netpoll(new_slave)) {
2057                         slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2058                         res = -EBUSY;
2059                         goto err_detach;
2060                 }
2061         }
2062 #endif
2063
2064         if (!(bond_dev->features & NETIF_F_LRO))
2065                 dev_disable_lro(slave_dev);
2066
2067         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2068                                          new_slave);
2069         if (res) {
2070                 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2071                 goto err_detach;
2072         }
2073
2074         res = bond_master_upper_dev_link(bond, new_slave, extack);
2075         if (res) {
2076                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2077                 goto err_unregister;
2078         }
2079
2080         bond_lower_state_changed(new_slave);
2081
2082         res = bond_sysfs_slave_add(new_slave);
2083         if (res) {
2084                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2085                 goto err_upper_unlink;
2086         }
2087
2088         /* If the mode uses primary, then the following is handled by
2089          * bond_change_active_slave().
2090          */
2091         if (!bond_uses_primary(bond)) {
2092                 /* set promiscuity level to new slave */
2093                 if (bond_dev->flags & IFF_PROMISC) {
2094                         res = dev_set_promiscuity(slave_dev, 1);
2095                         if (res)
2096                                 goto err_sysfs_del;
2097                 }
2098
2099                 /* set allmulti level to new slave */
2100                 if (bond_dev->flags & IFF_ALLMULTI) {
2101                         res = dev_set_allmulti(slave_dev, 1);
2102                         if (res) {
2103                                 if (bond_dev->flags & IFF_PROMISC)
2104                                         dev_set_promiscuity(slave_dev, -1);
2105                                 goto err_sysfs_del;
2106                         }
2107                 }
2108
2109                 netif_addr_lock_bh(bond_dev);
2110                 dev_mc_sync_multiple(slave_dev, bond_dev);
2111                 dev_uc_sync_multiple(slave_dev, bond_dev);
2112                 netif_addr_unlock_bh(bond_dev);
2113
2114                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2115                         /* add lacpdu mc addr to mc list */
2116                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
2117
2118                         dev_mc_add(slave_dev, lacpdu_multicast);
2119                 }
2120         }
2121
2122         bond->slave_cnt++;
2123         bond_compute_features(bond);
2124         bond_set_carrier(bond);
2125
2126         if (bond_uses_primary(bond)) {
2127                 block_netpoll_tx();
2128                 bond_select_active_slave(bond);
2129                 unblock_netpoll_tx();
2130         }
2131
2132         if (bond_mode_can_use_xmit_hash(bond))
2133                 bond_update_slave_arr(bond, NULL);
2134
2135
2136         slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2137                    bond_is_active_slave(new_slave) ? "an active" : "a backup",
2138                    new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2139
2140         /* enslave is successful */
2141         bond_queue_slave_event(new_slave);
2142         return 0;
2143
2144 /* Undo stages on error */
2145 err_sysfs_del:
2146         bond_sysfs_slave_del(new_slave);
2147
2148 err_upper_unlink:
2149         bond_upper_dev_unlink(bond, new_slave);
2150
2151 err_unregister:
2152         netdev_rx_handler_unregister(slave_dev);
2153
2154 err_detach:
2155         vlan_vids_del_by_dev(slave_dev, bond_dev);
2156         if (rcu_access_pointer(bond->primary_slave) == new_slave)
2157                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2158         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2159                 block_netpoll_tx();
2160                 bond_change_active_slave(bond, NULL);
2161                 bond_select_active_slave(bond);
2162                 unblock_netpoll_tx();
2163         }
2164         /* either primary_slave or curr_active_slave might've changed */
2165         synchronize_rcu();
2166         slave_disable_netpoll(new_slave);
2167
2168 err_close:
2169         if (!netif_is_bond_master(slave_dev))
2170                 slave_dev->priv_flags &= ~IFF_BONDING;
2171         dev_close(slave_dev);
2172
2173 err_restore_mac:
2174         slave_dev->flags &= ~IFF_SLAVE;
2175         if (!bond->params.fail_over_mac ||
2176             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2177                 /* XXX TODO - fom follow mode needs to change master's
2178                  * MAC if this slave's MAC is in use by the bond, or at
2179                  * least print a warning.
2180                  */
2181                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2182                                   new_slave->dev->addr_len);
2183                 ss.ss_family = slave_dev->type;
2184                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2185         }
2186
2187 err_restore_mtu:
2188         dev_set_mtu(slave_dev, new_slave->original_mtu);
2189
2190 err_free:
2191         kobject_put(&new_slave->kobj);
2192
2193 err_undo_flags:
2194         /* Enslave of first slave has failed and we need to fix master's mac */
2195         if (!bond_has_slaves(bond)) {
2196                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2197                                             slave_dev->dev_addr))
2198                         eth_hw_addr_random(bond_dev);
2199                 if (bond_dev->type != ARPHRD_ETHER) {
2200                         dev_close(bond_dev);
2201                         ether_setup(bond_dev);
2202                         bond_dev->flags |= IFF_MASTER;
2203                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2204                 }
2205         }
2206
2207         return res;
2208 }
2209
2210 /* Try to release the slave device <slave> from the bond device <master>
2211  * It is legal to access curr_active_slave without a lock because all the function
2212  * is RTNL-locked. If "all" is true it means that the function is being called
2213  * while destroying a bond interface and all slaves are being released.
2214  *
2215  * The rules for slave state should be:
2216  *   for Active/Backup:
2217  *     Active stays on all backups go down
2218  *   for Bonded connections:
2219  *     The first up interface should be left on and all others downed.
2220  */
2221 static int __bond_release_one(struct net_device *bond_dev,
2222                               struct net_device *slave_dev,
2223                               bool all, bool unregister)
2224 {
2225         struct bonding *bond = netdev_priv(bond_dev);
2226         struct slave *slave, *oldcurrent;
2227         struct sockaddr_storage ss;
2228         int old_flags = bond_dev->flags;
2229         netdev_features_t old_features = bond_dev->features;
2230
2231         /* slave is not a slave or master is not master of this slave */
2232         if (!(slave_dev->flags & IFF_SLAVE) ||
2233             !netdev_has_upper_dev(slave_dev, bond_dev)) {
2234                 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2235                 return -EINVAL;
2236         }
2237
2238         block_netpoll_tx();
2239
2240         slave = bond_get_slave_by_dev(bond, slave_dev);
2241         if (!slave) {
2242                 /* not a slave of this bond */
2243                 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2244                 unblock_netpoll_tx();
2245                 return -EINVAL;
2246         }
2247
2248         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2249
2250         bond_sysfs_slave_del(slave);
2251
2252         /* recompute stats just before removing the slave */
2253         bond_get_stats(bond->dev, &bond->bond_stats);
2254
2255         bond_upper_dev_unlink(bond, slave);
2256         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2257          * for this slave anymore.
2258          */
2259         netdev_rx_handler_unregister(slave_dev);
2260
2261         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2262                 bond_3ad_unbind_slave(slave);
2263
2264         if (bond_mode_can_use_xmit_hash(bond))
2265                 bond_update_slave_arr(bond, slave);
2266
2267         slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2268                     bond_is_active_slave(slave) ? "active" : "backup");
2269
2270         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2271
2272         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2273
2274         if (!all && (!bond->params.fail_over_mac ||
2275                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2276                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2277                     bond_has_slaves(bond))
2278                         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",
2279                                    slave->perm_hwaddr);
2280         }
2281
2282         if (rtnl_dereference(bond->primary_slave) == slave)
2283                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2284
2285         if (oldcurrent == slave)
2286                 bond_change_active_slave(bond, NULL);
2287
2288         if (bond_is_lb(bond)) {
2289                 /* Must be called only after the slave has been
2290                  * detached from the list and the curr_active_slave
2291                  * has been cleared (if our_slave == old_current),
2292                  * but before a new active slave is selected.
2293                  */
2294                 bond_alb_deinit_slave(bond, slave);
2295         }
2296
2297         if (all) {
2298                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2299         } else if (oldcurrent == slave) {
2300                 /* Note that we hold RTNL over this sequence, so there
2301                  * is no concern that another slave add/remove event
2302                  * will interfere.
2303                  */
2304                 bond_select_active_slave(bond);
2305         }
2306
2307         if (!bond_has_slaves(bond)) {
2308                 bond_set_carrier(bond);
2309                 eth_hw_addr_random(bond_dev);
2310         }
2311
2312         unblock_netpoll_tx();
2313         synchronize_rcu();
2314         bond->slave_cnt--;
2315
2316         if (!bond_has_slaves(bond)) {
2317                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2318                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2319         }
2320
2321         bond_compute_features(bond);
2322         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2323             (old_features & NETIF_F_VLAN_CHALLENGED))
2324                 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2325
2326         vlan_vids_del_by_dev(slave_dev, bond_dev);
2327
2328         /* If the mode uses primary, then this case was handled above by
2329          * bond_change_active_slave(..., NULL)
2330          */
2331         if (!bond_uses_primary(bond)) {
2332                 /* unset promiscuity level from slave
2333                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
2334                  * of the IFF_PROMISC flag in the bond_dev, but we need the
2335                  * value of that flag before that change, as that was the value
2336                  * when this slave was attached, so we cache at the start of the
2337                  * function and use it here. Same goes for ALLMULTI below
2338                  */
2339                 if (old_flags & IFF_PROMISC)
2340                         dev_set_promiscuity(slave_dev, -1);
2341
2342                 /* unset allmulti level from slave */
2343                 if (old_flags & IFF_ALLMULTI)
2344                         dev_set_allmulti(slave_dev, -1);
2345
2346                 bond_hw_addr_flush(bond_dev, slave_dev);
2347         }
2348
2349         slave_disable_netpoll(slave);
2350
2351         /* close slave before restoring its mac address */
2352         dev_close(slave_dev);
2353
2354         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2355             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2356                 /* restore original ("permanent") mac address */
2357                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2358                                   slave->dev->addr_len);
2359                 ss.ss_family = slave_dev->type;
2360                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2361         }
2362
2363         if (unregister)
2364                 __dev_set_mtu(slave_dev, slave->original_mtu);
2365         else
2366                 dev_set_mtu(slave_dev, slave->original_mtu);
2367
2368         if (!netif_is_bond_master(slave_dev))
2369                 slave_dev->priv_flags &= ~IFF_BONDING;
2370
2371         kobject_put(&slave->kobj);
2372
2373         return 0;
2374 }
2375
2376 /* A wrapper used because of ndo_del_link */
2377 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2378 {
2379         return __bond_release_one(bond_dev, slave_dev, false, false);
2380 }
2381
2382 /* First release a slave and then destroy the bond if no more slaves are left.
2383  * Must be under rtnl_lock when this function is called.
2384  */
2385 static int bond_release_and_destroy(struct net_device *bond_dev,
2386                                     struct net_device *slave_dev)
2387 {
2388         struct bonding *bond = netdev_priv(bond_dev);
2389         int ret;
2390
2391         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2392         if (ret == 0 && !bond_has_slaves(bond) &&
2393             bond_dev->reg_state != NETREG_UNREGISTERING) {
2394                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2395                 netdev_info(bond_dev, "Destroying bond\n");
2396                 bond_remove_proc_entry(bond);
2397                 unregister_netdevice(bond_dev);
2398         }
2399         return ret;
2400 }
2401
2402 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2403 {
2404         struct bonding *bond = netdev_priv(bond_dev);
2405
2406         bond_fill_ifbond(bond, info);
2407 }
2408
2409 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2410 {
2411         struct bonding *bond = netdev_priv(bond_dev);
2412         struct list_head *iter;
2413         int i = 0, res = -ENODEV;
2414         struct slave *slave;
2415
2416         bond_for_each_slave(bond, slave, iter) {
2417                 if (i++ == (int)info->slave_id) {
2418                         res = 0;
2419                         bond_fill_ifslave(slave, info);
2420                         break;
2421                 }
2422         }
2423
2424         return res;
2425 }
2426
2427 /*-------------------------------- Monitoring -------------------------------*/
2428
2429 /* called with rcu_read_lock() */
2430 static int bond_miimon_inspect(struct bonding *bond)
2431 {
2432         int link_state, commit = 0;
2433         struct list_head *iter;
2434         struct slave *slave;
2435         bool ignore_updelay;
2436
2437         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2438
2439         bond_for_each_slave_rcu(bond, slave, iter) {
2440                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2441
2442                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2443
2444                 switch (slave->link) {
2445                 case BOND_LINK_UP:
2446                         if (link_state)
2447                                 continue;
2448
2449                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2450                         commit++;
2451                         slave->delay = bond->params.downdelay;
2452                         if (slave->delay) {
2453                                 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2454                                            (BOND_MODE(bond) ==
2455                                             BOND_MODE_ACTIVEBACKUP) ?
2456                                             (bond_is_active_slave(slave) ?
2457                                              "active " : "backup ") : "",
2458                                            bond->params.downdelay * bond->params.miimon);
2459                         }
2460                         fallthrough;
2461                 case BOND_LINK_FAIL:
2462                         if (link_state) {
2463                                 /* recovered before downdelay expired */
2464                                 bond_propose_link_state(slave, BOND_LINK_UP);
2465                                 slave->last_link_up = jiffies;
2466                                 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2467                                            (bond->params.downdelay - slave->delay) *
2468                                            bond->params.miimon);
2469                                 commit++;
2470                                 continue;
2471                         }
2472
2473                         if (slave->delay <= 0) {
2474                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2475                                 commit++;
2476                                 continue;
2477                         }
2478
2479                         slave->delay--;
2480                         break;
2481
2482                 case BOND_LINK_DOWN:
2483                         if (!link_state)
2484                                 continue;
2485
2486                         bond_propose_link_state(slave, BOND_LINK_BACK);
2487                         commit++;
2488                         slave->delay = bond->params.updelay;
2489
2490                         if (slave->delay) {
2491                                 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2492                                            ignore_updelay ? 0 :
2493                                            bond->params.updelay *
2494                                            bond->params.miimon);
2495                         }
2496                         fallthrough;
2497                 case BOND_LINK_BACK:
2498                         if (!link_state) {
2499                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2500                                 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2501                                            (bond->params.updelay - slave->delay) *
2502                                            bond->params.miimon);
2503                                 commit++;
2504                                 continue;
2505                         }
2506
2507                         if (ignore_updelay)
2508                                 slave->delay = 0;
2509
2510                         if (slave->delay <= 0) {
2511                                 bond_propose_link_state(slave, BOND_LINK_UP);
2512                                 commit++;
2513                                 ignore_updelay = false;
2514                                 continue;
2515                         }
2516
2517                         slave->delay--;
2518                         break;
2519                 }
2520         }
2521
2522         return commit;
2523 }
2524
2525 static void bond_miimon_link_change(struct bonding *bond,
2526                                     struct slave *slave,
2527                                     char link)
2528 {
2529         switch (BOND_MODE(bond)) {
2530         case BOND_MODE_8023AD:
2531                 bond_3ad_handle_link_change(slave, link);
2532                 break;
2533         case BOND_MODE_TLB:
2534         case BOND_MODE_ALB:
2535                 bond_alb_handle_link_change(bond, slave, link);
2536                 break;
2537         case BOND_MODE_XOR:
2538                 bond_update_slave_arr(bond, NULL);
2539                 break;
2540         }
2541 }
2542
2543 static void bond_miimon_commit(struct bonding *bond)
2544 {
2545         struct list_head *iter;
2546         struct slave *slave, *primary;
2547
2548         bond_for_each_slave(bond, slave, iter) {
2549                 switch (slave->link_new_state) {
2550                 case BOND_LINK_NOCHANGE:
2551                         /* For 802.3ad mode, check current slave speed and
2552                          * duplex again in case its port was disabled after
2553                          * invalid speed/duplex reporting but recovered before
2554                          * link monitoring could make a decision on the actual
2555                          * link status
2556                          */
2557                         if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2558                             slave->link == BOND_LINK_UP)
2559                                 bond_3ad_adapter_speed_duplex_changed(slave);
2560                         continue;
2561
2562                 case BOND_LINK_UP:
2563                         if (bond_update_speed_duplex(slave) &&
2564                             bond_needs_speed_duplex(bond)) {
2565                                 slave->link = BOND_LINK_DOWN;
2566                                 if (net_ratelimit())
2567                                         slave_warn(bond->dev, slave->dev,
2568                                                    "failed to get link speed/duplex\n");
2569                                 continue;
2570                         }
2571                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2572                                                   BOND_SLAVE_NOTIFY_NOW);
2573                         slave->last_link_up = jiffies;
2574
2575                         primary = rtnl_dereference(bond->primary_slave);
2576                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2577                                 /* prevent it from being the active one */
2578                                 bond_set_backup_slave(slave);
2579                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2580                                 /* make it immediately active */
2581                                 bond_set_active_slave(slave);
2582                         }
2583
2584                         slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2585                                    slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2586                                    slave->duplex ? "full" : "half");
2587
2588                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2589
2590                         if (!bond->curr_active_slave || slave == primary)
2591                                 goto do_failover;
2592
2593                         continue;
2594
2595                 case BOND_LINK_DOWN:
2596                         if (slave->link_failure_count < UINT_MAX)
2597                                 slave->link_failure_count++;
2598
2599                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2600                                                   BOND_SLAVE_NOTIFY_NOW);
2601
2602                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2603                             BOND_MODE(bond) == BOND_MODE_8023AD)
2604                                 bond_set_slave_inactive_flags(slave,
2605                                                               BOND_SLAVE_NOTIFY_NOW);
2606
2607                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2608
2609                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2610
2611                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2612                                 goto do_failover;
2613
2614                         continue;
2615
2616                 default:
2617                         slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2618                                   slave->link_new_state);
2619                         bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2620
2621                         continue;
2622                 }
2623
2624 do_failover:
2625                 block_netpoll_tx();
2626                 bond_select_active_slave(bond);
2627                 unblock_netpoll_tx();
2628         }
2629
2630         bond_set_carrier(bond);
2631 }
2632
2633 /* bond_mii_monitor
2634  *
2635  * Really a wrapper that splits the mii monitor into two phases: an
2636  * inspection, then (if inspection indicates something needs to be done)
2637  * an acquisition of appropriate locks followed by a commit phase to
2638  * implement whatever link state changes are indicated.
2639  */
2640 static void bond_mii_monitor(struct work_struct *work)
2641 {
2642         struct bonding *bond = container_of(work, struct bonding,
2643                                             mii_work.work);
2644         bool should_notify_peers = false;
2645         bool commit;
2646         unsigned long delay;
2647         struct slave *slave;
2648         struct list_head *iter;
2649
2650         delay = msecs_to_jiffies(bond->params.miimon);
2651
2652         if (!bond_has_slaves(bond))
2653                 goto re_arm;
2654
2655         rcu_read_lock();
2656         should_notify_peers = bond_should_notify_peers(bond);
2657         commit = !!bond_miimon_inspect(bond);
2658         if (bond->send_peer_notif) {
2659                 rcu_read_unlock();
2660                 if (rtnl_trylock()) {
2661                         bond->send_peer_notif--;
2662                         rtnl_unlock();
2663                 }
2664         } else {
2665                 rcu_read_unlock();
2666         }
2667
2668         if (commit) {
2669                 /* Race avoidance with bond_close cancel of workqueue */
2670                 if (!rtnl_trylock()) {
2671                         delay = 1;
2672                         should_notify_peers = false;
2673                         goto re_arm;
2674                 }
2675
2676                 bond_for_each_slave(bond, slave, iter) {
2677                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2678                 }
2679                 bond_miimon_commit(bond);
2680
2681                 rtnl_unlock();  /* might sleep, hold no other locks */
2682         }
2683
2684 re_arm:
2685         if (bond->params.miimon)
2686                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2687
2688         if (should_notify_peers) {
2689                 if (!rtnl_trylock())
2690                         return;
2691                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2692                 rtnl_unlock();
2693         }
2694 }
2695
2696 static int bond_upper_dev_walk(struct net_device *upper,
2697                                struct netdev_nested_priv *priv)
2698 {
2699         __be32 ip = *(__be32 *)priv->data;
2700
2701         return ip == bond_confirm_addr(upper, 0, ip);
2702 }
2703
2704 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2705 {
2706         struct netdev_nested_priv priv = {
2707                 .data = (void *)&ip,
2708         };
2709         bool ret = false;
2710
2711         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2712                 return true;
2713
2714         rcu_read_lock();
2715         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2716                 ret = true;
2717         rcu_read_unlock();
2718
2719         return ret;
2720 }
2721
2722 /* We go to the (large) trouble of VLAN tagging ARP frames because
2723  * switches in VLAN mode (especially if ports are configured as
2724  * "native" to a VLAN) might not pass non-tagged frames.
2725  */
2726 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2727                           __be32 src_ip, struct bond_vlan_tag *tags)
2728 {
2729         struct sk_buff *skb;
2730         struct bond_vlan_tag *outer_tag = tags;
2731         struct net_device *slave_dev = slave->dev;
2732         struct net_device *bond_dev = slave->bond->dev;
2733
2734         slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2735                   arp_op, &dest_ip, &src_ip);
2736
2737         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2738                          NULL, slave_dev->dev_addr, NULL);
2739
2740         if (!skb) {
2741                 net_err_ratelimited("ARP packet allocation failed\n");
2742                 return;
2743         }
2744
2745         if (!tags || tags->vlan_proto == VLAN_N_VID)
2746                 goto xmit;
2747
2748         tags++;
2749
2750         /* Go through all the tags backwards and add them to the packet */
2751         while (tags->vlan_proto != VLAN_N_VID) {
2752                 if (!tags->vlan_id) {
2753                         tags++;
2754                         continue;
2755                 }
2756
2757                 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2758                           ntohs(outer_tag->vlan_proto), tags->vlan_id);
2759                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2760                                                 tags->vlan_id);
2761                 if (!skb) {
2762                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2763                         return;
2764                 }
2765
2766                 tags++;
2767         }
2768         /* Set the outer tag */
2769         if (outer_tag->vlan_id) {
2770                 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2771                           ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2772                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2773                                        outer_tag->vlan_id);
2774         }
2775
2776 xmit:
2777         arp_xmit(skb);
2778 }
2779
2780 /* Validate the device path between the @start_dev and the @end_dev.
2781  * The path is valid if the @end_dev is reachable through device
2782  * stacking.
2783  * When the path is validated, collect any vlan information in the
2784  * path.
2785  */
2786 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2787                                               struct net_device *end_dev,
2788                                               int level)
2789 {
2790         struct bond_vlan_tag *tags;
2791         struct net_device *upper;
2792         struct list_head  *iter;
2793
2794         if (start_dev == end_dev) {
2795                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2796                 if (!tags)
2797                         return ERR_PTR(-ENOMEM);
2798                 tags[level].vlan_proto = VLAN_N_VID;
2799                 return tags;
2800         }
2801
2802         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2803                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2804                 if (IS_ERR_OR_NULL(tags)) {
2805                         if (IS_ERR(tags))
2806                                 return tags;
2807                         continue;
2808                 }
2809                 if (is_vlan_dev(upper)) {
2810                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2811                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2812                 }
2813
2814                 return tags;
2815         }
2816
2817         return NULL;
2818 }
2819
2820 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2821 {
2822         struct rtable *rt;
2823         struct bond_vlan_tag *tags;
2824         __be32 *targets = bond->params.arp_targets, addr;
2825         int i;
2826
2827         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2828                 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2829                           __func__, &targets[i]);
2830                 tags = NULL;
2831
2832                 /* Find out through which dev should the packet go */
2833                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2834                                      RTO_ONLINK, 0);
2835                 if (IS_ERR(rt)) {
2836                         /* there's no route to target - try to send arp
2837                          * probe to generate any traffic (arp_validate=0)
2838                          */
2839                         if (bond->params.arp_validate)
2840                                 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2841                                                      bond->dev->name,
2842                                                      &targets[i]);
2843                         bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2844                                       0, tags);
2845                         continue;
2846                 }
2847
2848                 /* bond device itself */
2849                 if (rt->dst.dev == bond->dev)
2850                         goto found;
2851
2852                 rcu_read_lock();
2853                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2854                 rcu_read_unlock();
2855
2856                 if (!IS_ERR_OR_NULL(tags))
2857                         goto found;
2858
2859                 /* Not our device - skip */
2860                 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2861                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2862
2863                 ip_rt_put(rt);
2864                 continue;
2865
2866 found:
2867                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2868                 ip_rt_put(rt);
2869                 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2870                 kfree(tags);
2871         }
2872 }
2873
2874 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2875 {
2876         int i;
2877
2878         if (!sip || !bond_has_this_ip(bond, tip)) {
2879                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2880                            __func__, &sip, &tip);
2881                 return;
2882         }
2883
2884         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2885         if (i == -1) {
2886                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2887                            __func__, &sip);
2888                 return;
2889         }
2890         slave->last_rx = jiffies;
2891         slave->target_last_arp_rx[i] = jiffies;
2892 }
2893
2894 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2895                  struct slave *slave)
2896 {
2897         struct arphdr *arp = (struct arphdr *)skb->data;
2898         struct slave *curr_active_slave, *curr_arp_slave;
2899         unsigned char *arp_ptr;
2900         __be32 sip, tip;
2901         int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2902         unsigned int alen;
2903
2904         if (!slave_do_arp_validate(bond, slave)) {
2905                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2906                     !slave_do_arp_validate_only(bond))
2907                         slave->last_rx = jiffies;
2908                 return RX_HANDLER_ANOTHER;
2909         } else if (!is_arp) {
2910                 return RX_HANDLER_ANOTHER;
2911         }
2912
2913         alen = arp_hdr_len(bond->dev);
2914
2915         slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2916                    __func__, skb->dev->name);
2917
2918         if (alen > skb_headlen(skb)) {
2919                 arp = kmalloc(alen, GFP_ATOMIC);
2920                 if (!arp)
2921                         goto out_unlock;
2922                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2923                         goto out_unlock;
2924         }
2925
2926         if (arp->ar_hln != bond->dev->addr_len ||
2927             skb->pkt_type == PACKET_OTHERHOST ||
2928             skb->pkt_type == PACKET_LOOPBACK ||
2929             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2930             arp->ar_pro != htons(ETH_P_IP) ||
2931             arp->ar_pln != 4)
2932                 goto out_unlock;
2933
2934         arp_ptr = (unsigned char *)(arp + 1);
2935         arp_ptr += bond->dev->addr_len;
2936         memcpy(&sip, arp_ptr, 4);
2937         arp_ptr += 4 + bond->dev->addr_len;
2938         memcpy(&tip, arp_ptr, 4);
2939
2940         slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2941                   __func__, slave->dev->name, bond_slave_state(slave),
2942                   bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2943                   &sip, &tip);
2944
2945         curr_active_slave = rcu_dereference(bond->curr_active_slave);
2946         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2947
2948         /* We 'trust' the received ARP enough to validate it if:
2949          *
2950          * (a) the slave receiving the ARP is active (which includes the
2951          * current ARP slave, if any), or
2952          *
2953          * (b) the receiving slave isn't active, but there is a currently
2954          * active slave and it received valid arp reply(s) after it became
2955          * the currently active slave, or
2956          *
2957          * (c) there is an ARP slave that sent an ARP during the prior ARP
2958          * interval, and we receive an ARP reply on any slave.  We accept
2959          * these because switch FDB update delays may deliver the ARP
2960          * reply to a slave other than the sender of the ARP request.
2961          *
2962          * Note: for (b), backup slaves are receiving the broadcast ARP
2963          * request, not a reply.  This request passes from the sending
2964          * slave through the L2 switch(es) to the receiving slave.  Since
2965          * this is checking the request, sip/tip are swapped for
2966          * validation.
2967          *
2968          * This is done to avoid endless looping when we can't reach the
2969          * arp_ip_target and fool ourselves with our own arp requests.
2970          */
2971         if (bond_is_active_slave(slave))
2972                 bond_validate_arp(bond, slave, sip, tip);
2973         else if (curr_active_slave &&
2974                  time_after(slave_last_rx(bond, curr_active_slave),
2975                             curr_active_slave->last_link_up))
2976                 bond_validate_arp(bond, slave, tip, sip);
2977         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2978                  bond_time_in_interval(bond,
2979                                        dev_trans_start(curr_arp_slave->dev), 1))
2980                 bond_validate_arp(bond, slave, sip, tip);
2981
2982 out_unlock:
2983         if (arp != (struct arphdr *)skb->data)
2984                 kfree(arp);
2985         return RX_HANDLER_ANOTHER;
2986 }
2987
2988 /* function to verify if we're in the arp_interval timeslice, returns true if
2989  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2990  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2991  */
2992 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2993                                   int mod)
2994 {
2995         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2996
2997         return time_in_range(jiffies,
2998                              last_act - delta_in_ticks,
2999                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
3000 }
3001
3002 /* This function is called regularly to monitor each slave's link
3003  * ensuring that traffic is being sent and received when arp monitoring
3004  * is used in load-balancing mode. if the adapter has been dormant, then an
3005  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3006  * arp monitoring in active backup mode.
3007  */
3008 static void bond_loadbalance_arp_mon(struct bonding *bond)
3009 {
3010         struct slave *slave, *oldcurrent;
3011         struct list_head *iter;
3012         int do_failover = 0, slave_state_changed = 0;
3013
3014         if (!bond_has_slaves(bond))
3015                 goto re_arm;
3016
3017         rcu_read_lock();
3018
3019         oldcurrent = rcu_dereference(bond->curr_active_slave);
3020         /* see if any of the previous devices are up now (i.e. they have
3021          * xmt and rcv traffic). the curr_active_slave does not come into
3022          * the picture unless it is null. also, slave->last_link_up is not
3023          * needed here because we send an arp on each slave and give a slave
3024          * as long as it needs to get the tx/rx within the delta.
3025          * TODO: what about up/down delay in arp mode? it wasn't here before
3026          *       so it can wait
3027          */
3028         bond_for_each_slave_rcu(bond, slave, iter) {
3029                 unsigned long trans_start = dev_trans_start(slave->dev);
3030
3031                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3032
3033                 if (slave->link != BOND_LINK_UP) {
3034                         if (bond_time_in_interval(bond, trans_start, 1) &&
3035                             bond_time_in_interval(bond, slave->last_rx, 1)) {
3036
3037                                 bond_propose_link_state(slave, BOND_LINK_UP);
3038                                 slave_state_changed = 1;
3039
3040                                 /* primary_slave has no meaning in round-robin
3041                                  * mode. the window of a slave being up and
3042                                  * curr_active_slave being null after enslaving
3043                                  * is closed.
3044                                  */
3045                                 if (!oldcurrent) {
3046                                         slave_info(bond->dev, slave->dev, "link status definitely up\n");
3047                                         do_failover = 1;
3048                                 } else {
3049                                         slave_info(bond->dev, slave->dev, "interface is now up\n");
3050                                 }
3051                         }
3052                 } else {
3053                         /* slave->link == BOND_LINK_UP */
3054
3055                         /* not all switches will respond to an arp request
3056                          * when the source ip is 0, so don't take the link down
3057                          * if we don't know our ip yet
3058                          */
3059                         if (!bond_time_in_interval(bond, trans_start, 2) ||
3060                             !bond_time_in_interval(bond, slave->last_rx, 2)) {
3061
3062                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
3063                                 slave_state_changed = 1;
3064
3065                                 if (slave->link_failure_count < UINT_MAX)
3066                                         slave->link_failure_count++;
3067
3068                                 slave_info(bond->dev, slave->dev, "interface is now down\n");
3069
3070                                 if (slave == oldcurrent)
3071                                         do_failover = 1;
3072                         }
3073                 }
3074
3075                 /* note: if switch is in round-robin mode, all links
3076                  * must tx arp to ensure all links rx an arp - otherwise
3077                  * links may oscillate or not come up at all; if switch is
3078                  * in something like xor mode, there is nothing we can
3079                  * do - all replies will be rx'ed on same link causing slaves
3080                  * to be unstable during low/no traffic periods
3081                  */
3082                 if (bond_slave_is_up(slave))
3083                         bond_arp_send_all(bond, slave);
3084         }
3085
3086         rcu_read_unlock();
3087
3088         if (do_failover || slave_state_changed) {
3089                 if (!rtnl_trylock())
3090                         goto re_arm;
3091
3092                 bond_for_each_slave(bond, slave, iter) {
3093                         if (slave->link_new_state != BOND_LINK_NOCHANGE)
3094                                 slave->link = slave->link_new_state;
3095                 }
3096
3097                 if (slave_state_changed) {
3098                         bond_slave_state_change(bond);
3099                         if (BOND_MODE(bond) == BOND_MODE_XOR)
3100                                 bond_update_slave_arr(bond, NULL);
3101                 }
3102                 if (do_failover) {
3103                         block_netpoll_tx();
3104                         bond_select_active_slave(bond);
3105                         unblock_netpoll_tx();
3106                 }
3107                 rtnl_unlock();
3108         }
3109
3110 re_arm:
3111         if (bond->params.arp_interval)
3112                 queue_delayed_work(bond->wq, &bond->arp_work,
3113                                    msecs_to_jiffies(bond->params.arp_interval));
3114 }
3115
3116 /* Called to inspect slaves for active-backup mode ARP monitor link state
3117  * changes.  Sets proposed link state in slaves to specify what action
3118  * should take place for the slave.  Returns 0 if no changes are found, >0
3119  * if changes to link states must be committed.
3120  *
3121  * Called with rcu_read_lock held.
3122  */
3123 static int bond_ab_arp_inspect(struct bonding *bond)
3124 {
3125         unsigned long trans_start, last_rx;
3126         struct list_head *iter;
3127         struct slave *slave;
3128         int commit = 0;
3129
3130         bond_for_each_slave_rcu(bond, slave, iter) {
3131                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3132                 last_rx = slave_last_rx(bond, slave);
3133
3134                 if (slave->link != BOND_LINK_UP) {
3135                         if (bond_time_in_interval(bond, last_rx, 1)) {
3136                                 bond_propose_link_state(slave, BOND_LINK_UP);
3137                                 commit++;
3138                         } else if (slave->link == BOND_LINK_BACK) {
3139                                 bond_propose_link_state(slave, BOND_LINK_FAIL);
3140                                 commit++;
3141                         }
3142                         continue;
3143                 }
3144
3145                 /* Give slaves 2*delta after being enslaved or made
3146                  * active.  This avoids bouncing, as the last receive
3147                  * times need a full ARP monitor cycle to be updated.
3148                  */
3149                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3150                         continue;
3151
3152                 /* Backup slave is down if:
3153                  * - No current_arp_slave AND
3154                  * - more than 3*delta since last receive AND
3155                  * - the bond has an IP address
3156                  *
3157                  * Note: a non-null current_arp_slave indicates
3158                  * the curr_active_slave went down and we are
3159                  * searching for a new one; under this condition
3160                  * we only take the curr_active_slave down - this
3161                  * gives each slave a chance to tx/rx traffic
3162                  * before being taken out
3163                  */
3164                 if (!bond_is_active_slave(slave) &&
3165                     !rcu_access_pointer(bond->current_arp_slave) &&
3166                     !bond_time_in_interval(bond, last_rx, 3)) {
3167                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3168                         commit++;
3169                 }
3170
3171                 /* Active slave is down if:
3172                  * - more than 2*delta since transmitting OR
3173                  * - (more than 2*delta since receive AND
3174                  *    the bond has an IP address)
3175                  */
3176                 trans_start = dev_trans_start(slave->dev);
3177                 if (bond_is_active_slave(slave) &&
3178                     (!bond_time_in_interval(bond, trans_start, 2) ||
3179                      !bond_time_in_interval(bond, last_rx, 2))) {
3180                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3181                         commit++;
3182                 }
3183         }
3184
3185         return commit;
3186 }
3187
3188 /* Called to commit link state changes noted by inspection step of
3189  * active-backup mode ARP monitor.
3190  *
3191  * Called with RTNL hold.
3192  */
3193 static void bond_ab_arp_commit(struct bonding *bond)
3194 {
3195         unsigned long trans_start;
3196         struct list_head *iter;
3197         struct slave *slave;
3198
3199         bond_for_each_slave(bond, slave, iter) {
3200                 switch (slave->link_new_state) {
3201                 case BOND_LINK_NOCHANGE:
3202                         continue;
3203
3204                 case BOND_LINK_UP:
3205                         trans_start = dev_trans_start(slave->dev);
3206                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
3207                             (!rtnl_dereference(bond->curr_active_slave) &&
3208                              bond_time_in_interval(bond, trans_start, 1))) {
3209                                 struct slave *current_arp_slave;
3210
3211                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3212                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
3213                                                           BOND_SLAVE_NOTIFY_NOW);
3214                                 if (current_arp_slave) {
3215                                         bond_set_slave_inactive_flags(
3216                                                 current_arp_slave,
3217                                                 BOND_SLAVE_NOTIFY_NOW);
3218                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3219                                 }
3220
3221                                 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3222
3223                                 if (!rtnl_dereference(bond->curr_active_slave) ||
3224                                     slave == rtnl_dereference(bond->primary_slave))
3225                                         goto do_failover;
3226
3227                         }
3228
3229                         continue;
3230
3231                 case BOND_LINK_DOWN:
3232                         if (slave->link_failure_count < UINT_MAX)
3233                                 slave->link_failure_count++;
3234
3235                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3236                                                   BOND_SLAVE_NOTIFY_NOW);
3237                         bond_set_slave_inactive_flags(slave,
3238                                                       BOND_SLAVE_NOTIFY_NOW);
3239
3240                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3241
3242                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
3243                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3244                                 goto do_failover;
3245                         }
3246
3247                         continue;
3248
3249                 case BOND_LINK_FAIL:
3250                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3251                                                   BOND_SLAVE_NOTIFY_NOW);
3252                         bond_set_slave_inactive_flags(slave,
3253                                                       BOND_SLAVE_NOTIFY_NOW);
3254
3255                         /* A slave has just been enslaved and has become
3256                          * the current active slave.
3257                          */
3258                         if (rtnl_dereference(bond->curr_active_slave))
3259                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3260                         continue;
3261
3262                 default:
3263                         slave_err(bond->dev, slave->dev,
3264                                   "impossible: link_new_state %d on slave\n",
3265                                   slave->link_new_state);
3266                         continue;
3267                 }
3268
3269 do_failover:
3270                 block_netpoll_tx();
3271                 bond_select_active_slave(bond);
3272                 unblock_netpoll_tx();
3273         }
3274
3275         bond_set_carrier(bond);
3276 }
3277
3278 /* Send ARP probes for active-backup mode ARP monitor.
3279  *
3280  * Called with rcu_read_lock held.
3281  */
3282 static bool bond_ab_arp_probe(struct bonding *bond)
3283 {
3284         struct slave *slave, *before = NULL, *new_slave = NULL,
3285                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3286                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3287         struct list_head *iter;
3288         bool found = false;
3289         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3290
3291         if (curr_arp_slave && curr_active_slave)
3292                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3293                             curr_arp_slave->dev->name,
3294                             curr_active_slave->dev->name);
3295
3296         if (curr_active_slave) {
3297                 bond_arp_send_all(bond, curr_active_slave);
3298                 return should_notify_rtnl;
3299         }
3300
3301         /* if we don't have a curr_active_slave, search for the next available
3302          * backup slave from the current_arp_slave and make it the candidate
3303          * for becoming the curr_active_slave
3304          */
3305
3306         if (!curr_arp_slave) {
3307                 curr_arp_slave = bond_first_slave_rcu(bond);
3308                 if (!curr_arp_slave)
3309                         return should_notify_rtnl;
3310         }
3311
3312         bond_for_each_slave_rcu(bond, slave, iter) {
3313                 if (!found && !before && bond_slave_is_up(slave))
3314                         before = slave;
3315
3316                 if (found && !new_slave && bond_slave_is_up(slave))
3317                         new_slave = slave;
3318                 /* if the link state is up at this point, we
3319                  * mark it down - this can happen if we have
3320                  * simultaneous link failures and
3321                  * reselect_active_interface doesn't make this
3322                  * one the current slave so it is still marked
3323                  * up when it is actually down
3324                  */
3325                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3326                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3327                                                   BOND_SLAVE_NOTIFY_LATER);
3328                         if (slave->link_failure_count < UINT_MAX)
3329                                 slave->link_failure_count++;
3330
3331                         bond_set_slave_inactive_flags(slave,
3332                                                       BOND_SLAVE_NOTIFY_LATER);
3333
3334                         slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3335                 }
3336                 if (slave == curr_arp_slave)
3337                         found = true;
3338         }
3339
3340         if (!new_slave && before)
3341                 new_slave = before;
3342
3343         if (!new_slave)
3344                 goto check_state;
3345
3346         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3347                                   BOND_SLAVE_NOTIFY_LATER);
3348         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3349         bond_arp_send_all(bond, new_slave);
3350         new_slave->last_link_up = jiffies;
3351         rcu_assign_pointer(bond->current_arp_slave, new_slave);
3352
3353 check_state:
3354         bond_for_each_slave_rcu(bond, slave, iter) {
3355                 if (slave->should_notify || slave->should_notify_link) {
3356                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3357                         break;
3358                 }
3359         }
3360         return should_notify_rtnl;
3361 }
3362
3363 static void bond_activebackup_arp_mon(struct bonding *bond)
3364 {
3365         bool should_notify_peers = false;
3366         bool should_notify_rtnl = false;
3367         int delta_in_ticks;
3368
3369         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3370
3371         if (!bond_has_slaves(bond))
3372                 goto re_arm;
3373
3374         rcu_read_lock();
3375
3376         should_notify_peers = bond_should_notify_peers(bond);
3377
3378         if (bond_ab_arp_inspect(bond)) {
3379                 rcu_read_unlock();
3380
3381                 /* Race avoidance with bond_close flush of workqueue */
3382                 if (!rtnl_trylock()) {
3383                         delta_in_ticks = 1;
3384                         should_notify_peers = false;
3385                         goto re_arm;
3386                 }
3387
3388                 bond_ab_arp_commit(bond);
3389
3390                 rtnl_unlock();
3391                 rcu_read_lock();
3392         }
3393
3394         should_notify_rtnl = bond_ab_arp_probe(bond);
3395         rcu_read_unlock();
3396
3397 re_arm:
3398         if (bond->params.arp_interval)
3399                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3400
3401         if (should_notify_peers || should_notify_rtnl) {
3402                 if (!rtnl_trylock())
3403                         return;
3404
3405                 if (should_notify_peers)
3406                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3407                                                  bond->dev);
3408                 if (should_notify_rtnl) {
3409                         bond_slave_state_notify(bond);
3410                         bond_slave_link_notify(bond);
3411                 }
3412
3413                 rtnl_unlock();
3414         }
3415 }
3416
3417 static void bond_arp_monitor(struct work_struct *work)
3418 {
3419         struct bonding *bond = container_of(work, struct bonding,
3420                                             arp_work.work);
3421
3422         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3423                 bond_activebackup_arp_mon(bond);
3424         else
3425                 bond_loadbalance_arp_mon(bond);
3426 }
3427
3428 /*-------------------------- netdev event handling --------------------------*/
3429
3430 /* Change device name */
3431 static int bond_event_changename(struct bonding *bond)
3432 {
3433         bond_remove_proc_entry(bond);
3434         bond_create_proc_entry(bond);
3435
3436         bond_debug_reregister(bond);
3437
3438         return NOTIFY_DONE;
3439 }
3440
3441 static int bond_master_netdev_event(unsigned long event,
3442                                     struct net_device *bond_dev)
3443 {
3444         struct bonding *event_bond = netdev_priv(bond_dev);
3445
3446         netdev_dbg(bond_dev, "%s called\n", __func__);
3447
3448         switch (event) {
3449         case NETDEV_CHANGENAME:
3450                 return bond_event_changename(event_bond);
3451         case NETDEV_UNREGISTER:
3452                 bond_remove_proc_entry(event_bond);
3453 #ifdef CONFIG_XFRM_OFFLOAD
3454                 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3455 #endif /* CONFIG_XFRM_OFFLOAD */
3456                 break;
3457         case NETDEV_REGISTER:
3458                 bond_create_proc_entry(event_bond);
3459                 break;
3460         default:
3461                 break;
3462         }
3463
3464         return NOTIFY_DONE;
3465 }
3466
3467 static int bond_slave_netdev_event(unsigned long event,
3468                                    struct net_device *slave_dev)
3469 {
3470         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3471         struct bonding *bond;
3472         struct net_device *bond_dev;
3473
3474         /* A netdev event can be generated while enslaving a device
3475          * before netdev_rx_handler_register is called in which case
3476          * slave will be NULL
3477          */
3478         if (!slave) {
3479                 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3480                 return NOTIFY_DONE;
3481         }
3482
3483         bond_dev = slave->bond->dev;
3484         bond = slave->bond;
3485         primary = rtnl_dereference(bond->primary_slave);
3486
3487         slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3488
3489         switch (event) {
3490         case NETDEV_UNREGISTER:
3491                 if (bond_dev->type != ARPHRD_ETHER)
3492                         bond_release_and_destroy(bond_dev, slave_dev);
3493                 else
3494                         __bond_release_one(bond_dev, slave_dev, false, true);
3495                 break;
3496         case NETDEV_UP:
3497         case NETDEV_CHANGE:
3498                 /* For 802.3ad mode only:
3499                  * Getting invalid Speed/Duplex values here will put slave
3500                  * in weird state. Mark it as link-fail if the link was
3501                  * previously up or link-down if it hasn't yet come up, and
3502                  * let link-monitoring (miimon) set it right when correct
3503                  * speeds/duplex are available.
3504                  */
3505                 if (bond_update_speed_duplex(slave) &&
3506                     BOND_MODE(bond) == BOND_MODE_8023AD) {
3507                         if (slave->last_link_up)
3508                                 slave->link = BOND_LINK_FAIL;
3509                         else
3510                                 slave->link = BOND_LINK_DOWN;
3511                 }
3512
3513                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3514                         bond_3ad_adapter_speed_duplex_changed(slave);
3515                 fallthrough;
3516         case NETDEV_DOWN:
3517                 /* Refresh slave-array if applicable!
3518                  * If the setup does not use miimon or arpmon (mode-specific!),
3519                  * then these events will not cause the slave-array to be
3520                  * refreshed. This will cause xmit to use a slave that is not
3521                  * usable. Avoid such situation by refeshing the array at these
3522                  * events. If these (miimon/arpmon) parameters are configured
3523                  * then array gets refreshed twice and that should be fine!
3524                  */
3525                 if (bond_mode_can_use_xmit_hash(bond))
3526                         bond_update_slave_arr(bond, NULL);
3527                 break;
3528         case NETDEV_CHANGEMTU:
3529                 /* TODO: Should slaves be allowed to
3530                  * independently alter their MTU?  For
3531                  * an active-backup bond, slaves need
3532                  * not be the same type of device, so
3533                  * MTUs may vary.  For other modes,
3534                  * slaves arguably should have the
3535                  * same MTUs. To do this, we'd need to
3536                  * take over the slave's change_mtu
3537                  * function for the duration of their
3538                  * servitude.
3539                  */
3540                 break;
3541         case NETDEV_CHANGENAME:
3542                 /* we don't care if we don't have primary set */
3543                 if (!bond_uses_primary(bond) ||
3544                     !bond->params.primary[0])
3545                         break;
3546
3547                 if (slave == primary) {
3548                         /* slave's name changed - he's no longer primary */
3549                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3550                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3551                         /* we have a new primary slave */
3552                         rcu_assign_pointer(bond->primary_slave, slave);
3553                 } else { /* we didn't change primary - exit */
3554                         break;
3555                 }
3556
3557                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3558                             primary ? slave_dev->name : "none");
3559
3560                 block_netpoll_tx();
3561                 bond_select_active_slave(bond);
3562                 unblock_netpoll_tx();
3563                 break;
3564         case NETDEV_FEAT_CHANGE:
3565                 bond_compute_features(bond);
3566                 break;
3567         case NETDEV_RESEND_IGMP:
3568                 /* Propagate to master device */
3569                 call_netdevice_notifiers(event, slave->bond->dev);
3570                 break;
3571         default:
3572                 break;
3573         }
3574
3575         return NOTIFY_DONE;
3576 }
3577
3578 /* bond_netdev_event: handle netdev notifier chain events.
3579  *
3580  * This function receives events for the netdev chain.  The caller (an
3581  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3582  * locks for us to safely manipulate the slave devices (RTNL lock,
3583  * dev_probe_lock).
3584  */
3585 static int bond_netdev_event(struct notifier_block *this,
3586                              unsigned long event, void *ptr)
3587 {
3588         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3589
3590         netdev_dbg(event_dev, "%s received %s\n",
3591                    __func__, netdev_cmd_to_name(event));
3592
3593         if (!(event_dev->priv_flags & IFF_BONDING))
3594                 return NOTIFY_DONE;
3595
3596         if (event_dev->flags & IFF_MASTER) {
3597                 int ret;
3598
3599                 ret = bond_master_netdev_event(event, event_dev);
3600                 if (ret != NOTIFY_DONE)
3601                         return ret;
3602         }
3603
3604         if (event_dev->flags & IFF_SLAVE)
3605                 return bond_slave_netdev_event(event, event_dev);
3606
3607         return NOTIFY_DONE;
3608 }
3609
3610 static struct notifier_block bond_netdev_notifier = {
3611         .notifier_call = bond_netdev_event,
3612 };
3613
3614 /*---------------------------- Hashing Policies -----------------------------*/
3615
3616 /* L2 hash helper */
3617 static inline u32 bond_eth_hash(struct sk_buff *skb)
3618 {
3619         struct ethhdr *ep, hdr_tmp;
3620
3621         ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3622         if (ep)
3623                 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3624         return 0;
3625 }
3626
3627 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk,
3628                          int *noff, int *proto, bool l34)
3629 {
3630         const struct ipv6hdr *iph6;
3631         const struct iphdr *iph;
3632
3633         if (skb->protocol == htons(ETH_P_IP)) {
3634                 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph))))
3635                         return false;
3636                 iph = (const struct iphdr *)(skb->data + *noff);
3637                 iph_to_flow_copy_v4addrs(fk, iph);
3638                 *noff += iph->ihl << 2;
3639                 if (!ip_is_fragment(iph))
3640                         *proto = iph->protocol;
3641         } else if (skb->protocol == htons(ETH_P_IPV6)) {
3642                 if (unlikely(!pskb_may_pull(skb, *noff + sizeof(*iph6))))
3643                         return false;
3644                 iph6 = (const struct ipv6hdr *)(skb->data + *noff);
3645                 iph_to_flow_copy_v6addrs(fk, iph6);
3646                 *noff += sizeof(*iph6);
3647                 *proto = iph6->nexthdr;
3648         } else {
3649                 return false;
3650         }
3651
3652         if (l34 && *proto >= 0)
3653                 fk->ports.ports = skb_flow_get_ports(skb, *noff, *proto);
3654
3655         return true;
3656 }
3657
3658 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb)
3659 {
3660         struct ethhdr *mac_hdr = (struct ethhdr *)skb_mac_header(skb);
3661         u32 srcmac_vendor = 0, srcmac_dev = 0;
3662         u16 vlan;
3663         int i;
3664
3665         for (i = 0; i < 3; i++)
3666                 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
3667
3668         for (i = 3; i < ETH_ALEN; i++)
3669                 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
3670
3671         if (!skb_vlan_tag_present(skb))
3672                 return srcmac_vendor ^ srcmac_dev;
3673
3674         vlan = skb_vlan_tag_get(skb);
3675
3676         return vlan ^ srcmac_vendor ^ srcmac_dev;
3677 }
3678
3679 /* Extract the appropriate headers based on bond's xmit policy */
3680 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3681                               struct flow_keys *fk)
3682 {
3683         bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
3684         int noff, proto = -1;
3685
3686         switch (bond->params.xmit_policy) {
3687         case BOND_XMIT_POLICY_ENCAP23:
3688         case BOND_XMIT_POLICY_ENCAP34:
3689                 memset(fk, 0, sizeof(*fk));
3690                 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3691                                           fk, NULL, 0, 0, 0, 0);
3692         default:
3693                 break;
3694         }
3695
3696         fk->ports.ports = 0;
3697         memset(&fk->icmp, 0, sizeof(fk->icmp));
3698         noff = skb_network_offset(skb);
3699         if (!bond_flow_ip(skb, fk, &noff, &proto, l34))
3700                 return false;
3701
3702         /* ICMP error packets contains at least 8 bytes of the header
3703          * of the packet which generated the error. Use this information
3704          * to correlate ICMP error packets within the same flow which
3705          * generated the error.
3706          */
3707         if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6) {
3708                 skb_flow_get_icmp_tci(skb, &fk->icmp, skb->data,
3709                                       skb_transport_offset(skb),
3710                                       skb_headlen(skb));
3711                 if (proto == IPPROTO_ICMP) {
3712                         if (!icmp_is_err(fk->icmp.type))
3713                                 return true;
3714
3715                         noff += sizeof(struct icmphdr);
3716                 } else if (proto == IPPROTO_ICMPV6) {
3717                         if (!icmpv6_is_err(fk->icmp.type))
3718                                 return true;
3719
3720                         noff += sizeof(struct icmp6hdr);
3721                 }
3722                 return bond_flow_ip(skb, fk, &noff, &proto, l34);
3723         }
3724
3725         return true;
3726 }
3727
3728 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow)
3729 {
3730         hash ^= (__force u32)flow_get_u32_dst(flow) ^
3731                 (__force u32)flow_get_u32_src(flow);
3732         hash ^= (hash >> 16);
3733         hash ^= (hash >> 8);
3734         /* discard lowest hash bit to deal with the common even ports pattern */
3735         return hash >> 1;
3736 }
3737
3738 /**
3739  * bond_xmit_hash - generate a hash value based on the xmit policy
3740  * @bond: bonding device
3741  * @skb: buffer to use for headers
3742  *
3743  * This function will extract the necessary headers from the skb buffer and use
3744  * them to generate a hash based on the xmit_policy set in the bonding device
3745  */
3746 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3747 {
3748         struct flow_keys flow;
3749         u32 hash;
3750
3751         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3752             skb->l4_hash)
3753                 return skb->hash;
3754
3755         if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
3756                 return bond_vlan_srcmac_hash(skb);
3757
3758         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3759             !bond_flow_dissect(bond, skb, &flow))
3760                 return bond_eth_hash(skb);
3761
3762         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3763             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
3764                 hash = bond_eth_hash(skb);
3765         } else {
3766                 if (flow.icmp.id)
3767                         memcpy(&hash, &flow.icmp, sizeof(hash));
3768                 else
3769                         memcpy(&hash, &flow.ports.ports, sizeof(hash));
3770         }
3771
3772         return bond_ip_hash(hash, &flow);
3773 }
3774
3775 /*-------------------------- Device entry points ----------------------------*/
3776
3777 void bond_work_init_all(struct bonding *bond)
3778 {
3779         INIT_DELAYED_WORK(&bond->mcast_work,
3780                           bond_resend_igmp_join_requests_delayed);
3781         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3782         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3783         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3784         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3785         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3786 }
3787
3788 static void bond_work_cancel_all(struct bonding *bond)
3789 {
3790         cancel_delayed_work_sync(&bond->mii_work);
3791         cancel_delayed_work_sync(&bond->arp_work);
3792         cancel_delayed_work_sync(&bond->alb_work);
3793         cancel_delayed_work_sync(&bond->ad_work);
3794         cancel_delayed_work_sync(&bond->mcast_work);
3795         cancel_delayed_work_sync(&bond->slave_arr_work);
3796 }
3797
3798 static int bond_open(struct net_device *bond_dev)
3799 {
3800         struct bonding *bond = netdev_priv(bond_dev);
3801         struct list_head *iter;
3802         struct slave *slave;
3803
3804         /* reset slave->backup and slave->inactive */
3805         if (bond_has_slaves(bond)) {
3806                 bond_for_each_slave(bond, slave, iter) {
3807                         if (bond_uses_primary(bond) &&
3808                             slave != rcu_access_pointer(bond->curr_active_slave)) {
3809                                 bond_set_slave_inactive_flags(slave,
3810                                                               BOND_SLAVE_NOTIFY_NOW);
3811                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3812                                 bond_set_slave_active_flags(slave,
3813                                                             BOND_SLAVE_NOTIFY_NOW);
3814                         }
3815                 }
3816         }
3817
3818         if (bond_is_lb(bond)) {
3819                 /* bond_alb_initialize must be called before the timer
3820                  * is started.
3821                  */
3822                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3823                         return -ENOMEM;
3824                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3825                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
3826         }
3827
3828         if (bond->params.miimon)  /* link check interval, in milliseconds. */
3829                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3830
3831         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3832                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3833                 bond->recv_probe = bond_arp_rcv;
3834         }
3835
3836         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3837                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3838                 /* register to receive LACPDUs */
3839                 bond->recv_probe = bond_3ad_lacpdu_recv;
3840                 bond_3ad_initiate_agg_selection(bond, 1);
3841         }
3842
3843         if (bond_mode_can_use_xmit_hash(bond))
3844                 bond_update_slave_arr(bond, NULL);
3845
3846         return 0;
3847 }
3848
3849 static int bond_close(struct net_device *bond_dev)
3850 {
3851         struct bonding *bond = netdev_priv(bond_dev);
3852
3853         bond_work_cancel_all(bond);
3854         bond->send_peer_notif = 0;
3855         if (bond_is_lb(bond))
3856                 bond_alb_deinitialize(bond);
3857         bond->recv_probe = NULL;
3858
3859         return 0;
3860 }
3861
3862 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3863  * that some drivers can provide 32bit values only.
3864  */
3865 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3866                             const struct rtnl_link_stats64 *_new,
3867                             const struct rtnl_link_stats64 *_old)
3868 {
3869         const u64 *new = (const u64 *)_new;
3870         const u64 *old = (const u64 *)_old;
3871         u64 *res = (u64 *)_res;
3872         int i;
3873
3874         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3875                 u64 nv = new[i];
3876                 u64 ov = old[i];
3877                 s64 delta = nv - ov;
3878
3879                 /* detects if this particular field is 32bit only */
3880                 if (((nv | ov) >> 32) == 0)
3881                         delta = (s64)(s32)((u32)nv - (u32)ov);
3882
3883                 /* filter anomalies, some drivers reset their stats
3884                  * at down/up events.
3885                  */
3886                 if (delta > 0)
3887                         res[i] += delta;
3888         }
3889 }
3890
3891 #ifdef CONFIG_LOCKDEP
3892 static int bond_get_lowest_level_rcu(struct net_device *dev)
3893 {
3894         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
3895         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
3896         int cur = 0, max = 0;
3897
3898         now = dev;
3899         iter = &dev->adj_list.lower;
3900
3901         while (1) {
3902                 next = NULL;
3903                 while (1) {
3904                         ldev = netdev_next_lower_dev_rcu(now, &iter);
3905                         if (!ldev)
3906                                 break;
3907
3908                         next = ldev;
3909                         niter = &ldev->adj_list.lower;
3910                         dev_stack[cur] = now;
3911                         iter_stack[cur++] = iter;
3912                         if (max <= cur)
3913                                 max = cur;
3914                         break;
3915                 }
3916
3917                 if (!next) {
3918                         if (!cur)
3919                                 return max;
3920                         next = dev_stack[--cur];
3921                         niter = iter_stack[cur];
3922                 }
3923
3924                 now = next;
3925                 iter = niter;
3926         }
3927
3928         return max;
3929 }
3930 #endif
3931
3932 static void bond_get_stats(struct net_device *bond_dev,
3933                            struct rtnl_link_stats64 *stats)
3934 {
3935         struct bonding *bond = netdev_priv(bond_dev);
3936         struct rtnl_link_stats64 temp;
3937         struct list_head *iter;
3938         struct slave *slave;
3939         int nest_level = 0;
3940
3941
3942         rcu_read_lock();
3943 #ifdef CONFIG_LOCKDEP
3944         nest_level = bond_get_lowest_level_rcu(bond_dev);
3945 #endif
3946
3947         spin_lock_nested(&bond->stats_lock, nest_level);
3948         memcpy(stats, &bond->bond_stats, sizeof(*stats));
3949
3950         bond_for_each_slave_rcu(bond, slave, iter) {
3951                 const struct rtnl_link_stats64 *new =
3952                         dev_get_stats(slave->dev, &temp);
3953
3954                 bond_fold_stats(stats, new, &slave->slave_stats);
3955
3956                 /* save off the slave stats for the next run */
3957                 memcpy(&slave->slave_stats, new, sizeof(*new));
3958         }
3959
3960         memcpy(&bond->bond_stats, stats, sizeof(*stats));
3961         spin_unlock(&bond->stats_lock);
3962         rcu_read_unlock();
3963 }
3964
3965 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3966 {
3967         struct bonding *bond = netdev_priv(bond_dev);
3968         struct net_device *slave_dev = NULL;
3969         struct ifbond k_binfo;
3970         struct ifbond __user *u_binfo = NULL;
3971         struct ifslave k_sinfo;
3972         struct ifslave __user *u_sinfo = NULL;
3973         struct mii_ioctl_data *mii = NULL;
3974         struct bond_opt_value newval;
3975         struct net *net;
3976         int res = 0;
3977
3978         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3979
3980         switch (cmd) {
3981         case SIOCGMIIPHY:
3982                 mii = if_mii(ifr);
3983                 if (!mii)
3984                         return -EINVAL;
3985
3986                 mii->phy_id = 0;
3987                 fallthrough;
3988         case SIOCGMIIREG:
3989                 /* We do this again just in case we were called by SIOCGMIIREG
3990                  * instead of SIOCGMIIPHY.
3991                  */
3992                 mii = if_mii(ifr);
3993                 if (!mii)
3994                         return -EINVAL;
3995
3996                 if (mii->reg_num == 1) {
3997                         mii->val_out = 0;
3998                         if (netif_carrier_ok(bond->dev))
3999                                 mii->val_out = BMSR_LSTATUS;
4000                 }
4001
4002                 return 0;
4003         case BOND_INFO_QUERY_OLD:
4004         case SIOCBONDINFOQUERY:
4005                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4006
4007                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4008                         return -EFAULT;
4009
4010                 bond_info_query(bond_dev, &k_binfo);
4011                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4012                         return -EFAULT;
4013
4014                 return 0;
4015         case BOND_SLAVE_INFO_QUERY_OLD:
4016         case SIOCBONDSLAVEINFOQUERY:
4017                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4018
4019                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4020                         return -EFAULT;
4021
4022                 res = bond_slave_info_query(bond_dev, &k_sinfo);
4023                 if (res == 0 &&
4024                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4025                         return -EFAULT;
4026
4027                 return res;
4028         default:
4029                 break;
4030         }
4031
4032         net = dev_net(bond_dev);
4033
4034         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4035                 return -EPERM;
4036
4037         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4038
4039         slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4040
4041         if (!slave_dev)
4042                 return -ENODEV;
4043
4044         switch (cmd) {
4045         case BOND_ENSLAVE_OLD:
4046         case SIOCBONDENSLAVE:
4047                 res = bond_enslave(bond_dev, slave_dev, NULL);
4048                 break;
4049         case BOND_RELEASE_OLD:
4050         case SIOCBONDRELEASE:
4051                 res = bond_release(bond_dev, slave_dev);
4052                 break;
4053         case BOND_SETHWADDR_OLD:
4054         case SIOCBONDSETHWADDR:
4055                 res = bond_set_dev_addr(bond_dev, slave_dev);
4056                 break;
4057         case BOND_CHANGE_ACTIVE_OLD:
4058         case SIOCBONDCHANGEACTIVE:
4059                 bond_opt_initstr(&newval, slave_dev->name);
4060                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4061                                             &newval);
4062                 break;
4063         default:
4064                 res = -EOPNOTSUPP;
4065         }
4066
4067         return res;
4068 }
4069
4070 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4071 {
4072         struct bonding *bond = netdev_priv(bond_dev);
4073
4074         if (change & IFF_PROMISC)
4075                 bond_set_promiscuity(bond,
4076                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
4077
4078         if (change & IFF_ALLMULTI)
4079                 bond_set_allmulti(bond,
4080                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4081 }
4082
4083 static void bond_set_rx_mode(struct net_device *bond_dev)
4084 {
4085         struct bonding *bond = netdev_priv(bond_dev);
4086         struct list_head *iter;
4087         struct slave *slave;
4088
4089         rcu_read_lock();
4090         if (bond_uses_primary(bond)) {
4091                 slave = rcu_dereference(bond->curr_active_slave);
4092                 if (slave) {
4093                         dev_uc_sync(slave->dev, bond_dev);
4094                         dev_mc_sync(slave->dev, bond_dev);
4095                 }
4096         } else {
4097                 bond_for_each_slave_rcu(bond, slave, iter) {
4098                         dev_uc_sync_multiple(slave->dev, bond_dev);
4099                         dev_mc_sync_multiple(slave->dev, bond_dev);
4100                 }
4101         }
4102         rcu_read_unlock();
4103 }
4104
4105 static int bond_neigh_init(struct neighbour *n)
4106 {
4107         struct bonding *bond = netdev_priv(n->dev);
4108         const struct net_device_ops *slave_ops;
4109         struct neigh_parms parms;
4110         struct slave *slave;
4111         int ret = 0;
4112
4113         rcu_read_lock();
4114         slave = bond_first_slave_rcu(bond);
4115         if (!slave)
4116                 goto out;
4117         slave_ops = slave->dev->netdev_ops;
4118         if (!slave_ops->ndo_neigh_setup)
4119                 goto out;
4120
4121         /* TODO: find another way [1] to implement this.
4122          * Passing a zeroed structure is fragile,
4123          * but at least we do not pass garbage.
4124          *
4125          * [1] One way would be that ndo_neigh_setup() never touch
4126          *     struct neigh_parms, but propagate the new neigh_setup()
4127          *     back to ___neigh_create() / neigh_parms_alloc()
4128          */
4129         memset(&parms, 0, sizeof(parms));
4130         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4131
4132         if (ret)
4133                 goto out;
4134
4135         if (parms.neigh_setup)
4136                 ret = parms.neigh_setup(n);
4137 out:
4138         rcu_read_unlock();
4139         return ret;
4140 }
4141
4142 /* The bonding ndo_neigh_setup is called at init time beofre any
4143  * slave exists. So we must declare proxy setup function which will
4144  * be used at run time to resolve the actual slave neigh param setup.
4145  *
4146  * It's also called by master devices (such as vlans) to setup their
4147  * underlying devices. In that case - do nothing, we're already set up from
4148  * our init.
4149  */
4150 static int bond_neigh_setup(struct net_device *dev,
4151                             struct neigh_parms *parms)
4152 {
4153         /* modify only our neigh_parms */
4154         if (parms->dev == dev)
4155                 parms->neigh_setup = bond_neigh_init;
4156
4157         return 0;
4158 }
4159
4160 /* Change the MTU of all of a master's slaves to match the master */
4161 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4162 {
4163         struct bonding *bond = netdev_priv(bond_dev);
4164         struct slave *slave, *rollback_slave;
4165         struct list_head *iter;
4166         int res = 0;
4167
4168         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4169
4170         bond_for_each_slave(bond, slave, iter) {
4171                 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4172                            slave, slave->dev->netdev_ops->ndo_change_mtu);
4173
4174                 res = dev_set_mtu(slave->dev, new_mtu);
4175
4176                 if (res) {
4177                         /* If we failed to set the slave's mtu to the new value
4178                          * we must abort the operation even in ACTIVE_BACKUP
4179                          * mode, because if we allow the backup slaves to have
4180                          * different mtu values than the active slave we'll
4181                          * need to change their mtu when doing a failover. That
4182                          * means changing their mtu from timer context, which
4183                          * is probably not a good idea.
4184                          */
4185                         slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4186                                   res, new_mtu);
4187                         goto unwind;
4188                 }
4189         }
4190
4191         bond_dev->mtu = new_mtu;
4192
4193         return 0;
4194
4195 unwind:
4196         /* unwind from head to the slave that failed */
4197         bond_for_each_slave(bond, rollback_slave, iter) {
4198                 int tmp_res;
4199
4200                 if (rollback_slave == slave)
4201                         break;
4202
4203                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4204                 if (tmp_res)
4205                         slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4206                                   tmp_res);
4207         }
4208
4209         return res;
4210 }
4211
4212 /* Change HW address
4213  *
4214  * Note that many devices must be down to change the HW address, and
4215  * downing the master releases all slaves.  We can make bonds full of
4216  * bonding devices to test this, however.
4217  */
4218 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4219 {
4220         struct bonding *bond = netdev_priv(bond_dev);
4221         struct slave *slave, *rollback_slave;
4222         struct sockaddr_storage *ss = addr, tmp_ss;
4223         struct list_head *iter;
4224         int res = 0;
4225
4226         if (BOND_MODE(bond) == BOND_MODE_ALB)
4227                 return bond_alb_set_mac_address(bond_dev, addr);
4228
4229
4230         netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4231
4232         /* If fail_over_mac is enabled, do nothing and return success.
4233          * Returning an error causes ifenslave to fail.
4234          */
4235         if (bond->params.fail_over_mac &&
4236             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4237                 return 0;
4238
4239         if (!is_valid_ether_addr(ss->__data))
4240                 return -EADDRNOTAVAIL;
4241
4242         bond_for_each_slave(bond, slave, iter) {
4243                 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4244                           __func__, slave);
4245                 res = dev_set_mac_address(slave->dev, addr, NULL);
4246                 if (res) {
4247                         /* TODO: consider downing the slave
4248                          * and retry ?
4249                          * User should expect communications
4250                          * breakage anyway until ARP finish
4251                          * updating, so...
4252                          */
4253                         slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4254                                   __func__, res);
4255                         goto unwind;
4256                 }
4257         }
4258
4259         /* success */
4260         memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
4261         return 0;
4262
4263 unwind:
4264         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4265         tmp_ss.ss_family = bond_dev->type;
4266
4267         /* unwind from head to the slave that failed */
4268         bond_for_each_slave(bond, rollback_slave, iter) {
4269                 int tmp_res;
4270
4271                 if (rollback_slave == slave)
4272                         break;
4273
4274                 tmp_res = dev_set_mac_address(rollback_slave->dev,
4275                                               (struct sockaddr *)&tmp_ss, NULL);
4276                 if (tmp_res) {
4277                         slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4278                                    __func__, tmp_res);
4279                 }
4280         }
4281
4282         return res;
4283 }
4284
4285 /**
4286  * bond_get_slave_by_id - get xmit slave with slave_id
4287  * @bond: bonding device that is transmitting
4288  * @slave_id: slave id up to slave_cnt-1 through which to transmit
4289  *
4290  * This function tries to get slave with slave_id but in case
4291  * it fails, it tries to find the first available slave for transmission.
4292  */
4293 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4294                                           int slave_id)
4295 {
4296         struct list_head *iter;
4297         struct slave *slave;
4298         int i = slave_id;
4299
4300         /* Here we start from the slave with slave_id */
4301         bond_for_each_slave_rcu(bond, slave, iter) {
4302                 if (--i < 0) {
4303                         if (bond_slave_can_tx(slave))
4304                                 return slave;
4305                 }
4306         }
4307
4308         /* Here we start from the first slave up to slave_id */
4309         i = slave_id;
4310         bond_for_each_slave_rcu(bond, slave, iter) {
4311                 if (--i < 0)
4312                         break;
4313                 if (bond_slave_can_tx(slave))
4314                         return slave;
4315         }
4316         /* no slave that can tx has been found */
4317         return NULL;
4318 }
4319
4320 /**
4321  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4322  * @bond: bonding device to use
4323  *
4324  * Based on the value of the bonding device's packets_per_slave parameter
4325  * this function generates a slave id, which is usually used as the next
4326  * slave to transmit through.
4327  */
4328 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4329 {
4330         u32 slave_id;
4331         struct reciprocal_value reciprocal_packets_per_slave;
4332         int packets_per_slave = bond->params.packets_per_slave;
4333
4334         switch (packets_per_slave) {
4335         case 0:
4336                 slave_id = prandom_u32();
4337                 break;
4338         case 1:
4339                 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4340                 break;
4341         default:
4342                 reciprocal_packets_per_slave =
4343                         bond->params.reciprocal_packets_per_slave;
4344                 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4345                 slave_id = reciprocal_divide(slave_id,
4346                                              reciprocal_packets_per_slave);
4347                 break;
4348         }
4349
4350         return slave_id;
4351 }
4352
4353 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4354                                                     struct sk_buff *skb)
4355 {
4356         struct slave *slave;
4357         int slave_cnt;
4358         u32 slave_id;
4359
4360         /* Start with the curr_active_slave that joined the bond as the
4361          * default for sending IGMP traffic.  For failover purposes one
4362          * needs to maintain some consistency for the interface that will
4363          * send the join/membership reports.  The curr_active_slave found
4364          * will send all of this type of traffic.
4365          */
4366         if (skb->protocol == htons(ETH_P_IP)) {
4367                 int noff = skb_network_offset(skb);
4368                 struct iphdr *iph;
4369
4370                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4371                         goto non_igmp;
4372
4373                 iph = ip_hdr(skb);
4374                 if (iph->protocol == IPPROTO_IGMP) {
4375                         slave = rcu_dereference(bond->curr_active_slave);
4376                         if (slave)
4377                                 return slave;
4378                         return bond_get_slave_by_id(bond, 0);
4379                 }
4380         }
4381
4382 non_igmp:
4383         slave_cnt = READ_ONCE(bond->slave_cnt);
4384         if (likely(slave_cnt)) {
4385                 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4386                 return bond_get_slave_by_id(bond, slave_id);
4387         }
4388         return NULL;
4389 }
4390
4391 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4392                                         struct net_device *bond_dev)
4393 {
4394         struct bonding *bond = netdev_priv(bond_dev);
4395         struct slave *slave;
4396
4397         slave = bond_xmit_roundrobin_slave_get(bond, skb);
4398         if (likely(slave))
4399                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4400
4401         return bond_tx_drop(bond_dev, skb);
4402 }
4403
4404 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond,
4405                                                       struct sk_buff *skb)
4406 {
4407         return rcu_dereference(bond->curr_active_slave);
4408 }
4409
4410 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4411  * the bond has a usable interface.
4412  */
4413 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4414                                           struct net_device *bond_dev)
4415 {
4416         struct bonding *bond = netdev_priv(bond_dev);
4417         struct slave *slave;
4418
4419         slave = bond_xmit_activebackup_slave_get(bond, skb);
4420         if (slave)
4421                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4422
4423         return bond_tx_drop(bond_dev, skb);
4424 }
4425
4426 /* Use this to update slave_array when (a) it's not appropriate to update
4427  * slave_array right away (note that update_slave_array() may sleep)
4428  * and / or (b) RTNL is not held.
4429  */
4430 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4431 {
4432         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4433 }
4434
4435 /* Slave array work handler. Holds only RTNL */
4436 static void bond_slave_arr_handler(struct work_struct *work)
4437 {
4438         struct bonding *bond = container_of(work, struct bonding,
4439                                             slave_arr_work.work);
4440         int ret;
4441
4442         if (!rtnl_trylock())
4443                 goto err;
4444
4445         ret = bond_update_slave_arr(bond, NULL);
4446         rtnl_unlock();
4447         if (ret) {
4448                 pr_warn_ratelimited("Failed to update slave array from WT\n");
4449                 goto err;
4450         }
4451         return;
4452
4453 err:
4454         bond_slave_arr_work_rearm(bond, 1);
4455 }
4456
4457 static void bond_skip_slave(struct bond_up_slave *slaves,
4458                             struct slave *skipslave)
4459 {
4460         int idx;
4461
4462         /* Rare situation where caller has asked to skip a specific
4463          * slave but allocation failed (most likely!). BTW this is
4464          * only possible when the call is initiated from
4465          * __bond_release_one(). In this situation; overwrite the
4466          * skipslave entry in the array with the last entry from the
4467          * array to avoid a situation where the xmit path may choose
4468          * this to-be-skipped slave to send a packet out.
4469          */
4470         for (idx = 0; slaves && idx < slaves->count; idx++) {
4471                 if (skipslave == slaves->arr[idx]) {
4472                         slaves->arr[idx] =
4473                                 slaves->arr[slaves->count - 1];
4474                         slaves->count--;
4475                         break;
4476                 }
4477         }
4478 }
4479
4480 static void bond_set_slave_arr(struct bonding *bond,
4481                                struct bond_up_slave *usable_slaves,
4482                                struct bond_up_slave *all_slaves)
4483 {
4484         struct bond_up_slave *usable, *all;
4485
4486         usable = rtnl_dereference(bond->usable_slaves);
4487         rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4488         kfree_rcu(usable, rcu);
4489
4490         all = rtnl_dereference(bond->all_slaves);
4491         rcu_assign_pointer(bond->all_slaves, all_slaves);
4492         kfree_rcu(all, rcu);
4493 }
4494
4495 static void bond_reset_slave_arr(struct bonding *bond)
4496 {
4497         struct bond_up_slave *usable, *all;
4498
4499         usable = rtnl_dereference(bond->usable_slaves);
4500         if (usable) {
4501                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4502                 kfree_rcu(usable, rcu);
4503         }
4504
4505         all = rtnl_dereference(bond->all_slaves);
4506         if (all) {
4507                 RCU_INIT_POINTER(bond->all_slaves, NULL);
4508                 kfree_rcu(all, rcu);
4509         }
4510 }
4511
4512 /* Build the usable slaves array in control path for modes that use xmit-hash
4513  * to determine the slave interface -
4514  * (a) BOND_MODE_8023AD
4515  * (b) BOND_MODE_XOR
4516  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4517  *
4518  * The caller is expected to hold RTNL only and NO other lock!
4519  */
4520 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4521 {
4522         struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
4523         struct slave *slave;
4524         struct list_head *iter;
4525         int agg_id = 0;
4526         int ret = 0;
4527
4528         might_sleep();
4529
4530         usable_slaves = kzalloc(struct_size(usable_slaves, arr,
4531                                             bond->slave_cnt), GFP_KERNEL);
4532         all_slaves = kzalloc(struct_size(all_slaves, arr,
4533                                          bond->slave_cnt), GFP_KERNEL);
4534         if (!usable_slaves || !all_slaves) {
4535                 ret = -ENOMEM;
4536                 goto out;
4537         }
4538         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4539                 struct ad_info ad_info;
4540
4541                 spin_lock_bh(&bond->mode_lock);
4542                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4543                         spin_unlock_bh(&bond->mode_lock);
4544                         pr_debug("bond_3ad_get_active_agg_info failed\n");
4545                         /* No active aggragator means it's not safe to use
4546                          * the previous array.
4547                          */
4548                         bond_reset_slave_arr(bond);
4549                         goto out;
4550                 }
4551                 spin_unlock_bh(&bond->mode_lock);
4552                 agg_id = ad_info.aggregator_id;
4553         }
4554         bond_for_each_slave(bond, slave, iter) {
4555                 if (skipslave == slave)
4556                         continue;
4557
4558                 all_slaves->arr[all_slaves->count++] = slave;
4559                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4560                         struct aggregator *agg;
4561
4562                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
4563                         if (!agg || agg->aggregator_identifier != agg_id)
4564                                 continue;
4565                 }
4566                 if (!bond_slave_can_tx(slave))
4567                         continue;
4568
4569                 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4570                           usable_slaves->count);
4571
4572                 usable_slaves->arr[usable_slaves->count++] = slave;
4573         }
4574
4575         bond_set_slave_arr(bond, usable_slaves, all_slaves);
4576         return ret;
4577 out:
4578         if (ret != 0 && skipslave) {
4579                 bond_skip_slave(rtnl_dereference(bond->all_slaves),
4580                                 skipslave);
4581                 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
4582                                 skipslave);
4583         }
4584         kfree_rcu(all_slaves, rcu);
4585         kfree_rcu(usable_slaves, rcu);
4586
4587         return ret;
4588 }
4589
4590 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
4591                                                  struct sk_buff *skb,
4592                                                  struct bond_up_slave *slaves)
4593 {
4594         struct slave *slave;
4595         unsigned int count;
4596         u32 hash;
4597
4598         hash = bond_xmit_hash(bond, skb);
4599         count = slaves ? READ_ONCE(slaves->count) : 0;
4600         if (unlikely(!count))
4601                 return NULL;
4602
4603         slave = slaves->arr[hash % count];
4604         return slave;
4605 }
4606
4607 /* Use this Xmit function for 3AD as well as XOR modes. The current
4608  * usable slave array is formed in the control path. The xmit function
4609  * just calculates hash and sends the packet out.
4610  */
4611 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4612                                      struct net_device *dev)
4613 {
4614         struct bonding *bond = netdev_priv(dev);
4615         struct bond_up_slave *slaves;
4616         struct slave *slave;
4617
4618         slaves = rcu_dereference(bond->usable_slaves);
4619         slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4620         if (likely(slave))
4621                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4622
4623         return bond_tx_drop(dev, skb);
4624 }
4625
4626 /* in broadcast mode, we send everything to all usable interfaces. */
4627 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4628                                        struct net_device *bond_dev)
4629 {
4630         struct bonding *bond = netdev_priv(bond_dev);
4631         struct slave *slave = NULL;
4632         struct list_head *iter;
4633
4634         bond_for_each_slave_rcu(bond, slave, iter) {
4635                 if (bond_is_last_slave(bond, slave))
4636                         break;
4637                 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4638                         struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4639
4640                         if (!skb2) {
4641                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4642                                                     bond_dev->name, __func__);
4643                                 continue;
4644                         }
4645                         bond_dev_queue_xmit(bond, skb2, slave->dev);
4646                 }
4647         }
4648         if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4649                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4650
4651         return bond_tx_drop(bond_dev, skb);
4652 }
4653
4654 /*------------------------- Device initialization ---------------------------*/
4655
4656 /* Lookup the slave that corresponds to a qid */
4657 static inline int bond_slave_override(struct bonding *bond,
4658                                       struct sk_buff *skb)
4659 {
4660         struct slave *slave = NULL;
4661         struct list_head *iter;
4662
4663         if (!skb_rx_queue_recorded(skb))
4664                 return 1;
4665
4666         /* Find out if any slaves have the same mapping as this skb. */
4667         bond_for_each_slave_rcu(bond, slave, iter) {
4668                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4669                         if (bond_slave_is_up(slave) &&
4670                             slave->link == BOND_LINK_UP) {
4671                                 bond_dev_queue_xmit(bond, skb, slave->dev);
4672                                 return 0;
4673                         }
4674                         /* If the slave isn't UP, use default transmit policy. */
4675                         break;
4676                 }
4677         }
4678
4679         return 1;
4680 }
4681
4682
4683 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4684                              struct net_device *sb_dev)
4685 {
4686         /* This helper function exists to help dev_pick_tx get the correct
4687          * destination queue.  Using a helper function skips a call to
4688          * skb_tx_hash and will put the skbs in the queue we expect on their
4689          * way down to the bonding driver.
4690          */
4691         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4692
4693         /* Save the original txq to restore before passing to the driver */
4694         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4695
4696         if (unlikely(txq >= dev->real_num_tx_queues)) {
4697                 do {
4698                         txq -= dev->real_num_tx_queues;
4699                 } while (txq >= dev->real_num_tx_queues);
4700         }
4701         return txq;
4702 }
4703
4704 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
4705                                               struct sk_buff *skb,
4706                                               bool all_slaves)
4707 {
4708         struct bonding *bond = netdev_priv(master_dev);
4709         struct bond_up_slave *slaves;
4710         struct slave *slave = NULL;
4711
4712         switch (BOND_MODE(bond)) {
4713         case BOND_MODE_ROUNDROBIN:
4714                 slave = bond_xmit_roundrobin_slave_get(bond, skb);
4715                 break;
4716         case BOND_MODE_ACTIVEBACKUP:
4717                 slave = bond_xmit_activebackup_slave_get(bond, skb);
4718                 break;
4719         case BOND_MODE_8023AD:
4720         case BOND_MODE_XOR:
4721                 if (all_slaves)
4722                         slaves = rcu_dereference(bond->all_slaves);
4723                 else
4724                         slaves = rcu_dereference(bond->usable_slaves);
4725                 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4726                 break;
4727         case BOND_MODE_BROADCAST:
4728                 break;
4729         case BOND_MODE_ALB:
4730                 slave = bond_xmit_alb_slave_get(bond, skb);
4731                 break;
4732         case BOND_MODE_TLB:
4733                 slave = bond_xmit_tlb_slave_get(bond, skb);
4734                 break;
4735         default:
4736                 /* Should never happen, mode already checked */
4737                 WARN_ONCE(true, "Unknown bonding mode");
4738                 break;
4739         }
4740
4741         if (slave)
4742                 return slave->dev;
4743         return NULL;
4744 }
4745
4746 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
4747 {
4748         switch (sk->sk_family) {
4749 #if IS_ENABLED(CONFIG_IPV6)
4750         case AF_INET6:
4751                 if (sk->sk_ipv6only ||
4752                     ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
4753                         flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
4754                         flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
4755                         flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
4756                         break;
4757                 }
4758                 fallthrough;
4759 #endif
4760         default: /* AF_INET */
4761                 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
4762                 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
4763                 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
4764                 break;
4765         }
4766
4767         flow->ports.src = inet_sk(sk)->inet_sport;
4768         flow->ports.dst = inet_sk(sk)->inet_dport;
4769 }
4770
4771 /**
4772  * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
4773  * @sk: socket to use for headers
4774  *
4775  * This function will extract the necessary field from the socket and use
4776  * them to generate a hash based on the LAYER34 xmit_policy.
4777  * Assumes that sk is a TCP or UDP socket.
4778  */
4779 static u32 bond_sk_hash_l34(struct sock *sk)
4780 {
4781         struct flow_keys flow;
4782         u32 hash;
4783
4784         bond_sk_to_flow(sk, &flow);
4785
4786         /* L4 */
4787         memcpy(&hash, &flow.ports.ports, sizeof(hash));
4788         /* L3 */
4789         return bond_ip_hash(hash, &flow);
4790 }
4791
4792 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
4793                                                   struct sock *sk)
4794 {
4795         struct bond_up_slave *slaves;
4796         struct slave *slave;
4797         unsigned int count;
4798         u32 hash;
4799
4800         slaves = rcu_dereference(bond->usable_slaves);
4801         count = slaves ? READ_ONCE(slaves->count) : 0;
4802         if (unlikely(!count))
4803                 return NULL;
4804
4805         hash = bond_sk_hash_l34(sk);
4806         slave = slaves->arr[hash % count];
4807
4808         return slave->dev;
4809 }
4810
4811 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
4812                                                 struct sock *sk)
4813 {
4814         struct bonding *bond = netdev_priv(dev);
4815         struct net_device *lower = NULL;
4816
4817         rcu_read_lock();
4818         if (bond_sk_check(bond))
4819                 lower = __bond_sk_get_lower_dev(bond, sk);
4820         rcu_read_unlock();
4821
4822         return lower;
4823 }
4824
4825 #if IS_ENABLED(CONFIG_TLS_DEVICE)
4826 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
4827                                         struct net_device *dev)
4828 {
4829         if (likely(bond_get_slave_by_dev(bond, tls_get_ctx(skb->sk)->netdev)))
4830                 return bond_dev_queue_xmit(bond, skb, tls_get_ctx(skb->sk)->netdev);
4831         return bond_tx_drop(dev, skb);
4832 }
4833 #endif
4834
4835 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4836 {
4837         struct bonding *bond = netdev_priv(dev);
4838
4839         if (bond_should_override_tx_queue(bond) &&
4840             !bond_slave_override(bond, skb))
4841                 return NETDEV_TX_OK;
4842
4843 #if IS_ENABLED(CONFIG_TLS_DEVICE)
4844         if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
4845                 return bond_tls_device_xmit(bond, skb, dev);
4846 #endif
4847
4848         switch (BOND_MODE(bond)) {
4849         case BOND_MODE_ROUNDROBIN:
4850                 return bond_xmit_roundrobin(skb, dev);
4851         case BOND_MODE_ACTIVEBACKUP:
4852                 return bond_xmit_activebackup(skb, dev);
4853         case BOND_MODE_8023AD:
4854         case BOND_MODE_XOR:
4855                 return bond_3ad_xor_xmit(skb, dev);
4856         case BOND_MODE_BROADCAST:
4857                 return bond_xmit_broadcast(skb, dev);
4858         case BOND_MODE_ALB:
4859                 return bond_alb_xmit(skb, dev);
4860         case BOND_MODE_TLB:
4861                 return bond_tlb_xmit(skb, dev);
4862         default:
4863                 /* Should never happen, mode already checked */
4864                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4865                 WARN_ON_ONCE(1);
4866                 return bond_tx_drop(dev, skb);
4867         }
4868 }
4869
4870 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4871 {
4872         struct bonding *bond = netdev_priv(dev);
4873         netdev_tx_t ret = NETDEV_TX_OK;
4874
4875         /* If we risk deadlock from transmitting this in the
4876          * netpoll path, tell netpoll to queue the frame for later tx
4877          */
4878         if (unlikely(is_netpoll_tx_blocked(dev)))
4879                 return NETDEV_TX_BUSY;
4880
4881         rcu_read_lock();
4882         if (bond_has_slaves(bond))
4883                 ret = __bond_start_xmit(skb, dev);
4884         else
4885                 ret = bond_tx_drop(dev, skb);
4886         rcu_read_unlock();
4887
4888         return ret;
4889 }
4890
4891 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
4892 {
4893         if (speed == 0 || speed == SPEED_UNKNOWN)
4894                 speed = slave->speed;
4895         else
4896                 speed = min(speed, slave->speed);
4897
4898         return speed;
4899 }
4900
4901 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4902                                            struct ethtool_link_ksettings *cmd)
4903 {
4904         struct bonding *bond = netdev_priv(bond_dev);
4905         struct list_head *iter;
4906         struct slave *slave;
4907         u32 speed = 0;
4908
4909         cmd->base.duplex = DUPLEX_UNKNOWN;
4910         cmd->base.port = PORT_OTHER;
4911
4912         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4913          * do not need to check mode.  Though link speed might not represent
4914          * the true receive or transmit bandwidth (not all modes are symmetric)
4915          * this is an accurate maximum.
4916          */
4917         bond_for_each_slave(bond, slave, iter) {
4918                 if (bond_slave_can_tx(slave)) {
4919                         if (slave->speed != SPEED_UNKNOWN) {
4920                                 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
4921                                         speed = bond_mode_bcast_speed(slave,
4922                                                                       speed);
4923                                 else
4924                                         speed += slave->speed;
4925                         }
4926                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4927                             slave->duplex != DUPLEX_UNKNOWN)
4928                                 cmd->base.duplex = slave->duplex;
4929                 }
4930         }
4931         cmd->base.speed = speed ? : SPEED_UNKNOWN;
4932
4933         return 0;
4934 }
4935
4936 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4937                                      struct ethtool_drvinfo *drvinfo)
4938 {
4939         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4940         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4941                  BOND_ABI_VERSION);
4942 }
4943
4944 static const struct ethtool_ops bond_ethtool_ops = {
4945         .get_drvinfo            = bond_ethtool_get_drvinfo,
4946         .get_link               = ethtool_op_get_link,
4947         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
4948 };
4949
4950 static const struct net_device_ops bond_netdev_ops = {
4951         .ndo_init               = bond_init,
4952         .ndo_uninit             = bond_uninit,
4953         .ndo_open               = bond_open,
4954         .ndo_stop               = bond_close,
4955         .ndo_start_xmit         = bond_start_xmit,
4956         .ndo_select_queue       = bond_select_queue,
4957         .ndo_get_stats64        = bond_get_stats,
4958         .ndo_do_ioctl           = bond_do_ioctl,
4959         .ndo_change_rx_flags    = bond_change_rx_flags,
4960         .ndo_set_rx_mode        = bond_set_rx_mode,
4961         .ndo_change_mtu         = bond_change_mtu,
4962         .ndo_set_mac_address    = bond_set_mac_address,
4963         .ndo_neigh_setup        = bond_neigh_setup,
4964         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4965         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4966 #ifdef CONFIG_NET_POLL_CONTROLLER
4967         .ndo_netpoll_setup      = bond_netpoll_setup,
4968         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4969         .ndo_poll_controller    = bond_poll_controller,
4970 #endif
4971         .ndo_add_slave          = bond_enslave,
4972         .ndo_del_slave          = bond_release,
4973         .ndo_fix_features       = bond_fix_features,
4974         .ndo_features_check     = passthru_features_check,
4975         .ndo_get_xmit_slave     = bond_xmit_get_slave,
4976         .ndo_sk_get_lower_dev   = bond_sk_get_lower_dev,
4977 };
4978
4979 static const struct device_type bond_type = {
4980         .name = "bond",
4981 };
4982
4983 static void bond_destructor(struct net_device *bond_dev)
4984 {
4985         struct bonding *bond = netdev_priv(bond_dev);
4986
4987         if (bond->wq)
4988                 destroy_workqueue(bond->wq);
4989
4990         if (bond->rr_tx_counter)
4991                 free_percpu(bond->rr_tx_counter);
4992 }
4993
4994 void bond_setup(struct net_device *bond_dev)
4995 {
4996         struct bonding *bond = netdev_priv(bond_dev);
4997
4998         spin_lock_init(&bond->mode_lock);
4999         bond->params = bonding_defaults;
5000
5001         /* Initialize pointers */
5002         bond->dev = bond_dev;
5003
5004         /* Initialize the device entry points */
5005         ether_setup(bond_dev);
5006         bond_dev->max_mtu = ETH_MAX_MTU;
5007         bond_dev->netdev_ops = &bond_netdev_ops;
5008         bond_dev->ethtool_ops = &bond_ethtool_ops;
5009
5010         bond_dev->needs_free_netdev = true;
5011         bond_dev->priv_destructor = bond_destructor;
5012
5013         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5014
5015         /* Initialize the device options */
5016         bond_dev->flags |= IFF_MASTER;
5017         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5018         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5019
5020 #ifdef CONFIG_XFRM_OFFLOAD
5021         /* set up xfrm device ops (only supported in active-backup right now) */
5022         bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5023         INIT_LIST_HEAD(&bond->ipsec_list);
5024         spin_lock_init(&bond->ipsec_lock);
5025 #endif /* CONFIG_XFRM_OFFLOAD */
5026
5027         /* don't acquire bond device's netif_tx_lock when transmitting */
5028         bond_dev->features |= NETIF_F_LLTX;
5029
5030         /* By default, we declare the bond to be fully
5031          * VLAN hardware accelerated capable. Special
5032          * care is taken in the various xmit functions
5033          * when there are slaves that are not hw accel
5034          * capable
5035          */
5036
5037         /* Don't allow bond devices to change network namespaces. */
5038         bond_dev->features |= NETIF_F_NETNS_LOCAL;
5039
5040         bond_dev->hw_features = BOND_VLAN_FEATURES |
5041                                 NETIF_F_HW_VLAN_CTAG_RX |
5042                                 NETIF_F_HW_VLAN_CTAG_FILTER;
5043
5044         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5045         bond_dev->features |= bond_dev->hw_features;
5046         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5047 #ifdef CONFIG_XFRM_OFFLOAD
5048         bond_dev->hw_features |= BOND_XFRM_FEATURES;
5049         /* Only enable XFRM features if this is an active-backup config */
5050         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5051                 bond_dev->features |= BOND_XFRM_FEATURES;
5052 #endif /* CONFIG_XFRM_OFFLOAD */
5053 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5054         if (bond_sk_check(bond))
5055                 bond_dev->features |= BOND_TLS_FEATURES;
5056 #endif
5057 }
5058
5059 /* Destroy a bonding device.
5060  * Must be under rtnl_lock when this function is called.
5061  */
5062 static void bond_uninit(struct net_device *bond_dev)
5063 {
5064         struct bonding *bond = netdev_priv(bond_dev);
5065         struct bond_up_slave *usable, *all;
5066         struct list_head *iter;
5067         struct slave *slave;
5068
5069         bond_netpoll_cleanup(bond_dev);
5070
5071         /* Release the bonded slaves */
5072         bond_for_each_slave(bond, slave, iter)
5073                 __bond_release_one(bond_dev, slave->dev, true, true);
5074         netdev_info(bond_dev, "Released all slaves\n");
5075
5076         usable = rtnl_dereference(bond->usable_slaves);
5077         if (usable) {
5078                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5079                 kfree_rcu(usable, rcu);
5080         }
5081
5082         all = rtnl_dereference(bond->all_slaves);
5083         if (all) {
5084                 RCU_INIT_POINTER(bond->all_slaves, NULL);
5085                 kfree_rcu(all, rcu);
5086         }
5087
5088         list_del(&bond->bond_list);
5089
5090         bond_debug_unregister(bond);
5091 }
5092
5093 /*------------------------- Module initialization ---------------------------*/
5094
5095 static int bond_check_params(struct bond_params *params)
5096 {
5097         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5098         struct bond_opt_value newval;
5099         const struct bond_opt_value *valptr;
5100         int arp_all_targets_value = 0;
5101         u16 ad_actor_sys_prio = 0;
5102         u16 ad_user_port_key = 0;
5103         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5104         int arp_ip_count;
5105         int bond_mode   = BOND_MODE_ROUNDROBIN;
5106         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5107         int lacp_fast = 0;
5108         int tlb_dynamic_lb;
5109
5110         /* Convert string parameters. */
5111         if (mode) {
5112                 bond_opt_initstr(&newval, mode);
5113                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5114                 if (!valptr) {
5115                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5116                         return -EINVAL;
5117                 }
5118                 bond_mode = valptr->value;
5119         }
5120
5121         if (xmit_hash_policy) {
5122                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5123                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
5124                     bond_mode == BOND_MODE_BROADCAST) {
5125                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5126                                 bond_mode_name(bond_mode));
5127                 } else {
5128                         bond_opt_initstr(&newval, xmit_hash_policy);
5129                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5130                                                 &newval);
5131                         if (!valptr) {
5132                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5133                                        xmit_hash_policy);
5134                                 return -EINVAL;
5135                         }
5136                         xmit_hashtype = valptr->value;
5137                 }
5138         }
5139
5140         if (lacp_rate) {
5141                 if (bond_mode != BOND_MODE_8023AD) {
5142                         pr_info("lacp_rate param is irrelevant in mode %s\n",
5143                                 bond_mode_name(bond_mode));
5144                 } else {
5145                         bond_opt_initstr(&newval, lacp_rate);
5146                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5147                                                 &newval);
5148                         if (!valptr) {
5149                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
5150                                        lacp_rate);
5151                                 return -EINVAL;
5152                         }
5153                         lacp_fast = valptr->value;
5154                 }
5155         }
5156
5157         if (ad_select) {
5158                 bond_opt_initstr(&newval, ad_select);
5159                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
5160                                         &newval);
5161                 if (!valptr) {
5162                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
5163                         return -EINVAL;
5164                 }
5165                 params->ad_select = valptr->value;
5166                 if (bond_mode != BOND_MODE_8023AD)
5167                         pr_warn("ad_select param only affects 802.3ad mode\n");
5168         } else {
5169                 params->ad_select = BOND_AD_STABLE;
5170         }
5171
5172         if (max_bonds < 0) {
5173                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
5174                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
5175                 max_bonds = BOND_DEFAULT_MAX_BONDS;
5176         }
5177
5178         if (miimon < 0) {
5179                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5180                         miimon, INT_MAX);
5181                 miimon = 0;
5182         }
5183
5184         if (updelay < 0) {
5185                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5186                         updelay, INT_MAX);
5187                 updelay = 0;
5188         }
5189
5190         if (downdelay < 0) {
5191                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5192                         downdelay, INT_MAX);
5193                 downdelay = 0;
5194         }
5195
5196         if ((use_carrier != 0) && (use_carrier != 1)) {
5197                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5198                         use_carrier);
5199                 use_carrier = 1;
5200         }
5201
5202         if (num_peer_notif < 0 || num_peer_notif > 255) {
5203                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5204                         num_peer_notif);
5205                 num_peer_notif = 1;
5206         }
5207
5208         /* reset values for 802.3ad/TLB/ALB */
5209         if (!bond_mode_uses_arp(bond_mode)) {
5210                 if (!miimon) {
5211                         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");
5212                         pr_warn("Forcing miimon to 100msec\n");
5213                         miimon = BOND_DEFAULT_MIIMON;
5214                 }
5215         }
5216
5217         if (tx_queues < 1 || tx_queues > 255) {
5218                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
5219                         tx_queues, BOND_DEFAULT_TX_QUEUES);
5220                 tx_queues = BOND_DEFAULT_TX_QUEUES;
5221         }
5222
5223         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
5224                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
5225                         all_slaves_active);
5226                 all_slaves_active = 0;
5227         }
5228
5229         if (resend_igmp < 0 || resend_igmp > 255) {
5230                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
5231                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
5232                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
5233         }
5234
5235         bond_opt_initval(&newval, packets_per_slave);
5236         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
5237                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
5238                         packets_per_slave, USHRT_MAX);
5239                 packets_per_slave = 1;
5240         }
5241
5242         if (bond_mode == BOND_MODE_ALB) {
5243                 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",
5244                           updelay);
5245         }
5246
5247         if (!miimon) {
5248                 if (updelay || downdelay) {
5249                         /* just warn the user the up/down delay will have
5250                          * no effect since miimon is zero...
5251                          */
5252                         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",
5253                                 updelay, downdelay);
5254                 }
5255         } else {
5256                 /* don't allow arp monitoring */
5257                 if (arp_interval) {
5258                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
5259                                 miimon, arp_interval);
5260                         arp_interval = 0;
5261                 }
5262
5263                 if ((updelay % miimon) != 0) {
5264                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5265                                 updelay, miimon, (updelay / miimon) * miimon);
5266                 }
5267
5268                 updelay /= miimon;
5269
5270                 if ((downdelay % miimon) != 0) {
5271                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5272                                 downdelay, miimon,
5273                                 (downdelay / miimon) * miimon);
5274                 }
5275
5276                 downdelay /= miimon;
5277         }
5278
5279         if (arp_interval < 0) {
5280                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5281                         arp_interval, INT_MAX);
5282                 arp_interval = 0;
5283         }
5284
5285         for (arp_ip_count = 0, i = 0;
5286              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
5287                 __be32 ip;
5288
5289                 /* not a complete check, but good enough to catch mistakes */
5290                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
5291                     !bond_is_ip_target_ok(ip)) {
5292                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5293                                 arp_ip_target[i]);
5294                         arp_interval = 0;
5295                 } else {
5296                         if (bond_get_targets_ip(arp_target, ip) == -1)
5297                                 arp_target[arp_ip_count++] = ip;
5298                         else
5299                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
5300                                         &ip);
5301                 }
5302         }
5303
5304         if (arp_interval && !arp_ip_count) {
5305                 /* don't allow arping if no arp_ip_target given... */
5306                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5307                         arp_interval);
5308                 arp_interval = 0;
5309         }
5310
5311         if (arp_validate) {
5312                 if (!arp_interval) {
5313                         pr_err("arp_validate requires arp_interval\n");
5314                         return -EINVAL;
5315                 }
5316
5317                 bond_opt_initstr(&newval, arp_validate);
5318                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
5319                                         &newval);
5320                 if (!valptr) {
5321                         pr_err("Error: invalid arp_validate \"%s\"\n",
5322                                arp_validate);
5323                         return -EINVAL;
5324                 }
5325                 arp_validate_value = valptr->value;
5326         } else {
5327                 arp_validate_value = 0;
5328         }
5329
5330         if (arp_all_targets) {
5331                 bond_opt_initstr(&newval, arp_all_targets);
5332                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
5333                                         &newval);
5334                 if (!valptr) {
5335                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
5336                                arp_all_targets);
5337                         arp_all_targets_value = 0;
5338                 } else {
5339                         arp_all_targets_value = valptr->value;
5340                 }
5341         }
5342
5343         if (miimon) {
5344                 pr_info("MII link monitoring set to %d ms\n", miimon);
5345         } else if (arp_interval) {
5346                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
5347                                           arp_validate_value);
5348                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5349                         arp_interval, valptr->string, arp_ip_count);
5350
5351                 for (i = 0; i < arp_ip_count; i++)
5352                         pr_cont(" %s", arp_ip_target[i]);
5353
5354                 pr_cont("\n");
5355
5356         } else if (max_bonds) {
5357                 /* miimon and arp_interval not set, we need one so things
5358                  * work as expected, see bonding.txt for details
5359                  */
5360                 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");
5361         }
5362
5363         if (primary && !bond_mode_uses_primary(bond_mode)) {
5364                 /* currently, using a primary only makes sense
5365                  * in active backup, TLB or ALB modes
5366                  */
5367                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
5368                         primary, bond_mode_name(bond_mode));
5369                 primary = NULL;
5370         }
5371
5372         if (primary && primary_reselect) {
5373                 bond_opt_initstr(&newval, primary_reselect);
5374                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
5375                                         &newval);
5376                 if (!valptr) {
5377                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5378                                primary_reselect);
5379                         return -EINVAL;
5380                 }
5381                 primary_reselect_value = valptr->value;
5382         } else {
5383                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5384         }
5385
5386         if (fail_over_mac) {
5387                 bond_opt_initstr(&newval, fail_over_mac);
5388                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
5389                                         &newval);
5390                 if (!valptr) {
5391                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5392                                fail_over_mac);
5393                         return -EINVAL;
5394                 }
5395                 fail_over_mac_value = valptr->value;
5396                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5397                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
5398         } else {
5399                 fail_over_mac_value = BOND_FOM_NONE;
5400         }
5401
5402         bond_opt_initstr(&newval, "default");
5403         valptr = bond_opt_parse(
5404                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
5405                                      &newval);
5406         if (!valptr) {
5407                 pr_err("Error: No ad_actor_sys_prio default value");
5408                 return -EINVAL;
5409         }
5410         ad_actor_sys_prio = valptr->value;
5411
5412         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
5413                                 &newval);
5414         if (!valptr) {
5415                 pr_err("Error: No ad_user_port_key default value");
5416                 return -EINVAL;
5417         }
5418         ad_user_port_key = valptr->value;
5419
5420         bond_opt_initstr(&newval, "default");
5421         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
5422         if (!valptr) {
5423                 pr_err("Error: No tlb_dynamic_lb default value");
5424                 return -EINVAL;
5425         }
5426         tlb_dynamic_lb = valptr->value;
5427
5428         if (lp_interval == 0) {
5429                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
5430                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
5431                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
5432         }
5433
5434         /* fill params struct with the proper values */
5435         params->mode = bond_mode;
5436         params->xmit_policy = xmit_hashtype;
5437         params->miimon = miimon;
5438         params->num_peer_notif = num_peer_notif;
5439         params->arp_interval = arp_interval;
5440         params->arp_validate = arp_validate_value;
5441         params->arp_all_targets = arp_all_targets_value;
5442         params->updelay = updelay;
5443         params->downdelay = downdelay;
5444         params->peer_notif_delay = 0;
5445         params->use_carrier = use_carrier;
5446         params->lacp_fast = lacp_fast;
5447         params->primary[0] = 0;
5448         params->primary_reselect = primary_reselect_value;
5449         params->fail_over_mac = fail_over_mac_value;
5450         params->tx_queues = tx_queues;
5451         params->all_slaves_active = all_slaves_active;
5452         params->resend_igmp = resend_igmp;
5453         params->min_links = min_links;
5454         params->lp_interval = lp_interval;
5455         params->packets_per_slave = packets_per_slave;
5456         params->tlb_dynamic_lb = tlb_dynamic_lb;
5457         params->ad_actor_sys_prio = ad_actor_sys_prio;
5458         eth_zero_addr(params->ad_actor_system);
5459         params->ad_user_port_key = ad_user_port_key;
5460         if (packets_per_slave > 0) {
5461                 params->reciprocal_packets_per_slave =
5462                         reciprocal_value(packets_per_slave);
5463         } else {
5464                 /* reciprocal_packets_per_slave is unused if
5465                  * packets_per_slave is 0 or 1, just initialize it
5466                  */
5467                 params->reciprocal_packets_per_slave =
5468                         (struct reciprocal_value) { 0 };
5469         }
5470
5471         if (primary)
5472                 strscpy_pad(params->primary, primary, sizeof(params->primary));
5473
5474         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5475
5476         return 0;
5477 }
5478
5479 /* Called from registration process */
5480 static int bond_init(struct net_device *bond_dev)
5481 {
5482         struct bonding *bond = netdev_priv(bond_dev);
5483         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5484
5485         netdev_dbg(bond_dev, "Begin bond_init\n");
5486
5487         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
5488         if (!bond->wq)
5489                 return -ENOMEM;
5490
5491         if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN) {
5492                 bond->rr_tx_counter = alloc_percpu(u32);
5493                 if (!bond->rr_tx_counter) {
5494                         destroy_workqueue(bond->wq);
5495                         bond->wq = NULL;
5496                         return -ENOMEM;
5497                 }
5498         }
5499
5500         spin_lock_init(&bond->stats_lock);
5501         netdev_lockdep_set_classes(bond_dev);
5502
5503         list_add_tail(&bond->bond_list, &bn->dev_list);
5504
5505         bond_prepare_sysfs_group(bond);
5506
5507         bond_debug_register(bond);
5508
5509         /* Ensure valid dev_addr */
5510         if (is_zero_ether_addr(bond_dev->dev_addr) &&
5511             bond_dev->addr_assign_type == NET_ADDR_PERM)
5512                 eth_hw_addr_random(bond_dev);
5513
5514         return 0;
5515 }
5516
5517 unsigned int bond_get_num_tx_queues(void)
5518 {
5519         return tx_queues;
5520 }
5521
5522 /* Create a new bond based on the specified name and bonding parameters.
5523  * If name is NULL, obtain a suitable "bond%d" name for us.
5524  * Caller must NOT hold rtnl_lock; we need to release it here before we
5525  * set up our sysfs entries.
5526  */
5527 int bond_create(struct net *net, const char *name)
5528 {
5529         struct net_device *bond_dev;
5530         struct bonding *bond;
5531         struct alb_bond_info *bond_info;
5532         int res;
5533
5534         rtnl_lock();
5535
5536         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
5537                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
5538                                    bond_setup, tx_queues);
5539         if (!bond_dev) {
5540                 pr_err("%s: eek! can't alloc netdev!\n", name);
5541                 rtnl_unlock();
5542                 return -ENOMEM;
5543         }
5544
5545         /*
5546          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
5547          * It is set to 0 by default which is wrong.
5548          */
5549         bond = netdev_priv(bond_dev);
5550         bond_info = &(BOND_ALB_INFO(bond));
5551         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
5552
5553         dev_net_set(bond_dev, net);
5554         bond_dev->rtnl_link_ops = &bond_link_ops;
5555
5556         res = register_netdevice(bond_dev);
5557         if (res < 0) {
5558                 free_netdev(bond_dev);
5559                 rtnl_unlock();
5560
5561                 return res;
5562         }
5563
5564         netif_carrier_off(bond_dev);
5565
5566         bond_work_init_all(bond);
5567
5568         rtnl_unlock();
5569         return 0;
5570 }
5571
5572 static int __net_init bond_net_init(struct net *net)
5573 {
5574         struct bond_net *bn = net_generic(net, bond_net_id);
5575
5576         bn->net = net;
5577         INIT_LIST_HEAD(&bn->dev_list);
5578
5579         bond_create_proc_dir(bn);
5580         bond_create_sysfs(bn);
5581
5582         return 0;
5583 }
5584
5585 static void __net_exit bond_net_exit(struct net *net)
5586 {
5587         struct bond_net *bn = net_generic(net, bond_net_id);
5588         struct bonding *bond, *tmp_bond;
5589         LIST_HEAD(list);
5590
5591         bond_destroy_sysfs(bn);
5592
5593         /* Kill off any bonds created after unregistering bond rtnl ops */
5594         rtnl_lock();
5595         list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
5596                 unregister_netdevice_queue(bond->dev, &list);
5597         unregister_netdevice_many(&list);
5598         rtnl_unlock();
5599
5600         bond_destroy_proc_dir(bn);
5601 }
5602
5603 static struct pernet_operations bond_net_ops = {
5604         .init = bond_net_init,
5605         .exit = bond_net_exit,
5606         .id   = &bond_net_id,
5607         .size = sizeof(struct bond_net),
5608 };
5609
5610 static int __init bonding_init(void)
5611 {
5612         int i;
5613         int res;
5614
5615         res = bond_check_params(&bonding_defaults);
5616         if (res)
5617                 goto out;
5618
5619         res = register_pernet_subsys(&bond_net_ops);
5620         if (res)
5621                 goto out;
5622
5623         res = bond_netlink_init();
5624         if (res)
5625                 goto err_link;
5626
5627         bond_create_debugfs();
5628
5629         for (i = 0; i < max_bonds; i++) {
5630                 res = bond_create(&init_net, NULL);
5631                 if (res)
5632                         goto err;
5633         }
5634
5635         skb_flow_dissector_init(&flow_keys_bonding,
5636                                 flow_keys_bonding_keys,
5637                                 ARRAY_SIZE(flow_keys_bonding_keys));
5638
5639         register_netdevice_notifier(&bond_netdev_notifier);
5640 out:
5641         return res;
5642 err:
5643         bond_destroy_debugfs();
5644         bond_netlink_fini();
5645 err_link:
5646         unregister_pernet_subsys(&bond_net_ops);
5647         goto out;
5648
5649 }
5650
5651 static void __exit bonding_exit(void)
5652 {
5653         unregister_netdevice_notifier(&bond_netdev_notifier);
5654
5655         bond_destroy_debugfs();
5656
5657         bond_netlink_fini();
5658         unregister_pernet_subsys(&bond_net_ops);
5659
5660 #ifdef CONFIG_NET_POLL_CONTROLLER
5661         /* Make sure we don't have an imbalance on our netpoll blocking */
5662         WARN_ON(atomic_read(&netpoll_block_tx));
5663 #endif
5664 }
5665
5666 module_init(bonding_init);
5667 module_exit(bonding_exit);
5668 MODULE_LICENSE("GPL");
5669 MODULE_DESCRIPTION(DRV_DESCRIPTION);
5670 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");