linux/drivers/net/bonding/bond_main.c
<<
>>
Prefs
   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
  96static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
  97static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
  98static int num_peer_notif = 1;
  99static int miimon;
 100static int updelay;
 101static int downdelay;
 102static int use_carrier  = 1;
 103static char *mode;
 104static char *primary;
 105static char *primary_reselect;
 106static char *lacp_rate;
 107static int min_links;
 108static char *ad_select;
 109static char *xmit_hash_policy;
 110static int arp_interval;
 111static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 112static char *arp_validate;
 113static char *arp_all_targets;
 114static char *fail_over_mac;
 115static int all_slaves_active;
 116static struct bond_params bonding_defaults;
 117static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 118static int packets_per_slave = 1;
 119static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 120
 121module_param(max_bonds, int, 0);
 122MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 123module_param(tx_queues, int, 0);
 124MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 125module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 126MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 127                               "failover event (alias of num_unsol_na)");
 128module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 129MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 130                               "failover event (alias of num_grat_arp)");
 131module_param(miimon, int, 0);
 132MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 133module_param(updelay, int, 0);
 134MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 135module_param(downdelay, int, 0);
 136MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 137                            "in milliseconds");
 138module_param(use_carrier, int, 0);
 139MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 140                              "0 for off, 1 for on (default)");
 141module_param(mode, charp, 0);
 142MODULE_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");
 146module_param(primary, charp, 0);
 147MODULE_PARM_DESC(primary, "Primary network device to use");
 148module_param(primary_reselect, charp, 0);
 149MODULE_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");
 156module_param(lacp_rate, charp, 0);
 157MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 158                            "0 for slow, 1 for fast");
 159module_param(ad_select, charp, 0);
 160MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
 161                            "0 for stable (default), 1 for bandwidth, "
 162                            "2 for count");
 163module_param(min_links, int, 0);
 164MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 165
 166module_param(xmit_hash_policy, charp, 0);
 167MODULE_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");
 171module_param(arp_interval, int, 0);
 172MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 173module_param_array(arp_ip_target, charp, NULL, 0);
 174MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 175module_param(arp_validate, charp, 0);
 176MODULE_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");
 179module_param(arp_all_targets, charp, 0);
 180MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 181module_param(fail_over_mac, charp, 0);
 182MODULE_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");
 185module_param(all_slaves_active, int, 0);
 186MODULE_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.");
 189module_param(resend_igmp, int, 0);
 190MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 191                              "link failure");
 192module_param(packets_per_slave, int, 0);
 193MODULE_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.");
 196module_param(lp_interval, uint, 0);
 197MODULE_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
 204atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 205#endif
 206
 207unsigned int bond_net_id __read_mostly;
 208
 209static 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
 252static struct flow_dissector flow_keys_bonding __read_mostly;
 253
 254/*-------------------------- Forward declarations ---------------------------*/
 255
 256static int bond_init(struct net_device *bond_dev);
 257static void bond_uninit(struct net_device *bond_dev);
 258static void bond_get_stats(struct net_device *bond_dev,
 259                           struct rtnl_link_stats64 *stats);
 260static void bond_slave_arr_handler(struct work_struct *work);
 261static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 262                                  int mod);
 263static void bond_netdev_notify_work(struct work_struct *work);
 264
 265/*---------------------------- General routines -----------------------------*/
 266
 267const 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 */
 292netdev_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
 307bool 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
 320static bool bond_xdp_check(struct bonding *bond)
 321{
 322        switch (BOND_MODE(bond)) {
 323        case BOND_MODE_ROUNDROBIN:
 324        case BOND_MODE_ACTIVEBACKUP:
 325                return true;
 326        case BOND_MODE_8023AD:
 327        case BOND_MODE_XOR:
 328                /* vlan+srcmac is not supported with XDP as in most cases the 802.1q
 329                 * payload is not in the packet due to hardware offload.
 330                 */
 331                if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC)
 332                        return true;
 333                fallthrough;
 334        default:
 335                return false;
 336        }
 337}
 338
 339/*---------------------------------- VLAN -----------------------------------*/
 340
 341/* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 342 * We don't protect the slave list iteration with a lock because:
 343 * a. This operation is performed in IOCTL context,
 344 * b. The operation is protected by the RTNL semaphore in the 8021q code,
 345 * c. Holding a lock with BH disabled while directly calling a base driver
 346 *    entry point is generally a BAD idea.
 347 *
 348 * The design of synchronization/protection for this operation in the 8021q
 349 * module is good for one or more VLAN devices over a single physical device
 350 * and cannot be extended for a teaming solution like bonding, so there is a
 351 * potential race condition here where a net device from the vlan group might
 352 * be referenced (either by a base driver or the 8021q code) while it is being
 353 * removed from the system. However, it turns out we're not making matters
 354 * worse, and if it works for regular VLAN usage it will work here too.
 355*/
 356
 357/**
 358 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 359 * @bond_dev: bonding net device that got called
 360 * @proto: network protocol ID
 361 * @vid: vlan id being added
 362 */
 363static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 364                                __be16 proto, u16 vid)
 365{
 366        struct bonding *bond = netdev_priv(bond_dev);
 367        struct slave *slave, *rollback_slave;
 368        struct list_head *iter;
 369        int res;
 370
 371        bond_for_each_slave(bond, slave, iter) {
 372                res = vlan_vid_add(slave->dev, proto, vid);
 373                if (res)
 374                        goto unwind;
 375        }
 376
 377        return 0;
 378
 379unwind:
 380        /* unwind to the slave that failed */
 381        bond_for_each_slave(bond, rollback_slave, iter) {
 382                if (rollback_slave == slave)
 383                        break;
 384
 385                vlan_vid_del(rollback_slave->dev, proto, vid);
 386        }
 387
 388        return res;
 389}
 390
 391/**
 392 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 393 * @bond_dev: bonding net device that got called
 394 * @proto: network protocol ID
 395 * @vid: vlan id being removed
 396 */
 397static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 398                                 __be16 proto, u16 vid)
 399{
 400        struct bonding *bond = netdev_priv(bond_dev);
 401        struct list_head *iter;
 402        struct slave *slave;
 403
 404        bond_for_each_slave(bond, slave, iter)
 405                vlan_vid_del(slave->dev, proto, vid);
 406
 407        if (bond_is_lb(bond))
 408                bond_alb_clear_vlan(bond, vid);
 409
 410        return 0;
 411}
 412
 413/*---------------------------------- XFRM -----------------------------------*/
 414
 415#ifdef CONFIG_XFRM_OFFLOAD
 416/**
 417 * bond_ipsec_add_sa - program device with a security association
 418 * @xs: pointer to transformer state struct
 419 **/
 420static int bond_ipsec_add_sa(struct xfrm_state *xs)
 421{
 422        struct net_device *bond_dev = xs->xso.dev;
 423        struct bond_ipsec *ipsec;
 424        struct bonding *bond;
 425        struct slave *slave;
 426        int err;
 427
 428        if (!bond_dev)
 429                return -EINVAL;
 430
 431        rcu_read_lock();
 432        bond = netdev_priv(bond_dev);
 433        slave = rcu_dereference(bond->curr_active_slave);
 434        if (!slave) {
 435                rcu_read_unlock();
 436                return -ENODEV;
 437        }
 438
 439        if (!slave->dev->xfrmdev_ops ||
 440            !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
 441            netif_is_bond_master(slave->dev)) {
 442                slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
 443                rcu_read_unlock();
 444                return -EINVAL;
 445        }
 446
 447        ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
 448        if (!ipsec) {
 449                rcu_read_unlock();
 450                return -ENOMEM;
 451        }
 452        xs->xso.real_dev = slave->dev;
 453
 454        err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
 455        if (!err) {
 456                ipsec->xs = xs;
 457                INIT_LIST_HEAD(&ipsec->list);
 458                spin_lock_bh(&bond->ipsec_lock);
 459                list_add(&ipsec->list, &bond->ipsec_list);
 460                spin_unlock_bh(&bond->ipsec_lock);
 461        } else {
 462                kfree(ipsec);
 463        }
 464        rcu_read_unlock();
 465        return err;
 466}
 467
 468static void bond_ipsec_add_sa_all(struct bonding *bond)
 469{
 470        struct net_device *bond_dev = bond->dev;
 471        struct bond_ipsec *ipsec;
 472        struct slave *slave;
 473
 474        rcu_read_lock();
 475        slave = rcu_dereference(bond->curr_active_slave);
 476        if (!slave)
 477                goto out;
 478
 479        if (!slave->dev->xfrmdev_ops ||
 480            !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
 481            netif_is_bond_master(slave->dev)) {
 482                spin_lock_bh(&bond->ipsec_lock);
 483                if (!list_empty(&bond->ipsec_list))
 484                        slave_warn(bond_dev, slave->dev,
 485                                   "%s: no slave xdo_dev_state_add\n",
 486                                   __func__);
 487                spin_unlock_bh(&bond->ipsec_lock);
 488                goto out;
 489        }
 490
 491        spin_lock_bh(&bond->ipsec_lock);
 492        list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 493                ipsec->xs->xso.real_dev = slave->dev;
 494                if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) {
 495                        slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
 496                        ipsec->xs->xso.real_dev = NULL;
 497                }
 498        }
 499        spin_unlock_bh(&bond->ipsec_lock);
 500out:
 501        rcu_read_unlock();
 502}
 503
 504/**
 505 * bond_ipsec_del_sa - clear out this specific SA
 506 * @xs: pointer to transformer state struct
 507 **/
 508static void bond_ipsec_del_sa(struct xfrm_state *xs)
 509{
 510        struct net_device *bond_dev = xs->xso.dev;
 511        struct bond_ipsec *ipsec;
 512        struct bonding *bond;
 513        struct slave *slave;
 514
 515        if (!bond_dev)
 516                return;
 517
 518        rcu_read_lock();
 519        bond = netdev_priv(bond_dev);
 520        slave = rcu_dereference(bond->curr_active_slave);
 521
 522        if (!slave)
 523                goto out;
 524
 525        if (!xs->xso.real_dev)
 526                goto out;
 527
 528        WARN_ON(xs->xso.real_dev != slave->dev);
 529
 530        if (!slave->dev->xfrmdev_ops ||
 531            !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
 532            netif_is_bond_master(slave->dev)) {
 533                slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
 534                goto out;
 535        }
 536
 537        slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
 538out:
 539        spin_lock_bh(&bond->ipsec_lock);
 540        list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 541                if (ipsec->xs == xs) {
 542                        list_del(&ipsec->list);
 543                        kfree(ipsec);
 544                        break;
 545                }
 546        }
 547        spin_unlock_bh(&bond->ipsec_lock);
 548        rcu_read_unlock();
 549}
 550
 551static void bond_ipsec_del_sa_all(struct bonding *bond)
 552{
 553        struct net_device *bond_dev = bond->dev;
 554        struct bond_ipsec *ipsec;
 555        struct slave *slave;
 556
 557        rcu_read_lock();
 558        slave = rcu_dereference(bond->curr_active_slave);
 559        if (!slave) {
 560                rcu_read_unlock();
 561                return;
 562        }
 563
 564        spin_lock_bh(&bond->ipsec_lock);
 565        list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 566                if (!ipsec->xs->xso.real_dev)
 567                        continue;
 568
 569                if (!slave->dev->xfrmdev_ops ||
 570                    !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
 571                    netif_is_bond_master(slave->dev)) {
 572                        slave_warn(bond_dev, slave->dev,
 573                                   "%s: no slave xdo_dev_state_delete\n",
 574                                   __func__);
 575                } else {
 576                        slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
 577                }
 578                ipsec->xs->xso.real_dev = NULL;
 579        }
 580        spin_unlock_bh(&bond->ipsec_lock);
 581        rcu_read_unlock();
 582}
 583
 584/**
 585 * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
 586 * @skb: current data packet
 587 * @xs: pointer to transformer state struct
 588 **/
 589static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
 590{
 591        struct net_device *bond_dev = xs->xso.dev;
 592        struct net_device *real_dev;
 593        struct slave *curr_active;
 594        struct bonding *bond;
 595        int err;
 596
 597        bond = netdev_priv(bond_dev);
 598        rcu_read_lock();
 599        curr_active = rcu_dereference(bond->curr_active_slave);
 600        real_dev = curr_active->dev;
 601
 602        if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
 603                err = false;
 604                goto out;
 605        }
 606
 607        if (!xs->xso.real_dev) {
 608                err = false;
 609                goto out;
 610        }
 611
 612        if (!real_dev->xfrmdev_ops ||
 613            !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
 614            netif_is_bond_master(real_dev)) {
 615                err = false;
 616                goto out;
 617        }
 618
 619        err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
 620out:
 621        rcu_read_unlock();
 622        return err;
 623}
 624
 625static const struct xfrmdev_ops bond_xfrmdev_ops = {
 626        .xdo_dev_state_add = bond_ipsec_add_sa,
 627        .xdo_dev_state_delete = bond_ipsec_del_sa,
 628        .xdo_dev_offload_ok = bond_ipsec_offload_ok,
 629};
 630#endif /* CONFIG_XFRM_OFFLOAD */
 631
 632/*------------------------------- Link status -------------------------------*/
 633
 634/* Set the carrier state for the master according to the state of its
 635 * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 636 * do special 802.3ad magic.
 637 *
 638 * Returns zero if carrier state does not change, nonzero if it does.
 639 */
 640int bond_set_carrier(struct bonding *bond)
 641{
 642        struct list_head *iter;
 643        struct slave *slave;
 644
 645        if (!bond_has_slaves(bond))
 646                goto down;
 647
 648        if (BOND_MODE(bond) == BOND_MODE_8023AD)
 649                return bond_3ad_set_carrier(bond);
 650
 651        bond_for_each_slave(bond, slave, iter) {
 652                if (slave->link == BOND_LINK_UP) {
 653                        if (!netif_carrier_ok(bond->dev)) {
 654                                netif_carrier_on(bond->dev);
 655                                return 1;
 656                        }
 657                        return 0;
 658                }
 659        }
 660
 661down:
 662        if (netif_carrier_ok(bond->dev)) {
 663                netif_carrier_off(bond->dev);
 664                return 1;
 665        }
 666        return 0;
 667}
 668
 669/* Get link speed and duplex from the slave's base driver
 670 * using ethtool. If for some reason the call fails or the
 671 * values are invalid, set speed and duplex to -1,
 672 * and return. Return 1 if speed or duplex settings are
 673 * UNKNOWN; 0 otherwise.
 674 */
 675static int bond_update_speed_duplex(struct slave *slave)
 676{
 677        struct net_device *slave_dev = slave->dev;
 678        struct ethtool_link_ksettings ecmd;
 679        int res;
 680
 681        slave->speed = SPEED_UNKNOWN;
 682        slave->duplex = DUPLEX_UNKNOWN;
 683
 684        res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 685        if (res < 0)
 686                return 1;
 687        if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 688                return 1;
 689        switch (ecmd.base.duplex) {
 690        case DUPLEX_FULL:
 691        case DUPLEX_HALF:
 692                break;
 693        default:
 694                return 1;
 695        }
 696
 697        slave->speed = ecmd.base.speed;
 698        slave->duplex = ecmd.base.duplex;
 699
 700        return 0;
 701}
 702
 703const char *bond_slave_link_status(s8 link)
 704{
 705        switch (link) {
 706        case BOND_LINK_UP:
 707                return "up";
 708        case BOND_LINK_FAIL:
 709                return "going down";
 710        case BOND_LINK_DOWN:
 711                return "down";
 712        case BOND_LINK_BACK:
 713                return "going back";
 714        default:
 715                return "unknown";
 716        }
 717}
 718
 719/* if <dev> supports MII link status reporting, check its link status.
 720 *
 721 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 722 * depending upon the setting of the use_carrier parameter.
 723 *
 724 * Return either BMSR_LSTATUS, meaning that the link is up (or we
 725 * can't tell and just pretend it is), or 0, meaning that the link is
 726 * down.
 727 *
 728 * If reporting is non-zero, instead of faking link up, return -1 if
 729 * both ETHTOOL and MII ioctls fail (meaning the device does not
 730 * support them).  If use_carrier is set, return whatever it says.
 731 * It'd be nice if there was a good way to tell if a driver supports
 732 * netif_carrier, but there really isn't.
 733 */
 734static int bond_check_dev_link(struct bonding *bond,
 735                               struct net_device *slave_dev, int reporting)
 736{
 737        const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 738        int (*ioctl)(struct net_device *, struct ifreq *, int);
 739        struct ifreq ifr;
 740        struct mii_ioctl_data *mii;
 741
 742        if (!reporting && !netif_running(slave_dev))
 743                return 0;
 744
 745        if (bond->params.use_carrier)
 746                return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 747
 748        /* Try to get link status using Ethtool first. */
 749        if (slave_dev->ethtool_ops->get_link)
 750                return slave_dev->ethtool_ops->get_link(slave_dev) ?
 751                        BMSR_LSTATUS : 0;
 752
 753        /* Ethtool can't be used, fallback to MII ioctls. */
 754        ioctl = slave_ops->ndo_eth_ioctl;
 755        if (ioctl) {
 756                /* TODO: set pointer to correct ioctl on a per team member
 757                 *       bases to make this more efficient. that is, once
 758                 *       we determine the correct ioctl, we will always
 759                 *       call it and not the others for that team
 760                 *       member.
 761                 */
 762
 763                /* We cannot assume that SIOCGMIIPHY will also read a
 764                 * register; not all network drivers (e.g., e100)
 765                 * support that.
 766                 */
 767
 768                /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 769                strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 770                mii = if_mii(&ifr);
 771                if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 772                        mii->reg_num = MII_BMSR;
 773                        if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 774                                return mii->val_out & BMSR_LSTATUS;
 775                }
 776        }
 777
 778        /* If reporting, report that either there's no ndo_eth_ioctl,
 779         * or both SIOCGMIIREG and get_link failed (meaning that we
 780         * cannot report link status).  If not reporting, pretend
 781         * we're ok.
 782         */
 783        return reporting ? -1 : BMSR_LSTATUS;
 784}
 785
 786/*----------------------------- Multicast list ------------------------------*/
 787
 788/* Push the promiscuity flag down to appropriate slaves */
 789static int bond_set_promiscuity(struct bonding *bond, int inc)
 790{
 791        struct list_head *iter;
 792        int err = 0;
 793
 794        if (bond_uses_primary(bond)) {
 795                struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 796
 797                if (curr_active)
 798                        err = dev_set_promiscuity(curr_active->dev, inc);
 799        } else {
 800                struct slave *slave;
 801
 802                bond_for_each_slave(bond, slave, iter) {
 803                        err = dev_set_promiscuity(slave->dev, inc);
 804                        if (err)
 805                                return err;
 806                }
 807        }
 808        return err;
 809}
 810
 811/* Push the allmulti flag down to all slaves */
 812static int bond_set_allmulti(struct bonding *bond, int inc)
 813{
 814        struct list_head *iter;
 815        int err = 0;
 816
 817        if (bond_uses_primary(bond)) {
 818                struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 819
 820                if (curr_active)
 821                        err = dev_set_allmulti(curr_active->dev, inc);
 822        } else {
 823                struct slave *slave;
 824
 825                bond_for_each_slave(bond, slave, iter) {
 826                        err = dev_set_allmulti(slave->dev, inc);
 827                        if (err)
 828                                return err;
 829                }
 830        }
 831        return err;
 832}
 833
 834/* Retrieve the list of registered multicast addresses for the bonding
 835 * device and retransmit an IGMP JOIN request to the current active
 836 * slave.
 837 */
 838static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 839{
 840        struct bonding *bond = container_of(work, struct bonding,
 841                                            mcast_work.work);
 842
 843        if (!rtnl_trylock()) {
 844                queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 845                return;
 846        }
 847        call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 848
 849        if (bond->igmp_retrans > 1) {
 850                bond->igmp_retrans--;
 851                queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 852        }
 853        rtnl_unlock();
 854}
 855
 856/* Flush bond's hardware addresses from slave */
 857static void bond_hw_addr_flush(struct net_device *bond_dev,
 858                               struct net_device *slave_dev)
 859{
 860        struct bonding *bond = netdev_priv(bond_dev);
 861
 862        dev_uc_unsync(slave_dev, bond_dev);
 863        dev_mc_unsync(slave_dev, bond_dev);
 864
 865        if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 866                /* del lacpdu mc addr from mc list */
 867                u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 868
 869                dev_mc_del(slave_dev, lacpdu_multicast);
 870        }
 871}
 872
 873/*--------------------------- Active slave change ---------------------------*/
 874
 875/* Update the hardware address list and promisc/allmulti for the new and
 876 * old active slaves (if any).  Modes that are not using primary keep all
 877 * slaves up date at all times; only the modes that use primary need to call
 878 * this function to swap these settings during a failover.
 879 */
 880static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 881                              struct slave *old_active)
 882{
 883        if (old_active) {
 884                if (bond->dev->flags & IFF_PROMISC)
 885                        dev_set_promiscuity(old_active->dev, -1);
 886
 887                if (bond->dev->flags & IFF_ALLMULTI)
 888                        dev_set_allmulti(old_active->dev, -1);
 889
 890                bond_hw_addr_flush(bond->dev, old_active->dev);
 891        }
 892
 893        if (new_active) {
 894                /* FIXME: Signal errors upstream. */
 895                if (bond->dev->flags & IFF_PROMISC)
 896                        dev_set_promiscuity(new_active->dev, 1);
 897
 898                if (bond->dev->flags & IFF_ALLMULTI)
 899                        dev_set_allmulti(new_active->dev, 1);
 900
 901                netif_addr_lock_bh(bond->dev);
 902                dev_uc_sync(new_active->dev, bond->dev);
 903                dev_mc_sync(new_active->dev, bond->dev);
 904                netif_addr_unlock_bh(bond->dev);
 905        }
 906}
 907
 908/**
 909 * bond_set_dev_addr - clone slave's address to bond
 910 * @bond_dev: bond net device
 911 * @slave_dev: slave net device
 912 *
 913 * Should be called with RTNL held.
 914 */
 915static int bond_set_dev_addr(struct net_device *bond_dev,
 916                             struct net_device *slave_dev)
 917{
 918        int err;
 919
 920        slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 921                  bond_dev, slave_dev, slave_dev->addr_len);
 922        err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
 923        if (err)
 924                return err;
 925
 926        memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
 927        bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 928        call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 929        return 0;
 930}
 931
 932static struct slave *bond_get_old_active(struct bonding *bond,
 933                                         struct slave *new_active)
 934{
 935        struct slave *slave;
 936        struct list_head *iter;
 937
 938        bond_for_each_slave(bond, slave, iter) {
 939                if (slave == new_active)
 940                        continue;
 941
 942                if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 943                        return slave;
 944        }
 945
 946        return NULL;
 947}
 948
 949/* bond_do_fail_over_mac
 950 *
 951 * Perform special MAC address swapping for fail_over_mac settings
 952 *
 953 * Called with RTNL
 954 */
 955static void bond_do_fail_over_mac(struct bonding *bond,
 956                                  struct slave *new_active,
 957                                  struct slave *old_active)
 958{
 959        u8 tmp_mac[MAX_ADDR_LEN];
 960        struct sockaddr_storage ss;
 961        int rv;
 962
 963        switch (bond->params.fail_over_mac) {
 964        case BOND_FOM_ACTIVE:
 965                if (new_active) {
 966                        rv = bond_set_dev_addr(bond->dev, new_active->dev);
 967                        if (rv)
 968                                slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
 969                                          -rv);
 970                }
 971                break;
 972        case BOND_FOM_FOLLOW:
 973                /* if new_active && old_active, swap them
 974                 * if just old_active, do nothing (going to no active slave)
 975                 * if just new_active, set new_active to bond's MAC
 976                 */
 977                if (!new_active)
 978                        return;
 979
 980                if (!old_active)
 981                        old_active = bond_get_old_active(bond, new_active);
 982
 983                if (old_active) {
 984                        bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 985                                          new_active->dev->addr_len);
 986                        bond_hw_addr_copy(ss.__data,
 987                                          old_active->dev->dev_addr,
 988                                          old_active->dev->addr_len);
 989                        ss.ss_family = new_active->dev->type;
 990                } else {
 991                        bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 992                                          bond->dev->addr_len);
 993                        ss.ss_family = bond->dev->type;
 994                }
 995
 996                rv = dev_set_mac_address(new_active->dev,
 997                                         (struct sockaddr *)&ss, NULL);
 998                if (rv) {
 999                        slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1000                                  -rv);
1001                        goto out;
1002                }
1003
1004                if (!old_active)
1005                        goto out;
1006
1007                bond_hw_addr_copy(ss.__data, tmp_mac,
1008                                  new_active->dev->addr_len);
1009                ss.ss_family = old_active->dev->type;
1010
1011                rv = dev_set_mac_address(old_active->dev,
1012                                         (struct sockaddr *)&ss, NULL);
1013                if (rv)
1014                        slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1015                                  -rv);
1016out:
1017                break;
1018        default:
1019                netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1020                           bond->params.fail_over_mac);
1021                break;
1022        }
1023
1024}
1025
1026static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1027{
1028        struct slave *prim = rtnl_dereference(bond->primary_slave);
1029        struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1030
1031        if (!prim || prim->link != BOND_LINK_UP) {
1032                if (!curr || curr->link != BOND_LINK_UP)
1033                        return NULL;
1034                return curr;
1035        }
1036
1037        if (bond->force_primary) {
1038                bond->force_primary = false;
1039                return prim;
1040        }
1041
1042        if (!curr || curr->link != BOND_LINK_UP)
1043                return prim;
1044
1045        /* At this point, prim and curr are both up */
1046        switch (bond->params.primary_reselect) {
1047        case BOND_PRI_RESELECT_ALWAYS:
1048                return prim;
1049        case BOND_PRI_RESELECT_BETTER:
1050                if (prim->speed < curr->speed)
1051                        return curr;
1052                if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1053                        return curr;
1054                return prim;
1055        case BOND_PRI_RESELECT_FAILURE:
1056                return curr;
1057        default:
1058                netdev_err(bond->dev, "impossible primary_reselect %d\n",
1059                           bond->params.primary_reselect);
1060                return curr;
1061        }
1062}
1063
1064/**
1065 * bond_find_best_slave - select the best available slave to be the active one
1066 * @bond: our bonding struct
1067 */
1068static struct slave *bond_find_best_slave(struct bonding *bond)
1069{
1070        struct slave *slave, *bestslave = NULL;
1071        struct list_head *iter;
1072        int mintime = bond->params.updelay;
1073
1074        slave = bond_choose_primary_or_current(bond);
1075        if (slave)
1076                return slave;
1077
1078        bond_for_each_slave(bond, slave, iter) {
1079                if (slave->link == BOND_LINK_UP)
1080                        return slave;
1081                if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1082                    slave->delay < mintime) {
1083                        mintime = slave->delay;
1084                        bestslave = slave;
1085                }
1086        }
1087
1088        return bestslave;
1089}
1090
1091static bool bond_should_notify_peers(struct bonding *bond)
1092{
1093        struct slave *slave;
1094
1095        rcu_read_lock();
1096        slave = rcu_dereference(bond->curr_active_slave);
1097        rcu_read_unlock();
1098
1099        netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1100                   slave ? slave->dev->name : "NULL");
1101
1102        if (!slave || !bond->send_peer_notif ||
1103            bond->send_peer_notif %
1104            max(1, bond->params.peer_notif_delay) != 0 ||
1105            !netif_carrier_ok(bond->dev) ||
1106            test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1107                return false;
1108
1109        return true;
1110}
1111
1112/**
1113 * bond_change_active_slave - change the active slave into the specified one
1114 * @bond: our bonding struct
1115 * @new_active: the new slave to make the active one
1116 *
1117 * Set the new slave to the bond's settings and unset them on the old
1118 * curr_active_slave.
1119 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1120 *
1121 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1122 * because it is apparently the best available slave we have, even though its
1123 * updelay hasn't timed out yet.
1124 *
1125 * Caller must hold RTNL.
1126 */
1127void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1128{
1129        struct slave *old_active;
1130
1131        ASSERT_RTNL();
1132
1133        old_active = rtnl_dereference(bond->curr_active_slave);
1134
1135        if (old_active == new_active)
1136                return;
1137
1138#ifdef CONFIG_XFRM_OFFLOAD
1139        bond_ipsec_del_sa_all(bond);
1140#endif /* CONFIG_XFRM_OFFLOAD */
1141
1142        if (new_active) {
1143                new_active->last_link_up = jiffies;
1144
1145                if (new_active->link == BOND_LINK_BACK) {
1146                        if (bond_uses_primary(bond)) {
1147                                slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1148                                           (bond->params.updelay - new_active->delay) * bond->params.miimon);
1149                        }
1150
1151                        new_active->delay = 0;
1152                        bond_set_slave_link_state(new_active, BOND_LINK_UP,
1153                                                  BOND_SLAVE_NOTIFY_NOW);
1154
1155                        if (BOND_MODE(bond) == BOND_MODE_8023AD)
1156                                bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1157
1158                        if (bond_is_lb(bond))
1159                                bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1160                } else {
1161                        if (bond_uses_primary(bond))
1162                                slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1163                }
1164        }
1165
1166        if (bond_uses_primary(bond))
1167                bond_hw_addr_swap(bond, new_active, old_active);
1168
1169        if (bond_is_lb(bond)) {
1170                bond_alb_handle_active_change(bond, new_active);
1171                if (old_active)
1172                        bond_set_slave_inactive_flags(old_active,
1173                                                      BOND_SLAVE_NOTIFY_NOW);
1174                if (new_active)
1175                        bond_set_slave_active_flags(new_active,
1176                                                    BOND_SLAVE_NOTIFY_NOW);
1177        } else {
1178                rcu_assign_pointer(bond->curr_active_slave, new_active);
1179        }
1180
1181        if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1182                if (old_active)
1183                        bond_set_slave_inactive_flags(old_active,
1184                                                      BOND_SLAVE_NOTIFY_NOW);
1185
1186                if (new_active) {
1187                        bool should_notify_peers = false;
1188
1189                        bond_set_slave_active_flags(new_active,
1190                                                    BOND_SLAVE_NOTIFY_NOW);
1191
1192                        if (bond->params.fail_over_mac)
1193                                bond_do_fail_over_mac(bond, new_active,
1194                                                      old_active);
1195
1196                        if (netif_running(bond->dev)) {
1197                                bond->send_peer_notif =
1198                                        bond->params.num_peer_notif *
1199                                        max(1, bond->params.peer_notif_delay);
1200                                should_notify_peers =
1201                                        bond_should_notify_peers(bond);
1202                        }
1203
1204                        call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1205                        if (should_notify_peers) {
1206                                bond->send_peer_notif--;
1207                                call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1208                                                         bond->dev);
1209                        }
1210                }
1211        }
1212
1213#ifdef CONFIG_XFRM_OFFLOAD
1214        bond_ipsec_add_sa_all(bond);
1215#endif /* CONFIG_XFRM_OFFLOAD */
1216
1217        /* resend IGMP joins since active slave has changed or
1218         * all were sent on curr_active_slave.
1219         * resend only if bond is brought up with the affected
1220         * bonding modes and the retransmission is enabled
1221         */
1222        if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1223            ((bond_uses_primary(bond) && new_active) ||
1224             BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1225                bond->igmp_retrans = bond->params.resend_igmp;
1226                queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1227        }
1228}
1229
1230/**
1231 * bond_select_active_slave - select a new active slave, if needed
1232 * @bond: our bonding struct
1233 *
1234 * This functions should be called when one of the following occurs:
1235 * - The old curr_active_slave has been released or lost its link.
1236 * - The primary_slave has got its link back.
1237 * - A slave has got its link back and there's no old curr_active_slave.
1238 *
1239 * Caller must hold RTNL.
1240 */
1241void bond_select_active_slave(struct bonding *bond)
1242{
1243        struct slave *best_slave;
1244        int rv;
1245
1246        ASSERT_RTNL();
1247
1248        best_slave = bond_find_best_slave(bond);
1249        if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1250                bond_change_active_slave(bond, best_slave);
1251                rv = bond_set_carrier(bond);
1252                if (!rv)
1253                        return;
1254
1255                if (netif_carrier_ok(bond->dev))
1256                        netdev_info(bond->dev, "active interface up!\n");
1257                else
1258                        netdev_info(bond->dev, "now running without any active interface!\n");
1259        }
1260}
1261
1262#ifdef CONFIG_NET_POLL_CONTROLLER
1263static inline int slave_enable_netpoll(struct slave *slave)
1264{
1265        struct netpoll *np;
1266        int err = 0;
1267
1268        np = kzalloc(sizeof(*np), GFP_KERNEL);
1269        err = -ENOMEM;
1270        if (!np)
1271                goto out;
1272
1273        err = __netpoll_setup(np, slave->dev);
1274        if (err) {
1275                kfree(np);
1276                goto out;
1277        }
1278        slave->np = np;
1279out:
1280        return err;
1281}
1282static inline void slave_disable_netpoll(struct slave *slave)
1283{
1284        struct netpoll *np = slave->np;
1285
1286        if (!np)
1287                return;
1288
1289        slave->np = NULL;
1290
1291        __netpoll_free(np);
1292}
1293
1294static void bond_poll_controller(struct net_device *bond_dev)
1295{
1296        struct bonding *bond = netdev_priv(bond_dev);
1297        struct slave *slave = NULL;
1298        struct list_head *iter;
1299        struct ad_info ad_info;
1300
1301        if (BOND_MODE(bond) == BOND_MODE_8023AD)
1302                if (bond_3ad_get_active_agg_info(bond, &ad_info))
1303                        return;
1304
1305        bond_for_each_slave_rcu(bond, slave, iter) {
1306                if (!bond_slave_is_up(slave))
1307                        continue;
1308
1309                if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1310                        struct aggregator *agg =
1311                            SLAVE_AD_INFO(slave)->port.aggregator;
1312
1313                        if (agg &&
1314                            agg->aggregator_identifier != ad_info.aggregator_id)
1315                                continue;
1316                }
1317
1318                netpoll_poll_dev(slave->dev);
1319        }
1320}
1321
1322static void bond_netpoll_cleanup(struct net_device *bond_dev)
1323{
1324        struct bonding *bond = netdev_priv(bond_dev);
1325        struct list_head *iter;
1326        struct slave *slave;
1327
1328        bond_for_each_slave(bond, slave, iter)
1329                if (bond_slave_is_up(slave))
1330                        slave_disable_netpoll(slave);
1331}
1332
1333static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1334{
1335        struct bonding *bond = netdev_priv(dev);
1336        struct list_head *iter;
1337        struct slave *slave;
1338        int err = 0;
1339
1340        bond_for_each_slave(bond, slave, iter) {
1341                err = slave_enable_netpoll(slave);
1342                if (err) {
1343                        bond_netpoll_cleanup(dev);
1344                        break;
1345                }
1346        }
1347        return err;
1348}
1349#else
1350static inline int slave_enable_netpoll(struct slave *slave)
1351{
1352        return 0;
1353}
1354static inline void slave_disable_netpoll(struct slave *slave)
1355{
1356}
1357static void bond_netpoll_cleanup(struct net_device *bond_dev)
1358{
1359}
1360#endif
1361
1362/*---------------------------------- IOCTL ----------------------------------*/
1363
1364static netdev_features_t bond_fix_features(struct net_device *dev,
1365                                           netdev_features_t features)
1366{
1367        struct bonding *bond = netdev_priv(dev);
1368        struct list_head *iter;
1369        netdev_features_t mask;
1370        struct slave *slave;
1371
1372#if IS_ENABLED(CONFIG_TLS_DEVICE)
1373        if (bond_sk_check(bond))
1374                features |= BOND_TLS_FEATURES;
1375        else
1376                features &= ~BOND_TLS_FEATURES;
1377#endif
1378
1379        mask = features;
1380
1381        features &= ~NETIF_F_ONE_FOR_ALL;
1382        features |= NETIF_F_ALL_FOR_ALL;
1383
1384        bond_for_each_slave(bond, slave, iter) {
1385                features = netdev_increment_features(features,
1386                                                     slave->dev->features,
1387                                                     mask);
1388        }
1389        features = netdev_add_tso_features(features, mask);
1390
1391        return features;
1392}
1393
1394#define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1395                                 NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1396                                 NETIF_F_HIGHDMA | NETIF_F_LRO)
1397
1398#define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1399                                 NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1400
1401#define BOND_MPLS_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1402                                 NETIF_F_GSO_SOFTWARE)
1403
1404
1405static void bond_compute_features(struct bonding *bond)
1406{
1407        unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1408                                        IFF_XMIT_DST_RELEASE_PERM;
1409        netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1410        netdev_features_t enc_features  = BOND_ENC_FEATURES;
1411#ifdef CONFIG_XFRM_OFFLOAD
1412        netdev_features_t xfrm_features  = BOND_XFRM_FEATURES;
1413#endif /* CONFIG_XFRM_OFFLOAD */
1414        netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
1415        struct net_device *bond_dev = bond->dev;
1416        struct list_head *iter;
1417        struct slave *slave;
1418        unsigned short max_hard_header_len = ETH_HLEN;
1419        unsigned int gso_max_size = GSO_MAX_SIZE;
1420        u16 gso_max_segs = GSO_MAX_SEGS;
1421
1422        if (!bond_has_slaves(bond))
1423                goto done;
1424        vlan_features &= NETIF_F_ALL_FOR_ALL;
1425        mpls_features &= NETIF_F_ALL_FOR_ALL;
1426
1427        bond_for_each_slave(bond, slave, iter) {
1428                vlan_features = netdev_increment_features(vlan_features,
1429                        slave->dev->vlan_features, BOND_VLAN_FEATURES);
1430
1431                enc_features = netdev_increment_features(enc_features,
1432                                                         slave->dev->hw_enc_features,
1433                                                         BOND_ENC_FEATURES);
1434
1435#ifdef CONFIG_XFRM_OFFLOAD
1436                xfrm_features = netdev_increment_features(xfrm_features,
1437                                                          slave->dev->hw_enc_features,
1438                                                          BOND_XFRM_FEATURES);
1439#endif /* CONFIG_XFRM_OFFLOAD */
1440
1441                mpls_features = netdev_increment_features(mpls_features,
1442                                                          slave->dev->mpls_features,
1443                                                          BOND_MPLS_FEATURES);
1444
1445                dst_release_flag &= slave->dev->priv_flags;
1446                if (slave->dev->hard_header_len > max_hard_header_len)
1447                        max_hard_header_len = slave->dev->hard_header_len;
1448
1449                gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1450                gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1451        }
1452        bond_dev->hard_header_len = max_hard_header_len;
1453
1454done:
1455        bond_dev->vlan_features = vlan_features;
1456        bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1457                                    NETIF_F_HW_VLAN_CTAG_TX |
1458                                    NETIF_F_HW_VLAN_STAG_TX;
1459#ifdef CONFIG_XFRM_OFFLOAD
1460        bond_dev->hw_enc_features |= xfrm_features;
1461#endif /* CONFIG_XFRM_OFFLOAD */
1462        bond_dev->mpls_features = mpls_features;
1463        bond_dev->gso_max_segs = gso_max_segs;
1464        netif_set_gso_max_size(bond_dev, gso_max_size);
1465
1466        bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1467        if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1468            dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1469                bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1470
1471        netdev_change_features(bond_dev);
1472}
1473
1474static void bond_setup_by_slave(struct net_device *bond_dev,
1475                                struct net_device *slave_dev)
1476{
1477        bond_dev->header_ops        = slave_dev->header_ops;
1478
1479        bond_dev->type              = slave_dev->type;
1480        bond_dev->hard_header_len   = slave_dev->hard_header_len;
1481        bond_dev->needed_headroom   = slave_dev->needed_headroom;
1482        bond_dev->addr_len          = slave_dev->addr_len;
1483
1484        memcpy(bond_dev->broadcast, slave_dev->broadcast,
1485                slave_dev->addr_len);
1486}
1487
1488/* On bonding slaves other than the currently active slave, suppress
1489 * duplicates except for alb non-mcast/bcast.
1490 */
1491static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1492                                            struct slave *slave,
1493                                            struct bonding *bond)
1494{
1495        if (bond_is_slave_inactive(slave)) {
1496                if (BOND_MODE(bond) == BOND_MODE_ALB &&
1497                    skb->pkt_type != PACKET_BROADCAST &&
1498                    skb->pkt_type != PACKET_MULTICAST)
1499                        return false;
1500                return true;
1501        }
1502        return false;
1503}
1504
1505static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1506{
1507        struct sk_buff *skb = *pskb;
1508        struct slave *slave;
1509        struct bonding *bond;
1510        int (*recv_probe)(const struct sk_buff *, struct bonding *,
1511                          struct slave *);
1512        int ret = RX_HANDLER_ANOTHER;
1513
1514        skb = skb_share_check(skb, GFP_ATOMIC);
1515        if (unlikely(!skb))
1516                return RX_HANDLER_CONSUMED;
1517
1518        *pskb = skb;
1519
1520        slave = bond_slave_get_rcu(skb->dev);
1521        bond = slave->bond;
1522
1523        recv_probe = READ_ONCE(bond->recv_probe);
1524        if (recv_probe) {
1525                ret = recv_probe(skb, bond, slave);
1526                if (ret == RX_HANDLER_CONSUMED) {
1527                        consume_skb(skb);
1528                        return ret;
1529                }
1530        }
1531
1532        /*
1533         * For packets determined by bond_should_deliver_exact_match() call to
1534         * be suppressed we want to make an exception for link-local packets.
1535         * This is necessary for e.g. LLDP daemons to be able to monitor
1536         * inactive slave links without being forced to bind to them
1537         * explicitly.
1538         *
1539         * At the same time, packets that are passed to the bonding master
1540         * (including link-local ones) can have their originating interface
1541         * determined via PACKET_ORIGDEV socket option.
1542         */
1543        if (bond_should_deliver_exact_match(skb, slave, bond)) {
1544                if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1545                        return RX_HANDLER_PASS;
1546                return RX_HANDLER_EXACT;
1547        }
1548
1549        skb->dev = bond->dev;
1550
1551        if (BOND_MODE(bond) == BOND_MODE_ALB &&
1552            netif_is_bridge_port(bond->dev) &&
1553            skb->pkt_type == PACKET_HOST) {
1554
1555                if (unlikely(skb_cow_head(skb,
1556                                          skb->data - skb_mac_header(skb)))) {
1557                        kfree_skb(skb);
1558                        return RX_HANDLER_CONSUMED;
1559                }
1560                bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1561                                  bond->dev->addr_len);
1562        }
1563
1564        return ret;
1565}
1566
1567static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1568{
1569        switch (BOND_MODE(bond)) {
1570        case BOND_MODE_ROUNDROBIN:
1571                return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1572        case BOND_MODE_ACTIVEBACKUP:
1573                return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1574        case BOND_MODE_BROADCAST:
1575                return NETDEV_LAG_TX_TYPE_BROADCAST;
1576        case BOND_MODE_XOR:
1577        case BOND_MODE_8023AD:
1578                return NETDEV_LAG_TX_TYPE_HASH;
1579        default:
1580                return NETDEV_LAG_TX_TYPE_UNKNOWN;
1581        }
1582}
1583
1584static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1585                                               enum netdev_lag_tx_type type)
1586{
1587        if (type != NETDEV_LAG_TX_TYPE_HASH)
1588                return NETDEV_LAG_HASH_NONE;
1589
1590        switch (bond->params.xmit_policy) {
1591        case BOND_XMIT_POLICY_LAYER2:
1592                return NETDEV_LAG_HASH_L2;
1593        case BOND_XMIT_POLICY_LAYER34:
1594                return NETDEV_LAG_HASH_L34;
1595        case BOND_XMIT_POLICY_LAYER23:
1596                return NETDEV_LAG_HASH_L23;
1597        case BOND_XMIT_POLICY_ENCAP23:
1598                return NETDEV_LAG_HASH_E23;
1599        case BOND_XMIT_POLICY_ENCAP34:
1600                return NETDEV_LAG_HASH_E34;
1601        case BOND_XMIT_POLICY_VLAN_SRCMAC:
1602                return NETDEV_LAG_HASH_VLAN_SRCMAC;
1603        default:
1604                return NETDEV_LAG_HASH_UNKNOWN;
1605        }
1606}
1607
1608static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1609                                      struct netlink_ext_ack *extack)
1610{
1611        struct netdev_lag_upper_info lag_upper_info;
1612        enum netdev_lag_tx_type type;
1613
1614        type = bond_lag_tx_type(bond);
1615        lag_upper_info.tx_type = type;
1616        lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1617
1618        return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1619                                            &lag_upper_info, extack);
1620}
1621
1622static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1623{
1624        netdev_upper_dev_unlink(slave->dev, bond->dev);
1625        slave->dev->flags &= ~IFF_SLAVE;
1626}
1627
1628static void slave_kobj_release(struct kobject *kobj)
1629{
1630        struct slave *slave = to_slave(kobj);
1631        struct bonding *bond = bond_get_bond_by_slave(slave);
1632
1633        cancel_delayed_work_sync(&slave->notify_work);
1634        if (BOND_MODE(bond) == BOND_MODE_8023AD)
1635                kfree(SLAVE_AD_INFO(slave));
1636
1637        kfree(slave);
1638}
1639
1640static struct kobj_type slave_ktype = {
1641        .release = slave_kobj_release,
1642#ifdef CONFIG_SYSFS
1643        .sysfs_ops = &slave_sysfs_ops,
1644#endif
1645};
1646
1647static int bond_kobj_init(struct slave *slave)
1648{
1649        int err;
1650
1651        err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1652                                   &(slave->dev->dev.kobj), "bonding_slave");
1653        if (err)
1654                kobject_put(&slave->kobj);
1655
1656        return err;
1657}
1658
1659static struct slave *bond_alloc_slave(struct bonding *bond,
1660                                      struct net_device *slave_dev)
1661{
1662        struct slave *slave = NULL;
1663
1664        slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1665        if (!slave)
1666                return NULL;
1667
1668        slave->bond = bond;
1669        slave->dev = slave_dev;
1670        INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1671
1672        if (bond_kobj_init(slave))
1673                return NULL;
1674
1675        if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1676                SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1677                                               GFP_KERNEL);
1678                if (!SLAVE_AD_INFO(slave)) {
1679                        kobject_put(&slave->kobj);
1680                        return NULL;
1681                }
1682        }
1683
1684        return slave;
1685}
1686
1687static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1688{
1689        info->bond_mode = BOND_MODE(bond);
1690        info->miimon = bond->params.miimon;
1691        info->num_slaves = bond->slave_cnt;
1692}
1693
1694static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1695{
1696        strcpy(info->slave_name, slave->dev->name);
1697        info->link = slave->link;
1698        info->state = bond_slave_state(slave);
1699        info->link_failure_count = slave->link_failure_count;
1700}
1701
1702static void bond_netdev_notify_work(struct work_struct *_work)
1703{
1704        struct slave *slave = container_of(_work, struct slave,
1705                                           notify_work.work);
1706
1707        if (rtnl_trylock()) {
1708                struct netdev_bonding_info binfo;
1709
1710                bond_fill_ifslave(slave, &binfo.slave);
1711                bond_fill_ifbond(slave->bond, &binfo.master);
1712                netdev_bonding_info_change(slave->dev, &binfo);
1713                rtnl_unlock();
1714        } else {
1715                queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1716        }
1717}
1718
1719void bond_queue_slave_event(struct slave *slave)
1720{
1721        queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1722}
1723
1724void bond_lower_state_changed(struct slave *slave)
1725{
1726        struct netdev_lag_lower_state_info info;
1727
1728        info.link_up = slave->link == BOND_LINK_UP ||
1729                       slave->link == BOND_LINK_FAIL;
1730        info.tx_enabled = bond_is_active_slave(slave);
1731        netdev_lower_state_changed(slave->dev, &info);
1732}
1733
1734#define BOND_NL_ERR(bond_dev, extack, errmsg) do {              \
1735        if (extack)                                             \
1736                NL_SET_ERR_MSG(extack, errmsg);                 \
1737        else                                                    \
1738                netdev_err(bond_dev, "Error: %s\n", errmsg);    \
1739} while (0)
1740
1741#define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do {          \
1742        if (extack)                                                     \
1743                NL_SET_ERR_MSG(extack, errmsg);                         \
1744        else                                                            \
1745                slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg);  \
1746} while (0)
1747
1748/* enslave device <slave> to bond device <master> */
1749int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1750                 struct netlink_ext_ack *extack)
1751{
1752        struct bonding *bond = netdev_priv(bond_dev);
1753        const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1754        struct slave *new_slave = NULL, *prev_slave;
1755        struct sockaddr_storage ss;
1756        int link_reporting;
1757        int res = 0, i;
1758
1759        if (slave_dev->flags & IFF_MASTER &&
1760            !netif_is_bond_master(slave_dev)) {
1761                BOND_NL_ERR(bond_dev, extack,
1762                            "Device type (master device) cannot be enslaved");
1763                return -EPERM;
1764        }
1765
1766        if (!bond->params.use_carrier &&
1767            slave_dev->ethtool_ops->get_link == NULL &&
1768            slave_ops->ndo_eth_ioctl == NULL) {
1769                slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1770        }
1771
1772        /* already in-use? */
1773        if (netdev_is_rx_handler_busy(slave_dev)) {
1774                SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1775                             "Device is in use and cannot be enslaved");
1776                return -EBUSY;
1777        }
1778
1779        if (bond_dev == slave_dev) {
1780                BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1781                return -EPERM;
1782        }
1783
1784        /* vlan challenged mutual exclusion */
1785        /* no need to lock since we're protected by rtnl_lock */
1786        if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1787                slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1788                if (vlan_uses_dev(bond_dev)) {
1789                        SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1790                                     "Can not enslave VLAN challenged device to VLAN enabled bond");
1791                        return -EPERM;
1792                } else {
1793                        slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1794                }
1795        } else {
1796                slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1797        }
1798
1799        if (slave_dev->features & NETIF_F_HW_ESP)
1800                slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1801
1802        /* Old ifenslave binaries are no longer supported.  These can
1803         * be identified with moderate accuracy by the state of the slave:
1804         * the current ifenslave will set the interface down prior to
1805         * enslaving it; the old ifenslave will not.
1806         */
1807        if (slave_dev->flags & IFF_UP) {
1808                SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1809                             "Device can not be enslaved while up");
1810                return -EPERM;
1811        }
1812
1813        /* set bonding device ether type by slave - bonding netdevices are
1814         * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1815         * there is a need to override some of the type dependent attribs/funcs.
1816         *
1817         * bond ether type mutual exclusion - don't allow slaves of dissimilar
1818         * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1819         */
1820        if (!bond_has_slaves(bond)) {
1821                if (bond_dev->type != slave_dev->type) {
1822                        slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1823                                  bond_dev->type, slave_dev->type);
1824
1825                        res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1826                                                       bond_dev);
1827                        res = notifier_to_errno(res);
1828                        if (res) {
1829                                slave_err(bond_dev, slave_dev, "refused to change device type\n");
1830                                return -EBUSY;
1831                        }
1832
1833                        /* Flush unicast and multicast addresses */
1834                        dev_uc_flush(bond_dev);
1835                        dev_mc_flush(bond_dev);
1836
1837                        if (slave_dev->type != ARPHRD_ETHER)
1838                                bond_setup_by_slave(bond_dev, slave_dev);
1839                        else {
1840                                ether_setup(bond_dev);
1841                                bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1842                        }
1843
1844                        call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1845                                                 bond_dev);
1846                }
1847        } else if (bond_dev->type != slave_dev->type) {
1848                SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1849                             "Device type is different from other slaves");
1850                return -EINVAL;
1851        }
1852
1853        if (slave_dev->type == ARPHRD_INFINIBAND &&
1854            BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1855                SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1856                             "Only active-backup mode is supported for infiniband slaves");
1857                res = -EOPNOTSUPP;
1858                goto err_undo_flags;
1859        }
1860
1861        if (!slave_ops->ndo_set_mac_address ||
1862            slave_dev->type == ARPHRD_INFINIBAND) {
1863                slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1864                if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1865                    bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1866                        if (!bond_has_slaves(bond)) {
1867                                bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1868                                slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1869                        } else {
1870                                SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1871                                             "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1872                                res = -EOPNOTSUPP;
1873                                goto err_undo_flags;
1874                        }
1875                }
1876        }
1877
1878        call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1879
1880        /* If this is the first slave, then we need to set the master's hardware
1881         * address to be the same as the slave's.
1882         */
1883        if (!bond_has_slaves(bond) &&
1884            bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1885                res = bond_set_dev_addr(bond->dev, slave_dev);
1886                if (res)
1887                        goto err_undo_flags;
1888        }
1889
1890        new_slave = bond_alloc_slave(bond, slave_dev);
1891        if (!new_slave) {
1892                res = -ENOMEM;
1893                goto err_undo_flags;
1894        }
1895
1896        /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1897         * is set via sysfs or module option if desired.
1898         */
1899        new_slave->queue_id = 0;
1900
1901        /* Save slave's original mtu and then set it to match the bond */
1902        new_slave->original_mtu = slave_dev->mtu;
1903        res = dev_set_mtu(slave_dev, bond->dev->mtu);
1904        if (res) {
1905                slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1906                goto err_free;
1907        }
1908
1909        /* Save slave's original ("permanent") mac address for modes
1910         * that need it, and for restoring it upon release, and then
1911         * set it to the master's address
1912         */
1913        bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1914                          slave_dev->addr_len);
1915
1916        if (!bond->params.fail_over_mac ||
1917            BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1918                /* Set slave to master's mac address.  The application already
1919                 * set the master's mac address to that of the first slave
1920                 */
1921                memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1922                ss.ss_family = slave_dev->type;
1923                res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1924                                          extack);
1925                if (res) {
1926                        slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1927                        goto err_restore_mtu;
1928                }
1929        }
1930
1931        /* set slave flag before open to prevent IPv6 addrconf */
1932        slave_dev->flags |= IFF_SLAVE;
1933
1934        /* open the slave since the application closed it */
1935        res = dev_open(slave_dev, extack);
1936        if (res) {
1937                slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1938                goto err_restore_mac;
1939        }
1940
1941        slave_dev->priv_flags |= IFF_BONDING;
1942        /* initialize slave stats */
1943        dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1944
1945        if (bond_is_lb(bond)) {
1946                /* bond_alb_init_slave() must be called before all other stages since
1947                 * it might fail and we do not want to have to undo everything
1948                 */
1949                res = bond_alb_init_slave(bond, new_slave);
1950                if (res)
1951                        goto err_close;
1952        }
1953
1954        res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1955        if (res) {
1956                slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1957                goto err_close;
1958        }
1959
1960        prev_slave = bond_last_slave(bond);
1961
1962        new_slave->delay = 0;
1963        new_slave->link_failure_count = 0;
1964
1965        if (bond_update_speed_duplex(new_slave) &&
1966            bond_needs_speed_duplex(bond))
1967                new_slave->link = BOND_LINK_DOWN;
1968
1969        new_slave->last_rx = jiffies -
1970                (msecs_to_jiffies(bond->params.arp_interval) + 1);
1971        for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1972                new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1973
1974        if (bond->params.miimon && !bond->params.use_carrier) {
1975                link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1976
1977                if ((link_reporting == -1) && !bond->params.arp_interval) {
1978                        /* miimon is set but a bonded network driver
1979                         * does not support ETHTOOL/MII and
1980                         * arp_interval is not set.  Note: if
1981                         * use_carrier is enabled, we will never go
1982                         * here (because netif_carrier is always
1983                         * supported); thus, we don't need to change
1984                         * the messages for netif_carrier.
1985                         */
1986                        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");
1987                } else if (link_reporting == -1) {
1988                        /* unable get link status using mii/ethtool */
1989                        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");
1990                }
1991        }
1992
1993        /* check for initial state */
1994        new_slave->link = BOND_LINK_NOCHANGE;
1995        if (bond->params.miimon) {
1996                if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1997                        if (bond->params.updelay) {
1998                                bond_set_slave_link_state(new_slave,
1999                                                          BOND_LINK_BACK,
2000                                                          BOND_SLAVE_NOTIFY_NOW);
2001                                new_slave->delay = bond->params.updelay;
2002                        } else {
2003                                bond_set_slave_link_state(new_slave,
2004                                                          BOND_LINK_UP,
2005                                                          BOND_SLAVE_NOTIFY_NOW);
2006                        }
2007                } else {
2008                        bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2009                                                  BOND_SLAVE_NOTIFY_NOW);
2010                }
2011        } else if (bond->params.arp_interval) {
2012                bond_set_slave_link_state(new_slave,
2013                                          (netif_carrier_ok(slave_dev) ?
2014                                          BOND_LINK_UP : BOND_LINK_DOWN),
2015                                          BOND_SLAVE_NOTIFY_NOW);
2016        } else {
2017                bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2018                                          BOND_SLAVE_NOTIFY_NOW);
2019        }
2020
2021        if (new_slave->link != BOND_LINK_DOWN)
2022                new_slave->last_link_up = jiffies;
2023        slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2024                  new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2025                  (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2026
2027        if (bond_uses_primary(bond) && bond->params.primary[0]) {
2028                /* if there is a primary slave, remember it */
2029                if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2030                        rcu_assign_pointer(bond->primary_slave, new_slave);
2031                        bond->force_primary = true;
2032                }
2033        }
2034
2035        switch (BOND_MODE(bond)) {
2036        case BOND_MODE_ACTIVEBACKUP:
2037                bond_set_slave_inactive_flags(new_slave,
2038                                              BOND_SLAVE_NOTIFY_NOW);
2039                break;
2040        case BOND_MODE_8023AD:
2041                /* in 802.3ad mode, the internal mechanism
2042                 * will activate the slaves in the selected
2043                 * aggregator
2044                 */
2045                bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2046                /* if this is the first slave */
2047                if (!prev_slave) {
2048                        SLAVE_AD_INFO(new_slave)->id = 1;
2049                        /* Initialize AD with the number of times that the AD timer is called in 1 second
2050                         * can be called only after the mac address of the bond is set
2051                         */
2052                        bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
2053                } else {
2054                        SLAVE_AD_INFO(new_slave)->id =
2055                                SLAVE_AD_INFO(prev_slave)->id + 1;
2056                }
2057
2058                bond_3ad_bind_slave(new_slave);
2059                break;
2060        case BOND_MODE_TLB:
2061        case BOND_MODE_ALB:
2062                bond_set_active_slave(new_slave);
2063                bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2064                break;
2065        default:
2066                slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2067
2068                /* always active in trunk mode */
2069                bond_set_active_slave(new_slave);
2070
2071                /* In trunking mode there is little meaning to curr_active_slave
2072                 * anyway (it holds no special properties of the bond device),
2073                 * so we can change it without calling change_active_interface()
2074                 */
2075                if (!rcu_access_pointer(bond->curr_active_slave) &&
2076                    new_slave->link == BOND_LINK_UP)
2077                        rcu_assign_pointer(bond->curr_active_slave, new_slave);
2078
2079                break;
2080        } /* switch(bond_mode) */
2081
2082#ifdef CONFIG_NET_POLL_CONTROLLER
2083        if (bond->dev->npinfo) {
2084                if (slave_enable_netpoll(new_slave)) {
2085                        slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2086                        res = -EBUSY;
2087                        goto err_detach;
2088                }
2089        }
2090#endif
2091
2092        if (!(bond_dev->features & NETIF_F_LRO))
2093                dev_disable_lro(slave_dev);
2094
2095        res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2096                                         new_slave);
2097        if (res) {
2098                slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2099                goto err_detach;
2100        }
2101
2102        res = bond_master_upper_dev_link(bond, new_slave, extack);
2103        if (res) {
2104                slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2105                goto err_unregister;
2106        }
2107
2108        bond_lower_state_changed(new_slave);
2109
2110        res = bond_sysfs_slave_add(new_slave);
2111        if (res) {
2112                slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2113                goto err_upper_unlink;
2114        }
2115
2116        /* If the mode uses primary, then the following is handled by
2117         * bond_change_active_slave().
2118         */
2119        if (!bond_uses_primary(bond)) {
2120                /* set promiscuity level to new slave */
2121                if (bond_dev->flags & IFF_PROMISC) {
2122                        res = dev_set_promiscuity(slave_dev, 1);
2123                        if (res)
2124                                goto err_sysfs_del;
2125                }
2126
2127                /* set allmulti level to new slave */
2128                if (bond_dev->flags & IFF_ALLMULTI) {
2129                        res = dev_set_allmulti(slave_dev, 1);
2130                        if (res) {
2131                                if (bond_dev->flags & IFF_PROMISC)
2132                                        dev_set_promiscuity(slave_dev, -1);
2133                                goto err_sysfs_del;
2134                        }
2135                }
2136
2137                netif_addr_lock_bh(bond_dev);
2138                dev_mc_sync_multiple(slave_dev, bond_dev);
2139                dev_uc_sync_multiple(slave_dev, bond_dev);
2140                netif_addr_unlock_bh(bond_dev);
2141
2142                if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2143                        /* add lacpdu mc addr to mc list */
2144                        u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
2145
2146                        dev_mc_add(slave_dev, lacpdu_multicast);
2147                }
2148        }
2149
2150        bond->slave_cnt++;
2151        bond_compute_features(bond);
2152        bond_set_carrier(bond);
2153
2154        if (bond_uses_primary(bond)) {
2155                block_netpoll_tx();
2156                bond_select_active_slave(bond);
2157                unblock_netpoll_tx();
2158        }
2159
2160        if (bond_mode_can_use_xmit_hash(bond))
2161                bond_update_slave_arr(bond, NULL);
2162
2163
2164        if (!slave_dev->netdev_ops->ndo_bpf ||
2165            !slave_dev->netdev_ops->ndo_xdp_xmit) {
2166                if (bond->xdp_prog) {
2167                        SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2168                                     "Slave does not support XDP");
2169                        res = -EOPNOTSUPP;
2170                        goto err_sysfs_del;
2171                }
2172        } else if (bond->xdp_prog) {
2173                struct netdev_bpf xdp = {
2174                        .command = XDP_SETUP_PROG,
2175                        .flags   = 0,
2176                        .prog    = bond->xdp_prog,
2177                        .extack  = extack,
2178                };
2179
2180                if (dev_xdp_prog_count(slave_dev) > 0) {
2181                        SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2182                                     "Slave has XDP program loaded, please unload before enslaving");
2183                        res = -EOPNOTSUPP;
2184                        goto err_sysfs_del;
2185                }
2186
2187                res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2188                if (res < 0) {
2189                        /* ndo_bpf() sets extack error message */
2190                        slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2191                        goto err_sysfs_del;
2192                }
2193                if (bond->xdp_prog)
2194                        bpf_prog_inc(bond->xdp_prog);
2195        }
2196
2197        slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2198                   bond_is_active_slave(new_slave) ? "an active" : "a backup",
2199                   new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2200
2201        /* enslave is successful */
2202        bond_queue_slave_event(new_slave);
2203        return 0;
2204
2205/* Undo stages on error */
2206err_sysfs_del:
2207        bond_sysfs_slave_del(new_slave);
2208
2209err_upper_unlink:
2210        bond_upper_dev_unlink(bond, new_slave);
2211
2212err_unregister:
2213        netdev_rx_handler_unregister(slave_dev);
2214
2215err_detach:
2216        vlan_vids_del_by_dev(slave_dev, bond_dev);
2217        if (rcu_access_pointer(bond->primary_slave) == new_slave)
2218                RCU_INIT_POINTER(bond->primary_slave, NULL);
2219        if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2220                block_netpoll_tx();
2221                bond_change_active_slave(bond, NULL);
2222                bond_select_active_slave(bond);
2223                unblock_netpoll_tx();
2224        }
2225        /* either primary_slave or curr_active_slave might've changed */
2226        synchronize_rcu();
2227        slave_disable_netpoll(new_slave);
2228
2229err_close:
2230        if (!netif_is_bond_master(slave_dev))
2231                slave_dev->priv_flags &= ~IFF_BONDING;
2232        dev_close(slave_dev);
2233
2234err_restore_mac:
2235        slave_dev->flags &= ~IFF_SLAVE;
2236        if (!bond->params.fail_over_mac ||
2237            BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2238                /* XXX TODO - fom follow mode needs to change master's
2239                 * MAC if this slave's MAC is in use by the bond, or at
2240                 * least print a warning.
2241                 */
2242                bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2243                                  new_slave->dev->addr_len);
2244                ss.ss_family = slave_dev->type;
2245                dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2246        }
2247
2248err_restore_mtu:
2249        dev_set_mtu(slave_dev, new_slave->original_mtu);
2250
2251err_free:
2252        kobject_put(&new_slave->kobj);
2253
2254err_undo_flags:
2255        /* Enslave of first slave has failed and we need to fix master's mac */
2256        if (!bond_has_slaves(bond)) {
2257                if (ether_addr_equal_64bits(bond_dev->dev_addr,
2258                                            slave_dev->dev_addr))
2259                        eth_hw_addr_random(bond_dev);
2260                if (bond_dev->type != ARPHRD_ETHER) {
2261                        dev_close(bond_dev);
2262                        ether_setup(bond_dev);
2263                        bond_dev->flags |= IFF_MASTER;
2264                        bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2265                }
2266        }
2267
2268        return res;
2269}
2270
2271/* Try to release the slave device <slave> from the bond device <master>
2272 * It is legal to access curr_active_slave without a lock because all the function
2273 * is RTNL-locked. If "all" is true it means that the function is being called
2274 * while destroying a bond interface and all slaves are being released.
2275 *
2276 * The rules for slave state should be:
2277 *   for Active/Backup:
2278 *     Active stays on all backups go down
2279 *   for Bonded connections:
2280 *     The first up interface should be left on and all others downed.
2281 */
2282static int __bond_release_one(struct net_device *bond_dev,
2283                              struct net_device *slave_dev,
2284                              bool all, bool unregister)
2285{
2286        struct bonding *bond = netdev_priv(bond_dev);
2287        struct slave *slave, *oldcurrent;
2288        struct sockaddr_storage ss;
2289        int old_flags = bond_dev->flags;
2290        netdev_features_t old_features = bond_dev->features;
2291
2292        /* slave is not a slave or master is not master of this slave */
2293        if (!(slave_dev->flags & IFF_SLAVE) ||
2294            !netdev_has_upper_dev(slave_dev, bond_dev)) {
2295                slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2296                return -EINVAL;
2297        }
2298
2299        block_netpoll_tx();
2300
2301        slave = bond_get_slave_by_dev(bond, slave_dev);
2302        if (!slave) {
2303                /* not a slave of this bond */
2304                slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2305                unblock_netpoll_tx();
2306                return -EINVAL;
2307        }
2308
2309        bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2310
2311        bond_sysfs_slave_del(slave);
2312
2313        /* recompute stats just before removing the slave */
2314        bond_get_stats(bond->dev, &bond->bond_stats);
2315
2316        if (bond->xdp_prog) {
2317                struct netdev_bpf xdp = {
2318                        .command = XDP_SETUP_PROG,
2319                        .flags   = 0,
2320                        .prog    = NULL,
2321                        .extack  = NULL,
2322                };
2323                if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2324                        slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2325        }
2326
2327        /* unregister rx_handler early so bond_handle_frame wouldn't be called
2328         * for this slave anymore.
2329         */
2330        netdev_rx_handler_unregister(slave_dev);
2331
2332        if (BOND_MODE(bond) == BOND_MODE_8023AD)
2333                bond_3ad_unbind_slave(slave);
2334
2335        bond_upper_dev_unlink(bond, slave);
2336
2337        if (bond_mode_can_use_xmit_hash(bond))
2338                bond_update_slave_arr(bond, slave);
2339
2340        slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2341                    bond_is_active_slave(slave) ? "active" : "backup");
2342
2343        oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2344
2345        RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2346
2347        if (!all && (!bond->params.fail_over_mac ||
2348                     BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2349                if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2350                    bond_has_slaves(bond))
2351                        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",
2352                                   slave->perm_hwaddr);
2353        }
2354
2355        if (rtnl_dereference(bond->primary_slave) == slave)
2356                RCU_INIT_POINTER(bond->primary_slave, NULL);
2357
2358        if (oldcurrent == slave)
2359                bond_change_active_slave(bond, NULL);
2360
2361        if (bond_is_lb(bond)) {
2362                /* Must be called only after the slave has been
2363                 * detached from the list and the curr_active_slave
2364                 * has been cleared (if our_slave == old_current),
2365                 * but before a new active slave is selected.
2366                 */
2367                bond_alb_deinit_slave(bond, slave);
2368        }
2369
2370        if (all) {
2371                RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2372        } else if (oldcurrent == slave) {
2373                /* Note that we hold RTNL over this sequence, so there
2374                 * is no concern that another slave add/remove event
2375                 * will interfere.
2376                 */
2377                bond_select_active_slave(bond);
2378        }
2379
2380        if (!bond_has_slaves(bond)) {
2381                bond_set_carrier(bond);
2382                eth_hw_addr_random(bond_dev);
2383        }
2384
2385        unblock_netpoll_tx();
2386        synchronize_rcu();
2387        bond->slave_cnt--;
2388
2389        if (!bond_has_slaves(bond)) {
2390                call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2391                call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2392        }
2393
2394        bond_compute_features(bond);
2395        if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2396            (old_features & NETIF_F_VLAN_CHALLENGED))
2397                slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2398
2399        vlan_vids_del_by_dev(slave_dev, bond_dev);
2400
2401        /* If the mode uses primary, then this case was handled above by
2402         * bond_change_active_slave(..., NULL)
2403         */
2404        if (!bond_uses_primary(bond)) {
2405                /* unset promiscuity level from slave
2406                 * NOTE: The NETDEV_CHANGEADDR call above may change the value
2407                 * of the IFF_PROMISC flag in the bond_dev, but we need the
2408                 * value of that flag before that change, as that was the value
2409                 * when this slave was attached, so we cache at the start of the
2410                 * function and use it here. Same goes for ALLMULTI below
2411                 */
2412                if (old_flags & IFF_PROMISC)
2413                        dev_set_promiscuity(slave_dev, -1);
2414
2415                /* unset allmulti level from slave */
2416                if (old_flags & IFF_ALLMULTI)
2417                        dev_set_allmulti(slave_dev, -1);
2418
2419                bond_hw_addr_flush(bond_dev, slave_dev);
2420        }
2421
2422        slave_disable_netpoll(slave);
2423
2424        /* close slave before restoring its mac address */
2425        dev_close(slave_dev);
2426
2427        if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2428            BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2429                /* restore original ("permanent") mac address */
2430                bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2431                                  slave->dev->addr_len);
2432                ss.ss_family = slave_dev->type;
2433                dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2434        }
2435
2436        if (unregister)
2437                __dev_set_mtu(slave_dev, slave->original_mtu);
2438        else
2439                dev_set_mtu(slave_dev, slave->original_mtu);
2440
2441        if (!netif_is_bond_master(slave_dev))
2442                slave_dev->priv_flags &= ~IFF_BONDING;
2443
2444        kobject_put(&slave->kobj);
2445
2446        return 0;
2447}
2448
2449/* A wrapper used because of ndo_del_link */
2450int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2451{
2452        return __bond_release_one(bond_dev, slave_dev, false, false);
2453}
2454
2455/* First release a slave and then destroy the bond if no more slaves are left.
2456 * Must be under rtnl_lock when this function is called.
2457 */
2458static int bond_release_and_destroy(struct net_device *bond_dev,
2459                                    struct net_device *slave_dev)
2460{
2461        struct bonding *bond = netdev_priv(bond_dev);
2462        int ret;
2463
2464        ret = __bond_release_one(bond_dev, slave_dev, false, true);
2465        if (ret == 0 && !bond_has_slaves(bond) &&
2466            bond_dev->reg_state != NETREG_UNREGISTERING) {
2467                bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2468                netdev_info(bond_dev, "Destroying bond\n");
2469                bond_remove_proc_entry(bond);
2470                unregister_netdevice(bond_dev);
2471        }
2472        return ret;
2473}
2474
2475static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2476{
2477        struct bonding *bond = netdev_priv(bond_dev);
2478
2479        bond_fill_ifbond(bond, info);
2480}
2481
2482static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2483{
2484        struct bonding *bond = netdev_priv(bond_dev);
2485        struct list_head *iter;
2486        int i = 0, res = -ENODEV;
2487        struct slave *slave;
2488
2489        bond_for_each_slave(bond, slave, iter) {
2490                if (i++ == (int)info->slave_id) {
2491                        res = 0;
2492                        bond_fill_ifslave(slave, info);
2493                        break;
2494                }
2495        }
2496
2497        return res;
2498}
2499
2500/*-------------------------------- Monitoring -------------------------------*/
2501
2502/* called with rcu_read_lock() */
2503static int bond_miimon_inspect(struct bonding *bond)
2504{
2505        int link_state, commit = 0;
2506        struct list_head *iter;
2507        struct slave *slave;
2508        bool ignore_updelay;
2509
2510        ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2511
2512        bond_for_each_slave_rcu(bond, slave, iter) {
2513                bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2514
2515                link_state = bond_check_dev_link(bond, slave->dev, 0);
2516
2517                switch (slave->link) {
2518                case BOND_LINK_UP:
2519                        if (link_state)
2520                                continue;
2521
2522                        bond_propose_link_state(slave, BOND_LINK_FAIL);
2523                        commit++;
2524                        slave->delay = bond->params.downdelay;
2525                        if (slave->delay) {
2526                                slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2527                                           (BOND_MODE(bond) ==
2528                                            BOND_MODE_ACTIVEBACKUP) ?
2529                                            (bond_is_active_slave(slave) ?
2530                                             "active " : "backup ") : "",
2531                                           bond->params.downdelay * bond->params.miimon);
2532                        }
2533                        fallthrough;
2534                case BOND_LINK_FAIL:
2535                        if (link_state) {
2536                                /* recovered before downdelay expired */
2537                                bond_propose_link_state(slave, BOND_LINK_UP);
2538                                slave->last_link_up = jiffies;
2539                                slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2540                                           (bond->params.downdelay - slave->delay) *
2541                                           bond->params.miimon);
2542                                commit++;
2543                                continue;
2544                        }
2545
2546                        if (slave->delay <= 0) {
2547                                bond_propose_link_state(slave, BOND_LINK_DOWN);
2548                                commit++;
2549                                continue;
2550                        }
2551
2552                        slave->delay--;
2553                        break;
2554
2555                case BOND_LINK_DOWN:
2556                        if (!link_state)
2557                                continue;
2558
2559                        bond_propose_link_state(slave, BOND_LINK_BACK);
2560                        commit++;
2561                        slave->delay = bond->params.updelay;
2562
2563                        if (slave->delay) {
2564                                slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2565                                           ignore_updelay ? 0 :
2566                                           bond->params.updelay *
2567                                           bond->params.miimon);
2568                        }
2569                        fallthrough;
2570                case BOND_LINK_BACK:
2571                        if (!link_state) {
2572                                bond_propose_link_state(slave, BOND_LINK_DOWN);
2573                                slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2574                                           (bond->params.updelay - slave->delay) *
2575                                           bond->params.miimon);
2576                                commit++;
2577                                continue;
2578                        }
2579
2580                        if (ignore_updelay)
2581                                slave->delay = 0;
2582
2583                        if (slave->delay <= 0) {
2584                                bond_propose_link_state(slave, BOND_LINK_UP);
2585                                commit++;
2586                                ignore_updelay = false;
2587                                continue;
2588                        }
2589
2590                        slave->delay--;
2591                        break;
2592                }
2593        }
2594
2595        return commit;
2596}
2597
2598static void bond_miimon_link_change(struct bonding *bond,
2599                                    struct slave *slave,
2600                                    char link)
2601{
2602        switch (BOND_MODE(bond)) {
2603        case BOND_MODE_8023AD:
2604                bond_3ad_handle_link_change(slave, link);
2605                break;
2606        case BOND_MODE_TLB:
2607        case BOND_MODE_ALB:
2608                bond_alb_handle_link_change(bond, slave, link);
2609                break;
2610        case BOND_MODE_XOR:
2611                bond_update_slave_arr(bond, NULL);
2612                break;
2613        }
2614}
2615
2616static void bond_miimon_commit(struct bonding *bond)
2617{
2618        struct list_head *iter;
2619        struct slave *slave, *primary;
2620
2621        bond_for_each_slave(bond, slave, iter) {
2622                switch (slave->link_new_state) {
2623                case BOND_LINK_NOCHANGE:
2624                        /* For 802.3ad mode, check current slave speed and
2625                         * duplex again in case its port was disabled after
2626                         * invalid speed/duplex reporting but recovered before
2627                         * link monitoring could make a decision on the actual
2628                         * link status
2629                         */
2630                        if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2631                            slave->link == BOND_LINK_UP)
2632                                bond_3ad_adapter_speed_duplex_changed(slave);
2633                        continue;
2634
2635                case BOND_LINK_UP:
2636                        if (bond_update_speed_duplex(slave) &&
2637                            bond_needs_speed_duplex(bond)) {
2638                                slave->link = BOND_LINK_DOWN;
2639                                if (net_ratelimit())
2640                                        slave_warn(bond->dev, slave->dev,
2641                                                   "failed to get link speed/duplex\n");
2642                                continue;
2643                        }
2644                        bond_set_slave_link_state(slave, BOND_LINK_UP,
2645                                                  BOND_SLAVE_NOTIFY_NOW);
2646                        slave->last_link_up = jiffies;
2647
2648                        primary = rtnl_dereference(bond->primary_slave);
2649                        if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2650                                /* prevent it from being the active one */
2651                                bond_set_backup_slave(slave);
2652                        } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2653                                /* make it immediately active */
2654                                bond_set_active_slave(slave);
2655                        }
2656
2657                        slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2658                                   slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2659                                   slave->duplex ? "full" : "half");
2660
2661                        bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2662
2663                        if (!bond->curr_active_slave || slave == primary)
2664                                goto do_failover;
2665
2666                        continue;
2667
2668                case BOND_LINK_DOWN:
2669                        if (slave->link_failure_count < UINT_MAX)
2670                                slave->link_failure_count++;
2671
2672                        bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2673                                                  BOND_SLAVE_NOTIFY_NOW);
2674
2675                        if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2676                            BOND_MODE(bond) == BOND_MODE_8023AD)
2677                                bond_set_slave_inactive_flags(slave,
2678                                                              BOND_SLAVE_NOTIFY_NOW);
2679
2680                        slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2681
2682                        bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2683
2684                        if (slave == rcu_access_pointer(bond->curr_active_slave))
2685                                goto do_failover;
2686
2687                        continue;
2688
2689                default:
2690                        slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2691                                  slave->link_new_state);
2692                        bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2693
2694                        continue;
2695                }
2696
2697do_failover:
2698                block_netpoll_tx();
2699                bond_select_active_slave(bond);
2700                unblock_netpoll_tx();
2701        }
2702
2703        bond_set_carrier(bond);
2704}
2705
2706/* bond_mii_monitor
2707 *
2708 * Really a wrapper that splits the mii monitor into two phases: an
2709 * inspection, then (if inspection indicates something needs to be done)
2710 * an acquisition of appropriate locks followed by a commit phase to
2711 * implement whatever link state changes are indicated.
2712 */
2713static void bond_mii_monitor(struct work_struct *work)
2714{
2715        struct bonding *bond = container_of(work, struct bonding,
2716                                            mii_work.work);
2717        bool should_notify_peers = false;
2718        bool commit;
2719        unsigned long delay;
2720        struct slave *slave;
2721        struct list_head *iter;
2722
2723        delay = msecs_to_jiffies(bond->params.miimon);
2724
2725        if (!bond_has_slaves(bond))
2726                goto re_arm;
2727
2728        rcu_read_lock();
2729        should_notify_peers = bond_should_notify_peers(bond);
2730        commit = !!bond_miimon_inspect(bond);
2731        if (bond->send_peer_notif) {
2732                rcu_read_unlock();
2733                if (rtnl_trylock()) {
2734                        bond->send_peer_notif--;
2735                        rtnl_unlock();
2736                }
2737        } else {
2738                rcu_read_unlock();
2739        }
2740
2741        if (commit) {
2742                /* Race avoidance with bond_close cancel of workqueue */
2743                if (!rtnl_trylock()) {
2744                        delay = 1;
2745                        should_notify_peers = false;
2746                        goto re_arm;
2747                }
2748
2749                bond_for_each_slave(bond, slave, iter) {
2750                        bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2751                }
2752                bond_miimon_commit(bond);
2753
2754                rtnl_unlock();  /* might sleep, hold no other locks */
2755        }
2756
2757re_arm:
2758        if (bond->params.miimon)
2759                queue_delayed_work(bond->wq, &bond->mii_work, delay);
2760
2761        if (should_notify_peers) {
2762                if (!rtnl_trylock())
2763                        return;
2764                call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2765                rtnl_unlock();
2766        }
2767}
2768
2769static int bond_upper_dev_walk(struct net_device *upper,
2770                               struct netdev_nested_priv *priv)
2771{
2772        __be32 ip = *(__be32 *)priv->data;
2773
2774        return ip == bond_confirm_addr(upper, 0, ip);
2775}
2776
2777static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2778{
2779        struct netdev_nested_priv priv = {
2780                .data = (void *)&ip,
2781        };
2782        bool ret = false;
2783
2784        if (ip == bond_confirm_addr(bond->dev, 0, ip))
2785                return true;
2786
2787        rcu_read_lock();
2788        if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2789                ret = true;
2790        rcu_read_unlock();
2791
2792        return ret;
2793}
2794
2795/* We go to the (large) trouble of VLAN tagging ARP frames because
2796 * switches in VLAN mode (especially if ports are configured as
2797 * "native" to a VLAN) might not pass non-tagged frames.
2798 */
2799static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2800                          __be32 src_ip, struct bond_vlan_tag *tags)
2801{
2802        struct sk_buff *skb;
2803        struct bond_vlan_tag *outer_tag = tags;
2804        struct net_device *slave_dev = slave->dev;
2805        struct net_device *bond_dev = slave->bond->dev;
2806
2807        slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2808                  arp_op, &dest_ip, &src_ip);
2809
2810        skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2811                         NULL, slave_dev->dev_addr, NULL);
2812
2813        if (!skb) {
2814                net_err_ratelimited("ARP packet allocation failed\n");
2815                return;
2816        }
2817
2818        if (!tags || tags->vlan_proto == VLAN_N_VID)
2819                goto xmit;
2820
2821        tags++;
2822
2823        /* Go through all the tags backwards and add them to the packet */
2824        while (tags->vlan_proto != VLAN_N_VID) {
2825                if (!tags->vlan_id) {
2826                        tags++;
2827                        continue;
2828                }
2829
2830                slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2831                          ntohs(outer_tag->vlan_proto), tags->vlan_id);
2832                skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2833                                                tags->vlan_id);
2834                if (!skb) {
2835                        net_err_ratelimited("failed to insert inner VLAN tag\n");
2836                        return;
2837                }
2838
2839                tags++;
2840        }
2841        /* Set the outer tag */
2842        if (outer_tag->vlan_id) {
2843                slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2844                          ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2845                __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2846                                       outer_tag->vlan_id);
2847        }
2848
2849xmit:
2850        arp_xmit(skb);
2851}
2852
2853/* Validate the device path between the @start_dev and the @end_dev.
2854 * The path is valid if the @end_dev is reachable through device
2855 * stacking.
2856 * When the path is validated, collect any vlan information in the
2857 * path.
2858 */
2859struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2860                                              struct net_device *end_dev,
2861                                              int level)
2862{
2863        struct bond_vlan_tag *tags;
2864        struct net_device *upper;
2865        struct list_head  *iter;
2866
2867        if (start_dev == end_dev) {
2868                tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2869                if (!tags)
2870                        return ERR_PTR(-ENOMEM);
2871                tags[level].vlan_proto = VLAN_N_VID;
2872                return tags;
2873        }
2874
2875        netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2876                tags = bond_verify_device_path(upper, end_dev, level + 1);
2877                if (IS_ERR_OR_NULL(tags)) {
2878                        if (IS_ERR(tags))
2879                                return tags;
2880                        continue;
2881                }
2882                if (is_vlan_dev(upper)) {
2883                        tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2884                        tags[level].vlan_id = vlan_dev_vlan_id(upper);
2885                }
2886
2887                return tags;
2888        }
2889
2890        return NULL;
2891}
2892
2893static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2894{
2895        struct rtable *rt;
2896        struct bond_vlan_tag *tags;
2897        __be32 *targets = bond->params.arp_targets, addr;
2898        int i;
2899
2900        for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2901                slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2902                          __func__, &targets[i]);
2903                tags = NULL;
2904
2905                /* Find out through which dev should the packet go */
2906                rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2907                                     RTO_ONLINK, 0);
2908                if (IS_ERR(rt)) {
2909                        /* there's no route to target - try to send arp
2910                         * probe to generate any traffic (arp_validate=0)
2911                         */
2912                        if (bond->params.arp_validate)
2913                                pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2914                                             bond->dev->name,
2915                                             &targets[i]);
2916                        bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2917                                      0, tags);
2918                        continue;
2919                }
2920
2921                /* bond device itself */
2922                if (rt->dst.dev == bond->dev)
2923                        goto found;
2924
2925                rcu_read_lock();
2926                tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2927                rcu_read_unlock();
2928
2929                if (!IS_ERR_OR_NULL(tags))
2930                        goto found;
2931
2932                /* Not our device - skip */
2933                slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2934                           &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2935
2936                ip_rt_put(rt);
2937                continue;
2938
2939found:
2940                addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2941                ip_rt_put(rt);
2942                bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2943                kfree(tags);
2944        }
2945}
2946
2947static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2948{
2949        int i;
2950
2951        if (!sip || !bond_has_this_ip(bond, tip)) {
2952                slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2953                           __func__, &sip, &tip);
2954                return;
2955        }
2956
2957        i = bond_get_targets_ip(bond->params.arp_targets, sip);
2958        if (i == -1) {
2959                slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2960                           __func__, &sip);
2961                return;
2962        }
2963        slave->last_rx = jiffies;
2964        slave->target_last_arp_rx[i] = jiffies;
2965}
2966
2967int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2968                 struct slave *slave)
2969{
2970        struct arphdr *arp = (struct arphdr *)skb->data;
2971        struct slave *curr_active_slave, *curr_arp_slave;
2972        unsigned char *arp_ptr;
2973        __be32 sip, tip;
2974        int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2975        unsigned int alen;
2976
2977        if (!slave_do_arp_validate(bond, slave)) {
2978                if ((slave_do_arp_validate_only(bond) && is_arp) ||
2979                    !slave_do_arp_validate_only(bond))
2980                        slave->last_rx = jiffies;
2981                return RX_HANDLER_ANOTHER;
2982        } else if (!is_arp) {
2983                return RX_HANDLER_ANOTHER;
2984        }
2985
2986        alen = arp_hdr_len(bond->dev);
2987
2988        slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
2989                   __func__, skb->dev->name);
2990
2991        if (alen > skb_headlen(skb)) {
2992                arp = kmalloc(alen, GFP_ATOMIC);
2993                if (!arp)
2994                        goto out_unlock;
2995                if (skb_copy_bits(skb, 0, arp, alen) < 0)
2996                        goto out_unlock;
2997        }
2998
2999        if (arp->ar_hln != bond->dev->addr_len ||
3000            skb->pkt_type == PACKET_OTHERHOST ||
3001            skb->pkt_type == PACKET_LOOPBACK ||
3002            arp->ar_hrd != htons(ARPHRD_ETHER) ||
3003            arp->ar_pro != htons(ETH_P_IP) ||
3004            arp->ar_pln != 4)
3005                goto out_unlock;
3006
3007        arp_ptr = (unsigned char *)(arp + 1);
3008        arp_ptr += bond->dev->addr_len;
3009        memcpy(&sip, arp_ptr, 4);
3010        arp_ptr += 4 + bond->dev->addr_len;
3011        memcpy(&tip, arp_ptr, 4);
3012
3013        slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3014                  __func__, slave->dev->name, bond_slave_state(slave),
3015                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3016                  &sip, &tip);
3017
3018        curr_active_slave = rcu_dereference(bond->curr_active_slave);
3019        curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3020
3021        /* We 'trust' the received ARP enough to validate it if:
3022         *
3023         * (a) the slave receiving the ARP is active (which includes the
3024         * current ARP slave, if any), or
3025         *
3026         * (b) the receiving slave isn't active, but there is a currently
3027         * active slave and it received valid arp reply(s) after it became
3028         * the currently active slave, or
3029         *
3030         * (c) there is an ARP slave that sent an ARP during the prior ARP
3031         * interval, and we receive an ARP reply on any slave.  We accept
3032         * these because switch FDB update delays may deliver the ARP
3033         * reply to a slave other than the sender of the ARP request.
3034         *
3035         * Note: for (b), backup slaves are receiving the broadcast ARP
3036         * request, not a reply.  This request passes from the sending
3037         * slave through the L2 switch(es) to the receiving slave.  Since
3038         * this is checking the request, sip/tip are swapped for
3039         * validation.
3040         *
3041         * This is done to avoid endless looping when we can't reach the
3042         * arp_ip_target and fool ourselves with our own arp requests.
3043         */
3044        if (bond_is_active_slave(slave))
3045                bond_validate_arp(bond, slave, sip, tip);
3046        else if (curr_active_slave &&
3047                 time_after(slave_last_rx(bond, curr_active_slave),
3048                            curr_active_slave->last_link_up))
3049                bond_validate_arp(bond, slave, tip, sip);
3050        else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3051                 bond_time_in_interval(bond,
3052                                       dev_trans_start(curr_arp_slave->dev), 1))
3053                bond_validate_arp(bond, slave, sip, tip);
3054
3055out_unlock:
3056        if (arp != (struct arphdr *)skb->data)
3057                kfree(arp);
3058        return RX_HANDLER_ANOTHER;
3059}
3060
3061/* function to verify if we're in the arp_interval timeslice, returns true if
3062 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3063 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3064 */
3065static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3066                                  int mod)
3067{
3068        int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3069
3070        return time_in_range(jiffies,
3071                             last_act - delta_in_ticks,
3072                             last_act + mod * delta_in_ticks + delta_in_ticks/2);
3073}
3074
3075/* This function is called regularly to monitor each slave's link
3076 * ensuring that traffic is being sent and received when arp monitoring
3077 * is used in load-balancing mode. if the adapter has been dormant, then an
3078 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3079 * arp monitoring in active backup mode.
3080 */
3081static void bond_loadbalance_arp_mon(struct bonding *bond)
3082{
3083        struct slave *slave, *oldcurrent;
3084        struct list_head *iter;
3085        int do_failover = 0, slave_state_changed = 0;
3086
3087        if (!bond_has_slaves(bond))
3088                goto re_arm;
3089
3090        rcu_read_lock();
3091
3092        oldcurrent = rcu_dereference(bond->curr_active_slave);
3093        /* see if any of the previous devices are up now (i.e. they have
3094         * xmt and rcv traffic). the curr_active_slave does not come into
3095         * the picture unless it is null. also, slave->last_link_up is not
3096         * needed here because we send an arp on each slave and give a slave
3097         * as long as it needs to get the tx/rx within the delta.
3098         * TODO: what about up/down delay in arp mode? it wasn't here before
3099         *       so it can wait
3100         */
3101        bond_for_each_slave_rcu(bond, slave, iter) {
3102                unsigned long trans_start = dev_trans_start(slave->dev);
3103
3104                bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3105
3106                if (slave->link != BOND_LINK_UP) {
3107                        if (bond_time_in_interval(bond, trans_start, 1) &&
3108                            bond_time_in_interval(bond, slave->last_rx, 1)) {
3109
3110                                bond_propose_link_state(slave, BOND_LINK_UP);
3111                                slave_state_changed = 1;
3112
3113                                /* primary_slave has no meaning in round-robin
3114                                 * mode. the window of a slave being up and
3115                                 * curr_active_slave being null after enslaving
3116                                 * is closed.
3117                                 */
3118                                if (!oldcurrent) {
3119                                        slave_info(bond->dev, slave->dev, "link status definitely up\n");
3120                                        do_failover = 1;
3121                                } else {
3122                                        slave_info(bond->dev, slave->dev, "interface is now up\n");
3123                                }
3124                        }
3125                } else {
3126                        /* slave->link == BOND_LINK_UP */
3127
3128                        /* not all switches will respond to an arp request
3129                         * when the source ip is 0, so don't take the link down
3130                         * if we don't know our ip yet
3131                         */
3132                        if (!bond_time_in_interval(bond, trans_start, 2) ||
3133                            !bond_time_in_interval(bond, slave->last_rx, 2)) {
3134
3135                                bond_propose_link_state(slave, BOND_LINK_DOWN);
3136                                slave_state_changed = 1;
3137
3138                                if (slave->link_failure_count < UINT_MAX)
3139                                        slave->link_failure_count++;
3140
3141                                slave_info(bond->dev, slave->dev, "interface is now down\n");
3142
3143                                if (slave == oldcurrent)
3144                                        do_failover = 1;
3145                        }
3146                }
3147
3148                /* note: if switch is in round-robin mode, all links
3149                 * must tx arp to ensure all links rx an arp - otherwise
3150                 * links may oscillate or not come up at all; if switch is
3151                 * in something like xor mode, there is nothing we can
3152                 * do - all replies will be rx'ed on same link causing slaves
3153                 * to be unstable during low/no traffic periods
3154                 */
3155                if (bond_slave_is_up(slave))
3156                        bond_arp_send_all(bond, slave);
3157        }
3158
3159        rcu_read_unlock();
3160
3161        if (do_failover || slave_state_changed) {
3162                if (!rtnl_trylock())
3163                        goto re_arm;
3164
3165                bond_for_each_slave(bond, slave, iter) {
3166                        if (slave->link_new_state != BOND_LINK_NOCHANGE)
3167                                slave->link = slave->link_new_state;
3168                }
3169
3170                if (slave_state_changed) {
3171                        bond_slave_state_change(bond);
3172                        if (BOND_MODE(bond) == BOND_MODE_XOR)
3173                                bond_update_slave_arr(bond, NULL);
3174                }
3175                if (do_failover) {
3176                        block_netpoll_tx();
3177                        bond_select_active_slave(bond);
3178                        unblock_netpoll_tx();
3179                }
3180                rtnl_unlock();
3181        }
3182
3183re_arm:
3184        if (bond->params.arp_interval)
3185                queue_delayed_work(bond->wq, &bond->arp_work,
3186                                   msecs_to_jiffies(bond->params.arp_interval));
3187}
3188
3189/* Called to inspect slaves for active-backup mode ARP monitor link state
3190 * changes.  Sets proposed link state in slaves to specify what action
3191 * should take place for the slave.  Returns 0 if no changes are found, >0
3192 * if changes to link states must be committed.
3193 *
3194 * Called with rcu_read_lock held.
3195 */
3196static int bond_ab_arp_inspect(struct bonding *bond)
3197{
3198        unsigned long trans_start, last_rx;
3199        struct list_head *iter;
3200        struct slave *slave;
3201        int commit = 0;
3202
3203        bond_for_each_slave_rcu(bond, slave, iter) {
3204                bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3205                last_rx = slave_last_rx(bond, slave);
3206
3207                if (slave->link != BOND_LINK_UP) {
3208                        if (bond_time_in_interval(bond, last_rx, 1)) {
3209                                bond_propose_link_state(slave, BOND_LINK_UP);
3210                                commit++;
3211                        } else if (slave->link == BOND_LINK_BACK) {
3212                                bond_propose_link_state(slave, BOND_LINK_FAIL);
3213                                commit++;
3214                        }
3215                        continue;
3216                }
3217
3218                /* Give slaves 2*delta after being enslaved or made
3219                 * active.  This avoids bouncing, as the last receive
3220                 * times need a full ARP monitor cycle to be updated.
3221                 */
3222                if (bond_time_in_interval(bond, slave->last_link_up, 2))
3223                        continue;
3224
3225                /* Backup slave is down if:
3226                 * - No current_arp_slave AND
3227                 * - more than 3*delta since last receive AND
3228                 * - the bond has an IP address
3229                 *
3230                 * Note: a non-null current_arp_slave indicates
3231                 * the curr_active_slave went down and we are
3232                 * searching for a new one; under this condition
3233                 * we only take the curr_active_slave down - this
3234                 * gives each slave a chance to tx/rx traffic
3235                 * before being taken out
3236                 */
3237                if (!bond_is_active_slave(slave) &&
3238                    !rcu_access_pointer(bond->current_arp_slave) &&
3239                    !bond_time_in_interval(bond, last_rx, 3)) {
3240                        bond_propose_link_state(slave, BOND_LINK_DOWN);
3241                        commit++;
3242                }
3243
3244                /* Active slave is down if:
3245                 * - more than 2*delta since transmitting OR
3246                 * - (more than 2*delta since receive AND
3247                 *    the bond has an IP address)
3248                 */
3249                trans_start = dev_trans_start(slave->dev);
3250                if (bond_is_active_slave(slave) &&
3251                    (!bond_time_in_interval(bond, trans_start, 2) ||
3252                     !bond_time_in_interval(bond, last_rx, 2))) {
3253                        bond_propose_link_state(slave, BOND_LINK_DOWN);
3254                        commit++;
3255                }
3256        }
3257
3258        return commit;
3259}
3260
3261/* Called to commit link state changes noted by inspection step of
3262 * active-backup mode ARP monitor.
3263 *
3264 * Called with RTNL hold.
3265 */
3266static void bond_ab_arp_commit(struct bonding *bond)
3267{
3268        unsigned long trans_start;
3269        struct list_head *iter;
3270        struct slave *slave;
3271
3272        bond_for_each_slave(bond, slave, iter) {
3273                switch (slave->link_new_state) {
3274                case BOND_LINK_NOCHANGE:
3275                        continue;
3276
3277                case BOND_LINK_UP:
3278                        trans_start = dev_trans_start(slave->dev);
3279                        if (rtnl_dereference(bond->curr_active_slave) != slave ||
3280                            (!rtnl_dereference(bond->curr_active_slave) &&
3281                             bond_time_in_interval(bond, trans_start, 1))) {
3282                                struct slave *current_arp_slave;
3283
3284                                current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3285                                bond_set_slave_link_state(slave, BOND_LINK_UP,
3286                                                          BOND_SLAVE_NOTIFY_NOW);
3287                                if (current_arp_slave) {
3288                                        bond_set_slave_inactive_flags(
3289                                                current_arp_slave,
3290                                                BOND_SLAVE_NOTIFY_NOW);
3291                                        RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3292                                }
3293
3294                                slave_info(bond->dev, slave->dev, "link status definitely up\n");
3295
3296                                if (!rtnl_dereference(bond->curr_active_slave) ||
3297                                    slave == rtnl_dereference(bond->primary_slave))
3298                                        goto do_failover;
3299
3300                        }
3301
3302                        continue;
3303
3304                case BOND_LINK_DOWN:
3305                        if (slave->link_failure_count < UINT_MAX)
3306                                slave->link_failure_count++;
3307
3308                        bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3309                                                  BOND_SLAVE_NOTIFY_NOW);
3310                        bond_set_slave_inactive_flags(slave,
3311                                                      BOND_SLAVE_NOTIFY_NOW);
3312
3313                        slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3314
3315                        if (slave == rtnl_dereference(bond->curr_active_slave)) {
3316                                RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3317                                goto do_failover;
3318                        }
3319
3320                        continue;
3321
3322                case BOND_LINK_FAIL:
3323                        bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3324                                                  BOND_SLAVE_NOTIFY_NOW);
3325                        bond_set_slave_inactive_flags(slave,
3326                                                      BOND_SLAVE_NOTIFY_NOW);
3327
3328                        /* A slave has just been enslaved and has become
3329                         * the current active slave.
3330                         */
3331                        if (rtnl_dereference(bond->curr_active_slave))
3332                                RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3333                        continue;
3334
3335                default:
3336                        slave_err(bond->dev, slave->dev,
3337                                  "impossible: link_new_state %d on slave\n",
3338                                  slave->link_new_state);
3339                        continue;
3340                }
3341
3342do_failover:
3343                block_netpoll_tx();
3344                bond_select_active_slave(bond);
3345                unblock_netpoll_tx();
3346        }
3347
3348        bond_set_carrier(bond);
3349}
3350
3351/* Send ARP probes for active-backup mode ARP monitor.
3352 *
3353 * Called with rcu_read_lock held.
3354 */
3355static bool bond_ab_arp_probe(struct bonding *bond)
3356{
3357        struct slave *slave, *before = NULL, *new_slave = NULL,
3358                     *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3359                     *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3360        struct list_head *iter;
3361        bool found = false;
3362        bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3363
3364        if (curr_arp_slave && curr_active_slave)
3365                netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3366                            curr_arp_slave->dev->name,
3367                            curr_active_slave->dev->name);
3368
3369        if (curr_active_slave) {
3370                bond_arp_send_all(bond, curr_active_slave);
3371                return should_notify_rtnl;
3372        }
3373
3374        /* if we don't have a curr_active_slave, search for the next available
3375         * backup slave from the current_arp_slave and make it the candidate
3376         * for becoming the curr_active_slave
3377         */
3378
3379        if (!curr_arp_slave) {
3380                curr_arp_slave = bond_first_slave_rcu(bond);
3381                if (!curr_arp_slave)
3382                        return should_notify_rtnl;
3383        }
3384
3385        bond_for_each_slave_rcu(bond, slave, iter) {
3386                if (!found && !before && bond_slave_is_up(slave))
3387                        before = slave;
3388
3389                if (found && !new_slave && bond_slave_is_up(slave))
3390                        new_slave = slave;
3391                /* if the link state is up at this point, we
3392                 * mark it down - this can happen if we have
3393                 * simultaneous link failures and
3394                 * reselect_active_interface doesn't make this
3395                 * one the current slave so it is still marked
3396                 * up when it is actually down
3397                 */
3398                if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3399                        bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3400                                                  BOND_SLAVE_NOTIFY_LATER);
3401                        if (slave->link_failure_count < UINT_MAX)
3402                                slave->link_failure_count++;
3403
3404                        bond_set_slave_inactive_flags(slave,
3405                                                      BOND_SLAVE_NOTIFY_LATER);
3406
3407                        slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3408                }
3409                if (slave == curr_arp_slave)
3410                        found = true;
3411        }
3412
3413        if (!new_slave && before)
3414                new_slave = before;
3415
3416        if (!new_slave)
3417                goto check_state;
3418
3419        bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3420                                  BOND_SLAVE_NOTIFY_LATER);
3421        bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3422        bond_arp_send_all(bond, new_slave);
3423        new_slave->last_link_up = jiffies;
3424        rcu_assign_pointer(bond->current_arp_slave, new_slave);
3425
3426check_state:
3427        bond_for_each_slave_rcu(bond, slave, iter) {
3428                if (slave->should_notify || slave->should_notify_link) {
3429                        should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3430                        break;
3431                }
3432        }
3433        return should_notify_rtnl;
3434}
3435
3436static void bond_activebackup_arp_mon(struct bonding *bond)
3437{
3438        bool should_notify_peers = false;
3439        bool should_notify_rtnl = false;
3440        int delta_in_ticks;
3441
3442        delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3443
3444        if (!bond_has_slaves(bond))
3445                goto re_arm;
3446
3447        rcu_read_lock();
3448
3449        should_notify_peers = bond_should_notify_peers(bond);
3450
3451        if (bond_ab_arp_inspect(bond)) {
3452                rcu_read_unlock();
3453
3454                /* Race avoidance with bond_close flush of workqueue */
3455                if (!rtnl_trylock()) {
3456                        delta_in_ticks = 1;
3457                        should_notify_peers = false;
3458                        goto re_arm;
3459                }
3460
3461                bond_ab_arp_commit(bond);
3462
3463                rtnl_unlock();
3464                rcu_read_lock();
3465        }
3466
3467        should_notify_rtnl = bond_ab_arp_probe(bond);
3468        rcu_read_unlock();
3469
3470re_arm:
3471        if (bond->params.arp_interval)
3472                queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3473
3474        if (should_notify_peers || should_notify_rtnl) {
3475                if (!rtnl_trylock())
3476                        return;
3477
3478                if (should_notify_peers)
3479                        call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3480                                                 bond->dev);
3481                if (should_notify_rtnl) {
3482                        bond_slave_state_notify(bond);
3483                        bond_slave_link_notify(bond);
3484                }
3485
3486                rtnl_unlock();
3487        }
3488}
3489
3490static void bond_arp_monitor(struct work_struct *work)
3491{
3492        struct bonding *bond = container_of(work, struct bonding,
3493                                            arp_work.work);
3494
3495        if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3496                bond_activebackup_arp_mon(bond);
3497        else
3498                bond_loadbalance_arp_mon(bond);
3499}
3500
3501/*-------------------------- netdev event handling --------------------------*/
3502
3503/* Change device name */
3504static int bond_event_changename(struct bonding *bond)
3505{
3506        bond_remove_proc_entry(bond);
3507        bond_create_proc_entry(bond);
3508
3509        bond_debug_reregister(bond);
3510
3511        return NOTIFY_DONE;
3512}
3513
3514static int bond_master_netdev_event(unsigned long event,
3515                                    struct net_device *bond_dev)
3516{
3517        struct bonding *event_bond = netdev_priv(bond_dev);
3518
3519        netdev_dbg(bond_dev, "%s called\n", __func__);
3520
3521        switch (event) {
3522        case NETDEV_CHANGENAME:
3523                return bond_event_changename(event_bond);
3524        case NETDEV_UNREGISTER:
3525                bond_remove_proc_entry(event_bond);
3526#ifdef CONFIG_XFRM_OFFLOAD
3527                xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3528#endif /* CONFIG_XFRM_OFFLOAD */
3529                break;
3530        case NETDEV_REGISTER:
3531                bond_create_proc_entry(event_bond);
3532                break;
3533        default:
3534                break;
3535        }
3536
3537        return NOTIFY_DONE;
3538}
3539
3540static int bond_slave_netdev_event(unsigned long event,
3541                                   struct net_device *slave_dev)
3542{
3543        struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3544        struct bonding *bond;
3545        struct net_device *bond_dev;
3546
3547        /* A netdev event can be generated while enslaving a device
3548         * before netdev_rx_handler_register is called in which case
3549         * slave will be NULL
3550         */
3551        if (!slave) {
3552                netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3553                return NOTIFY_DONE;
3554        }
3555
3556        bond_dev = slave->bond->dev;
3557        bond = slave->bond;
3558        primary = rtnl_dereference(bond->primary_slave);
3559
3560        slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3561
3562        switch (event) {
3563        case NETDEV_UNREGISTER:
3564                if (bond_dev->type != ARPHRD_ETHER)
3565                        bond_release_and_destroy(bond_dev, slave_dev);
3566                else
3567                        __bond_release_one(bond_dev, slave_dev, false, true);
3568                break;
3569        case NETDEV_UP:
3570        case NETDEV_CHANGE:
3571                /* For 802.3ad mode only:
3572                 * Getting invalid Speed/Duplex values here will put slave
3573                 * in weird state. Mark it as link-fail if the link was
3574                 * previously up or link-down if it hasn't yet come up, and
3575                 * let link-monitoring (miimon) set it right when correct
3576                 * speeds/duplex are available.
3577                 */
3578                if (bond_update_speed_duplex(slave) &&
3579                    BOND_MODE(bond) == BOND_MODE_8023AD) {
3580                        if (slave->last_link_up)
3581                                slave->link = BOND_LINK_FAIL;
3582                        else
3583                                slave->link = BOND_LINK_DOWN;
3584                }
3585
3586                if (BOND_MODE(bond) == BOND_MODE_8023AD)
3587                        bond_3ad_adapter_speed_duplex_changed(slave);
3588                fallthrough;
3589        case NETDEV_DOWN:
3590                /* Refresh slave-array if applicable!
3591                 * If the setup does not use miimon or arpmon (mode-specific!),
3592                 * then these events will not cause the slave-array to be
3593                 * refreshed. This will cause xmit to use a slave that is not
3594                 * usable. Avoid such situation by refeshing the array at these
3595                 * events. If these (miimon/arpmon) parameters are configured
3596                 * then array gets refreshed twice and that should be fine!
3597                 */
3598                if (bond_mode_can_use_xmit_hash(bond))
3599                        bond_update_slave_arr(bond, NULL);
3600                break;
3601        case NETDEV_CHANGEMTU:
3602                /* TODO: Should slaves be allowed to
3603                 * independently alter their MTU?  For
3604                 * an active-backup bond, slaves need
3605                 * not be the same type of device, so
3606                 * MTUs may vary.  For other modes,
3607                 * slaves arguably should have the
3608                 * same MTUs. To do this, we'd need to
3609                 * take over the slave's change_mtu
3610                 * function for the duration of their
3611                 * servitude.
3612                 */
3613                break;
3614        case NETDEV_CHANGENAME:
3615                /* we don't care if we don't have primary set */
3616                if (!bond_uses_primary(bond) ||
3617                    !bond->params.primary[0])
3618                        break;
3619
3620                if (slave == primary) {
3621                        /* slave's name changed - he's no longer primary */
3622                        RCU_INIT_POINTER(bond->primary_slave, NULL);
3623                } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3624                        /* we have a new primary slave */
3625                        rcu_assign_pointer(bond->primary_slave, slave);
3626                } else { /* we didn't change primary - exit */
3627                        break;
3628                }
3629
3630                netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3631                            primary ? slave_dev->name : "none");
3632
3633                block_netpoll_tx();
3634                bond_select_active_slave(bond);
3635                unblock_netpoll_tx();
3636                break;
3637        case NETDEV_FEAT_CHANGE:
3638                bond_compute_features(bond);
3639                break;
3640        case NETDEV_RESEND_IGMP:
3641                /* Propagate to master device */
3642                call_netdevice_notifiers(event, slave->bond->dev);
3643                break;
3644        default:
3645                break;
3646        }
3647
3648        return NOTIFY_DONE;
3649}
3650
3651/* bond_netdev_event: handle netdev notifier chain events.
3652 *
3653 * This function receives events for the netdev chain.  The caller (an
3654 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3655 * locks for us to safely manipulate the slave devices (RTNL lock,
3656 * dev_probe_lock).
3657 */
3658static int bond_netdev_event(struct notifier_block *this,
3659                             unsigned long event, void *ptr)
3660{
3661        struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3662
3663        netdev_dbg(event_dev, "%s received %s\n",
3664                   __func__, netdev_cmd_to_name(event));
3665
3666        if (!(event_dev->priv_flags & IFF_BONDING))
3667                return NOTIFY_DONE;
3668
3669        if (event_dev->flags & IFF_MASTER) {
3670                int ret;
3671
3672                ret = bond_master_netdev_event(event, event_dev);
3673                if (ret != NOTIFY_DONE)
3674                        return ret;
3675        }
3676
3677        if (event_dev->flags & IFF_SLAVE)
3678                return bond_slave_netdev_event(event, event_dev);
3679
3680        return NOTIFY_DONE;
3681}
3682
3683static struct notifier_block bond_netdev_notifier = {
3684        .notifier_call = bond_netdev_event,
3685};
3686
3687/*---------------------------- Hashing Policies -----------------------------*/
3688
3689/* Helper to access data in a packet, with or without a backing skb.
3690 * If skb is given the data is linearized if necessary via pskb_may_pull.
3691 */
3692static inline const void *bond_pull_data(struct sk_buff *skb,
3693                                         const void *data, int hlen, int n)
3694{
3695        if (likely(n <= hlen))
3696                return data;
3697        else if (skb && likely(pskb_may_pull(skb, n)))
3698                return skb->head;
3699
3700        return NULL;
3701}
3702
3703/* L2 hash helper */
3704static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
3705{
3706        struct ethhdr *ep;
3707
3708        data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
3709        if (!data)
3710                return 0;
3711
3712        ep = (struct ethhdr *)(data + mhoff);
3713        return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
3714}
3715
3716static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
3717                         int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
3718{
3719        const struct ipv6hdr *iph6;
3720        const struct iphdr *iph;
3721
3722        if (l2_proto == htons(ETH_P_IP)) {
3723                data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
3724                if (!data)
3725                        return false;
3726
3727                iph = (const struct iphdr *)(data + *nhoff);
3728                iph_to_flow_copy_v4addrs(fk, iph);
3729                *nhoff += iph->ihl << 2;
3730                if (!ip_is_fragment(iph))
3731                        *ip_proto = iph->protocol;
3732        } else if (l2_proto == htons(ETH_P_IPV6)) {
3733                data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
3734                if (!data)
3735                        return false;
3736
3737                iph6 = (const struct ipv6hdr *)(data + *nhoff);
3738                iph_to_flow_copy_v6addrs(fk, iph6);
3739                *nhoff += sizeof(*iph6);
3740                *ip_proto = iph6->nexthdr;
3741        } else {
3742                return false;
3743        }
3744
3745        if (l34 && *ip_proto >= 0)
3746                fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
3747
3748        return true;
3749}
3750
3751static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
3752{
3753        u32 srcmac_vendor = 0, srcmac_dev = 0;
3754        struct ethhdr *mac_hdr;
3755        u16 vlan = 0;
3756        int i;
3757
3758        data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
3759        if (!data)
3760                return 0;
3761        mac_hdr = (struct ethhdr *)(data + mhoff);
3762
3763        for (i = 0; i < 3; i++)
3764                srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
3765
3766        for (i = 3; i < ETH_ALEN; i++)
3767                srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
3768
3769        if (skb && skb_vlan_tag_present(skb))
3770                vlan = skb_vlan_tag_get(skb);
3771
3772        return vlan ^ srcmac_vendor ^ srcmac_dev;
3773}
3774
3775/* Extract the appropriate headers based on bond's xmit policy */
3776static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
3777                              __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
3778{
3779        bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
3780        int ip_proto = -1;
3781
3782        switch (bond->params.xmit_policy) {
3783        case BOND_XMIT_POLICY_ENCAP23:
3784        case BOND_XMIT_POLICY_ENCAP34:
3785                memset(fk, 0, sizeof(*fk));
3786                return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3787                                          fk, data, l2_proto, nhoff, hlen, 0);
3788        default:
3789                break;
3790        }
3791
3792        fk->ports.ports = 0;
3793        memset(&fk->icmp, 0, sizeof(fk->icmp));
3794        if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
3795                return false;
3796
3797        /* ICMP error packets contains at least 8 bytes of the header
3798         * of the packet which generated the error. Use this information
3799         * to correlate ICMP error packets within the same flow which
3800         * generated the error.
3801         */
3802        if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
3803                skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
3804                if (ip_proto == IPPROTO_ICMP) {
3805                        if (!icmp_is_err(fk->icmp.type))
3806                                return true;
3807
3808                        nhoff += sizeof(struct icmphdr);
3809                } else if (ip_proto == IPPROTO_ICMPV6) {
3810                        if (!icmpv6_is_err(fk->icmp.type))
3811                                return true;
3812
3813                        nhoff += sizeof(struct icmp6hdr);
3814                }
3815                return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
3816        }
3817
3818        return true;
3819}
3820
3821static u32 bond_ip_hash(u32 hash, struct flow_keys *flow)
3822{
3823        hash ^= (__force u32)flow_get_u32_dst(flow) ^
3824                (__force u32)flow_get_u32_src(flow);
3825        hash ^= (hash >> 16);
3826        hash ^= (hash >> 8);
3827        /* discard lowest hash bit to deal with the common even ports pattern */
3828        return hash >> 1;
3829}
3830
3831/* Generate hash based on xmit policy. If @skb is given it is used to linearize
3832 * the data as required, but this function can be used without it if the data is
3833 * known to be linear (e.g. with xdp_buff).
3834 */
3835static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
3836                            __be16 l2_proto, int mhoff, int nhoff, int hlen)
3837{
3838        struct flow_keys flow;
3839        u32 hash;
3840
3841        if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
3842                return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
3843
3844        if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3845            !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
3846                return bond_eth_hash(skb, data, mhoff, hlen);
3847
3848        if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3849            bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
3850                hash = bond_eth_hash(skb, data, mhoff, hlen);
3851        } else {
3852                if (flow.icmp.id)
3853                        memcpy(&hash, &flow.icmp, sizeof(hash));
3854                else
3855                        memcpy(&hash, &flow.ports.ports, sizeof(hash));
3856        }
3857
3858        return bond_ip_hash(hash, &flow);
3859}
3860
3861/**
3862 * bond_xmit_hash - generate a hash value based on the xmit policy
3863 * @bond: bonding device
3864 * @skb: buffer to use for headers
3865 *
3866 * This function will extract the necessary headers from the skb buffer and use
3867 * them to generate a hash based on the xmit_policy set in the bonding device
3868 */
3869u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3870{
3871        if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3872            skb->l4_hash)
3873                return skb->hash;
3874
3875        return __bond_xmit_hash(bond, skb, skb->head, skb->protocol,
3876                                skb->mac_header, skb->network_header,
3877                                skb_headlen(skb));
3878}
3879
3880/**
3881 * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
3882 * @bond: bonding device
3883 * @xdp: buffer to use for headers
3884 *
3885 * The XDP variant of bond_xmit_hash.
3886 */
3887static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
3888{
3889        struct ethhdr *eth;
3890
3891        if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
3892                return 0;
3893
3894        eth = (struct ethhdr *)xdp->data;
3895
3896        return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
3897                                sizeof(struct ethhdr), xdp->data_end - xdp->data);
3898}
3899
3900/*-------------------------- Device entry points ----------------------------*/
3901
3902void bond_work_init_all(struct bonding *bond)
3903{
3904        INIT_DELAYED_WORK(&bond->mcast_work,
3905                          bond_resend_igmp_join_requests_delayed);
3906        INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3907        INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3908        INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3909        INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3910        INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3911}
3912
3913static void bond_work_cancel_all(struct bonding *bond)
3914{
3915        cancel_delayed_work_sync(&bond->mii_work);
3916        cancel_delayed_work_sync(&bond->arp_work);
3917        cancel_delayed_work_sync(&bond->alb_work);
3918        cancel_delayed_work_sync(&bond->ad_work);
3919        cancel_delayed_work_sync(&bond->mcast_work);
3920        cancel_delayed_work_sync(&bond->slave_arr_work);
3921}
3922
3923static int bond_open(struct net_device *bond_dev)
3924{
3925        struct bonding *bond = netdev_priv(bond_dev);
3926        struct list_head *iter;
3927        struct slave *slave;
3928
3929        /* reset slave->backup and slave->inactive */
3930        if (bond_has_slaves(bond)) {
3931                bond_for_each_slave(bond, slave, iter) {
3932                        if (bond_uses_primary(bond) &&
3933                            slave != rcu_access_pointer(bond->curr_active_slave)) {
3934                                bond_set_slave_inactive_flags(slave,
3935                                                              BOND_SLAVE_NOTIFY_NOW);
3936                        } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3937                                bond_set_slave_active_flags(slave,
3938                                                            BOND_SLAVE_NOTIFY_NOW);
3939                        }
3940                }
3941        }
3942
3943        if (bond_is_lb(bond)) {
3944                /* bond_alb_initialize must be called before the timer
3945                 * is started.
3946                 */
3947                if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3948                        return -ENOMEM;
3949                if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3950                        queue_delayed_work(bond->wq, &bond->alb_work, 0);
3951        }
3952
3953        if (bond->params.miimon)  /* link check interval, in milliseconds. */
3954                queue_delayed_work(bond->wq, &bond->mii_work, 0);
3955
3956        if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3957                queue_delayed_work(bond->wq, &bond->arp_work, 0);
3958                bond->recv_probe = bond_arp_rcv;
3959        }
3960
3961        if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3962                queue_delayed_work(bond->wq, &bond->ad_work, 0);
3963                /* register to receive LACPDUs */
3964                bond->recv_probe = bond_3ad_lacpdu_recv;
3965                bond_3ad_initiate_agg_selection(bond, 1);
3966        }
3967
3968        if (bond_mode_can_use_xmit_hash(bond))
3969                bond_update_slave_arr(bond, NULL);
3970
3971        return 0;
3972}
3973
3974static int bond_close(struct net_device *bond_dev)
3975{
3976        struct bonding *bond = netdev_priv(bond_dev);
3977
3978        bond_work_cancel_all(bond);
3979        bond->send_peer_notif = 0;
3980        if (bond_is_lb(bond))
3981                bond_alb_deinitialize(bond);
3982        bond->recv_probe = NULL;
3983
3984        return 0;
3985}
3986
3987/* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3988 * that some drivers can provide 32bit values only.
3989 */
3990static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3991                            const struct rtnl_link_stats64 *_new,
3992                            const struct rtnl_link_stats64 *_old)
3993{
3994        const u64 *new = (const u64 *)_new;
3995        const u64 *old = (const u64 *)_old;
3996        u64 *res = (u64 *)_res;
3997        int i;
3998
3999        for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4000                u64 nv = new[i];
4001                u64 ov = old[i];
4002                s64 delta = nv - ov;
4003
4004                /* detects if this particular field is 32bit only */
4005                if (((nv | ov) >> 32) == 0)
4006                        delta = (s64)(s32)((u32)nv - (u32)ov);
4007
4008                /* filter anomalies, some drivers reset their stats
4009                 * at down/up events.
4010                 */
4011                if (delta > 0)
4012                        res[i] += delta;
4013        }
4014}
4015
4016#ifdef CONFIG_LOCKDEP
4017static int bond_get_lowest_level_rcu(struct net_device *dev)
4018{
4019        struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4020        struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4021        int cur = 0, max = 0;
4022
4023        now = dev;
4024        iter = &dev->adj_list.lower;
4025
4026        while (1) {
4027                next = NULL;
4028                while (1) {
4029                        ldev = netdev_next_lower_dev_rcu(now, &iter);
4030                        if (!ldev)
4031                                break;
4032
4033                        next = ldev;
4034                        niter = &ldev->adj_list.lower;
4035                        dev_stack[cur] = now;
4036                        iter_stack[cur++] = iter;
4037                        if (max <= cur)
4038                                max = cur;
4039                        break;
4040                }
4041
4042                if (!next) {
4043                        if (!cur)
4044                                return max;
4045                        next = dev_stack[--cur];
4046                        niter = iter_stack[cur];
4047                }
4048
4049                now = next;
4050                iter = niter;
4051        }
4052
4053        return max;
4054}
4055#endif
4056
4057static void bond_get_stats(struct net_device *bond_dev,
4058                           struct rtnl_link_stats64 *stats)
4059{
4060        struct bonding *bond = netdev_priv(bond_dev);
4061        struct rtnl_link_stats64 temp;
4062        struct list_head *iter;
4063        struct slave *slave;
4064        int nest_level = 0;
4065
4066
4067        rcu_read_lock();
4068#ifdef CONFIG_LOCKDEP
4069        nest_level = bond_get_lowest_level_rcu(bond_dev);
4070#endif
4071
4072        spin_lock_nested(&bond->stats_lock, nest_level);
4073        memcpy(stats, &bond->bond_stats, sizeof(*stats));
4074
4075        bond_for_each_slave_rcu(bond, slave, iter) {
4076                const struct rtnl_link_stats64 *new =
4077                        dev_get_stats(slave->dev, &temp);
4078
4079                bond_fold_stats(stats, new, &slave->slave_stats);
4080
4081                /* save off the slave stats for the next run */
4082                memcpy(&slave->slave_stats, new, sizeof(*new));
4083        }
4084
4085        memcpy(&bond->bond_stats, stats, sizeof(*stats));
4086        spin_unlock(&bond->stats_lock);
4087        rcu_read_unlock();
4088}
4089
4090static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4091{
4092        struct bonding *bond = netdev_priv(bond_dev);
4093        struct mii_ioctl_data *mii = NULL;
4094        int res;
4095
4096        netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4097
4098        switch (cmd) {
4099        case SIOCGMIIPHY:
4100                mii = if_mii(ifr);
4101                if (!mii)
4102                        return -EINVAL;
4103
4104                mii->phy_id = 0;
4105                fallthrough;
4106        case SIOCGMIIREG:
4107                /* We do this again just in case we were called by SIOCGMIIREG
4108                 * instead of SIOCGMIIPHY.
4109                 */
4110                mii = if_mii(ifr);
4111                if (!mii)
4112                        return -EINVAL;
4113
4114                if (mii->reg_num == 1) {
4115                        mii->val_out = 0;
4116                        if (netif_carrier_ok(bond->dev))
4117                                mii->val_out = BMSR_LSTATUS;
4118                }
4119
4120                return 0;
4121        default:
4122                res = -EOPNOTSUPP;
4123        }
4124
4125        return res;
4126}
4127
4128static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4129{
4130        struct bonding *bond = netdev_priv(bond_dev);
4131        struct net_device *slave_dev = NULL;
4132        struct ifbond k_binfo;
4133        struct ifbond __user *u_binfo = NULL;
4134        struct ifslave k_sinfo;
4135        struct ifslave __user *u_sinfo = NULL;
4136        struct bond_opt_value newval;
4137        struct net *net;
4138        int res = 0;
4139
4140        netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4141
4142        switch (cmd) {
4143        case SIOCBONDINFOQUERY:
4144                u_binfo = (struct ifbond __user *)ifr->ifr_data;
4145
4146                if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4147                        return -EFAULT;
4148
4149                bond_info_query(bond_dev, &k_binfo);
4150                if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4151                        return -EFAULT;
4152
4153                return 0;
4154        case SIOCBONDSLAVEINFOQUERY:
4155                u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4156
4157                if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4158                        return -EFAULT;
4159
4160                res = bond_slave_info_query(bond_dev, &k_sinfo);
4161                if (res == 0 &&
4162                    copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4163                        return -EFAULT;
4164
4165                return res;
4166        default:
4167                break;
4168        }
4169
4170        net = dev_net(bond_dev);
4171
4172        if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4173                return -EPERM;
4174
4175        slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4176
4177        slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4178
4179        if (!slave_dev)
4180                return -ENODEV;
4181
4182        switch (cmd) {
4183        case SIOCBONDENSLAVE:
4184                res = bond_enslave(bond_dev, slave_dev, NULL);
4185                break;
4186        case SIOCBONDRELEASE:
4187                res = bond_release(bond_dev, slave_dev);
4188                break;
4189        case SIOCBONDSETHWADDR:
4190                res = bond_set_dev_addr(bond_dev, slave_dev);
4191                break;
4192        case SIOCBONDCHANGEACTIVE:
4193                bond_opt_initstr(&newval, slave_dev->name);
4194                res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4195                                            &newval);
4196                break;
4197        default:
4198                res = -EOPNOTSUPP;
4199        }
4200
4201        return res;
4202}
4203
4204static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4205                               void __user *data, int cmd)
4206{
4207        struct ifreq ifrdata = { .ifr_data = data };
4208
4209        switch (cmd) {
4210        case BOND_INFO_QUERY_OLD:
4211                return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4212        case BOND_SLAVE_INFO_QUERY_OLD:
4213                return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4214        case BOND_ENSLAVE_OLD:
4215                return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4216        case BOND_RELEASE_OLD:
4217                return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4218        case BOND_SETHWADDR_OLD:
4219                return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4220        case BOND_CHANGE_ACTIVE_OLD:
4221                return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4222        }
4223
4224        return -EOPNOTSUPP;
4225}
4226
4227static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4228{
4229        struct bonding *bond = netdev_priv(bond_dev);
4230
4231        if (change & IFF_PROMISC)
4232                bond_set_promiscuity(bond,
4233                                     bond_dev->flags & IFF_PROMISC ? 1 : -1);
4234
4235        if (change & IFF_ALLMULTI)
4236                bond_set_allmulti(bond,
4237                                  bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4238}
4239
4240static void bond_set_rx_mode(struct net_device *bond_dev)
4241{
4242        struct bonding *bond = netdev_priv(bond_dev);
4243        struct list_head *iter;
4244        struct slave *slave;
4245
4246        rcu_read_lock();
4247        if (bond_uses_primary(bond)) {
4248                slave = rcu_dereference(bond->curr_active_slave);
4249                if (slave) {
4250                        dev_uc_sync(slave->dev, bond_dev);
4251                        dev_mc_sync(slave->dev, bond_dev);
4252                }
4253        } else {
4254                bond_for_each_slave_rcu(bond, slave, iter) {
4255                        dev_uc_sync_multiple(slave->dev, bond_dev);
4256                        dev_mc_sync_multiple(slave->dev, bond_dev);
4257                }
4258        }
4259        rcu_read_unlock();
4260}
4261
4262static int bond_neigh_init(struct neighbour *n)
4263{
4264        struct bonding *bond = netdev_priv(n->dev);
4265        const struct net_device_ops *slave_ops;
4266        struct neigh_parms parms;
4267        struct slave *slave;
4268        int ret = 0;
4269
4270        rcu_read_lock();
4271        slave = bond_first_slave_rcu(bond);
4272        if (!slave)
4273                goto out;
4274        slave_ops = slave->dev->netdev_ops;
4275        if (!slave_ops->ndo_neigh_setup)
4276                goto out;
4277
4278        /* TODO: find another way [1] to implement this.
4279         * Passing a zeroed structure is fragile,
4280         * but at least we do not pass garbage.
4281         *
4282         * [1] One way would be that ndo_neigh_setup() never touch
4283         *     struct neigh_parms, but propagate the new neigh_setup()
4284         *     back to ___neigh_create() / neigh_parms_alloc()
4285         */
4286        memset(&parms, 0, sizeof(parms));
4287        ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4288
4289        if (ret)
4290                goto out;
4291
4292        if (parms.neigh_setup)
4293                ret = parms.neigh_setup(n);
4294out:
4295        rcu_read_unlock();
4296        return ret;
4297}
4298
4299/* The bonding ndo_neigh_setup is called at init time beofre any
4300 * slave exists. So we must declare proxy setup function which will
4301 * be used at run time to resolve the actual slave neigh param setup.
4302 *
4303 * It's also called by master devices (such as vlans) to setup their
4304 * underlying devices. In that case - do nothing, we're already set up from
4305 * our init.
4306 */
4307static int bond_neigh_setup(struct net_device *dev,
4308                            struct neigh_parms *parms)
4309{
4310        /* modify only our neigh_parms */
4311        if (parms->dev == dev)
4312                parms->neigh_setup = bond_neigh_init;
4313
4314        return 0;
4315}
4316
4317/* Change the MTU of all of a master's slaves to match the master */
4318static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4319{
4320        struct bonding *bond = netdev_priv(bond_dev);
4321        struct slave *slave, *rollback_slave;
4322        struct list_head *iter;
4323        int res = 0;
4324
4325        netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4326
4327        bond_for_each_slave(bond, slave, iter) {
4328                slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4329                           slave, slave->dev->netdev_ops->ndo_change_mtu);
4330
4331                res = dev_set_mtu(slave->dev, new_mtu);
4332
4333                if (res) {
4334                        /* If we failed to set the slave's mtu to the new value
4335                         * we must abort the operation even in ACTIVE_BACKUP
4336                         * mode, because if we allow the backup slaves to have
4337                         * different mtu values than the active slave we'll
4338                         * need to change their mtu when doing a failover. That
4339                         * means changing their mtu from timer context, which
4340                         * is probably not a good idea.
4341                         */
4342                        slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4343                                  res, new_mtu);
4344                        goto unwind;
4345                }
4346        }
4347
4348        bond_dev->mtu = new_mtu;
4349
4350        return 0;
4351
4352unwind:
4353        /* unwind from head to the slave that failed */
4354        bond_for_each_slave(bond, rollback_slave, iter) {
4355                int tmp_res;
4356
4357                if (rollback_slave == slave)
4358                        break;
4359
4360                tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4361                if (tmp_res)
4362                        slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4363                                  tmp_res);
4364        }
4365
4366        return res;
4367}
4368
4369/* Change HW address
4370 *
4371 * Note that many devices must be down to change the HW address, and
4372 * downing the master releases all slaves.  We can make bonds full of
4373 * bonding devices to test this, however.
4374 */
4375static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4376{
4377        struct bonding *bond = netdev_priv(bond_dev);
4378        struct slave *slave, *rollback_slave;
4379        struct sockaddr_storage *ss = addr, tmp_ss;
4380        struct list_head *iter;
4381        int res = 0;
4382
4383        if (BOND_MODE(bond) == BOND_MODE_ALB)
4384                return bond_alb_set_mac_address(bond_dev, addr);
4385
4386
4387        netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4388
4389        /* If fail_over_mac is enabled, do nothing and return success.
4390         * Returning an error causes ifenslave to fail.
4391         */
4392        if (bond->params.fail_over_mac &&
4393            BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4394                return 0;
4395
4396        if (!is_valid_ether_addr(ss->__data))
4397                return -EADDRNOTAVAIL;
4398
4399        bond_for_each_slave(bond, slave, iter) {
4400                slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4401                          __func__, slave);
4402                res = dev_set_mac_address(slave->dev, addr, NULL);
4403                if (res) {
4404                        /* TODO: consider downing the slave
4405                         * and retry ?
4406                         * User should expect communications
4407                         * breakage anyway until ARP finish
4408                         * updating, so...
4409                         */
4410                        slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4411                                  __func__, res);
4412                        goto unwind;
4413                }
4414        }
4415
4416        /* success */
4417        memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
4418        return 0;
4419
4420unwind:
4421        memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4422        tmp_ss.ss_family = bond_dev->type;
4423
4424        /* unwind from head to the slave that failed */
4425        bond_for_each_slave(bond, rollback_slave, iter) {
4426                int tmp_res;
4427
4428                if (rollback_slave == slave)
4429                        break;
4430
4431                tmp_res = dev_set_mac_address(rollback_slave->dev,
4432                                              (struct sockaddr *)&tmp_ss, NULL);
4433                if (tmp_res) {
4434                        slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4435                                   __func__, tmp_res);
4436                }
4437        }
4438
4439        return res;
4440}
4441
4442/**
4443 * bond_get_slave_by_id - get xmit slave with slave_id
4444 * @bond: bonding device that is transmitting
4445 * @slave_id: slave id up to slave_cnt-1 through which to transmit
4446 *
4447 * This function tries to get slave with slave_id but in case
4448 * it fails, it tries to find the first available slave for transmission.
4449 */
4450static struct slave *bond_get_slave_by_id(struct bonding *bond,
4451                                          int slave_id)
4452{
4453        struct list_head *iter;
4454        struct slave *slave;
4455        int i = slave_id;
4456
4457        /* Here we start from the slave with slave_id */
4458        bond_for_each_slave_rcu(bond, slave, iter) {
4459                if (--i < 0) {
4460                        if (bond_slave_can_tx(slave))
4461                                return slave;
4462                }
4463        }
4464
4465        /* Here we start from the first slave up to slave_id */
4466        i = slave_id;
4467        bond_for_each_slave_rcu(bond, slave, iter) {
4468                if (--i < 0)
4469                        break;
4470                if (bond_slave_can_tx(slave))
4471                        return slave;
4472        }
4473        /* no slave that can tx has been found */
4474        return NULL;
4475}
4476
4477/**
4478 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4479 * @bond: bonding device to use
4480 *
4481 * Based on the value of the bonding device's packets_per_slave parameter
4482 * this function generates a slave id, which is usually used as the next
4483 * slave to transmit through.
4484 */
4485static u32 bond_rr_gen_slave_id(struct bonding *bond)
4486{
4487        u32 slave_id;
4488        struct reciprocal_value reciprocal_packets_per_slave;
4489        int packets_per_slave = bond->params.packets_per_slave;
4490
4491        switch (packets_per_slave) {
4492        case 0:
4493                slave_id = prandom_u32();
4494                break;
4495        case 1:
4496                slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4497                break;
4498        default:
4499                reciprocal_packets_per_slave =
4500                        bond->params.reciprocal_packets_per_slave;
4501                slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4502                slave_id = reciprocal_divide(slave_id,
4503                                             reciprocal_packets_per_slave);
4504                break;
4505        }
4506
4507        return slave_id;
4508}
4509
4510static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4511                                                    struct sk_buff *skb)
4512{
4513        struct slave *slave;
4514        int slave_cnt;
4515        u32 slave_id;
4516
4517        /* Start with the curr_active_slave that joined the bond as the
4518         * default for sending IGMP traffic.  For failover purposes one
4519         * needs to maintain some consistency for the interface that will
4520         * send the join/membership reports.  The curr_active_slave found
4521         * will send all of this type of traffic.
4522         */
4523        if (skb->protocol == htons(ETH_P_IP)) {
4524                int noff = skb_network_offset(skb);
4525                struct iphdr *iph;
4526
4527                if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4528                        goto non_igmp;
4529
4530                iph = ip_hdr(skb);
4531                if (iph->protocol == IPPROTO_IGMP) {
4532                        slave = rcu_dereference(bond->curr_active_slave);
4533                        if (slave)
4534                                return slave;
4535                        return bond_get_slave_by_id(bond, 0);
4536                }
4537        }
4538
4539non_igmp:
4540        slave_cnt = READ_ONCE(bond->slave_cnt);
4541        if (likely(slave_cnt)) {
4542                slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4543                return bond_get_slave_by_id(bond, slave_id);
4544        }
4545        return NULL;
4546}
4547
4548static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4549                                                        struct xdp_buff *xdp)
4550{
4551        struct slave *slave;
4552        int slave_cnt;
4553        u32 slave_id;
4554        const struct ethhdr *eth;
4555        void *data = xdp->data;
4556
4557        if (data + sizeof(struct ethhdr) > xdp->data_end)
4558                goto non_igmp;
4559
4560        eth = (struct ethhdr *)data;
4561        data += sizeof(struct ethhdr);
4562
4563        /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4564        if (eth->h_proto == htons(ETH_P_IP)) {
4565                const struct iphdr *iph;
4566
4567                if (data + sizeof(struct iphdr) > xdp->data_end)
4568                        goto non_igmp;
4569
4570                iph = (struct iphdr *)data;
4571
4572                if (iph->protocol == IPPROTO_IGMP) {
4573                        slave = rcu_dereference(bond->curr_active_slave);
4574                        if (slave)
4575                                return slave;
4576                        return bond_get_slave_by_id(bond, 0);
4577                }
4578        }
4579
4580non_igmp:
4581        slave_cnt = READ_ONCE(bond->slave_cnt);
4582        if (likely(slave_cnt)) {
4583                slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4584                return bond_get_slave_by_id(bond, slave_id);
4585        }
4586        return NULL;
4587}
4588
4589static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4590                                        struct net_device *bond_dev)
4591{
4592        struct bonding *bond = netdev_priv(bond_dev);
4593        struct slave *slave;
4594
4595        slave = bond_xmit_roundrobin_slave_get(bond, skb);
4596        if (likely(slave))
4597                return bond_dev_queue_xmit(bond, skb, slave->dev);
4598
4599        return bond_tx_drop(bond_dev, skb);
4600}
4601
4602static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4603{
4604        return rcu_dereference(bond->curr_active_slave);
4605}
4606
4607/* In active-backup mode, we know that bond->curr_active_slave is always valid if
4608 * the bond has a usable interface.
4609 */
4610static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4611                                          struct net_device *bond_dev)
4612{
4613        struct bonding *bond = netdev_priv(bond_dev);
4614        struct slave *slave;
4615
4616        slave = bond_xmit_activebackup_slave_get(bond);
4617        if (slave)
4618                return bond_dev_queue_xmit(bond, skb, slave->dev);
4619
4620        return bond_tx_drop(bond_dev, skb);
4621}
4622
4623/* Use this to update slave_array when (a) it's not appropriate to update
4624 * slave_array right away (note that update_slave_array() may sleep)
4625 * and / or (b) RTNL is not held.
4626 */
4627void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4628{
4629        queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4630}
4631
4632/* Slave array work handler. Holds only RTNL */
4633static void bond_slave_arr_handler(struct work_struct *work)
4634{
4635        struct bonding *bond = container_of(work, struct bonding,
4636                                            slave_arr_work.work);
4637        int ret;
4638
4639        if (!rtnl_trylock())
4640                goto err;
4641
4642        ret = bond_update_slave_arr(bond, NULL);
4643        rtnl_unlock();
4644        if (ret) {
4645                pr_warn_ratelimited("Failed to update slave array from WT\n");
4646                goto err;
4647        }
4648        return;
4649
4650err:
4651        bond_slave_arr_work_rearm(bond, 1);
4652}
4653
4654static void bond_skip_slave(struct bond_up_slave *slaves,
4655                            struct slave *skipslave)
4656{
4657        int idx;
4658
4659        /* Rare situation where caller has asked to skip a specific
4660         * slave but allocation failed (most likely!). BTW this is
4661         * only possible when the call is initiated from
4662         * __bond_release_one(). In this situation; overwrite the
4663         * skipslave entry in the array with the last entry from the
4664         * array to avoid a situation where the xmit path may choose
4665         * this to-be-skipped slave to send a packet out.
4666         */
4667        for (idx = 0; slaves && idx < slaves->count; idx++) {
4668                if (skipslave == slaves->arr[idx]) {
4669                        slaves->arr[idx] =
4670                                slaves->arr[slaves->count - 1];
4671                        slaves->count--;
4672                        break;
4673                }
4674        }
4675}
4676
4677static void bond_set_slave_arr(struct bonding *bond,
4678                               struct bond_up_slave *usable_slaves,
4679                               struct bond_up_slave *all_slaves)
4680{
4681        struct bond_up_slave *usable, *all;
4682
4683        usable = rtnl_dereference(bond->usable_slaves);
4684        rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4685        kfree_rcu(usable, rcu);
4686
4687        all = rtnl_dereference(bond->all_slaves);
4688        rcu_assign_pointer(bond->all_slaves, all_slaves);
4689        kfree_rcu(all, rcu);
4690}
4691
4692static void bond_reset_slave_arr(struct bonding *bond)
4693{
4694        struct bond_up_slave *usable, *all;
4695
4696        usable = rtnl_dereference(bond->usable_slaves);
4697        if (usable) {
4698                RCU_INIT_POINTER(bond->usable_slaves, NULL);
4699                kfree_rcu(usable, rcu);
4700        }
4701
4702        all = rtnl_dereference(bond->all_slaves);
4703        if (all) {
4704                RCU_INIT_POINTER(bond->all_slaves, NULL);
4705                kfree_rcu(all, rcu);
4706        }
4707}
4708
4709/* Build the usable slaves array in control path for modes that use xmit-hash
4710 * to determine the slave interface -
4711 * (a) BOND_MODE_8023AD
4712 * (b) BOND_MODE_XOR
4713 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4714 *
4715 * The caller is expected to hold RTNL only and NO other lock!
4716 */
4717int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4718{
4719        struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
4720        struct slave *slave;
4721        struct list_head *iter;
4722        int agg_id = 0;
4723        int ret = 0;
4724
4725        might_sleep();
4726
4727        usable_slaves = kzalloc(struct_size(usable_slaves, arr,
4728                                            bond->slave_cnt), GFP_KERNEL);
4729        all_slaves = kzalloc(struct_size(all_slaves, arr,
4730                                         bond->slave_cnt), GFP_KERNEL);
4731        if (!usable_slaves || !all_slaves) {
4732                ret = -ENOMEM;
4733                goto out;
4734        }
4735        if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4736                struct ad_info ad_info;
4737
4738                spin_lock_bh(&bond->mode_lock);
4739                if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4740                        spin_unlock_bh(&bond->mode_lock);
4741                        pr_debug("bond_3ad_get_active_agg_info failed\n");
4742                        /* No active aggragator means it's not safe to use
4743                         * the previous array.
4744                         */
4745                        bond_reset_slave_arr(bond);
4746                        goto out;
4747                }
4748                spin_unlock_bh(&bond->mode_lock);
4749                agg_id = ad_info.aggregator_id;
4750        }
4751        bond_for_each_slave(bond, slave, iter) {
4752                if (skipslave == slave)
4753                        continue;
4754
4755                all_slaves->arr[all_slaves->count++] = slave;
4756                if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4757                        struct aggregator *agg;
4758
4759                        agg = SLAVE_AD_INFO(slave)->port.aggregator;
4760                        if (!agg || agg->aggregator_identifier != agg_id)
4761                                continue;
4762                }
4763                if (!bond_slave_can_tx(slave))
4764                        continue;
4765
4766                slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
4767                          usable_slaves->count);
4768
4769                usable_slaves->arr[usable_slaves->count++] = slave;
4770        }
4771
4772        bond_set_slave_arr(bond, usable_slaves, all_slaves);
4773        return ret;
4774out:
4775        if (ret != 0 && skipslave) {
4776                bond_skip_slave(rtnl_dereference(bond->all_slaves),
4777                                skipslave);
4778                bond_skip_slave(rtnl_dereference(bond->usable_slaves),
4779                                skipslave);
4780        }
4781        kfree_rcu(all_slaves, rcu);
4782        kfree_rcu(usable_slaves, rcu);
4783
4784        return ret;
4785}
4786
4787static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
4788                                                 struct sk_buff *skb,
4789                                                 struct bond_up_slave *slaves)
4790{
4791        struct slave *slave;
4792        unsigned int count;
4793        u32 hash;
4794
4795        hash = bond_xmit_hash(bond, skb);
4796        count = slaves ? READ_ONCE(slaves->count) : 0;
4797        if (unlikely(!count))
4798                return NULL;
4799
4800        slave = slaves->arr[hash % count];
4801        return slave;
4802}
4803
4804static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
4805                                                     struct xdp_buff *xdp)
4806{
4807        struct bond_up_slave *slaves;
4808        unsigned int count;
4809        u32 hash;
4810
4811        hash = bond_xmit_hash_xdp(bond, xdp);
4812        slaves = rcu_dereference(bond->usable_slaves);
4813        count = slaves ? READ_ONCE(slaves->count) : 0;
4814        if (unlikely(!count))
4815                return NULL;
4816
4817        return slaves->arr[hash % count];
4818}
4819
4820/* Use this Xmit function for 3AD as well as XOR modes. The current
4821 * usable slave array is formed in the control path. The xmit function
4822 * just calculates hash and sends the packet out.
4823 */
4824static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4825                                     struct net_device *dev)
4826{
4827        struct bonding *bond = netdev_priv(dev);
4828        struct bond_up_slave *slaves;
4829        struct slave *slave;
4830
4831        slaves = rcu_dereference(bond->usable_slaves);
4832        slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4833        if (likely(slave))
4834                return bond_dev_queue_xmit(bond, skb, slave->dev);
4835
4836        return bond_tx_drop(dev, skb);
4837}
4838
4839/* in broadcast mode, we send everything to all usable interfaces. */
4840static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4841                                       struct net_device *bond_dev)
4842{
4843        struct bonding *bond = netdev_priv(bond_dev);
4844        struct slave *slave = NULL;
4845        struct list_head *iter;
4846
4847        bond_for_each_slave_rcu(bond, slave, iter) {
4848                if (bond_is_last_slave(bond, slave))
4849                        break;
4850                if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4851                        struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4852
4853                        if (!skb2) {
4854                                net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4855                                                    bond_dev->name, __func__);
4856                                continue;
4857                        }
4858                        bond_dev_queue_xmit(bond, skb2, slave->dev);
4859                }
4860        }
4861        if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4862                return bond_dev_queue_xmit(bond, skb, slave->dev);
4863
4864        return bond_tx_drop(bond_dev, skb);
4865}
4866
4867/*------------------------- Device initialization ---------------------------*/
4868
4869/* Lookup the slave that corresponds to a qid */
4870static inline int bond_slave_override(struct bonding *bond,
4871                                      struct sk_buff *skb)
4872{
4873        struct slave *slave = NULL;
4874        struct list_head *iter;
4875
4876        if (!skb_rx_queue_recorded(skb))
4877                return 1;
4878
4879        /* Find out if any slaves have the same mapping as this skb. */
4880        bond_for_each_slave_rcu(bond, slave, iter) {
4881                if (slave->queue_id == skb_get_queue_mapping(skb)) {
4882                        if (bond_slave_is_up(slave) &&
4883                            slave->link == BOND_LINK_UP) {
4884                                bond_dev_queue_xmit(bond, skb, slave->dev);
4885                                return 0;
4886                        }
4887                        /* If the slave isn't UP, use default transmit policy. */
4888                        break;
4889                }
4890        }
4891
4892        return 1;
4893}
4894
4895
4896static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4897                             struct net_device *sb_dev)
4898{
4899        /* This helper function exists to help dev_pick_tx get the correct
4900         * destination queue.  Using a helper function skips a call to
4901         * skb_tx_hash and will put the skbs in the queue we expect on their
4902         * way down to the bonding driver.
4903         */
4904        u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4905
4906        /* Save the original txq to restore before passing to the driver */
4907        qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4908
4909        if (unlikely(txq >= dev->real_num_tx_queues)) {
4910                do {
4911                        txq -= dev->real_num_tx_queues;
4912                } while (txq >= dev->real_num_tx_queues);
4913        }
4914        return txq;
4915}
4916
4917static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
4918                                              struct sk_buff *skb,
4919                                              bool all_slaves)
4920{
4921        struct bonding *bond = netdev_priv(master_dev);
4922        struct bond_up_slave *slaves;
4923        struct slave *slave = NULL;
4924
4925        switch (BOND_MODE(bond)) {
4926        case BOND_MODE_ROUNDROBIN:
4927                slave = bond_xmit_roundrobin_slave_get(bond, skb);
4928                break;
4929        case BOND_MODE_ACTIVEBACKUP:
4930                slave = bond_xmit_activebackup_slave_get(bond);
4931                break;
4932        case BOND_MODE_8023AD:
4933        case BOND_MODE_XOR:
4934                if (all_slaves)
4935                        slaves = rcu_dereference(bond->all_slaves);
4936                else
4937                        slaves = rcu_dereference(bond->usable_slaves);
4938                slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
4939                break;
4940        case BOND_MODE_BROADCAST:
4941                break;
4942        case BOND_MODE_ALB:
4943                slave = bond_xmit_alb_slave_get(bond, skb);
4944                break;
4945        case BOND_MODE_TLB:
4946                slave = bond_xmit_tlb_slave_get(bond, skb);
4947                break;
4948        default:
4949                /* Should never happen, mode already checked */
4950                WARN_ONCE(true, "Unknown bonding mode");
4951                break;
4952        }
4953
4954        if (slave)
4955                return slave->dev;
4956        return NULL;
4957}
4958
4959static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
4960{
4961        switch (sk->sk_family) {
4962#if IS_ENABLED(CONFIG_IPV6)
4963        case AF_INET6:
4964                if (sk->sk_ipv6only ||
4965                    ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
4966                        flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
4967                        flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
4968                        flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
4969                        break;
4970                }
4971                fallthrough;
4972#endif
4973        default: /* AF_INET */
4974                flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
4975                flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
4976                flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
4977                break;
4978        }
4979
4980        flow->ports.src = inet_sk(sk)->inet_sport;
4981        flow->ports.dst = inet_sk(sk)->inet_dport;
4982}
4983
4984/**
4985 * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
4986 * @sk: socket to use for headers
4987 *
4988 * This function will extract the necessary field from the socket and use
4989 * them to generate a hash based on the LAYER34 xmit_policy.
4990 * Assumes that sk is a TCP or UDP socket.
4991 */
4992static u32 bond_sk_hash_l34(struct sock *sk)
4993{
4994        struct flow_keys flow;
4995        u32 hash;
4996
4997        bond_sk_to_flow(sk, &flow);
4998
4999        /* L4 */
5000        memcpy(&hash, &flow.ports.ports, sizeof(hash));
5001        /* L3 */
5002        return bond_ip_hash(hash, &flow);
5003}
5004
5005static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5006                                                  struct sock *sk)
5007{
5008        struct bond_up_slave *slaves;
5009        struct slave *slave;
5010        unsigned int count;
5011        u32 hash;
5012
5013        slaves = rcu_dereference(bond->usable_slaves);
5014        count = slaves ? READ_ONCE(slaves->count) : 0;
5015        if (unlikely(!count))
5016                return NULL;
5017
5018        hash = bond_sk_hash_l34(sk);
5019        slave = slaves->arr[hash % count];
5020
5021        return slave->dev;
5022}
5023
5024static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5025                                                struct sock *sk)
5026{
5027        struct bonding *bond = netdev_priv(dev);
5028        struct net_device *lower = NULL;
5029
5030        rcu_read_lock();
5031        if (bond_sk_check(bond))
5032                lower = __bond_sk_get_lower_dev(bond, sk);
5033        rcu_read_unlock();
5034
5035        return lower;
5036}
5037
5038#if IS_ENABLED(CONFIG_TLS_DEVICE)
5039static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5040                                        struct net_device *dev)
5041{
5042        if (likely(bond_get_slave_by_dev(bond, tls_get_ctx(skb->sk)->netdev)))
5043                return bond_dev_queue_xmit(bond, skb, tls_get_ctx(skb->sk)->netdev);
5044        return bond_tx_drop(dev, skb);
5045}
5046#endif
5047
5048static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5049{
5050        struct bonding *bond = netdev_priv(dev);
5051
5052        if (bond_should_override_tx_queue(bond) &&
5053            !bond_slave_override(bond, skb))
5054                return NETDEV_TX_OK;
5055
5056#if IS_ENABLED(CONFIG_TLS_DEVICE)
5057        if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
5058                return bond_tls_device_xmit(bond, skb, dev);
5059#endif
5060
5061        switch (BOND_MODE(bond)) {
5062        case BOND_MODE_ROUNDROBIN:
5063                return bond_xmit_roundrobin(skb, dev);
5064        case BOND_MODE_ACTIVEBACKUP:
5065                return bond_xmit_activebackup(skb, dev);
5066        case BOND_MODE_8023AD:
5067        case BOND_MODE_XOR:
5068                return bond_3ad_xor_xmit(skb, dev);
5069        case BOND_MODE_BROADCAST:
5070                return bond_xmit_broadcast(skb, dev);
5071        case BOND_MODE_ALB:
5072                return bond_alb_xmit(skb, dev);
5073        case BOND_MODE_TLB:
5074                return bond_tlb_xmit(skb, dev);
5075        default:
5076                /* Should never happen, mode already checked */
5077                netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5078                WARN_ON_ONCE(1);
5079                return bond_tx_drop(dev, skb);
5080        }
5081}
5082
5083static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5084{
5085        struct bonding *bond = netdev_priv(dev);
5086        netdev_tx_t ret = NETDEV_TX_OK;
5087
5088        /* If we risk deadlock from transmitting this in the
5089         * netpoll path, tell netpoll to queue the frame for later tx
5090         */
5091        if (unlikely(is_netpoll_tx_blocked(dev)))
5092                return NETDEV_TX_BUSY;
5093
5094        rcu_read_lock();
5095        if (bond_has_slaves(bond))
5096                ret = __bond_start_xmit(skb, dev);
5097        else
5098                ret = bond_tx_drop(dev, skb);
5099        rcu_read_unlock();
5100
5101        return ret;
5102}
5103
5104static struct net_device *
5105bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5106{
5107        struct bonding *bond = netdev_priv(bond_dev);
5108        struct slave *slave;
5109
5110        /* Caller needs to hold rcu_read_lock() */
5111
5112        switch (BOND_MODE(bond)) {
5113        case BOND_MODE_ROUNDROBIN:
5114                slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5115                break;
5116
5117        case BOND_MODE_ACTIVEBACKUP:
5118                slave = bond_xmit_activebackup_slave_get(bond);
5119                break;
5120
5121        case BOND_MODE_8023AD:
5122        case BOND_MODE_XOR:
5123                slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5124                break;
5125
5126        default:
5127                /* Should never happen. Mode guarded by bond_xdp_check() */
5128                netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5129                WARN_ON_ONCE(1);
5130                return NULL;
5131        }
5132
5133        if (slave)
5134                return slave->dev;
5135
5136        return NULL;
5137}
5138
5139static int bond_xdp_xmit(struct net_device *bond_dev,
5140                         int n, struct xdp_frame **frames, u32 flags)
5141{
5142        int nxmit, err = -ENXIO;
5143
5144        rcu_read_lock();
5145
5146        for (nxmit = 0; nxmit < n; nxmit++) {
5147                struct xdp_frame *frame = frames[nxmit];
5148                struct xdp_frame *frames1[] = {frame};
5149                struct net_device *slave_dev;
5150                struct xdp_buff xdp;
5151
5152                xdp_convert_frame_to_buff(frame, &xdp);
5153
5154                slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5155                if (!slave_dev) {
5156                        err = -ENXIO;
5157                        break;
5158                }
5159
5160                err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5161                if (err < 1)
5162                        break;
5163        }
5164
5165        rcu_read_unlock();
5166
5167        /* If error happened on the first frame then we can pass the error up, otherwise
5168         * report the number of frames that were xmitted.
5169         */
5170        if (err < 0)
5171                return (nxmit == 0 ? err : nxmit);
5172
5173        return nxmit;
5174}
5175
5176static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5177                        struct netlink_ext_ack *extack)
5178{
5179        struct bonding *bond = netdev_priv(dev);
5180        struct list_head *iter;
5181        struct slave *slave, *rollback_slave;
5182        struct bpf_prog *old_prog;
5183        struct netdev_bpf xdp = {
5184                .command = XDP_SETUP_PROG,
5185                .flags   = 0,
5186                .prog    = prog,
5187                .extack  = extack,
5188        };
5189        int err;
5190
5191        ASSERT_RTNL();
5192
5193        if (!bond_xdp_check(bond))
5194                return -EOPNOTSUPP;
5195
5196        old_prog = bond->xdp_prog;
5197        bond->xdp_prog = prog;
5198
5199        bond_for_each_slave(bond, slave, iter) {
5200                struct net_device *slave_dev = slave->dev;
5201
5202                if (!slave_dev->netdev_ops->ndo_bpf ||
5203                    !slave_dev->netdev_ops->ndo_xdp_xmit) {
5204                        SLAVE_NL_ERR(dev, slave_dev, extack,
5205                                     "Slave device does not support XDP");
5206                        err = -EOPNOTSUPP;
5207                        goto err;
5208                }
5209
5210                if (dev_xdp_prog_count(slave_dev) > 0) {
5211                        SLAVE_NL_ERR(dev, slave_dev, extack,
5212                                     "Slave has XDP program loaded, please unload before enslaving");
5213                        err = -EOPNOTSUPP;
5214                        goto err;
5215                }
5216
5217                err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5218                if (err < 0) {
5219                        /* ndo_bpf() sets extack error message */
5220                        slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5221                        goto err;
5222                }
5223