drivers/net/bonding/bond_main.c

   1 /*
   2  * originally based on the dummy device.
   3  *
   4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
   5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
   6  *
   7  * bonding.c: an Ethernet Bonding driver
   8  *
   9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
  10  *      Cisco 5500
  11  *      Sun Trunking (Solaris)
  12  *      Alteon AceDirector Trunks
  13  *      Linux Bonding
  14  *      and probably many L2 switches ...
  15  *
  16  * How it works:
  17  *    ifconfig bond0 ipaddress netmask up
  18  *      will setup a network device, with an ip address.  No mac address
  19  *      will be assigned at this time.  The hw mac address will come from
  20  *      the first slave bonded to the channel.  All slaves will then use
  21  *      this hw mac address.
  22  *
  23  *    ifconfig bond0 down
  24  *         will release all slaves, marking them as down.
  25  *
  26  *    ifenslave bond0 eth0
  27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
  28  *      a: be used as initial mac address
  29  *      b: if a hw mac address already is there, eth0's hw mac address
  30  *         will then be set from bond0.
  31  *
  32  */
  33
  34 #include <linux/kernel.h>
  35 #include <linux/module.h>
  36 #include <linux/types.h>
  37 #include <linux/fcntl.h>
  38 #include <linux/filter.h>
  39 #include <linux/interrupt.h>
  40 #include <linux/ptrace.h>
  41 #include <linux/ioport.h>
  42 #include <linux/in.h>
  43 #include <net/ip.h>
  44 #include <linux/ip.h>
  45 #include <linux/icmp.h>
  46 #include <linux/icmpv6.h>
  47 #include <linux/tcp.h>
  48 #include <linux/udp.h>
  49 #include <linux/slab.h>
  50 #include <linux/string.h>
  51 #include <linux/init.h>
  52 #include <linux/timer.h>
  53 #include <linux/socket.h>
  54 #include <linux/ctype.h>
  55 #include <linux/inet.h>
  56 #include <linux/bitops.h>
  57 #include <linux/io.h>
  58 #include <asm/dma.h>
  59 #include <linux/uaccess.h>
  60 #include <linux/errno.h>
  61 #include <linux/netdevice.h>
  62 #include <linux/inetdevice.h>
  63 #include <linux/igmp.h>
  64 #include <linux/etherdevice.h>
  65 #include <linux/skbuff.h>
  66 #include <net/sock.h>
  67 #include <linux/rtnetlink.h>
  68 #include <linux/smp.h>
  69 #include <linux/if_ether.h>
  70 #include <net/arp.h>
  71 #include <linux/mii.h>
  72 #include <linux/ethtool.h>
  73 #include <linux/if_vlan.h>
  74 #include <linux/if_bonding.h>
  75 #include <linux/phy.h>
  76 #include <linux/jiffies.h>
  77 #include <linux/preempt.h>
  78 #include <net/route.h>
  79 #include <net/net_namespace.h>
  80 #include <net/netns/generic.h>
  81 #include <net/pkt_sched.h>
  82 #include <linux/rculist.h>
  83 #include <net/flow_dissector.h>
  84 #include <net/xfrm.h>
  85 #include <net/bonding.h>
  86 #include <net/bond_3ad.h>
  87 #include <net/bond_alb.h>
  88 #if IS_ENABLED(CONFIG_TLS_DEVICE)
  89 #include <net/tls.h>
  90 #endif
  91 #include <net/ip6_route.h>
  92
  93 #include "bonding_priv.h"
  94
  95 /*---------------------------- Module parameters ----------------------------*/
  96
  97 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
  98
  99 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
 100 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
 101 static int num_peer_notif = 1;
 102 static int miimon;
 103 static int updelay;
 104 static int downdelay;
 105 static int use_carrier  = 1;
 106 static char *mode;
 107 static char *primary;
 108 static char *primary_reselect;
 109 static char *lacp_rate;
 110 static int min_links;
 111 static char *ad_select;
 112 static char *xmit_hash_policy;
 113 static int arp_interval;
 114 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
 115 static char *arp_validate;
 116 static char *arp_all_targets;
 117 static char *fail_over_mac;
 118 static int all_slaves_active;
 119 static struct bond_params bonding_defaults;
 120 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
 121 static int packets_per_slave = 1;
 122 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
 123
 124 module_param(max_bonds, int, 0);
 125 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
 126 module_param(tx_queues, int, 0);
 127 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
 128 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
 129 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
 130                                "failover event (alias of num_unsol_na)");
 131 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
 132 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
 133                                "failover event (alias of num_grat_arp)");
 134 module_param(miimon, int, 0);
 135 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
 136 module_param(updelay, int, 0);
 137 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
 138 module_param(downdelay, int, 0);
 139 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
 140                             "in milliseconds");
 141 module_param(use_carrier, int, 0);
 142 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
 143                               "0 for off, 1 for on (default)");
 144 module_param(mode, charp, 0);
 145 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
 146                        "1 for active-backup, 2 for balance-xor, "
 147                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
 148                        "6 for balance-alb");
 149 module_param(primary, charp, 0);
 150 MODULE_PARM_DESC(primary, "Primary network device to use");
 151 module_param(primary_reselect, charp, 0);
 152 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
 153                                    "once it comes up; "
 154                                    "0 for always (default), "
 155                                    "1 for only if speed of primary is "
 156                                    "better, "
 157                                    "2 for only on active slave "
 158                                    "failure");
 159 module_param(lacp_rate, charp, 0);
 160 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
 161                             "0 for slow, 1 for fast");
 162 module_param(ad_select, charp, 0);
 163 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
 164                             "0 for stable (default), 1 for bandwidth, "
 165                             "2 for count");
 166 module_param(min_links, int, 0);
 167 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
 168
 169 module_param(xmit_hash_policy, charp, 0);
 170 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
 171                                    "0 for layer 2 (default), 1 for layer 3+4, "
 172                                    "2 for layer 2+3, 3 for encap layer 2+3, "
 173                                    "4 for encap layer 3+4, 5 for vlan+srcmac");
 174 module_param(arp_interval, int, 0);
 175 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
 176 module_param_array(arp_ip_target, charp, NULL, 0);
 177 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
 178 module_param(arp_validate, charp, 0);
 179 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
 180                                "0 for none (default), 1 for active, "
 181                                "2 for backup, 3 for all");
 182 module_param(arp_all_targets, charp, 0);
 183 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
 184 module_param(fail_over_mac, charp, 0);
 185 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
 186                                 "the same MAC; 0 for none (default), "
 187                                 "1 for active, 2 for follow");
 188 module_param(all_slaves_active, int, 0);
 189 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
 190                                      "by setting active flag for all slaves; "
 191                                      "0 for never (default), 1 for always.");
 192 module_param(resend_igmp, int, 0);
 193 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
 194                               "link failure");
 195 module_param(packets_per_slave, int, 0);
 196 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
 197                                     "mode; 0 for a random slave, 1 packet per "
 198                                     "slave (default), >1 packets per slave.");
 199 module_param(lp_interval, uint, 0);
 200 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
 201                               "the bonding driver sends learning packets to "
 202                               "each slaves peer switch. The default is 1.");
 203
 204 /*----------------------------- Global variables ----------------------------*/
 205
 206 #ifdef CONFIG_NET_POLL_CONTROLLER
 207 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
 208 #endif
 209
 210 unsigned int bond_net_id __read_mostly;
 211
 212 static const struct flow_dissector_key flow_keys_bonding_keys[] = {
 213         {
 214                 .key_id = FLOW_DISSECTOR_KEY_CONTROL,
 215                 .offset = offsetof(struct flow_keys, control),
 216         },
 217         {
 218                 .key_id = FLOW_DISSECTOR_KEY_BASIC,
 219                 .offset = offsetof(struct flow_keys, basic),
 220         },
 221         {
 222                 .key_id = FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 223                 .offset = offsetof(struct flow_keys, addrs.v4addrs),
 224         },
 225         {
 226                 .key_id = FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 227                 .offset = offsetof(struct flow_keys, addrs.v6addrs),
 228         },
 229         {
 230                 .key_id = FLOW_DISSECTOR_KEY_TIPC,
 231                 .offset = offsetof(struct flow_keys, addrs.tipckey),
 232         },
 233         {
 234                 .key_id = FLOW_DISSECTOR_KEY_PORTS,
 235                 .offset = offsetof(struct flow_keys, ports),
 236         },
 237         {
 238                 .key_id = FLOW_DISSECTOR_KEY_ICMP,
 239                 .offset = offsetof(struct flow_keys, icmp),
 240         },
 241         {
 242                 .key_id = FLOW_DISSECTOR_KEY_VLAN,
 243                 .offset = offsetof(struct flow_keys, vlan),
 244         },
 245         {
 246                 .key_id = FLOW_DISSECTOR_KEY_FLOW_LABEL,
 247                 .offset = offsetof(struct flow_keys, tags),
 248         },
 249         {
 250                 .key_id = FLOW_DISSECTOR_KEY_GRE_KEYID,
 251                 .offset = offsetof(struct flow_keys, keyid),
 252         },
 253 };
 254
 255 static struct flow_dissector flow_keys_bonding __read_mostly;
 256
 257 /*-------------------------- Forward declarations ---------------------------*/
 258
 259 static int bond_init(struct net_device *bond_dev);
 260 static void bond_uninit(struct net_device *bond_dev);
 261 static void bond_get_stats(struct net_device *bond_dev,
 262                            struct rtnl_link_stats64 *stats);
 263 static void bond_slave_arr_handler(struct work_struct *work);
 264 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
 265                                   int mod);
 266 static void bond_netdev_notify_work(struct work_struct *work);
 267
 268 /*---------------------------- General routines -----------------------------*/
 269
 270 const char *bond_mode_name(int mode)
 271 {
 272         static const char *names[] = {
 273                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
 274                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
 275                 [BOND_MODE_XOR] = "load balancing (xor)",
 276                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
 277                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
 278                 [BOND_MODE_TLB] = "transmit load balancing",
 279                 [BOND_MODE_ALB] = "adaptive load balancing",
 280         };
 281
 282         if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
 283                 return "unknown";
 284
 285         return names[mode];
 286 }
 287
 288 /**
 289  * bond_dev_queue_xmit - Prepare skb for xmit.
 290  *
 291  * @bond: bond device that got this skb for tx.
 292  * @skb: hw accel VLAN tagged skb to transmit
 293  * @slave_dev: slave that is supposed to xmit this skbuff
 294  */
 295 netdev_tx_t bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
 296                         struct net_device *slave_dev)
 297 {
 298         skb->dev = slave_dev;
 299
 300         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
 301                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
 302         skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 303
 304         if (unlikely(netpoll_tx_running(bond->dev)))
 305                 return bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
 306
 307         return dev_queue_xmit(skb);
 308 }
 309
 310 static bool bond_sk_check(struct bonding *bond)
 311 {
 312         switch (BOND_MODE(bond)) {
 313         case BOND_MODE_8023AD:
 314         case BOND_MODE_XOR:
 315                 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
 316                         return true;
 317                 fallthrough;
 318         default:
 319                 return false;
 320         }
 321 }
 322
 323 static bool bond_xdp_check(struct bonding *bond)
 324 {
 325         switch (BOND_MODE(bond)) {
 326         case BOND_MODE_ROUNDROBIN:
 327         case BOND_MODE_ACTIVEBACKUP:
 328                 return true;
 329         case BOND_MODE_8023AD:
 330         case BOND_MODE_XOR:
 331                 /* vlan+srcmac is not supported with XDP as in most cases the 802.1q
 332                  * payload is not in the packet due to hardware offload.
 333                  */
 334                 if (bond->params.xmit_policy != BOND_XMIT_POLICY_VLAN_SRCMAC)
 335                         return true;
 336                 fallthrough;
 337         default:
 338                 return false;
 339         }
 340 }
 341
 342 /*---------------------------------- VLAN -----------------------------------*/
 343
 344 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
 345  * We don't protect the slave list iteration with a lock because:
 346  * a. This operation is performed in IOCTL context,
 347  * b. The operation is protected by the RTNL semaphore in the 8021q code,
 348  * c. Holding a lock with BH disabled while directly calling a base driver
 349  *    entry point is generally a BAD idea.
 350  *
 351  * The design of synchronization/protection for this operation in the 8021q
 352  * module is good for one or more VLAN devices over a single physical device
 353  * and cannot be extended for a teaming solution like bonding, so there is a
 354  * potential race condition here where a net device from the vlan group might
 355  * be referenced (either by a base driver or the 8021q code) while it is being
 356  * removed from the system. However, it turns out we're not making matters
 357  * worse, and if it works for regular VLAN usage it will work here too.
 358 */
 359
 360 /**
 361  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
 362  * @bond_dev: bonding net device that got called
 363  * @proto: network protocol ID
 364  * @vid: vlan id being added
 365  */
 366 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
 367                                 __be16 proto, u16 vid)
 368 {
 369         struct bonding *bond = netdev_priv(bond_dev);
 370         struct slave *slave, *rollback_slave;
 371         struct list_head *iter;
 372         int res;
 373
 374         bond_for_each_slave(bond, slave, iter) {
 375                 res = vlan_vid_add(slave->dev, proto, vid);
 376                 if (res)
 377                         goto unwind;
 378         }
 379
 380         return 0;
 381
 382 unwind:
 383         /* unwind to the slave that failed */
 384         bond_for_each_slave(bond, rollback_slave, iter) {
 385                 if (rollback_slave == slave)
 386                         break;
 387
 388                 vlan_vid_del(rollback_slave->dev, proto, vid);
 389         }
 390
 391         return res;
 392 }
 393
 394 /**
 395  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
 396  * @bond_dev: bonding net device that got called
 397  * @proto: network protocol ID
 398  * @vid: vlan id being removed
 399  */
 400 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
 401                                  __be16 proto, u16 vid)
 402 {
 403         struct bonding *bond = netdev_priv(bond_dev);
 404         struct list_head *iter;
 405         struct slave *slave;
 406
 407         bond_for_each_slave(bond, slave, iter)
 408                 vlan_vid_del(slave->dev, proto, vid);
 409
 410         if (bond_is_lb(bond))
 411                 bond_alb_clear_vlan(bond, vid);
 412
 413         return 0;
 414 }
 415
 416 /*---------------------------------- XFRM -----------------------------------*/
 417
 418 #ifdef CONFIG_XFRM_OFFLOAD
 419 /**
 420  * bond_ipsec_add_sa - program device with a security association
 421  * @xs: pointer to transformer state struct
 422  * @extack: extack point to fill failure reason
 423  **/
 424 static int bond_ipsec_add_sa(struct xfrm_state *xs,
 425                              struct netlink_ext_ack *extack)
 426 {
 427         struct net_device *bond_dev = xs->xso.dev;
 428         struct bond_ipsec *ipsec;
 429         struct bonding *bond;
 430         struct slave *slave;
 431         int err;
 432
 433         if (!bond_dev)
 434                 return -EINVAL;
 435
 436         rcu_read_lock();
 437         bond = netdev_priv(bond_dev);
 438         slave = rcu_dereference(bond->curr_active_slave);
 439         if (!slave) {
 440                 rcu_read_unlock();
 441                 return -ENODEV;
 442         }
 443
 444         if (!slave->dev->xfrmdev_ops ||
 445             !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
 446             netif_is_bond_master(slave->dev)) {
 447                 NL_SET_ERR_MSG_MOD(extack, "Slave does not support ipsec offload");
 448                 rcu_read_unlock();
 449                 return -EINVAL;
 450         }
 451
 452         ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
 453         if (!ipsec) {
 454                 rcu_read_unlock();
 455                 return -ENOMEM;
 456         }
 457         xs->xso.real_dev = slave->dev;
 458
 459         err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs, extack);
 460         if (!err) {
 461                 ipsec->xs = xs;
 462                 INIT_LIST_HEAD(&ipsec->list);
 463                 spin_lock_bh(&bond->ipsec_lock);
 464                 list_add(&ipsec->list, &bond->ipsec_list);
 465                 spin_unlock_bh(&bond->ipsec_lock);
 466         } else {
 467                 kfree(ipsec);
 468         }
 469         rcu_read_unlock();
 470         return err;
 471 }
 472
 473 static void bond_ipsec_add_sa_all(struct bonding *bond)
 474 {
 475         struct net_device *bond_dev = bond->dev;
 476         struct bond_ipsec *ipsec;
 477         struct slave *slave;
 478
 479         rcu_read_lock();
 480         slave = rcu_dereference(bond->curr_active_slave);
 481         if (!slave)
 482                 goto out;
 483
 484         if (!slave->dev->xfrmdev_ops ||
 485             !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
 486             netif_is_bond_master(slave->dev)) {
 487                 spin_lock_bh(&bond->ipsec_lock);
 488                 if (!list_empty(&bond->ipsec_list))
 489                         slave_warn(bond_dev, slave->dev,
 490                                    "%s: no slave xdo_dev_state_add\n",
 491                                    __func__);
 492                 spin_unlock_bh(&bond->ipsec_lock);
 493                 goto out;
 494         }
 495
 496         spin_lock_bh(&bond->ipsec_lock);
 497         list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 498                 ipsec->xs->xso.real_dev = slave->dev;
 499                 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs, NULL)) {
 500                         slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
 501                         ipsec->xs->xso.real_dev = NULL;
 502                 }
 503         }
 504         spin_unlock_bh(&bond->ipsec_lock);
 505 out:
 506         rcu_read_unlock();
 507 }
 508
 509 /**
 510  * bond_ipsec_del_sa - clear out this specific SA
 511  * @xs: pointer to transformer state struct
 512  **/
 513 static void bond_ipsec_del_sa(struct xfrm_state *xs)
 514 {
 515         struct net_device *bond_dev = xs->xso.dev;
 516         struct bond_ipsec *ipsec;
 517         struct bonding *bond;
 518         struct slave *slave;
 519
 520         if (!bond_dev)
 521                 return;
 522
 523         rcu_read_lock();
 524         bond = netdev_priv(bond_dev);
 525         slave = rcu_dereference(bond->curr_active_slave);
 526
 527         if (!slave)
 528                 goto out;
 529
 530         if (!xs->xso.real_dev)
 531                 goto out;
 532
 533         WARN_ON(xs->xso.real_dev != slave->dev);
 534
 535         if (!slave->dev->xfrmdev_ops ||
 536             !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
 537             netif_is_bond_master(slave->dev)) {
 538                 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
 539                 goto out;
 540         }
 541
 542         slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
 543 out:
 544         spin_lock_bh(&bond->ipsec_lock);
 545         list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 546                 if (ipsec->xs == xs) {
 547                         list_del(&ipsec->list);
 548                         kfree(ipsec);
 549                         break;
 550                 }
 551         }
 552         spin_unlock_bh(&bond->ipsec_lock);
 553         rcu_read_unlock();
 554 }
 555
 556 static void bond_ipsec_del_sa_all(struct bonding *bond)
 557 {
 558         struct net_device *bond_dev = bond->dev;
 559         struct bond_ipsec *ipsec;
 560         struct slave *slave;
 561
 562         rcu_read_lock();
 563         slave = rcu_dereference(bond->curr_active_slave);
 564         if (!slave) {
 565                 rcu_read_unlock();
 566                 return;
 567         }
 568
 569         spin_lock_bh(&bond->ipsec_lock);
 570         list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 571                 if (!ipsec->xs->xso.real_dev)
 572                         continue;
 573
 574                 if (!slave->dev->xfrmdev_ops ||
 575                     !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
 576                     netif_is_bond_master(slave->dev)) {
 577                         slave_warn(bond_dev, slave->dev,
 578                                    "%s: no slave xdo_dev_state_delete\n",
 579                                    __func__);
 580                 } else {
 581                         slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
 582                 }
 583                 ipsec->xs->xso.real_dev = NULL;
 584         }
 585         spin_unlock_bh(&bond->ipsec_lock);
 586         rcu_read_unlock();
 587 }
 588
 589 /**
 590  * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
 591  * @skb: current data packet
 592  * @xs: pointer to transformer state struct
 593  **/
 594 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
 595 {
 596         struct net_device *bond_dev = xs->xso.dev;
 597         struct net_device *real_dev;
 598         struct slave *curr_active;
 599         struct bonding *bond;
 600         int err;
 601
 602         bond = netdev_priv(bond_dev);
 603         rcu_read_lock();
 604         curr_active = rcu_dereference(bond->curr_active_slave);
 605         real_dev = curr_active->dev;
 606
 607         if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
 608                 err = false;
 609                 goto out;
 610         }
 611
 612         if (!xs->xso.real_dev) {
 613                 err = false;
 614                 goto out;
 615         }
 616
 617         if (!real_dev->xfrmdev_ops ||
 618             !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
 619             netif_is_bond_master(real_dev)) {
 620                 err = false;
 621                 goto out;
 622         }
 623
 624         err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
 625 out:
 626         rcu_read_unlock();
 627         return err;
 628 }
 629
 630 static const struct xfrmdev_ops bond_xfrmdev_ops = {
 631         .xdo_dev_state_add = bond_ipsec_add_sa,
 632         .xdo_dev_state_delete = bond_ipsec_del_sa,
 633         .xdo_dev_offload_ok = bond_ipsec_offload_ok,
 634 };
 635 #endif /* CONFIG_XFRM_OFFLOAD */
 636
 637 /*------------------------------- Link status -------------------------------*/
 638
 639 /* Set the carrier state for the master according to the state of its
 640  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 641  * do special 802.3ad magic.
 642  *
 643  * Returns zero if carrier state does not change, nonzero if it does.
 644  */
 645 int bond_set_carrier(struct bonding *bond)
 646 {
 647         struct list_head *iter;
 648         struct slave *slave;
 649
 650         if (!bond_has_slaves(bond))
 651                 goto down;
 652
 653         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 654                 return bond_3ad_set_carrier(bond);
 655
 656         bond_for_each_slave(bond, slave, iter) {
 657                 if (slave->link == BOND_LINK_UP) {
 658                         if (!netif_carrier_ok(bond->dev)) {
 659                                 netif_carrier_on(bond->dev);
 660                                 return 1;
 661                         }
 662                         return 0;
 663                 }
 664         }
 665
 666 down:
 667         if (netif_carrier_ok(bond->dev)) {
 668                 netif_carrier_off(bond->dev);
 669                 return 1;
 670         }
 671         return 0;
 672 }
 673
 674 /* Get link speed and duplex from the slave's base driver
 675  * using ethtool. If for some reason the call fails or the
 676  * values are invalid, set speed and duplex to -1,
 677  * and return. Return 1 if speed or duplex settings are
 678  * UNKNOWN; 0 otherwise.
 679  */
 680 static int bond_update_speed_duplex(struct slave *slave)
 681 {
 682         struct net_device *slave_dev = slave->dev;
 683         struct ethtool_link_ksettings ecmd;
 684         int res;
 685
 686         slave->speed = SPEED_UNKNOWN;
 687         slave->duplex = DUPLEX_UNKNOWN;
 688
 689         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 690         if (res < 0)
 691                 return 1;
 692         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 693                 return 1;
 694         switch (ecmd.base.duplex) {
 695         case DUPLEX_FULL:
 696         case DUPLEX_HALF:
 697                 break;
 698         default:
 699                 return 1;
 700         }
 701
 702         slave->speed = ecmd.base.speed;
 703         slave->duplex = ecmd.base.duplex;
 704
 705         return 0;
 706 }
 707
 708 const char *bond_slave_link_status(s8 link)
 709 {
 710         switch (link) {
 711         case BOND_LINK_UP:
 712                 return "up";
 713         case BOND_LINK_FAIL:
 714                 return "going down";
 715         case BOND_LINK_DOWN:
 716                 return "down";
 717         case BOND_LINK_BACK:
 718                 return "going back";
 719         default:
 720                 return "unknown";
 721         }
 722 }
 723
 724 /* if <dev> supports MII link status reporting, check its link status.
 725  *
 726  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 727  * depending upon the setting of the use_carrier parameter.
 728  *
 729  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 730  * can't tell and just pretend it is), or 0, meaning that the link is
 731  * down.
 732  *
 733  * If reporting is non-zero, instead of faking link up, return -1 if
 734  * both ETHTOOL and MII ioctls fail (meaning the device does not
 735  * support them).  If use_carrier is set, return whatever it says.
 736  * It'd be nice if there was a good way to tell if a driver supports
 737  * netif_carrier, but there really isn't.
 738  */
 739 static int bond_check_dev_link(struct bonding *bond,
 740                                struct net_device *slave_dev, int reporting)
 741 {
 742         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 743         int (*ioctl)(struct net_device *, struct ifreq *, int);
 744         struct ifreq ifr;
 745         struct mii_ioctl_data *mii;
 746
 747         if (!reporting && !netif_running(slave_dev))
 748                 return 0;
 749
 750         if (bond->params.use_carrier)
 751                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 752
 753         /* Try to get link status using Ethtool first. */
 754         if (slave_dev->ethtool_ops->get_link)
 755                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 756                         BMSR_LSTATUS : 0;
 757
 758         /* Ethtool can't be used, fallback to MII ioctls. */
 759         ioctl = slave_ops->ndo_eth_ioctl;
 760         if (ioctl) {
 761                 /* TODO: set pointer to correct ioctl on a per team member
 762                  *       bases to make this more efficient. that is, once
 763                  *       we determine the correct ioctl, we will always
 764                  *       call it and not the others for that team
 765                  *       member.
 766                  */
 767
 768                 /* We cannot assume that SIOCGMIIPHY will also read a
 769                  * register; not all network drivers (e.g., e100)
 770                  * support that.
 771                  */
 772
 773                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 774                 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 775                 mii = if_mii(&ifr);
 776                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 777                         mii->reg_num = MII_BMSR;
 778                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 779                                 return mii->val_out & BMSR_LSTATUS;
 780                 }
 781         }
 782
 783         /* If reporting, report that either there's no ndo_eth_ioctl,
 784          * or both SIOCGMIIREG and get_link failed (meaning that we
 785          * cannot report link status).  If not reporting, pretend
 786          * we're ok.
 787          */
 788         return reporting ? -1 : BMSR_LSTATUS;
 789 }
 790
 791 /*----------------------------- Multicast list ------------------------------*/
 792
 793 /* Push the promiscuity flag down to appropriate slaves */
 794 static int bond_set_promiscuity(struct bonding *bond, int inc)
 795 {
 796         struct list_head *iter;
 797         int err = 0;
 798
 799         if (bond_uses_primary(bond)) {
 800                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 801
 802                 if (curr_active)
 803                         err = dev_set_promiscuity(curr_active->dev, inc);
 804         } else {
 805                 struct slave *slave;
 806
 807                 bond_for_each_slave(bond, slave, iter) {
 808                         err = dev_set_promiscuity(slave->dev, inc);
 809                         if (err)
 810                                 return err;
 811                 }
 812         }
 813         return err;
 814 }
 815
 816 /* Push the allmulti flag down to all slaves */
 817 static int bond_set_allmulti(struct bonding *bond, int inc)
 818 {
 819         struct list_head *iter;
 820         int err = 0;
 821
 822         if (bond_uses_primary(bond)) {
 823                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 824
 825                 if (curr_active)
 826                         err = dev_set_allmulti(curr_active->dev, inc);
 827         } else {
 828                 struct slave *slave;
 829
 830                 bond_for_each_slave(bond, slave, iter) {
 831                         err = dev_set_allmulti(slave->dev, inc);
 832                         if (err)
 833                                 return err;
 834                 }
 835         }
 836         return err;
 837 }
 838
 839 /* Retrieve the list of registered multicast addresses for the bonding
 840  * device and retransmit an IGMP JOIN request to the current active
 841  * slave.
 842  */
 843 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 844 {
 845         struct bonding *bond = container_of(work, struct bonding,
 846                                             mcast_work.work);
 847
 848         if (!rtnl_trylock()) {
 849                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 850                 return;
 851         }
 852         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 853
 854         if (bond->igmp_retrans > 1) {
 855                 bond->igmp_retrans--;
 856                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 857         }
 858         rtnl_unlock();
 859 }
 860
 861 /* Flush bond's hardware addresses from slave */
 862 static void bond_hw_addr_flush(struct net_device *bond_dev,
 863                                struct net_device *slave_dev)
 864 {
 865         struct bonding *bond = netdev_priv(bond_dev);
 866
 867         dev_uc_unsync(slave_dev, bond_dev);
 868         dev_mc_unsync(slave_dev, bond_dev);
 869
 870         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 871                 dev_mc_del(slave_dev, lacpdu_mcast_addr);
 872 }
 873
 874 /*--------------------------- Active slave change ---------------------------*/
 875
 876 /* Update the hardware address list and promisc/allmulti for the new and
 877  * old active slaves (if any).  Modes that are not using primary keep all
 878  * slaves up date at all times; only the modes that use primary need to call
 879  * this function to swap these settings during a failover.
 880  */
 881 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 882                               struct slave *old_active)
 883 {
 884         if (old_active) {
 885                 if (bond->dev->flags & IFF_PROMISC)
 886                         dev_set_promiscuity(old_active->dev, -1);
 887
 888                 if (bond->dev->flags & IFF_ALLMULTI)
 889                         dev_set_allmulti(old_active->dev, -1);
 890
 891                 if (bond->dev->flags & IFF_UP)
 892                         bond_hw_addr_flush(bond->dev, old_active->dev);
 893         }
 894
 895         if (new_active) {
 896                 /* FIXME: Signal errors upstream. */
 897                 if (bond->dev->flags & IFF_PROMISC)
 898                         dev_set_promiscuity(new_active->dev, 1);
 899
 900                 if (bond->dev->flags & IFF_ALLMULTI)
 901                         dev_set_allmulti(new_active->dev, 1);
 902
 903                 if (bond->dev->flags & IFF_UP) {
 904                         netif_addr_lock_bh(bond->dev);
 905                         dev_uc_sync(new_active->dev, bond->dev);
 906                         dev_mc_sync(new_active->dev, bond->dev);
 907                         netif_addr_unlock_bh(bond->dev);
 908                 }
 909         }
 910 }
 911
 912 /**
 913  * bond_set_dev_addr - clone slave's address to bond
 914  * @bond_dev: bond net device
 915  * @slave_dev: slave net device
 916  *
 917  * Should be called with RTNL held.
 918  */
 919 static int bond_set_dev_addr(struct net_device *bond_dev,
 920                              struct net_device *slave_dev)
 921 {
 922         int err;
 923
 924         slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 925                   bond_dev, slave_dev, slave_dev->addr_len);
 926         err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
 927         if (err)
 928                 return err;
 929
 930         __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len);
 931         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 932         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 933         return 0;
 934 }
 935
 936 static struct slave *bond_get_old_active(struct bonding *bond,
 937                                          struct slave *new_active)
 938 {
 939         struct slave *slave;
 940         struct list_head *iter;
 941
 942         bond_for_each_slave(bond, slave, iter) {
 943                 if (slave == new_active)
 944                         continue;
 945
 946                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 947                         return slave;
 948         }
 949
 950         return NULL;
 951 }
 952
 953 /* bond_do_fail_over_mac
 954  *
 955  * Perform special MAC address swapping for fail_over_mac settings
 956  *
 957  * Called with RTNL
 958  */
 959 static void bond_do_fail_over_mac(struct bonding *bond,
 960                                   struct slave *new_active,
 961                                   struct slave *old_active)
 962 {
 963         u8 tmp_mac[MAX_ADDR_LEN];
 964         struct sockaddr_storage ss;
 965         int rv;
 966
 967         switch (bond->params.fail_over_mac) {
 968         case BOND_FOM_ACTIVE:
 969                 if (new_active) {
 970                         rv = bond_set_dev_addr(bond->dev, new_active->dev);
 971                         if (rv)
 972                                 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
 973                                           -rv);
 974                 }
 975                 break;
 976         case BOND_FOM_FOLLOW:
 977                 /* if new_active && old_active, swap them
 978                  * if just old_active, do nothing (going to no active slave)
 979                  * if just new_active, set new_active to bond's MAC
 980                  */
 981                 if (!new_active)
 982                         return;
 983
 984                 if (!old_active)
 985                         old_active = bond_get_old_active(bond, new_active);
 986
 987                 if (old_active) {
 988                         bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 989                                           new_active->dev->addr_len);
 990                         bond_hw_addr_copy(ss.__data,
 991                                           old_active->dev->dev_addr,
 992                                           old_active->dev->addr_len);
 993                         ss.ss_family = new_active->dev->type;
 994                 } else {
 995                         bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 996                                           bond->dev->addr_len);
 997                         ss.ss_family = bond->dev->type;
 998                 }
 999
1000                 rv = dev_set_mac_address(new_active->dev,
1001                                          (struct sockaddr *)&ss, NULL);
1002                 if (rv) {
1003                         slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1004                                   -rv);
1005                         goto out;
1006                 }
1007
1008                 if (!old_active)
1009                         goto out;
1010
1011                 bond_hw_addr_copy(ss.__data, tmp_mac,
1012                                   new_active->dev->addr_len);
1013                 ss.ss_family = old_active->dev->type;
1014
1015                 rv = dev_set_mac_address(old_active->dev,
1016                                          (struct sockaddr *)&ss, NULL);
1017                 if (rv)
1018                         slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1019                                   -rv);
1020 out:
1021                 break;
1022         default:
1023                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1024                            bond->params.fail_over_mac);
1025                 break;
1026         }
1027
1028 }
1029
1030 /**
1031  * bond_choose_primary_or_current - select the primary or high priority slave
1032  * @bond: our bonding struct
1033  *
1034  * - Check if there is a primary link. If the primary link was set and is up,
1035  *   go on and do link reselection.
1036  *
1037  * - If primary link is not set or down, find the highest priority link.
1038  *   If the highest priority link is not current slave, set it as primary
1039  *   link and do link reselection.
1040  */
1041 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1042 {
1043         struct slave *prim = rtnl_dereference(bond->primary_slave);
1044         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1045         struct slave *slave, *hprio = NULL;
1046         struct list_head *iter;
1047
1048         if (!prim || prim->link != BOND_LINK_UP) {
1049                 bond_for_each_slave(bond, slave, iter) {
1050                         if (slave->link == BOND_LINK_UP) {
1051                                 hprio = hprio ?: slave;
1052                                 if (slave->prio > hprio->prio)
1053                                         hprio = slave;
1054                         }
1055                 }
1056
1057                 if (hprio && hprio != curr) {
1058                         prim = hprio;
1059                         goto link_reselect;
1060                 }
1061
1062                 if (!curr || curr->link != BOND_LINK_UP)
1063                         return NULL;
1064                 return curr;
1065         }
1066
1067         if (bond->force_primary) {
1068                 bond->force_primary = false;
1069                 return prim;
1070         }
1071
1072 link_reselect:
1073         if (!curr || curr->link != BOND_LINK_UP)
1074                 return prim;
1075
1076         /* At this point, prim and curr are both up */
1077         switch (bond->params.primary_reselect) {
1078         case BOND_PRI_RESELECT_ALWAYS:
1079                 return prim;
1080         case BOND_PRI_RESELECT_BETTER:
1081                 if (prim->speed < curr->speed)
1082                         return curr;
1083                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1084                         return curr;
1085                 return prim;
1086         case BOND_PRI_RESELECT_FAILURE:
1087                 return curr;
1088         default:
1089                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1090                            bond->params.primary_reselect);
1091                 return curr;
1092         }
1093 }
1094
1095 /**
1096  * bond_find_best_slave - select the best available slave to be the active one
1097  * @bond: our bonding struct
1098  */
1099 static struct slave *bond_find_best_slave(struct bonding *bond)
1100 {
1101         struct slave *slave, *bestslave = NULL;
1102         struct list_head *iter;
1103         int mintime = bond->params.updelay;
1104
1105         slave = bond_choose_primary_or_current(bond);
1106         if (slave)
1107                 return slave;
1108
1109         bond_for_each_slave(bond, slave, iter) {
1110                 if (slave->link == BOND_LINK_UP)
1111                         return slave;
1112                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1113                     slave->delay < mintime) {
1114                         mintime = slave->delay;
1115                         bestslave = slave;
1116                 }
1117         }
1118
1119         return bestslave;
1120 }
1121
1122 static bool bond_should_notify_peers(struct bonding *bond)
1123 {
1124         struct slave *slave;
1125
1126         rcu_read_lock();
1127         slave = rcu_dereference(bond->curr_active_slave);
1128         rcu_read_unlock();
1129
1130         if (!slave || !bond->send_peer_notif ||
1131             bond->send_peer_notif %
1132             max(1, bond->params.peer_notif_delay) != 0 ||
1133             !netif_carrier_ok(bond->dev) ||
1134             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1135                 return false;
1136
1137         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1138                    slave ? slave->dev->name : "NULL");
1139
1140         return true;
1141 }
1142
1143 /**
1144  * bond_change_active_slave - change the active slave into the specified one
1145  * @bond: our bonding struct
1146  * @new_active: the new slave to make the active one
1147  *
1148  * Set the new slave to the bond's settings and unset them on the old
1149  * curr_active_slave.
1150  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1151  *
1152  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1153  * because it is apparently the best available slave we have, even though its
1154  * updelay hasn't timed out yet.
1155  *
1156  * Caller must hold RTNL.
1157  */
1158 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1159 {
1160         struct slave *old_active;
1161
1162         ASSERT_RTNL();
1163
1164         old_active = rtnl_dereference(bond->curr_active_slave);
1165
1166         if (old_active == new_active)
1167                 return;
1168
1169 #ifdef CONFIG_XFRM_OFFLOAD
1170         bond_ipsec_del_sa_all(bond);
1171 #endif /* CONFIG_XFRM_OFFLOAD */
1172
1173         if (new_active) {
1174                 new_active->last_link_up = jiffies;
1175
1176                 if (new_active->link == BOND_LINK_BACK) {
1177                         if (bond_uses_primary(bond)) {
1178                                 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1179                                            (bond->params.updelay - new_active->delay) * bond->params.miimon);
1180                         }
1181
1182                         new_active->delay = 0;
1183                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
1184                                                   BOND_SLAVE_NOTIFY_NOW);
1185
1186                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1187                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1188
1189                         if (bond_is_lb(bond))
1190                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1191                 } else {
1192                         if (bond_uses_primary(bond))
1193                                 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1194                 }
1195         }
1196
1197         if (bond_uses_primary(bond))
1198                 bond_hw_addr_swap(bond, new_active, old_active);
1199
1200         if (bond_is_lb(bond)) {
1201                 bond_alb_handle_active_change(bond, new_active);
1202                 if (old_active)
1203                         bond_set_slave_inactive_flags(old_active,
1204                                                       BOND_SLAVE_NOTIFY_NOW);
1205                 if (new_active)
1206                         bond_set_slave_active_flags(new_active,
1207                                                     BOND_SLAVE_NOTIFY_NOW);
1208         } else {
1209                 rcu_assign_pointer(bond->curr_active_slave, new_active);
1210         }
1211
1212         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1213                 if (old_active)
1214                         bond_set_slave_inactive_flags(old_active,
1215                                                       BOND_SLAVE_NOTIFY_NOW);
1216
1217                 if (new_active) {
1218                         bool should_notify_peers = false;
1219
1220                         bond_set_slave_active_flags(new_active,
1221                                                     BOND_SLAVE_NOTIFY_NOW);
1222
1223                         if (bond->params.fail_over_mac)
1224                                 bond_do_fail_over_mac(bond, new_active,
1225                                                       old_active);
1226
1227                         if (netif_running(bond->dev)) {
1228                                 bond->send_peer_notif =
1229                                         bond->params.num_peer_notif *
1230                                         max(1, bond->params.peer_notif_delay);
1231                                 should_notify_peers =
1232                                         bond_should_notify_peers(bond);
1233                         }
1234
1235                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1236                         if (should_notify_peers) {
1237                                 bond->send_peer_notif--;
1238                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1239                                                          bond->dev);
1240                         }
1241                 }
1242         }
1243
1244 #ifdef CONFIG_XFRM_OFFLOAD
1245         bond_ipsec_add_sa_all(bond);
1246 #endif /* CONFIG_XFRM_OFFLOAD */
1247
1248         /* resend IGMP joins since active slave has changed or
1249          * all were sent on curr_active_slave.
1250          * resend only if bond is brought up with the affected
1251          * bonding modes and the retransmission is enabled
1252          */
1253         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1254             ((bond_uses_primary(bond) && new_active) ||
1255              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1256                 bond->igmp_retrans = bond->params.resend_igmp;
1257                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1258         }
1259 }
1260
1261 /**
1262  * bond_select_active_slave - select a new active slave, if needed
1263  * @bond: our bonding struct
1264  *
1265  * This functions should be called when one of the following occurs:
1266  * - The old curr_active_slave has been released or lost its link.
1267  * - The primary_slave has got its link back.
1268  * - A slave has got its link back and there's no old curr_active_slave.
1269  *
1270  * Caller must hold RTNL.
1271  */
1272 void bond_select_active_slave(struct bonding *bond)
1273 {
1274         struct slave *best_slave;
1275         int rv;
1276
1277         ASSERT_RTNL();
1278
1279         best_slave = bond_find_best_slave(bond);
1280         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1281                 bond_change_active_slave(bond, best_slave);
1282                 rv = bond_set_carrier(bond);
1283                 if (!rv)
1284                         return;
1285
1286                 if (netif_carrier_ok(bond->dev))
1287                         netdev_info(bond->dev, "active interface up!\n");
1288                 else
1289                         netdev_info(bond->dev, "now running without any active interface!\n");
1290         }
1291 }
1292
1293 #ifdef CONFIG_NET_POLL_CONTROLLER
1294 static inline int slave_enable_netpoll(struct slave *slave)
1295 {
1296         struct netpoll *np;
1297         int err = 0;
1298
1299         np = kzalloc(sizeof(*np), GFP_KERNEL);
1300         err = -ENOMEM;
1301         if (!np)
1302                 goto out;
1303
1304         err = __netpoll_setup(np, slave->dev);
1305         if (err) {
1306                 kfree(np);
1307                 goto out;
1308         }
1309         slave->np = np;
1310 out:
1311         return err;
1312 }
1313 static inline void slave_disable_netpoll(struct slave *slave)
1314 {
1315         struct netpoll *np = slave->np;
1316
1317         if (!np)
1318                 return;
1319
1320         slave->np = NULL;
1321
1322         __netpoll_free(np);
1323 }
1324
1325 static void bond_poll_controller(struct net_device *bond_dev)
1326 {
1327         struct bonding *bond = netdev_priv(bond_dev);
1328         struct slave *slave = NULL;
1329         struct list_head *iter;
1330         struct ad_info ad_info;
1331
1332         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1333                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1334                         return;
1335
1336         bond_for_each_slave_rcu(bond, slave, iter) {
1337                 if (!bond_slave_is_up(slave))
1338                         continue;
1339
1340                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1341                         struct aggregator *agg =
1342                             SLAVE_AD_INFO(slave)->port.aggregator;
1343
1344                         if (agg &&
1345                             agg->aggregator_identifier != ad_info.aggregator_id)
1346                                 continue;
1347                 }
1348
1349                 netpoll_poll_dev(slave->dev);
1350         }
1351 }
1352
1353 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1354 {
1355         struct bonding *bond = netdev_priv(bond_dev);
1356         struct list_head *iter;
1357         struct slave *slave;
1358
1359         bond_for_each_slave(bond, slave, iter)
1360                 if (bond_slave_is_up(slave))
1361                         slave_disable_netpoll(slave);
1362 }
1363
1364 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1365 {
1366         struct bonding *bond = netdev_priv(dev);
1367         struct list_head *iter;
1368         struct slave *slave;
1369         int err = 0;
1370
1371         bond_for_each_slave(bond, slave, iter) {
1372                 err = slave_enable_netpoll(slave);
1373                 if (err) {
1374                         bond_netpoll_cleanup(dev);
1375                         break;
1376                 }
1377         }
1378         return err;
1379 }
1380 #else
1381 static inline int slave_enable_netpoll(struct slave *slave)
1382 {
1383         return 0;
1384 }
1385 static inline void slave_disable_netpoll(struct slave *slave)
1386 {
1387 }
1388 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1389 {
1390 }
1391 #endif
1392
1393 /*---------------------------------- IOCTL ----------------------------------*/
1394
1395 static netdev_features_t bond_fix_features(struct net_device *dev,
1396                                            netdev_features_t features)
1397 {
1398         struct bonding *bond = netdev_priv(dev);
1399         struct list_head *iter;
1400         netdev_features_t mask;
1401         struct slave *slave;
1402
1403         mask = features;
1404
1405         features &= ~NETIF_F_ONE_FOR_ALL;
1406         features |= NETIF_F_ALL_FOR_ALL;
1407
1408         bond_for_each_slave(bond, slave, iter) {
1409                 features = netdev_increment_features(features,
1410                                                      slave->dev->features,
1411                                                      mask);
1412         }
1413         features = netdev_add_tso_features(features, mask);
1414
1415         return features;
1416 }
1417
1418 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1419                                  NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1420                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1421
1422 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1423                                  NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1424
1425 #define BOND_MPLS_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1426                                  NETIF_F_GSO_SOFTWARE)
1427
1428
1429 static void bond_compute_features(struct bonding *bond)
1430 {
1431         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1432                                         IFF_XMIT_DST_RELEASE_PERM;
1433         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1434         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1435 #ifdef CONFIG_XFRM_OFFLOAD
1436         netdev_features_t xfrm_features  = BOND_XFRM_FEATURES;
1437 #endif /* CONFIG_XFRM_OFFLOAD */
1438         netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
1439         struct net_device *bond_dev = bond->dev;
1440         struct list_head *iter;
1441         struct slave *slave;
1442         unsigned short max_hard_header_len = ETH_HLEN;
1443         unsigned int tso_max_size = TSO_MAX_SIZE;
1444         u16 tso_max_segs = TSO_MAX_SEGS;
1445
1446         if (!bond_has_slaves(bond))
1447                 goto done;
1448         vlan_features &= NETIF_F_ALL_FOR_ALL;
1449         mpls_features &= NETIF_F_ALL_FOR_ALL;
1450
1451         bond_for_each_slave(bond, slave, iter) {
1452                 vlan_features = netdev_increment_features(vlan_features,
1453                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1454
1455                 enc_features = netdev_increment_features(enc_features,
1456                                                          slave->dev->hw_enc_features,
1457                                                          BOND_ENC_FEATURES);
1458
1459 #ifdef CONFIG_XFRM_OFFLOAD
1460                 xfrm_features = netdev_increment_features(xfrm_features,
1461                                                           slave->dev->hw_enc_features,
1462                                                           BOND_XFRM_FEATURES);
1463 #endif /* CONFIG_XFRM_OFFLOAD */
1464
1465                 mpls_features = netdev_increment_features(mpls_features,
1466                                                           slave->dev->mpls_features,
1467                                                           BOND_MPLS_FEATURES);
1468
1469                 dst_release_flag &= slave->dev->priv_flags;
1470                 if (slave->dev->hard_header_len > max_hard_header_len)
1471                         max_hard_header_len = slave->dev->hard_header_len;
1472
1473                 tso_max_size = min(tso_max_size, slave->dev->tso_max_size);
1474                 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs);
1475         }
1476         bond_dev->hard_header_len = max_hard_header_len;
1477
1478 done:
1479         bond_dev->vlan_features = vlan_features;
1480         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1481                                     NETIF_F_HW_VLAN_CTAG_TX |
1482                                     NETIF_F_HW_VLAN_STAG_TX;
1483 #ifdef CONFIG_XFRM_OFFLOAD
1484         bond_dev->hw_enc_features |= xfrm_features;
1485 #endif /* CONFIG_XFRM_OFFLOAD */
1486         bond_dev->mpls_features = mpls_features;
1487         netif_set_tso_max_segs(bond_dev, tso_max_segs);
1488         netif_set_tso_max_size(bond_dev, tso_max_size);
1489
1490         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1491         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1492             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1493                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1494
1495         netdev_change_features(bond_dev);
1496 }
1497
1498 static void bond_setup_by_slave(struct net_device *bond_dev,
1499                                 struct net_device *slave_dev)
1500 {
1501         bond_dev->header_ops        = slave_dev->header_ops;
1502
1503         bond_dev->type              = slave_dev->type;
1504         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1505         bond_dev->needed_headroom   = slave_dev->needed_headroom;
1506         bond_dev->addr_len          = slave_dev->addr_len;
1507
1508         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1509                 slave_dev->addr_len);
1510 }
1511
1512 /* On bonding slaves other than the currently active slave, suppress
1513  * duplicates except for alb non-mcast/bcast.
1514  */
1515 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1516                                             struct slave *slave,
1517                                             struct bonding *bond)
1518 {
1519         if (bond_is_slave_inactive(slave)) {
1520                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1521                     skb->pkt_type != PACKET_BROADCAST &&
1522                     skb->pkt_type != PACKET_MULTICAST)
1523                         return false;
1524                 return true;
1525         }
1526         return false;
1527 }
1528
1529 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1530 {
1531         struct sk_buff *skb = *pskb;
1532         struct slave *slave;
1533         struct bonding *bond;
1534         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1535                           struct slave *);
1536         int ret = RX_HANDLER_ANOTHER;
1537
1538         skb = skb_share_check(skb, GFP_ATOMIC);
1539         if (unlikely(!skb))
1540                 return RX_HANDLER_CONSUMED;
1541
1542         *pskb = skb;
1543
1544         slave = bond_slave_get_rcu(skb->dev);
1545         bond = slave->bond;
1546
1547         recv_probe = READ_ONCE(bond->recv_probe);
1548         if (recv_probe) {
1549                 ret = recv_probe(skb, bond, slave);
1550                 if (ret == RX_HANDLER_CONSUMED) {
1551                         consume_skb(skb);
1552                         return ret;
1553                 }
1554         }
1555
1556         /*
1557          * For packets determined by bond_should_deliver_exact_match() call to
1558          * be suppressed we want to make an exception for link-local packets.
1559          * This is necessary for e.g. LLDP daemons to be able to monitor
1560          * inactive slave links without being forced to bind to them
1561          * explicitly.
1562          *
1563          * At the same time, packets that are passed to the bonding master
1564          * (including link-local ones) can have their originating interface
1565          * determined via PACKET_ORIGDEV socket option.
1566          */
1567         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1568                 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1569                         return RX_HANDLER_PASS;
1570                 return RX_HANDLER_EXACT;
1571         }
1572
1573         skb->dev = bond->dev;
1574
1575         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1576             netif_is_bridge_port(bond->dev) &&
1577             skb->pkt_type == PACKET_HOST) {
1578
1579                 if (unlikely(skb_cow_head(skb,
1580                                           skb->data - skb_mac_header(skb)))) {
1581                         kfree_skb(skb);
1582                         return RX_HANDLER_CONSUMED;
1583                 }
1584                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1585                                   bond->dev->addr_len);
1586         }
1587
1588         return ret;
1589 }
1590
1591 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1592 {
1593         switch (BOND_MODE(bond)) {
1594         case BOND_MODE_ROUNDROBIN:
1595                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1596         case BOND_MODE_ACTIVEBACKUP:
1597                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1598         case BOND_MODE_BROADCAST:
1599                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1600         case BOND_MODE_XOR:
1601         case BOND_MODE_8023AD:
1602                 return NETDEV_LAG_TX_TYPE_HASH;
1603         default:
1604                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1605         }
1606 }
1607
1608 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1609                                                enum netdev_lag_tx_type type)
1610 {
1611         if (type != NETDEV_LAG_TX_TYPE_HASH)
1612                 return NETDEV_LAG_HASH_NONE;
1613
1614         switch (bond->params.xmit_policy) {
1615         case BOND_XMIT_POLICY_LAYER2:
1616                 return NETDEV_LAG_HASH_L2;
1617         case BOND_XMIT_POLICY_LAYER34:
1618                 return NETDEV_LAG_HASH_L34;
1619         case BOND_XMIT_POLICY_LAYER23:
1620                 return NETDEV_LAG_HASH_L23;
1621         case BOND_XMIT_POLICY_ENCAP23:
1622                 return NETDEV_LAG_HASH_E23;
1623         case BOND_XMIT_POLICY_ENCAP34:
1624                 return NETDEV_LAG_HASH_E34;
1625         case BOND_XMIT_POLICY_VLAN_SRCMAC:
1626                 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1627         default:
1628                 return NETDEV_LAG_HASH_UNKNOWN;
1629         }
1630 }
1631
1632 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1633                                       struct netlink_ext_ack *extack)
1634 {
1635         struct netdev_lag_upper_info lag_upper_info;
1636         enum netdev_lag_tx_type type;
1637         int err;
1638
1639         type = bond_lag_tx_type(bond);
1640         lag_upper_info.tx_type = type;
1641         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1642
1643         err = netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1644                                            &lag_upper_info, extack);
1645         if (err)
1646                 return err;
1647
1648         slave->dev->flags |= IFF_SLAVE;
1649         return 0;
1650 }
1651
1652 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1653 {
1654         netdev_upper_dev_unlink(slave->dev, bond->dev);
1655         slave->dev->flags &= ~IFF_SLAVE;
1656 }
1657
1658 static void slave_kobj_release(struct kobject *kobj)
1659 {
1660         struct slave *slave = to_slave(kobj);
1661         struct bonding *bond = bond_get_bond_by_slave(slave);
1662
1663         cancel_delayed_work_sync(&slave->notify_work);
1664         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1665                 kfree(SLAVE_AD_INFO(slave));
1666
1667         kfree(slave);
1668 }
1669
1670 static struct kobj_type slave_ktype = {
1671         .release = slave_kobj_release,
1672 #ifdef CONFIG_SYSFS
1673         .sysfs_ops = &slave_sysfs_ops,
1674 #endif
1675 };
1676
1677 static int bond_kobj_init(struct slave *slave)
1678 {
1679         int err;
1680
1681         err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1682                                    &(slave->dev->dev.kobj), "bonding_slave");
1683         if (err)
1684                 kobject_put(&slave->kobj);
1685
1686         return err;
1687 }
1688
1689 static struct slave *bond_alloc_slave(struct bonding *bond,
1690                                       struct net_device *slave_dev)
1691 {
1692         struct slave *slave = NULL;
1693
1694         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1695         if (!slave)
1696                 return NULL;
1697
1698         slave->bond = bond;
1699         slave->dev = slave_dev;
1700         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1701
1702         if (bond_kobj_init(slave))
1703                 return NULL;
1704
1705         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1706                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1707                                                GFP_KERNEL);
1708                 if (!SLAVE_AD_INFO(slave)) {
1709                         kobject_put(&slave->kobj);
1710                         return NULL;
1711                 }
1712         }
1713
1714         return slave;
1715 }
1716
1717 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1718 {
1719         info->bond_mode = BOND_MODE(bond);
1720         info->miimon = bond->params.miimon;
1721         info->num_slaves = bond->slave_cnt;
1722 }
1723
1724 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1725 {
1726         strcpy(info->slave_name, slave->dev->name);
1727         info->link = slave->link;
1728         info->state = bond_slave_state(slave);
1729         info->link_failure_count = slave->link_failure_count;
1730 }
1731
1732 static void bond_netdev_notify_work(struct work_struct *_work)
1733 {
1734         struct slave *slave = container_of(_work, struct slave,
1735                                            notify_work.work);
1736
1737         if (rtnl_trylock()) {
1738                 struct netdev_bonding_info binfo;
1739
1740                 bond_fill_ifslave(slave, &binfo.slave);
1741                 bond_fill_ifbond(slave->bond, &binfo.master);
1742                 netdev_bonding_info_change(slave->dev, &binfo);
1743                 rtnl_unlock();
1744         } else {
1745                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1746         }
1747 }
1748
1749 void bond_queue_slave_event(struct slave *slave)
1750 {
1751         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1752 }
1753
1754 void bond_lower_state_changed(struct slave *slave)
1755 {
1756         struct netdev_lag_lower_state_info info;
1757
1758         info.link_up = slave->link == BOND_LINK_UP ||
1759                        slave->link == BOND_LINK_FAIL;
1760         info.tx_enabled = bond_is_active_slave(slave);
1761         netdev_lower_state_changed(slave->dev, &info);
1762 }
1763
1764 #define BOND_NL_ERR(bond_dev, extack, errmsg) do {              \
1765         if (extack)                                             \
1766                 NL_SET_ERR_MSG(extack, errmsg);                 \
1767         else                                                    \
1768                 netdev_err(bond_dev, "Error: %s\n", errmsg);    \
1769 } while (0)
1770
1771 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do {          \
1772         if (extack)                                                     \
1773                 NL_SET_ERR_MSG(extack, errmsg);                         \
1774         else                                                            \
1775                 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg);  \
1776 } while (0)
1777
1778 /* The bonding driver uses ether_setup() to convert a master bond device
1779  * to ARPHRD_ETHER, that resets the target netdevice's flags so we always
1780  * have to restore the IFF_MASTER flag, and only restore IFF_SLAVE if it was set
1781  */
1782 static void bond_ether_setup(struct net_device *bond_dev)
1783 {
1784         unsigned int slave_flag = bond_dev->flags & IFF_SLAVE;
1785
1786         ether_setup(bond_dev);
1787         bond_dev->flags |= IFF_MASTER | slave_flag;
1788         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1789 }
1790
1791 /* enslave device <slave> to bond device <master> */
1792 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1793                  struct netlink_ext_ack *extack)
1794 {
1795         struct bonding *bond = netdev_priv(bond_dev);
1796         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1797         struct slave *new_slave = NULL, *prev_slave;
1798         struct sockaddr_storage ss;
1799         int link_reporting;
1800         int res = 0, i;
1801
1802         if (slave_dev->flags & IFF_MASTER &&
1803             !netif_is_bond_master(slave_dev)) {
1804                 BOND_NL_ERR(bond_dev, extack,
1805                             "Device type (master device) cannot be enslaved");
1806                 return -EPERM;
1807         }
1808
1809         if (!bond->params.use_carrier &&
1810             slave_dev->ethtool_ops->get_link == NULL &&
1811             slave_ops->ndo_eth_ioctl == NULL) {
1812                 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1813         }
1814
1815         /* already in-use? */
1816         if (netdev_is_rx_handler_busy(slave_dev)) {
1817                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1818                              "Device is in use and cannot be enslaved");
1819                 return -EBUSY;
1820         }
1821
1822         if (bond_dev == slave_dev) {
1823                 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1824                 return -EPERM;
1825         }
1826
1827         /* vlan challenged mutual exclusion */
1828         /* no need to lock since we're protected by rtnl_lock */
1829         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1830                 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1831                 if (vlan_uses_dev(bond_dev)) {
1832                         SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1833                                      "Can not enslave VLAN challenged device to VLAN enabled bond");
1834                         return -EPERM;
1835                 } else {
1836                         slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1837                 }
1838         } else {
1839                 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1840         }
1841
1842         if (slave_dev->features & NETIF_F_HW_ESP)
1843                 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1844
1845         /* Old ifenslave binaries are no longer supported.  These can
1846          * be identified with moderate accuracy by the state of the slave:
1847          * the current ifenslave will set the interface down prior to
1848          * enslaving it; the old ifenslave will not.
1849          */
1850         if (slave_dev->flags & IFF_UP) {
1851                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1852                              "Device can not be enslaved while up");
1853                 return -EPERM;
1854         }
1855
1856         /* set bonding device ether type by slave - bonding netdevices are
1857          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1858          * there is a need to override some of the type dependent attribs/funcs.
1859          *
1860          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1861          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1862          */
1863         if (!bond_has_slaves(bond)) {
1864                 if (bond_dev->type != slave_dev->type) {
1865                         slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1866                                   bond_dev->type, slave_dev->type);
1867
1868                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1869                                                        bond_dev);
1870                         res = notifier_to_errno(res);
1871                         if (res) {
1872                                 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1873                                 return -EBUSY;
1874                         }
1875
1876                         /* Flush unicast and multicast addresses */
1877                         dev_uc_flush(bond_dev);
1878                         dev_mc_flush(bond_dev);
1879
1880                         if (slave_dev->type != ARPHRD_ETHER)
1881                                 bond_setup_by_slave(bond_dev, slave_dev);
1882                         else
1883                                 bond_ether_setup(bond_dev);
1884
1885                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1886                                                  bond_dev);
1887                 }
1888         } else if (bond_dev->type != slave_dev->type) {
1889                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1890                              "Device type is different from other slaves");
1891                 return -EINVAL;
1892         }
1893
1894         if (slave_dev->type == ARPHRD_INFINIBAND &&
1895             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1896                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1897                              "Only active-backup mode is supported for infiniband slaves");
1898                 res = -EOPNOTSUPP;
1899                 goto err_undo_flags;
1900         }
1901
1902         if (!slave_ops->ndo_set_mac_address ||
1903             slave_dev->type == ARPHRD_INFINIBAND) {
1904                 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1905                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1906                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1907                         if (!bond_has_slaves(bond)) {
1908                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1909                                 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1910                         } else {
1911                                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1912                                              "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1913                                 res = -EOPNOTSUPP;
1914                                 goto err_undo_flags;
1915                         }
1916                 }
1917         }
1918
1919         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1920
1921         /* If this is the first slave, then we need to set the master's hardware
1922          * address to be the same as the slave's.
1923          */
1924         if (!bond_has_slaves(bond) &&
1925             bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1926                 res = bond_set_dev_addr(bond->dev, slave_dev);
1927                 if (res)
1928                         goto err_undo_flags;
1929         }
1930
1931         new_slave = bond_alloc_slave(bond, slave_dev);
1932         if (!new_slave) {
1933                 res = -ENOMEM;
1934                 goto err_undo_flags;
1935         }
1936
1937         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1938          * is set via sysfs or module option if desired.
1939          */
1940         new_slave->queue_id = 0;
1941
1942         /* Save slave's original mtu and then set it to match the bond */
1943         new_slave->original_mtu = slave_dev->mtu;
1944         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1945         if (res) {
1946                 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1947                 goto err_free;
1948         }
1949
1950         /* Save slave's original ("permanent") mac address for modes
1951          * that need it, and for restoring it upon release, and then
1952          * set it to the master's address
1953          */
1954         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1955                           slave_dev->addr_len);
1956
1957         if (!bond->params.fail_over_mac ||
1958             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1959                 /* Set slave to master's mac address.  The application already
1960                  * set the master's mac address to that of the first slave
1961                  */
1962                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1963                 ss.ss_family = slave_dev->type;
1964                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1965                                           extack);
1966                 if (res) {
1967                         slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1968                         goto err_restore_mtu;
1969                 }
1970         }
1971
1972         /* set no_addrconf flag before open to prevent IPv6 addrconf */
1973         slave_dev->priv_flags |= IFF_NO_ADDRCONF;
1974
1975         /* open the slave since the application closed it */
1976         res = dev_open(slave_dev, extack);
1977         if (res) {
1978                 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1979                 goto err_restore_mac;
1980         }
1981
1982         slave_dev->priv_flags |= IFF_BONDING;
1983         /* initialize slave stats */
1984         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1985
1986         if (bond_is_lb(bond)) {
1987                 /* bond_alb_init_slave() must be called before all other stages since
1988                  * it might fail and we do not want to have to undo everything
1989                  */
1990                 res = bond_alb_init_slave(bond, new_slave);
1991                 if (res)
1992                         goto err_close;
1993         }
1994
1995         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1996         if (res) {
1997                 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1998                 goto err_close;
1999         }
2000
2001         prev_slave = bond_last_slave(bond);
2002
2003         new_slave->delay = 0;
2004         new_slave->link_failure_count = 0;
2005
2006         if (bond_update_speed_duplex(new_slave) &&
2007             bond_needs_speed_duplex(bond))
2008                 new_slave->link = BOND_LINK_DOWN;
2009
2010         new_slave->last_rx = jiffies -
2011                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
2012         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2013                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2014
2015         new_slave->last_tx = new_slave->last_rx;
2016
2017         if (bond->params.miimon && !bond->params.use_carrier) {
2018                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
2019
2020                 if ((link_reporting == -1) && !bond->params.arp_interval) {
2021                         /* miimon is set but a bonded network driver
2022                          * does not support ETHTOOL/MII and
2023                          * arp_interval is not set.  Note: if
2024                          * use_carrier is enabled, we will never go
2025                          * here (because netif_carrier is always
2026                          * supported); thus, we don't need to change
2027                          * the messages for netif_carrier.
2028                          */
2029                         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");
2030                 } else if (link_reporting == -1) {
2031                         /* unable get link status using mii/ethtool */
2032                         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");
2033                 }
2034         }
2035
2036         /* check for initial state */
2037         new_slave->link = BOND_LINK_NOCHANGE;
2038         if (bond->params.miimon) {
2039                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2040                         if (bond->params.updelay) {
2041                                 bond_set_slave_link_state(new_slave,
2042                                                           BOND_LINK_BACK,
2043                                                           BOND_SLAVE_NOTIFY_NOW);
2044                                 new_slave->delay = bond->params.updelay;
2045                         } else {
2046                                 bond_set_slave_link_state(new_slave,
2047                                                           BOND_LINK_UP,
2048                                                           BOND_SLAVE_NOTIFY_NOW);
2049                         }
2050                 } else {
2051                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2052                                                   BOND_SLAVE_NOTIFY_NOW);
2053                 }
2054         } else if (bond->params.arp_interval) {
2055                 bond_set_slave_link_state(new_slave,
2056                                           (netif_carrier_ok(slave_dev) ?
2057                                           BOND_LINK_UP : BOND_LINK_DOWN),
2058                                           BOND_SLAVE_NOTIFY_NOW);
2059         } else {
2060                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2061                                           BOND_SLAVE_NOTIFY_NOW);
2062         }
2063
2064         if (new_slave->link != BOND_LINK_DOWN)
2065                 new_slave->last_link_up = jiffies;
2066         slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2067                   new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2068                   (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2069
2070         if (bond_uses_primary(bond) && bond->params.primary[0]) {
2071                 /* if there is a primary slave, remember it */
2072                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2073                         rcu_assign_pointer(bond->primary_slave, new_slave);
2074                         bond->force_primary = true;
2075                 }
2076         }
2077
2078         switch (BOND_MODE(bond)) {
2079         case BOND_MODE_ACTIVEBACKUP:
2080                 bond_set_slave_inactive_flags(new_slave,
2081                                               BOND_SLAVE_NOTIFY_NOW);
2082                 break;
2083         case BOND_MODE_8023AD:
2084                 /* in 802.3ad mode, the internal mechanism
2085                  * will activate the slaves in the selected
2086                  * aggregator
2087                  */
2088                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2089                 /* if this is the first slave */
2090                 if (!prev_slave) {
2091                         SLAVE_AD_INFO(new_slave)->id = 1;
2092                         /* Initialize AD with the number of times that the AD timer is called in 1 second
2093                          * can be called only after the mac address of the bond is set
2094                          */
2095                         bond_3ad_initialize(bond);
2096                 } else {
2097                         SLAVE_AD_INFO(new_slave)->id =
2098                                 SLAVE_AD_INFO(prev_slave)->id + 1;
2099                 }
2100
2101                 bond_3ad_bind_slave(new_slave);
2102                 break;
2103         case BOND_MODE_TLB:
2104         case BOND_MODE_ALB:
2105                 bond_set_active_slave(new_slave);
2106                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2107                 break;
2108         default:
2109                 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2110
2111                 /* always active in trunk mode */
2112                 bond_set_active_slave(new_slave);
2113
2114                 /* In trunking mode there is little meaning to curr_active_slave
2115                  * anyway (it holds no special properties of the bond device),
2116                  * so we can change it without calling change_active_interface()
2117                  */
2118                 if (!rcu_access_pointer(bond->curr_active_slave) &&
2119                     new_slave->link == BOND_LINK_UP)
2120                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
2121
2122                 break;
2123         } /* switch(bond_mode) */
2124
2125 #ifdef CONFIG_NET_POLL_CONTROLLER
2126         if (bond->dev->npinfo) {
2127                 if (slave_enable_netpoll(new_slave)) {
2128                         slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2129                         res = -EBUSY;
2130                         goto err_detach;
2131                 }
2132         }
2133 #endif
2134
2135         if (!(bond_dev->features & NETIF_F_LRO))
2136                 dev_disable_lro(slave_dev);
2137
2138         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2139                                          new_slave);
2140         if (res) {
2141                 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2142                 goto err_detach;
2143         }
2144
2145         res = bond_master_upper_dev_link(bond, new_slave, extack);
2146         if (res) {
2147                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2148                 goto err_unregister;
2149         }
2150
2151         bond_lower_state_changed(new_slave);
2152
2153         res = bond_sysfs_slave_add(new_slave);
2154         if (res) {
2155                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2156                 goto err_upper_unlink;
2157         }
2158
2159         /* If the mode uses primary, then the following is handled by
2160          * bond_change_active_slave().
2161          */
2162         if (!bond_uses_primary(bond)) {
2163                 /* set promiscuity level to new slave */
2164                 if (bond_dev->flags & IFF_PROMISC) {
2165                         res = dev_set_promiscuity(slave_dev, 1);
2166                         if (res)
2167                                 goto err_sysfs_del;
2168                 }
2169
2170                 /* set allmulti level to new slave */
2171                 if (bond_dev->flags & IFF_ALLMULTI) {
2172                         res = dev_set_allmulti(slave_dev, 1);
2173                         if (res) {
2174                                 if (bond_dev->flags & IFF_PROMISC)
2175                                         dev_set_promiscuity(slave_dev, -1);
2176                                 goto err_sysfs_del;
2177                         }
2178                 }
2179
2180                 if (bond_dev->flags & IFF_UP) {
2181                         netif_addr_lock_bh(bond_dev);
2182                         dev_mc_sync_multiple(slave_dev, bond_dev);
2183                         dev_uc_sync_multiple(slave_dev, bond_dev);
2184                         netif_addr_unlock_bh(bond_dev);
2185
2186                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2187                                 dev_mc_add(slave_dev, lacpdu_mcast_addr);
2188                 }
2189         }
2190
2191         bond->slave_cnt++;
2192         bond_compute_features(bond);
2193         bond_set_carrier(bond);
2194
2195         if (bond_uses_primary(bond)) {
2196                 block_netpoll_tx();
2197                 bond_select_active_slave(bond);
2198                 unblock_netpoll_tx();
2199         }
2200
2201         if (bond_mode_can_use_xmit_hash(bond))
2202                 bond_update_slave_arr(bond, NULL);
2203
2204
2205         if (!slave_dev->netdev_ops->ndo_bpf ||
2206             !slave_dev->netdev_ops->ndo_xdp_xmit) {
2207                 if (bond->xdp_prog) {
2208                         SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2209                                      "Slave does not support XDP");
2210                         res = -EOPNOTSUPP;
2211                         goto err_sysfs_del;
2212                 }
2213         } else if (bond->xdp_prog) {
2214                 struct netdev_bpf xdp = {
2215                         .command = XDP_SETUP_PROG,
2216                         .flags   = 0,
2217                         .prog    = bond->xdp_prog,
2218                         .extack  = extack,
2219                 };
2220
2221                 if (dev_xdp_prog_count(slave_dev) > 0) {
2222                         SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2223                                      "Slave has XDP program loaded, please unload before enslaving");
2224                         res = -EOPNOTSUPP;
2225                         goto err_sysfs_del;
2226                 }
2227
2228                 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2229                 if (res < 0) {
2230                         /* ndo_bpf() sets extack error message */
2231                         slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2232                         goto err_sysfs_del;
2233                 }
2234                 if (bond->xdp_prog)
2235                         bpf_prog_inc(bond->xdp_prog);
2236         }
2237
2238         slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2239                    bond_is_active_slave(new_slave) ? "an active" : "a backup",
2240                    new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2241
2242         /* enslave is successful */
2243         bond_queue_slave_event(new_slave);
2244         return 0;
2245
2246 /* Undo stages on error */
2247 err_sysfs_del:
2248         bond_sysfs_slave_del(new_slave);
2249
2250 err_upper_unlink:
2251         bond_upper_dev_unlink(bond, new_slave);
2252
2253 err_unregister:
2254         netdev_rx_handler_unregister(slave_dev);
2255
2256 err_detach:
2257         vlan_vids_del_by_dev(slave_dev, bond_dev);
2258         if (rcu_access_pointer(bond->primary_slave) == new_slave)
2259                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2260         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2261                 block_netpoll_tx();
2262                 bond_change_active_slave(bond, NULL);
2263                 bond_select_active_slave(bond);
2264                 unblock_netpoll_tx();
2265         }
2266         /* either primary_slave or curr_active_slave might've changed */
2267         synchronize_rcu();
2268         slave_disable_netpoll(new_slave);
2269
2270 err_close:
2271         if (!netif_is_bond_master(slave_dev))
2272                 slave_dev->priv_flags &= ~IFF_BONDING;
2273         dev_close(slave_dev);
2274
2275 err_restore_mac:
2276         slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2277         if (!bond->params.fail_over_mac ||
2278             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2279                 /* XXX TODO - fom follow mode needs to change master's
2280                  * MAC if this slave's MAC is in use by the bond, or at
2281                  * least print a warning.
2282                  */
2283                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2284                                   new_slave->dev->addr_len);
2285                 ss.ss_family = slave_dev->type;
2286                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2287         }
2288
2289 err_restore_mtu:
2290         dev_set_mtu(slave_dev, new_slave->original_mtu);
2291
2292 err_free:
2293         kobject_put(&new_slave->kobj);
2294
2295 err_undo_flags:
2296         /* Enslave of first slave has failed and we need to fix master's mac */
2297         if (!bond_has_slaves(bond)) {
2298                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2299                                             slave_dev->dev_addr))
2300                         eth_hw_addr_random(bond_dev);
2301                 if (bond_dev->type != ARPHRD_ETHER) {
2302                         dev_close(bond_dev);
2303                         bond_ether_setup(bond_dev);
2304                 }
2305         }
2306
2307         return res;
2308 }
2309
2310 /* Try to release the slave device <slave> from the bond device <master>
2311  * It is legal to access curr_active_slave without a lock because all the function
2312  * is RTNL-locked. If "all" is true it means that the function is being called
2313  * while destroying a bond interface and all slaves are being released.
2314  *
2315  * The rules for slave state should be:
2316  *   for Active/Backup:
2317  *     Active stays on all backups go down
2318  *   for Bonded connections:
2319  *     The first up interface should be left on and all others downed.
2320  */
2321 static int __bond_release_one(struct net_device *bond_dev,
2322                               struct net_device *slave_dev,
2323                               bool all, bool unregister)
2324 {
2325         struct bonding *bond = netdev_priv(bond_dev);
2326         struct slave *slave, *oldcurrent;
2327         struct sockaddr_storage ss;
2328         int old_flags = bond_dev->flags;
2329         netdev_features_t old_features = bond_dev->features;
2330
2331         /* slave is not a slave or master is not master of this slave */
2332         if (!(slave_dev->flags & IFF_SLAVE) ||
2333             !netdev_has_upper_dev(slave_dev, bond_dev)) {
2334                 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2335                 return -EINVAL;
2336         }
2337
2338         block_netpoll_tx();
2339
2340         slave = bond_get_slave_by_dev(bond, slave_dev);
2341         if (!slave) {
2342                 /* not a slave of this bond */
2343                 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2344                 unblock_netpoll_tx();
2345                 return -EINVAL;
2346         }
2347
2348         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2349
2350         bond_sysfs_slave_del(slave);
2351
2352         /* recompute stats just before removing the slave */
2353         bond_get_stats(bond->dev, &bond->bond_stats);
2354
2355         if (bond->xdp_prog) {
2356                 struct netdev_bpf xdp = {
2357                         .command = XDP_SETUP_PROG,
2358                         .flags   = 0,
2359                         .prog    = NULL,
2360                         .extack  = NULL,
2361                 };
2362                 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2363                         slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2364         }
2365
2366         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2367          * for this slave anymore.
2368          */
2369         netdev_rx_handler_unregister(slave_dev);
2370
2371         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2372                 bond_3ad_unbind_slave(slave);
2373
2374         bond_upper_dev_unlink(bond, slave);
2375
2376         if (bond_mode_can_use_xmit_hash(bond))
2377                 bond_update_slave_arr(bond, slave);
2378
2379         slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2380                     bond_is_active_slave(slave) ? "active" : "backup");
2381
2382         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2383
2384         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2385
2386         if (!all && (!bond->params.fail_over_mac ||
2387                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2388                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2389                     bond_has_slaves(bond))
2390                         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",
2391                                    slave->perm_hwaddr);
2392         }
2393
2394         if (rtnl_dereference(bond->primary_slave) == slave)
2395                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2396
2397         if (oldcurrent == slave)
2398                 bond_change_active_slave(bond, NULL);
2399
2400         if (bond_is_lb(bond)) {
2401                 /* Must be called only after the slave has been
2402                  * detached from the list and the curr_active_slave
2403                  * has been cleared (if our_slave == old_current),
2404                  * but before a new active slave is selected.
2405                  */
2406                 bond_alb_deinit_slave(bond, slave);
2407         }
2408
2409         if (all) {
2410                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2411         } else if (oldcurrent == slave) {
2412                 /* Note that we hold RTNL over this sequence, so there
2413                  * is no concern that another slave add/remove event
2414                  * will interfere.
2415                  */
2416                 bond_select_active_slave(bond);
2417         }
2418
2419         bond_set_carrier(bond);
2420         if (!bond_has_slaves(bond))
2421                 eth_hw_addr_random(bond_dev);
2422
2423         unblock_netpoll_tx();
2424         synchronize_rcu();
2425         bond->slave_cnt--;
2426
2427         if (!bond_has_slaves(bond)) {
2428                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2429                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2430         }
2431
2432         bond_compute_features(bond);
2433         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2434             (old_features & NETIF_F_VLAN_CHALLENGED))
2435                 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2436
2437         vlan_vids_del_by_dev(slave_dev, bond_dev);
2438
2439         /* If the mode uses primary, then this case was handled above by
2440          * bond_change_active_slave(..., NULL)
2441          */
2442         if (!bond_uses_primary(bond)) {
2443                 /* unset promiscuity level from slave
2444                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
2445                  * of the IFF_PROMISC flag in the bond_dev, but we need the
2446                  * value of that flag before that change, as that was the value
2447                  * when this slave was attached, so we cache at the start of the
2448                  * function and use it here. Same goes for ALLMULTI below
2449                  */
2450                 if (old_flags & IFF_PROMISC)
2451                         dev_set_promiscuity(slave_dev, -1);
2452
2453                 /* unset allmulti level from slave */
2454                 if (old_flags & IFF_ALLMULTI)
2455                         dev_set_allmulti(slave_dev, -1);
2456
2457                 if (old_flags & IFF_UP)
2458                         bond_hw_addr_flush(bond_dev, slave_dev);
2459         }
2460
2461         slave_disable_netpoll(slave);
2462
2463         /* close slave before restoring its mac address */
2464         dev_close(slave_dev);
2465
2466         slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2467
2468         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2469             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2470                 /* restore original ("permanent") mac address */
2471                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2472                                   slave->dev->addr_len);
2473                 ss.ss_family = slave_dev->type;
2474                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2475         }
2476
2477         if (unregister)
2478                 __dev_set_mtu(slave_dev, slave->original_mtu);
2479         else
2480                 dev_set_mtu(slave_dev, slave->original_mtu);
2481
2482         if (!netif_is_bond_master(slave_dev))
2483                 slave_dev->priv_flags &= ~IFF_BONDING;
2484
2485         kobject_put(&slave->kobj);
2486
2487         return 0;
2488 }
2489
2490 /* A wrapper used because of ndo_del_link */
2491 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2492 {
2493         return __bond_release_one(bond_dev, slave_dev, false, false);
2494 }
2495
2496 /* First release a slave and then destroy the bond if no more slaves are left.
2497  * Must be under rtnl_lock when this function is called.
2498  */
2499 static int bond_release_and_destroy(struct net_device *bond_dev,
2500                                     struct net_device *slave_dev)
2501 {
2502         struct bonding *bond = netdev_priv(bond_dev);
2503         int ret;
2504
2505         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2506         if (ret == 0 && !bond_has_slaves(bond) &&
2507             bond_dev->reg_state != NETREG_UNREGISTERING) {
2508                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2509                 netdev_info(bond_dev, "Destroying bond\n");
2510                 bond_remove_proc_entry(bond);
2511                 unregister_netdevice(bond_dev);
2512         }
2513         return ret;
2514 }
2515
2516 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2517 {
2518         struct bonding *bond = netdev_priv(bond_dev);
2519
2520         bond_fill_ifbond(bond, info);
2521 }
2522
2523 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2524 {
2525         struct bonding *bond = netdev_priv(bond_dev);
2526         struct list_head *iter;
2527         int i = 0, res = -ENODEV;
2528         struct slave *slave;
2529
2530         bond_for_each_slave(bond, slave, iter) {
2531                 if (i++ == (int)info->slave_id) {
2532                         res = 0;
2533                         bond_fill_ifslave(slave, info);
2534                         break;
2535                 }
2536         }
2537
2538         return res;
2539 }
2540
2541 /*-------------------------------- Monitoring -------------------------------*/
2542
2543 /* called with rcu_read_lock() */
2544 static int bond_miimon_inspect(struct bonding *bond)
2545 {
2546         bool ignore_updelay = false;
2547         int link_state, commit = 0;
2548         struct list_head *iter;
2549         struct slave *slave;
2550
2551         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
2552                 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2553         } else {
2554                 struct bond_up_slave *usable_slaves;
2555
2556                 usable_slaves = rcu_dereference(bond->usable_slaves);
2557
2558                 if (usable_slaves && usable_slaves->count == 0)
2559                         ignore_updelay = true;
2560         }
2561
2562         bond_for_each_slave_rcu(bond, slave, iter) {
2563                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2564
2565                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2566
2567                 switch (slave->link) {
2568                 case BOND_LINK_UP:
2569                         if (link_state)
2570                                 continue;
2571
2572                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2573                         commit++;
2574                         slave->delay = bond->params.downdelay;
2575                         if (slave->delay) {
2576                                 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2577                                            (BOND_MODE(bond) ==
2578                                             BOND_MODE_ACTIVEBACKUP) ?
2579                                             (bond_is_active_slave(slave) ?
2580                                              "active " : "backup ") : "",
2581                                            bond->params.downdelay * bond->params.miimon);
2582                         }
2583                         fallthrough;
2584                 case BOND_LINK_FAIL:
2585                         if (link_state) {
2586                                 /* recovered before downdelay expired */
2587                                 bond_propose_link_state(slave, BOND_LINK_UP);
2588                                 slave->last_link_up = jiffies;
2589                                 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2590                                            (bond->params.downdelay - slave->delay) *
2591                                            bond->params.miimon);
2592                                 commit++;
2593                                 continue;
2594                         }
2595
2596                         if (slave->delay <= 0) {
2597                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2598                                 commit++;
2599                                 continue;
2600                         }
2601
2602                         slave->delay--;
2603                         break;
2604
2605                 case BOND_LINK_DOWN:
2606                         if (!link_state)
2607                                 continue;
2608
2609                         bond_propose_link_state(slave, BOND_LINK_BACK);
2610                         commit++;
2611                         slave->delay = bond->params.updelay;
2612
2613                         if (slave->delay) {
2614                                 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2615                                            ignore_updelay ? 0 :
2616                                            bond->params.updelay *
2617                                            bond->params.miimon);
2618                         }
2619                         fallthrough;
2620                 case BOND_LINK_BACK:
2621                         if (!link_state) {
2622                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2623                                 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2624                                            (bond->params.updelay - slave->delay) *
2625                                            bond->params.miimon);
2626                                 commit++;
2627                                 continue;
2628                         }
2629
2630                         if (ignore_updelay)
2631                                 slave->delay = 0;
2632
2633                         if (slave->delay <= 0) {
2634                                 bond_propose_link_state(slave, BOND_LINK_UP);
2635                                 commit++;
2636                                 ignore_updelay = false;
2637                                 continue;
2638                         }
2639
2640                         slave->delay--;
2641                         break;
2642                 }
2643         }
2644
2645         return commit;
2646 }
2647
2648 static void bond_miimon_link_change(struct bonding *bond,
2649                                     struct slave *slave,
2650                                     char link)
2651 {
2652         switch (BOND_MODE(bond)) {
2653         case BOND_MODE_8023AD:
2654                 bond_3ad_handle_link_change(slave, link);
2655                 break;
2656         case BOND_MODE_TLB:
2657         case BOND_MODE_ALB:
2658                 bond_alb_handle_link_change(bond, slave, link);
2659                 break;
2660         case BOND_MODE_XOR:
2661                 bond_update_slave_arr(bond, NULL);
2662                 break;
2663         }
2664 }
2665
2666 static void bond_miimon_commit(struct bonding *bond)
2667 {
2668         struct slave *slave, *primary, *active;
2669         bool do_failover = false;
2670         struct list_head *iter;
2671
2672         ASSERT_RTNL();
2673
2674         bond_for_each_slave(bond, slave, iter) {
2675                 switch (slave->link_new_state) {
2676                 case BOND_LINK_NOCHANGE:
2677                         /* For 802.3ad mode, check current slave speed and
2678                          * duplex again in case its port was disabled after
2679                          * invalid speed/duplex reporting but recovered before
2680                          * link monitoring could make a decision on the actual
2681                          * link status
2682                          */
2683                         if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2684                             slave->link == BOND_LINK_UP)
2685                                 bond_3ad_adapter_speed_duplex_changed(slave);
2686                         continue;
2687
2688                 case BOND_LINK_UP:
2689                         if (bond_update_speed_duplex(slave) &&
2690                             bond_needs_speed_duplex(bond)) {
2691                                 slave->link = BOND_LINK_DOWN;
2692                                 if (net_ratelimit())
2693                                         slave_warn(bond->dev, slave->dev,
2694                                                    "failed to get link speed/duplex\n");
2695                                 continue;
2696                         }
2697                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2698                                                   BOND_SLAVE_NOTIFY_NOW);
2699                         slave->last_link_up = jiffies;
2700
2701                         primary = rtnl_dereference(bond->primary_slave);
2702                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2703                                 /* prevent it from being the active one */
2704                                 bond_set_backup_slave(slave);
2705                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2706                                 /* make it immediately active */
2707                                 bond_set_active_slave(slave);
2708                         }
2709
2710                         slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2711                                    slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2712                                    slave->duplex ? "full" : "half");
2713
2714                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2715
2716                         active = rtnl_dereference(bond->curr_active_slave);
2717                         if (!active || slave == primary || slave->prio > active->prio)
2718                                 do_failover = true;
2719
2720                         continue;
2721
2722                 case BOND_LINK_DOWN:
2723                         if (slave->link_failure_count < UINT_MAX)
2724                                 slave->link_failure_count++;
2725
2726                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2727                                                   BOND_SLAVE_NOTIFY_NOW);
2728
2729                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2730                             BOND_MODE(bond) == BOND_MODE_8023AD)
2731                                 bond_set_slave_inactive_flags(slave,
2732                                                               BOND_SLAVE_NOTIFY_NOW);
2733
2734                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2735
2736                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2737
2738                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2739                                 do_failover = true;
2740
2741                         continue;
2742
2743                 default:
2744                         slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2745                                   slave->link_new_state);
2746                         bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2747
2748                         continue;
2749                 }
2750         }
2751
2752         if (do_failover) {
2753                 block_netpoll_tx();
2754                 bond_select_active_slave(bond);
2755                 unblock_netpoll_tx();
2756         }
2757
2758         bond_set_carrier(bond);
2759 }
2760
2761 /* bond_mii_monitor
2762  *
2763  * Really a wrapper that splits the mii monitor into two phases: an
2764  * inspection, then (if inspection indicates something needs to be done)
2765  * an acquisition of appropriate locks followed by a commit phase to
2766  * implement whatever link state changes are indicated.
2767  */
2768 static void bond_mii_monitor(struct work_struct *work)
2769 {
2770         struct bonding *bond = container_of(work, struct bonding,
2771                                             mii_work.work);
2772         bool should_notify_peers = false;
2773         bool commit;
2774         unsigned long delay;
2775         struct slave *slave;
2776         struct list_head *iter;
2777
2778         delay = msecs_to_jiffies(bond->params.miimon);
2779
2780         if (!bond_has_slaves(bond))
2781                 goto re_arm;
2782
2783         rcu_read_lock();
2784         should_notify_peers = bond_should_notify_peers(bond);
2785         commit = !!bond_miimon_inspect(bond);
2786         if (bond->send_peer_notif) {
2787                 rcu_read_unlock();
2788                 if (rtnl_trylock()) {
2789                         bond->send_peer_notif--;
2790                         rtnl_unlock();
2791                 }
2792         } else {
2793                 rcu_read_unlock();
2794         }
2795
2796         if (commit) {
2797                 /* Race avoidance with bond_close cancel of workqueue */
2798                 if (!rtnl_trylock()) {
2799                         delay = 1;
2800                         should_notify_peers = false;
2801                         goto re_arm;
2802                 }
2803
2804                 bond_for_each_slave(bond, slave, iter) {
2805                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2806                 }
2807                 bond_miimon_commit(bond);
2808
2809                 rtnl_unlock();  /* might sleep, hold no other locks */
2810         }
2811
2812 re_arm:
2813         if (bond->params.miimon)
2814                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2815
2816         if (should_notify_peers) {
2817                 if (!rtnl_trylock())
2818                         return;
2819                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2820                 rtnl_unlock();
2821         }
2822 }
2823
2824 static int bond_upper_dev_walk(struct net_device *upper,
2825                                struct netdev_nested_priv *priv)
2826 {
2827         __be32 ip = *(__be32 *)priv->data;
2828
2829         return ip == bond_confirm_addr(upper, 0, ip);
2830 }
2831
2832 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2833 {
2834         struct netdev_nested_priv priv = {
2835                 .data = (void *)&ip,
2836         };
2837         bool ret = false;
2838
2839         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2840                 return true;
2841
2842         rcu_read_lock();
2843         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2844                 ret = true;
2845         rcu_read_unlock();
2846
2847         return ret;
2848 }
2849
2850 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2851                              struct sk_buff *skb)
2852 {
2853         struct net_device *bond_dev = slave->bond->dev;
2854         struct net_device *slave_dev = slave->dev;
2855         struct bond_vlan_tag *outer_tag = tags;
2856
2857         if (!tags || tags->vlan_proto == VLAN_N_VID)
2858                 return true;
2859
2860         tags++;
2861
2862         /* Go through all the tags backwards and add them to the packet */
2863         while (tags->vlan_proto != VLAN_N_VID) {
2864                 if (!tags->vlan_id) {
2865                         tags++;
2866                         continue;
2867                 }
2868
2869                 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2870                           ntohs(outer_tag->vlan_proto), tags->vlan_id);
2871                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2872                                                 tags->vlan_id);
2873                 if (!skb) {
2874                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2875                         return false;
2876                 }
2877
2878                 tags++;
2879         }
2880         /* Set the outer tag */
2881         if (outer_tag->vlan_id) {
2882                 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2883                           ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2884                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2885                                        outer_tag->vlan_id);
2886         }
2887
2888         return true;
2889 }
2890
2891 /* We go to the (large) trouble of VLAN tagging ARP frames because
2892  * switches in VLAN mode (especially if ports are configured as
2893  * "native" to a VLAN) might not pass non-tagged frames.
2894  */
2895 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2896                           __be32 src_ip, struct bond_vlan_tag *tags)
2897 {
2898         struct net_device *bond_dev = slave->bond->dev;
2899         struct net_device *slave_dev = slave->dev;
2900         struct sk_buff *skb;
2901
2902         slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2903                   arp_op, &dest_ip, &src_ip);
2904
2905         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2906                          NULL, slave_dev->dev_addr, NULL);
2907
2908         if (!skb) {
2909                 net_err_ratelimited("ARP packet allocation failed\n");
2910                 return;
2911         }
2912
2913         if (bond_handle_vlan(slave, tags, skb)) {
2914                 slave_update_last_tx(slave);
2915                 arp_xmit(skb);
2916         }
2917
2918         return;
2919 }
2920
2921 /* Validate the device path between the @start_dev and the @end_dev.
2922  * The path is valid if the @end_dev is reachable through device
2923  * stacking.
2924  * When the path is validated, collect any vlan information in the
2925  * path.
2926  */
2927 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2928                                               struct net_device *end_dev,
2929                                               int level)
2930 {
2931         struct bond_vlan_tag *tags;
2932         struct net_device *upper;
2933         struct list_head  *iter;
2934
2935         if (start_dev == end_dev) {
2936                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2937                 if (!tags)
2938                         return ERR_PTR(-ENOMEM);
2939                 tags[level].vlan_proto = VLAN_N_VID;
2940                 return tags;
2941         }
2942
2943         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2944                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2945                 if (IS_ERR_OR_NULL(tags)) {
2946                         if (IS_ERR(tags))
2947                                 return tags;
2948                         continue;
2949                 }
2950                 if (is_vlan_dev(upper)) {
2951                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2952                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2953                 }
2954
2955                 return tags;
2956         }
2957
2958         return NULL;
2959 }
2960
2961 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2962 {
2963         struct rtable *rt;
2964         struct bond_vlan_tag *tags;
2965         __be32 *targets = bond->params.arp_targets, addr;
2966         int i;
2967
2968         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2969                 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2970                           __func__, &targets[i]);
2971                 tags = NULL;
2972
2973                 /* Find out through which dev should the packet go */
2974                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2975                                      RTO_ONLINK, 0);
2976                 if (IS_ERR(rt)) {
2977                         /* there's no route to target - try to send arp
2978                          * probe to generate any traffic (arp_validate=0)
2979                          */
2980                         if (bond->params.arp_validate)
2981                                 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2982                                              bond->dev->name,
2983                                              &targets[i]);
2984                         bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2985                                       0, tags);
2986                         continue;
2987                 }
2988
2989                 /* bond device itself */
2990                 if (rt->dst.dev == bond->dev)
2991                         goto found;
2992
2993                 rcu_read_lock();
2994                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2995                 rcu_read_unlock();
2996
2997                 if (!IS_ERR_OR_NULL(tags))
2998                         goto found;
2999
3000                 /* Not our device - skip */
3001                 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
3002                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
3003
3004                 ip_rt_put(rt);
3005                 continue;
3006
3007 found:
3008                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
3009                 ip_rt_put(rt);
3010                 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
3011                 kfree(tags);
3012         }
3013 }
3014
3015 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
3016 {
3017         int i;
3018
3019         if (!sip || !bond_has_this_ip(bond, tip)) {
3020                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
3021                            __func__, &sip, &tip);
3022                 return;
3023         }
3024
3025         i = bond_get_targets_ip(bond->params.arp_targets, sip);
3026         if (i == -1) {
3027                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3028                            __func__, &sip);
3029                 return;
3030         }
3031         slave->last_rx = jiffies;
3032         slave->target_last_arp_rx[i] = jiffies;
3033 }
3034
3035 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3036                         struct slave *slave)
3037 {
3038         struct arphdr *arp = (struct arphdr *)skb->data;
3039         struct slave *curr_active_slave, *curr_arp_slave;
3040         unsigned char *arp_ptr;
3041         __be32 sip, tip;
3042         unsigned int alen;
3043
3044         alen = arp_hdr_len(bond->dev);
3045
3046         if (alen > skb_headlen(skb)) {
3047                 arp = kmalloc(alen, GFP_ATOMIC);
3048                 if (!arp)
3049                         goto out_unlock;
3050                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
3051                         goto out_unlock;
3052         }
3053
3054         if (arp->ar_hln != bond->dev->addr_len ||
3055             skb->pkt_type == PACKET_OTHERHOST ||
3056             skb->pkt_type == PACKET_LOOPBACK ||
3057             arp->ar_hrd != htons(ARPHRD_ETHER) ||
3058             arp->ar_pro != htons(ETH_P_IP) ||
3059             arp->ar_pln != 4)
3060                 goto out_unlock;
3061
3062         arp_ptr = (unsigned char *)(arp + 1);
3063         arp_ptr += bond->dev->addr_len;
3064         memcpy(&sip, arp_ptr, 4);
3065         arp_ptr += 4 + bond->dev->addr_len;
3066         memcpy(&tip, arp_ptr, 4);
3067
3068         slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3069                   __func__, slave->dev->name, bond_slave_state(slave),
3070                   bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3071                   &sip, &tip);
3072
3073         curr_active_slave = rcu_dereference(bond->curr_active_slave);
3074         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3075
3076         /* We 'trust' the received ARP enough to validate it if:
3077          *
3078          * (a) the slave receiving the ARP is active (which includes the
3079          * current ARP slave, if any), or
3080          *
3081          * (b) the receiving slave isn't active, but there is a currently
3082          * active slave and it received valid arp reply(s) after it became
3083          * the currently active slave, or
3084          *
3085          * (c) there is an ARP slave that sent an ARP during the prior ARP
3086          * interval, and we receive an ARP reply on any slave.  We accept
3087          * these because switch FDB update delays may deliver the ARP
3088          * reply to a slave other than the sender of the ARP request.
3089          *
3090          * Note: for (b), backup slaves are receiving the broadcast ARP
3091          * request, not a reply.  This request passes from the sending
3092          * slave through the L2 switch(es) to the receiving slave.  Since
3093          * this is checking the request, sip/tip are swapped for
3094          * validation.
3095          *
3096          * This is done to avoid endless looping when we can't reach the
3097          * arp_ip_target and fool ourselves with our own arp requests.
3098          */
3099         if (bond_is_active_slave(slave))
3100                 bond_validate_arp(bond, slave, sip, tip);
3101         else if (curr_active_slave &&
3102                  time_after(slave_last_rx(bond, curr_active_slave),
3103                             curr_active_slave->last_link_up))
3104                 bond_validate_arp(bond, slave, tip, sip);
3105         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3106                  bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3107                 bond_validate_arp(bond, slave, sip, tip);
3108
3109 out_unlock:
3110         if (arp != (struct arphdr *)skb->data)
3111                 kfree(arp);
3112         return RX_HANDLER_ANOTHER;
3113 }
3114
3115 #if IS_ENABLED(CONFIG_IPV6)
3116 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3117                          const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3118 {
3119         struct net_device *bond_dev = slave->bond->dev;
3120         struct net_device *slave_dev = slave->dev;
3121         struct in6_addr mcaddr;
3122         struct sk_buff *skb;
3123
3124         slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3125                   daddr, saddr);
3126
3127         skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3128         if (!skb) {
3129                 net_err_ratelimited("NS packet allocation failed\n");
3130                 return;
3131         }
3132
3133         addrconf_addr_solict_mult(daddr, &mcaddr);
3134         if (bond_handle_vlan(slave, tags, skb)) {
3135                 slave_update_last_tx(slave);
3136                 ndisc_send_skb(skb, &mcaddr, saddr);
3137         }
3138 }
3139
3140 static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3141 {
3142         struct in6_addr *targets = bond->params.ns_targets;
3143         struct bond_vlan_tag *tags;
3144         struct dst_entry *dst;
3145         struct in6_addr saddr;
3146         struct flowi6 fl6;
3147         int i;
3148
3149         for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3150                 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3151                           __func__, &targets[i]);
3152                 tags = NULL;
3153
3154                 /* Find out through which dev should the packet go */
3155                 memset(&fl6, 0, sizeof(struct flowi6));
3156                 fl6.daddr = targets[i];
3157                 fl6.flowi6_oif = bond->dev->ifindex;
3158
3159                 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3160                 if (dst->error) {
3161                         dst_release(dst);
3162                         /* there's no route to target - try to send arp
3163                          * probe to generate any traffic (arp_validate=0)
3164                          */
3165                         if (bond->params.arp_validate)
3166                                 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3167                                              bond->dev->name,
3168                                              &targets[i]);
3169                         bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3170                         continue;
3171                 }
3172
3173                 /* bond device itself */
3174                 if (dst->dev == bond->dev)
3175                         goto found;
3176
3177                 rcu_read_lock();
3178                 tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3179                 rcu_read_unlock();
3180
3181                 if (!IS_ERR_OR_NULL(tags))
3182                         goto found;
3183
3184                 /* Not our device - skip */
3185                 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3186                           &targets[i], dst->dev ? dst->dev->name : "NULL");
3187
3188                 dst_release(dst);
3189                 continue;
3190
3191 found:
3192                 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3193                         bond_ns_send(slave, &targets[i], &saddr, tags);
3194                 else
3195                         bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3196
3197                 dst_release(dst);
3198                 kfree(tags);
3199         }
3200 }
3201
3202 static int bond_confirm_addr6(struct net_device *dev,
3203                               struct netdev_nested_priv *priv)
3204 {
3205         struct in6_addr *addr = (struct in6_addr *)priv->data;
3206
3207         return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3208 }
3209
3210 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3211 {
3212         struct netdev_nested_priv priv = {
3213                 .data = addr,
3214         };
3215         int ret = false;
3216
3217         if (bond_confirm_addr6(bond->dev, &priv))
3218                 return true;
3219
3220         rcu_read_lock();
3221         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3222                 ret = true;
3223         rcu_read_unlock();
3224
3225         return ret;
3226 }
3227
3228 static void bond_validate_na(struct bonding *bond, struct slave *slave,
3229                              struct in6_addr *saddr, struct in6_addr *daddr)
3230 {
3231         int i;
3232
3233         /* Ignore NAs that:
3234          * 1. Source address is unspecified address.
3235          * 2. Dest address is neither all-nodes multicast address nor
3236          *    exist on bond interface.
3237          */
3238         if (ipv6_addr_any(saddr) ||
3239             (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) &&
3240              !bond_has_this_ip6(bond, daddr))) {
3241                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3242                           __func__, saddr, daddr);
3243                 return;
3244         }
3245
3246         i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3247         if (i == -1) {
3248                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3249                           __func__, saddr);
3250                 return;
3251         }
3252         slave->last_rx = jiffies;
3253         slave->target_last_arp_rx[i] = jiffies;
3254 }
3255
3256 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3257                        struct slave *slave)
3258 {
3259         struct slave *curr_active_slave, *curr_arp_slave;
3260         struct in6_addr *saddr, *daddr;
3261         struct {
3262                 struct ipv6hdr ip6;
3263                 struct icmp6hdr icmp6;
3264         } *combined, _combined;
3265
3266         if (skb->pkt_type == PACKET_OTHERHOST ||
3267             skb->pkt_type == PACKET_LOOPBACK)
3268                 goto out;
3269
3270         combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined);
3271         if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP ||
3272             combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT)
3273                 goto out;
3274
3275         saddr = &combined->ip6.saddr;
3276         daddr = &combined->ip6.daddr;
3277
3278         slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3279                   __func__, slave->dev->name, bond_slave_state(slave),
3280                   bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3281                   saddr, daddr);
3282
3283         curr_active_slave = rcu_dereference(bond->curr_active_slave);
3284         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3285
3286         /* We 'trust' the received ARP enough to validate it if:
3287          * see bond_arp_rcv().
3288          */
3289         if (bond_is_active_slave(slave))
3290                 bond_validate_na(bond, slave, saddr, daddr);
3291         else if (curr_active_slave &&
3292                  time_after(slave_last_rx(bond, curr_active_slave),
3293                             curr_active_slave->last_link_up))
3294                 bond_validate_na(bond, slave, saddr, daddr);
3295         else if (curr_arp_slave &&
3296                  bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3297                 bond_validate_na(bond, slave, saddr, daddr);
3298
3299 out:
3300         return RX_HANDLER_ANOTHER;
3301 }
3302 #endif
3303
3304 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3305                       struct slave *slave)
3306 {
3307 #if IS_ENABLED(CONFIG_IPV6)
3308         bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3309 #endif
3310         bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3311
3312         slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3313                   __func__, skb->dev->name);
3314
3315         /* Use arp validate logic for both ARP and NS */
3316         if (!slave_do_arp_validate(bond, slave)) {
3317                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
3318 #if IS_ENABLED(CONFIG_IPV6)
3319                     (slave_do_arp_validate_only(bond) && is_ipv6) ||
3320 #endif
3321                     !slave_do_arp_validate_only(bond))
3322                         slave->last_rx = jiffies;
3323                 return RX_HANDLER_ANOTHER;
3324         } else if (is_arp) {
3325                 return bond_arp_rcv(skb, bond, slave);
3326 #if IS_ENABLED(CONFIG_IPV6)
3327         } else if (is_ipv6) {
3328                 return bond_na_rcv(skb, bond, slave);
3329 #endif
3330         } else {
3331                 return RX_HANDLER_ANOTHER;
3332         }
3333 }
3334
3335 static void bond_send_validate(struct bonding *bond, struct slave *slave)
3336 {
3337         bond_arp_send_all(bond, slave);
3338 #if IS_ENABLED(CONFIG_IPV6)
3339         bond_ns_send_all(bond, slave);
3340 #endif
3341 }
3342
3343 /* function to verify if we're in the arp_interval timeslice, returns true if
3344  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3345  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3346  */
3347 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3348                                   int mod)
3349 {
3350         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3351
3352         return time_in_range(jiffies,
3353                              last_act - delta_in_ticks,
3354                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
3355 }
3356
3357 /* This function is called regularly to monitor each slave's link
3358  * ensuring that traffic is being sent and received when arp monitoring
3359  * is used in load-balancing mode. if the adapter has been dormant, then an
3360  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3361  * arp monitoring in active backup mode.
3362  */
3363 static void bond_loadbalance_arp_mon(struct bonding *bond)
3364 {
3365         struct slave *slave, *oldcurrent;
3366         struct list_head *iter;
3367         int do_failover = 0, slave_state_changed = 0;
3368
3369         if (!bond_has_slaves(bond))
3370                 goto re_arm;
3371
3372         rcu_read_lock();
3373
3374         oldcurrent = rcu_dereference(bond->curr_active_slave);
3375         /* see if any of the previous devices are up now (i.e. they have
3376          * xmt and rcv traffic). the curr_active_slave does not come into
3377          * the picture unless it is null. also, slave->last_link_up is not
3378          * needed here because we send an arp on each slave and give a slave
3379          * as long as it needs to get the tx/rx within the delta.
3380          * TODO: what about up/down delay in arp mode? it wasn't here before
3381          *       so it can wait
3382          */
3383         bond_for_each_slave_rcu(bond, slave, iter) {
3384                 unsigned long last_tx = slave_last_tx(slave);
3385
3386                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3387
3388                 if (slave->link != BOND_LINK_UP) {
3389                         if (bond_time_in_interval(bond, last_tx, 1) &&
3390                             bond_time_in_interval(bond, slave->last_rx, 1)) {
3391
3392                                 bond_propose_link_state(slave, BOND_LINK_UP);
3393                                 slave_state_changed = 1;
3394
3395                                 /* primary_slave has no meaning in round-robin
3396                                  * mode. the window of a slave being up and
3397                                  * curr_active_slave being null after enslaving
3398                                  * is closed.
3399                                  */
3400                                 if (!oldcurrent) {
3401                                         slave_info(bond->dev, slave->dev, "link status definitely up\n");
3402                                         do_failover = 1;
3403                                 } else {
3404                                         slave_info(bond->dev, slave->dev, "interface is now up\n");
3405                                 }
3406                         }
3407                 } else {
3408                         /* slave->link == BOND_LINK_UP */
3409
3410                         /* not all switches will respond to an arp request
3411                          * when the source ip is 0, so don't take the link down
3412                          * if we don't know our ip yet
3413                          */
3414                         if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3415                             !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3416
3417                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
3418                                 slave_state_changed = 1;
3419
3420                                 if (slave->link_failure_count < UINT_MAX)
3421                                         slave->link_failure_count++;
3422
3423                                 slave_info(bond->dev, slave->dev, "interface is now down\n");
3424
3425                                 if (slave == oldcurrent)
3426                                         do_failover = 1;
3427                         }
3428                 }
3429
3430                 /* note: if switch is in round-robin mode, all links
3431                  * must tx arp to ensure all links rx an arp - otherwise
3432                  * links may oscillate or not come up at all; if switch is
3433                  * in something like xor mode, there is nothing we can
3434                  * do - all replies will be rx'ed on same link causing slaves
3435                  * to be unstable during low/no traffic periods
3436                  */
3437                 if (bond_slave_is_up(slave))
3438                         bond_send_validate(bond, slave);
3439         }
3440
3441         rcu_read_unlock();
3442
3443         if (do_failover || slave_state_changed) {
3444                 if (!rtnl_trylock())
3445                         goto re_arm;
3446
3447                 bond_for_each_slave(bond, slave, iter) {
3448                         if (slave->link_new_state != BOND_LINK_NOCHANGE)
3449                                 slave->link = slave->link_new_state;
3450                 }
3451
3452                 if (slave_state_changed) {
3453                         bond_slave_state_change(bond);
3454                         if (BOND_MODE(bond) == BOND_MODE_XOR)
3455                                 bond_update_slave_arr(bond, NULL);
3456                 }
3457                 if (do_failover) {
3458                         block_netpoll_tx();
3459                         bond_select_active_slave(bond);
3460                         unblock_netpoll_tx();
3461                 }
3462                 rtnl_unlock();
3463         }
3464
3465 re_arm:
3466         if (bond->params.arp_interval)
3467                 queue_delayed_work(bond->wq, &bond->arp_work,
3468                                    msecs_to_jiffies(bond->params.arp_interval));
3469 }
3470
3471 /* Called to inspect slaves for active-backup mode ARP monitor link state
3472  * changes.  Sets proposed link state in slaves to specify what action
3473  * should take place for the slave.  Returns 0 if no changes are found, >0
3474  * if changes to link states must be committed.
3475  *
3476  * Called with rcu_read_lock held.
3477  */
3478 static int bond_ab_arp_inspect(struct bonding *bond)
3479 {
3480         unsigned long last_tx, last_rx;
3481         struct list_head *iter;
3482         struct slave *slave;
3483         int commit = 0;
3484
3485         bond_for_each_slave_rcu(bond, slave, iter) {
3486                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3487                 last_rx = slave_last_rx(bond, slave);
3488
3489                 if (slave->link != BOND_LINK_UP) {
3490                         if (bond_time_in_interval(bond, last_rx, 1)) {
3491                                 bond_propose_link_state(slave, BOND_LINK_UP);
3492                                 commit++;
3493                         } else if (slave->link == BOND_LINK_BACK) {
3494                                 bond_propose_link_state(slave, BOND_LINK_FAIL);
3495                                 commit++;
3496                         }
3497                         continue;
3498                 }
3499
3500                 /* Give slaves 2*delta after being enslaved or made
3501                  * active.  This avoids bouncing, as the last receive
3502                  * times need a full ARP monitor cycle to be updated.
3503                  */
3504                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3505                         continue;
3506
3507                 /* Backup slave is down if:
3508                  * - No current_arp_slave AND
3509                  * - more than (missed_max+1)*delta since last receive AND
3510                  * - the bond has an IP address
3511                  *
3512                  * Note: a non-null current_arp_slave indicates
3513                  * the curr_active_slave went down and we are
3514                  * searching for a new one; under this condition
3515                  * we only take the curr_active_slave down - this
3516                  * gives each slave a chance to tx/rx traffic
3517                  * before being taken out
3518                  */
3519                 if (!bond_is_active_slave(slave) &&
3520                     !rcu_access_pointer(bond->current_arp_slave) &&
3521                     !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3522                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3523                         commit++;
3524                 }
3525
3526                 /* Active slave is down if:
3527                  * - more than missed_max*delta since transmitting OR
3528                  * - (more than missed_max*delta since receive AND
3529                  *    the bond has an IP address)
3530                  */
3531                 last_tx = slave_last_tx(slave);
3532                 if (bond_is_active_slave(slave) &&
3533                     (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3534                      !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3535                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3536                         commit++;
3537                 }
3538         }
3539
3540         return commit;
3541 }
3542
3543 /* Called to commit link state changes noted by inspection step of
3544  * active-backup mode ARP monitor.
3545  *
3546  * Called with RTNL hold.
3547  */
3548 static void bond_ab_arp_commit(struct bonding *bond)
3549 {
3550         bool do_failover = false;
3551         struct list_head *iter;
3552         unsigned long last_tx;
3553         struct slave *slave;
3554
3555         bond_for_each_slave(bond, slave, iter) {
3556                 switch (slave->link_new_state) {
3557                 case BOND_LINK_NOCHANGE:
3558                         continue;
3559
3560                 case BOND_LINK_UP:
3561                         last_tx = slave_last_tx(slave);
3562                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
3563                             (!rtnl_dereference(bond->curr_active_slave) &&
3564                              bond_time_in_interval(bond, last_tx, 1))) {
3565                                 struct slave *current_arp_slave;
3566
3567                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3568                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
3569                                                           BOND_SLAVE_NOTIFY_NOW);
3570                                 if (current_arp_slave) {
3571                                         bond_set_slave_inactive_flags(
3572                                                 current_arp_slave,
3573                                                 BOND_SLAVE_NOTIFY_NOW);
3574                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3575                                 }
3576
3577                                 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3578
3579                                 if (!rtnl_dereference(bond->curr_active_slave) ||
3580                                     slave == rtnl_dereference(bond->primary_slave) ||
3581                                     slave->prio > rtnl_dereference(bond->curr_active_slave)->prio)
3582                                         do_failover = true;
3583
3584                         }
3585
3586                         continue;
3587
3588                 case BOND_LINK_DOWN:
3589                         if (slave->link_failure_count < UINT_MAX)
3590                                 slave->link_failure_count++;
3591
3592                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3593                                                   BOND_SLAVE_NOTIFY_NOW);
3594                         bond_set_slave_inactive_flags(slave,
3595                                                       BOND_SLAVE_NOTIFY_NOW);
3596
3597                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3598
3599                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
3600                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3601                                 do_failover = true;
3602                         }
3603
3604                         continue;
3605
3606                 case BOND_LINK_FAIL:
3607                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3608                                                   BOND_SLAVE_NOTIFY_NOW);
3609                         bond_set_slave_inactive_flags(slave,
3610                                                       BOND_SLAVE_NOTIFY_NOW);
3611
3612                         /* A slave has just been enslaved and has become
3613                          * the current active slave.
3614                          */
3615                         if (rtnl_dereference(bond->curr_active_slave))
3616                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3617                         continue;
3618
3619                 default:
3620                         slave_err(bond->dev, slave->dev,
3621                                   "impossible: link_new_state %d on slave\n",
3622                                   slave->link_new_state);
3623                         continue;
3624                 }
3625         }
3626
3627         if (do_failover) {
3628                 block_netpoll_tx();
3629                 bond_select_active_slave(bond);
3630                 unblock_netpoll_tx();
3631         }
3632
3633         bond_set_carrier(bond);
3634 }
3635
3636 /* Send ARP probes for active-backup mode ARP monitor.
3637  *
3638  * Called with rcu_read_lock held.
3639  */
3640 static bool bond_ab_arp_probe(struct bonding *bond)
3641 {
3642         struct slave *slave, *before = NULL, *new_slave = NULL,
3643                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3644                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3645         struct list_head *iter;
3646         bool found = false;
3647         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3648
3649         if (curr_arp_slave && curr_active_slave)
3650                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3651                             curr_arp_slave->dev->name,
3652                             curr_active_slave->dev->name);
3653
3654         if (curr_active_slave) {
3655                 bond_send_validate(bond, curr_active_slave);
3656                 return should_notify_rtnl;
3657         }
3658
3659         /* if we don't have a curr_active_slave, search for the next available
3660          * backup slave from the current_arp_slave and make it the candidate
3661          * for becoming the curr_active_slave
3662          */
3663
3664         if (!curr_arp_slave) {
3665                 curr_arp_slave = bond_first_slave_rcu(bond);
3666                 if (!curr_arp_slave)
3667                         return should_notify_rtnl;
3668         }
3669
3670         bond_for_each_slave_rcu(bond, slave, iter) {
3671                 if (!found && !before && bond_slave_is_up(slave))
3672                         before = slave;
3673
3674                 if (found && !new_slave && bond_slave_is_up(slave))
3675                         new_slave = slave;
3676                 /* if the link state is up at this point, we
3677                  * mark it down - this can happen if we have
3678                  * simultaneous link failures and
3679                  * reselect_active_interface doesn't make this
3680                  * one the current slave so it is still marked
3681                  * up when it is actually down
3682                  */
3683                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3684                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3685                                                   BOND_SLAVE_NOTIFY_LATER);
3686                         if (slave->link_failure_count < UINT_MAX)
3687                                 slave->link_failure_count++;
3688
3689                         bond_set_slave_inactive_flags(slave,
3690                                                       BOND_SLAVE_NOTIFY_LATER);
3691
3692                         slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3693                 }
3694                 if (slave == curr_arp_slave)
3695                         found = true;
3696         }
3697
3698         if (!new_slave && before)
3699                 new_slave = before;
3700
3701         if (!new_slave)
3702                 goto check_state;
3703
3704         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3705                                   BOND_SLAVE_NOTIFY_LATER);
3706         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3707         bond_send_validate(bond, new_slave);
3708         new_slave->last_link_up = jiffies;
3709         rcu_assign_pointer(bond->current_arp_slave, new_slave);
3710
3711 check_state:
3712         bond_for_each_slave_rcu(bond, slave, iter) {
3713                 if (slave->should_notify || slave->should_notify_link) {
3714                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3715                         break;
3716                 }
3717         }
3718         return should_notify_rtnl;
3719 }
3720
3721 static void bond_activebackup_arp_mon(struct bonding *bond)
3722 {
3723         bool should_notify_peers = false;
3724         bool should_notify_rtnl = false;
3725         int delta_in_ticks;
3726
3727         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3728
3729         if (!bond_has_slaves(bond))
3730                 goto re_arm;
3731
3732         rcu_read_lock();
3733
3734         should_notify_peers = bond_should_notify_peers(bond);
3735
3736         if (bond_ab_arp_inspect(bond)) {
3737                 rcu_read_unlock();
3738
3739                 /* Race avoidance with bond_close flush of workqueue */
3740                 if (!rtnl_trylock()) {
3741                         delta_in_ticks = 1;
3742                         should_notify_peers = false;
3743                         goto re_arm;
3744                 }
3745
3746                 bond_ab_arp_commit(bond);
3747
3748                 rtnl_unlock();
3749                 rcu_read_lock();
3750         }
3751
3752         should_notify_rtnl = bond_ab_arp_probe(bond);
3753         rcu_read_unlock();
3754
3755 re_arm:
3756         if (bond->params.arp_interval)
3757                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3758
3759         if (should_notify_peers || should_notify_rtnl) {
3760                 if (!rtnl_trylock())
3761                         return;
3762
3763                 if (should_notify_peers) {
3764                         bond->send_peer_notif--;
3765                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3766                                                  bond->dev);
3767                 }
3768                 if (should_notify_rtnl) {
3769                         bond_slave_state_notify(bond);
3770                         bond_slave_link_notify(bond);
3771                 }
3772
3773                 rtnl_unlock();
3774         }
3775 }
3776
3777 static void bond_arp_monitor(struct work_struct *work)
3778 {
3779         struct bonding *bond = container_of(work, struct bonding,
3780                                             arp_work.work);
3781
3782         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3783                 bond_activebackup_arp_mon(bond);
3784         else
3785                 bond_loadbalance_arp_mon(bond);
3786 }
3787
3788 /*-------------------------- netdev event handling --------------------------*/
3789
3790 /* Change device name */
3791 static int bond_event_changename(struct bonding *bond)
3792 {
3793         bond_remove_proc_entry(bond);
3794         bond_create_proc_entry(bond);
3795
3796         bond_debug_reregister(bond);
3797
3798         return NOTIFY_DONE;
3799 }
3800
3801 static int bond_master_netdev_event(unsigned long event,
3802                                     struct net_device *bond_dev)
3803 {
3804         struct bonding *event_bond = netdev_priv(bond_dev);
3805
3806         netdev_dbg(bond_dev, "%s called\n", __func__);
3807
3808         switch (event) {
3809         case NETDEV_CHANGENAME:
3810                 return bond_event_changename(event_bond);
3811         case NETDEV_UNREGISTER:
3812                 bond_remove_proc_entry(event_bond);
3813 #ifdef CONFIG_XFRM_OFFLOAD
3814                 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3815 #endif /* CONFIG_XFRM_OFFLOAD */
3816                 break;
3817         case NETDEV_REGISTER:
3818                 bond_create_proc_entry(event_bond);
3819                 break;
3820         default:
3821                 break;
3822         }
3823
3824         return NOTIFY_DONE;
3825 }
3826
3827 static int bond_slave_netdev_event(unsigned long event,
3828                                    struct net_device *slave_dev)
3829 {
3830         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3831         struct bonding *bond;
3832         struct net_device *bond_dev;
3833
3834         /* A netdev event can be generated while enslaving a device
3835          * before netdev_rx_handler_register is called in which case
3836          * slave will be NULL
3837          */
3838         if (!slave) {
3839                 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3840                 return NOTIFY_DONE;
3841         }
3842
3843         bond_dev = slave->bond->dev;
3844         bond = slave->bond;
3845         primary = rtnl_dereference(bond->primary_slave);
3846
3847         slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3848
3849         switch (event) {
3850         case NETDEV_UNREGISTER:
3851                 if (bond_dev->type != ARPHRD_ETHER)
3852                         bond_release_and_destroy(bond_dev, slave_dev);
3853                 else
3854                         __bond_release_one(bond_dev, slave_dev, false, true);
3855                 break;
3856         case NETDEV_UP:
3857         case NETDEV_CHANGE:
3858                 /* For 802.3ad mode only:
3859                  * Getting invalid Speed/Duplex values here will put slave
3860                  * in weird state. Mark it as link-fail if the link was
3861                  * previously up or link-down if it hasn't yet come up, and
3862                  * let link-monitoring (miimon) set it right when correct
3863                  * speeds/duplex are available.
3864                  */
3865                 if (bond_update_speed_duplex(slave) &&
3866                     BOND_MODE(bond) == BOND_MODE_8023AD) {
3867                         if (slave->last_link_up)
3868                                 slave->link = BOND_LINK_FAIL;
3869                         else
3870                                 slave->link = BOND_LINK_DOWN;
3871                 }
3872
3873                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3874                         bond_3ad_adapter_speed_duplex_changed(slave);
3875                 fallthrough;
3876         case NETDEV_DOWN:
3877                 /* Refresh slave-array if applicable!
3878                  * If the setup does not use miimon or arpmon (mode-specific!),
3879                  * then these events will not cause the slave-array to be
3880                  * refreshed. This will cause xmit to use a slave that is not
3881                  * usable. Avoid such situation by refeshing the array at these
3882                  * events. If these (miimon/arpmon) parameters are configured
3883                  * then array gets refreshed twice and that should be fine!
3884                  */
3885                 if (bond_mode_can_use_xmit_hash(bond))
3886                         bond_update_slave_arr(bond, NULL);
3887                 break;
3888         case NETDEV_CHANGEMTU:
3889                 /* TODO: Should slaves be allowed to
3890                  * independently alter their MTU?  For
3891                  * an active-backup bond, slaves need
3892                  * not be the same type of device, so
3893                  * MTUs may vary.  For other modes,
3894                  * slaves arguably should have the
3895                  * same MTUs. To do this, we'd need to
3896                  * take over the slave's change_mtu
3897                  * function for the duration of their
3898                  * servitude.
3899                  */
3900                 break;
3901         case NETDEV_CHANGENAME:
3902                 /* we don't care if we don't have primary set */
3903                 if (!bond_uses_primary(bond) ||
3904                     !bond->params.primary[0])
3905                         break;
3906
3907                 if (slave == primary) {
3908                         /* slave's name changed - he's no longer primary */
3909                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3910                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3911                         /* we have a new primary slave */
3912                         rcu_assign_pointer(bond->primary_slave, slave);
3913                 } else { /* we didn't change primary - exit */
3914                         break;
3915                 }
3916
3917                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3918                             primary ? slave_dev->name : "none");
3919
3920                 block_netpoll_tx();
3921                 bond_select_active_slave(bond);
3922                 unblock_netpoll_tx();
3923                 break;
3924         case NETDEV_FEAT_CHANGE:
3925                 bond_compute_features(bond);
3926                 break;
3927         case NETDEV_RESEND_IGMP:
3928                 /* Propagate to master device */
3929                 call_netdevice_notifiers(event, slave->bond->dev);
3930                 break;
3931         default:
3932                 break;
3933         }
3934
3935         return NOTIFY_DONE;
3936 }
3937
3938 /* bond_netdev_event: handle netdev notifier chain events.
3939  *
3940  * This function receives events for the netdev chain.  The caller (an
3941  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3942  * locks for us to safely manipulate the slave devices (RTNL lock,
3943  * dev_probe_lock).
3944  */
3945 static int bond_netdev_event(struct notifier_block *this,
3946                              unsigned long event, void *ptr)
3947 {
3948         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3949
3950         netdev_dbg(event_dev, "%s received %s\n",
3951                    __func__, netdev_cmd_to_name(event));
3952
3953         if (!(event_dev->priv_flags & IFF_BONDING))
3954                 return NOTIFY_DONE;
3955
3956         if (event_dev->flags & IFF_MASTER) {
3957                 int ret;
3958
3959                 ret = bond_master_netdev_event(event, event_dev);
3960                 if (ret != NOTIFY_DONE)
3961                         return ret;
3962         }
3963
3964         if (event_dev->flags & IFF_SLAVE)
3965                 return bond_slave_netdev_event(event, event_dev);
3966
3967         return NOTIFY_DONE;
3968 }
3969
3970 static struct notifier_block bond_netdev_notifier = {
3971         .notifier_call = bond_netdev_event,
3972 };
3973
3974 /*---------------------------- Hashing Policies -----------------------------*/
3975
3976 /* Helper to access data in a packet, with or without a backing skb.
3977  * If skb is given the data is linearized if necessary via pskb_may_pull.
3978  */
3979 static inline const void *bond_pull_data(struct sk_buff *skb,
3980                                          const void *data, int hlen, int n)
3981 {
3982         if (likely(n <= hlen))
3983                 return data;
3984         else if (skb && likely(pskb_may_pull(skb, n)))
3985                 return skb->head;
3986
3987         return NULL;
3988 }
3989
3990 /* L2 hash helper */
3991 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
3992 {
3993         struct ethhdr *ep;
3994
3995         data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
3996         if (!data)
3997                 return 0;
3998
3999         ep = (struct ethhdr *)(data + mhoff);
4000         return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
4001 }
4002
4003 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
4004                          int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
4005 {
4006         const struct ipv6hdr *iph6;
4007         const struct iphdr *iph;
4008
4009         if (l2_proto == htons(ETH_P_IP)) {
4010                 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
4011                 if (!data)
4012                         return false;
4013
4014                 iph = (const struct iphdr *)(data + *nhoff);
4015                 iph_to_flow_copy_v4addrs(fk, iph);
4016                 *nhoff += iph->ihl << 2;
4017                 if (!ip_is_fragment(iph))
4018                         *ip_proto = iph->protocol;
4019         } else if (l2_proto == htons(ETH_P_IPV6)) {
4020                 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
4021                 if (!data)
4022                         return false;
4023
4024                 iph6 = (const struct ipv6hdr *)(data + *nhoff);
4025                 iph_to_flow_copy_v6addrs(fk, iph6);
4026                 *nhoff += sizeof(*iph6);
4027                 *ip_proto = iph6->nexthdr;
4028         } else {
4029                 return false;
4030         }
4031
4032         if (l34 && *ip_proto >= 0)
4033                 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
4034
4035         return true;
4036 }
4037
4038 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4039 {
4040         u32 srcmac_vendor = 0, srcmac_dev = 0;
4041         struct ethhdr *mac_hdr;
4042         u16 vlan = 0;
4043         int i;
4044
4045         data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4046         if (!data)
4047                 return 0;
4048         mac_hdr = (struct ethhdr *)(data + mhoff);
4049
4050         for (i = 0; i < 3; i++)
4051                 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4052
4053         for (i = 3; i < ETH_ALEN; i++)
4054                 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4055
4056         if (skb && skb_vlan_tag_present(skb))
4057                 vlan = skb_vlan_tag_get(skb);
4058
4059         return vlan ^ srcmac_vendor ^ srcmac_dev;
4060 }
4061
4062 /* Extract the appropriate headers based on bond's xmit policy */
4063 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4064                               __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4065 {
4066         bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4067         int ip_proto = -1;
4068
4069         switch (bond->params.xmit_policy) {
4070         case BOND_XMIT_POLICY_ENCAP23:
4071         case BOND_XMIT_POLICY_ENCAP34:
4072                 memset(fk, 0, sizeof(*fk));
4073                 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
4074                                           fk, data, l2_proto, nhoff, hlen, 0);
4075         default:
4076                 break;
4077         }
4078
4079         fk->ports.ports = 0;
4080         memset(&fk->icmp, 0, sizeof(fk->icmp));
4081         if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4082                 return false;
4083
4084         /* ICMP error packets contains at least 8 bytes of the header
4085          * of the packet which generated the error. Use this information
4086          * to correlate ICMP error packets within the same flow which
4087          * generated the error.
4088          */
4089         if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4090                 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4091                 if (ip_proto == IPPROTO_ICMP) {
4092                         if (!icmp_is_err(fk->icmp.type))
4093                                 return true;
4094
4095                         nhoff += sizeof(struct icmphdr);
4096                 } else if (ip_proto == IPPROTO_ICMPV6) {
4097                         if (!icmpv6_is_err(fk->icmp.type))
4098                                 return true;
4099
4100                         nhoff += sizeof(struct icmp6hdr);
4101                 }
4102                 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4103         }
4104
4105         return true;
4106 }
4107
4108 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4109 {
4110         hash ^= (__force u32)flow_get_u32_dst(flow) ^
4111                 (__force u32)flow_get_u32_src(flow);
4112         hash ^= (hash >> 16);
4113         hash ^= (hash >> 8);
4114
4115         /* discard lowest hash bit to deal with the common even ports pattern */
4116         if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4117                 xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4118                 return hash >> 1;
4119
4120         return hash;
4121 }
4122
4123 /* Generate hash based on xmit policy. If @skb is given it is used to linearize
4124  * the data as required, but this function can be used without it if the data is
4125  * known to be linear (e.g. with xdp_buff).
4126  */
4127 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4128                             __be16 l2_proto, int mhoff, int nhoff, int hlen)
4129 {
4130         struct flow_keys flow;
4131         u32 hash;
4132
4133         if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4134                 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4135
4136         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4137             !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4138                 return bond_eth_hash(skb, data, mhoff, hlen);
4139
4140         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4141             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4142                 hash = bond_eth_hash(skb, data, mhoff, hlen);
4143         } else {
4144                 if (flow.icmp.id)
4145                         memcpy(&hash, &flow.icmp, sizeof(hash));
4146                 else
4147                         memcpy(&hash, &flow.ports.ports, sizeof(hash));
4148         }
4149
4150         return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4151 }
4152
4153 /**
4154  * bond_xmit_hash - generate a hash value based on the xmit policy
4155  * @bond: bonding device
4156  * @skb: buffer to use for headers
4157  *
4158  * This function will extract the necessary headers from the skb buffer and use
4159  * them to generate a hash based on the xmit_policy set in the bonding device
4160  */
4161 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4162 {
4163         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4164             skb->l4_hash)
4165                 return skb->hash;
4166
4167         return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4168                                 skb_mac_offset(skb), skb_network_offset(skb),
4169                                 skb_headlen(skb));
4170 }
4171
4172 /**
4173  * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4174  * @bond: bonding device
4175  * @xdp: buffer to use for headers
4176  *
4177  * The XDP variant of bond_xmit_hash.
4178  */
4179 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4180 {
4181         struct ethhdr *eth;
4182
4183         if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4184                 return 0;
4185
4186         eth = (struct ethhdr *)xdp->data;
4187
4188         return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4189                                 sizeof(struct ethhdr), xdp->data_end - xdp->data);
4190 }
4191
4192 /*-------------------------- Device entry points ----------------------------*/
4193
4194 void bond_work_init_all(struct bonding *bond)
4195 {
4196         INIT_DELAYED_WORK(&bond->mcast_work,
4197                           bond_resend_igmp_join_requests_delayed);
4198         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4199         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4200         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4201         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4202         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4203 }
4204
4205 static void bond_work_cancel_all(struct bonding *bond)
4206 {
4207         cancel_delayed_work_sync(&bond->mii_work);
4208         cancel_delayed_work_sync(&bond->arp_work);
4209         cancel_delayed_work_sync(&bond->alb_work);
4210         cancel_delayed_work_sync(&bond->ad_work);
4211         cancel_delayed_work_sync(&bond->mcast_work);
4212         cancel_delayed_work_sync(&bond->slave_arr_work);
4213 }
4214
4215 static int bond_open(struct net_device *bond_dev)
4216 {
4217         struct bonding *bond = netdev_priv(bond_dev);
4218         struct list_head *iter;
4219         struct slave *slave;
4220
4221         if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) {
4222                 bond->rr_tx_counter = alloc_percpu(u32);
4223                 if (!bond->rr_tx_counter)
4224                         return -ENOMEM;
4225         }
4226
4227         /* reset slave->backup and slave->inactive */
4228         if (bond_has_slaves(bond)) {
4229                 bond_for_each_slave(bond, slave, iter) {
4230                         if (bond_uses_primary(bond) &&
4231                             slave != rcu_access_pointer(bond->curr_active_slave)) {
4232                                 bond_set_slave_inactive_flags(slave,
4233                                                               BOND_SLAVE_NOTIFY_NOW);
4234                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4235                                 bond_set_slave_active_flags(slave,
4236                                                             BOND_SLAVE_NOTIFY_NOW);
4237                         }
4238                 }
4239         }
4240
4241         if (bond_is_lb(bond)) {
4242                 /* bond_alb_initialize must be called before the timer
4243                  * is started.
4244                  */
4245                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4246                         return -ENOMEM;
4247                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4248                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
4249         }
4250
4251         if (bond->params.miimon)  /* link check interval, in milliseconds. */
4252                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
4253
4254         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
4255                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
4256                 bond->recv_probe = bond_rcv_validate;
4257         }
4258
4259         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4260                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
4261                 /* register to receive LACPDUs */
4262                 bond->recv_probe = bond_3ad_lacpdu_recv;
4263                 bond_3ad_initiate_agg_selection(bond, 1);
4264
4265                 bond_for_each_slave(bond, slave, iter)
4266                         dev_mc_add(slave->dev, lacpdu_mcast_addr);
4267         }
4268
4269         if (bond_mode_can_use_xmit_hash(bond))
4270                 bond_update_slave_arr(bond, NULL);
4271
4272         return 0;
4273 }
4274
4275 static int bond_close(struct net_device *bond_dev)
4276 {
4277         struct bonding *bond = netdev_priv(bond_dev);
4278         struct slave *slave;
4279
4280         bond_work_cancel_all(bond);
4281         bond->send_peer_notif = 0;
4282         if (bond_is_lb(bond))
4283                 bond_alb_deinitialize(bond);
4284         bond->recv_probe = NULL;
4285
4286         if (bond_uses_primary(bond)) {
4287                 rcu_read_lock();
4288                 slave = rcu_dereference(bond->curr_active_slave);
4289                 if (slave)
4290                         bond_hw_addr_flush(bond_dev, slave->dev);
4291                 rcu_read_unlock();
4292         } else {
4293                 struct list_head *iter;
4294
4295                 bond_for_each_slave(bond, slave, iter)
4296                         bond_hw_addr_flush(bond_dev, slave->dev);
4297         }
4298
4299         return 0;
4300 }
4301
4302 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4303  * that some drivers can provide 32bit values only.
4304  */
4305 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4306                             const struct rtnl_link_stats64 *_new,
4307                             const struct rtnl_link_stats64 *_old)
4308 {
4309         const u64 *new = (const u64 *)_new;
4310         const u64 *old = (const u64 *)_old;
4311         u64 *res = (u64 *)_res;
4312         int i;
4313
4314         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4315                 u64 nv = new[i];
4316                 u64 ov = old[i];
4317                 s64 delta = nv - ov;
4318
4319                 /* detects if this particular field is 32bit only */
4320                 if (((nv | ov) >> 32) == 0)
4321                         delta = (s64)(s32)((u32)nv - (u32)ov);
4322
4323                 /* filter anomalies, some drivers reset their stats
4324                  * at down/up events.
4325                  */
4326                 if (delta > 0)
4327                         res[i] += delta;
4328         }
4329 }
4330
4331 #ifdef CONFIG_LOCKDEP
4332 static int bond_get_lowest_level_rcu(struct net_device *dev)
4333 {
4334         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4335         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4336         int cur = 0, max = 0;
4337
4338         now = dev;
4339         iter = &dev->adj_list.lower;
4340
4341         while (1) {
4342                 next = NULL;
4343                 while (1) {
4344                         ldev = netdev_next_lower_dev_rcu(now, &iter);
4345                         if (!ldev)
4346                                 break;
4347
4348                         next = ldev;
4349                         niter = &ldev->adj_list.lower;
4350                         dev_stack[cur] = now;
4351                         iter_stack[cur++] = iter;
4352                         if (max <= cur)
4353                                 max = cur;
4354                         break;
4355                 }
4356
4357                 if (!next) {
4358                         if (!cur)
4359                                 return max;
4360                         next = dev_stack[--cur];
4361                         niter = iter_stack[cur];
4362                 }
4363
4364                 now = next;
4365                 iter = niter;
4366         }
4367
4368         return max;
4369 }
4370 #endif
4371
4372 static void bond_get_stats(struct net_device *bond_dev,
4373                            struct rtnl_link_stats64 *stats)
4374 {
4375         struct bonding *bond = netdev_priv(bond_dev);
4376         struct rtnl_link_stats64 temp;
4377         struct list_head *iter;
4378         struct slave *slave;
4379         int nest_level = 0;
4380
4381
4382         rcu_read_lock();
4383 #ifdef CONFIG_LOCKDEP
4384         nest_level = bond_get_lowest_level_rcu(bond_dev);
4385 #endif
4386
4387         spin_lock_nested(&bond->stats_lock, nest_level);
4388         memcpy(stats, &bond->bond_stats, sizeof(*stats));
4389
4390         bond_for_each_slave_rcu(bond, slave, iter) {
4391                 const struct rtnl_link_stats64 *new =
4392                         dev_get_stats(slave->dev, &temp);
4393
4394                 bond_fold_stats(stats, new, &slave->slave_stats);
4395
4396                 /* save off the slave stats for the next run */
4397                 memcpy(&slave->slave_stats, new, sizeof(*new));
4398         }
4399
4400         memcpy(&bond->bond_stats, stats, sizeof(*stats));
4401         spin_unlock(&bond->stats_lock);
4402         rcu_read_unlock();
4403 }
4404
4405 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4406 {
4407         struct bonding *bond = netdev_priv(bond_dev);
4408         struct mii_ioctl_data *mii = NULL;
4409         const struct net_device_ops *ops;
4410         struct net_device *real_dev;
4411         struct hwtstamp_config cfg;
4412         struct ifreq ifrr;
4413         int res = 0;
4414
4415         netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4416
4417         switch (cmd) {
4418         case SIOCGMIIPHY:
4419                 mii = if_mii(ifr);
4420                 if (!mii)
4421                         return -EINVAL;
4422
4423                 mii->phy_id = 0;
4424                 fallthrough;
4425         case SIOCGMIIREG:
4426                 /* We do this again just in case we were called by SIOCGMIIREG
4427                  * instead of SIOCGMIIPHY.
4428                  */
4429                 mii = if_mii(ifr);
4430                 if (!mii)
4431                         return -EINVAL;
4432
4433                 if (mii->reg_num == 1) {
4434                         mii->val_out = 0;
4435                         if (netif_carrier_ok(bond->dev))
4436                                 mii->val_out = BMSR_LSTATUS;
4437                 }
4438
4439                 break;
4440         case SIOCSHWTSTAMP:
4441                 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4442                         return -EFAULT;
4443
4444                 if (!(cfg.flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
4445                         return -EOPNOTSUPP;
4446
4447                 fallthrough;
4448         case SIOCGHWTSTAMP:
4449                 real_dev = bond_option_active_slave_get_rcu(bond);
4450                 if (!real_dev)
4451                         return -EOPNOTSUPP;
4452
4453                 strscpy_pad(ifrr.ifr_name, real_dev->name, IFNAMSIZ);
4454                 ifrr.ifr_ifru = ifr->ifr_ifru;
4455
4456                 ops = real_dev->netdev_ops;
4457                 if (netif_device_present(real_dev) && ops->ndo_eth_ioctl) {
4458                         res = ops->ndo_eth_ioctl(real_dev, &ifrr, cmd);
4459                         if (res)
4460                                 return res;
4461
4462                         ifr->ifr_ifru = ifrr.ifr_ifru;
4463                         if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4464                                 return -EFAULT;
4465
4466                         /* Set the BOND_PHC_INDEX flag to notify user space */
4467                         cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
4468
4469                         return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ?
4470                                 -EFAULT : 0;
4471                 }
4472                 fallthrough;
4473         default:
4474                 res = -EOPNOTSUPP;
4475         }
4476
4477         return res;
4478 }
4479
4480 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4481 {
4482         struct bonding *bond = netdev_priv(bond_dev);
4483         struct net_device *slave_dev = NULL;
4484         struct ifbond k_binfo;
4485         struct ifbond __user *u_binfo = NULL;
4486         struct ifslave k_sinfo;
4487         struct ifslave __user *u_sinfo = NULL;
4488         struct bond_opt_value newval;
4489         struct net *net;
4490         int res = 0;
4491
4492         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4493
4494         switch (cmd) {
4495         case SIOCBONDINFOQUERY:
4496                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4497
4498                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4499                         return -EFAULT;
4500
4501                 bond_info_query(bond_dev, &k_binfo);
4502                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4503                         return -EFAULT;
4504
4505                 return 0;
4506         case SIOCBONDSLAVEINFOQUERY:
4507                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4508
4509                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4510                         return -EFAULT;
4511
4512                 res = bond_slave_info_query(bond_dev, &k_sinfo);
4513                 if (res == 0 &&
4514                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4515                         return -EFAULT;
4516
4517                 return res;
4518         default:
4519                 break;
4520         }
4521
4522         net = dev_net(bond_dev);
4523
4524         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4525                 return -EPERM;
4526
4527         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4528
4529         slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4530
4531         if (!slave_dev)
4532                 return -ENODEV;
4533
4534         switch (cmd) {
4535         case SIOCBONDENSLAVE:
4536                 res = bond_enslave(bond_dev, slave_dev, NULL);
4537                 break;
4538         case SIOCBONDRELEASE:
4539                 res = bond_release(bond_dev, slave_dev);
4540                 break;
4541         case SIOCBONDSETHWADDR:
4542                 res = bond_set_dev_addr(bond_dev, slave_dev);
4543                 break;
4544         case SIOCBONDCHANGEACTIVE:
4545                 bond_opt_initstr(&newval, slave_dev->name);
4546                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4547                                             &newval);
4548                 break;
4549         default:
4550                 res = -EOPNOTSUPP;
4551         }
4552
4553         return res;
4554 }
4555
4556 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4557                                void __user *data, int cmd)
4558 {
4559         struct ifreq ifrdata = { .ifr_data = data };
4560
4561         switch (cmd) {
4562         case BOND_INFO_QUERY_OLD:
4563                 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4564         case BOND_SLAVE_INFO_QUERY_OLD:
4565                 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4566         case BOND_ENSLAVE_OLD:
4567                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4568         case BOND_RELEASE_OLD:
4569                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4570         case BOND_SETHWADDR_OLD:
4571                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4572         case BOND_CHANGE_ACTIVE_OLD:
4573                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4574         }
4575
4576         return -EOPNOTSUPP;
4577 }
4578
4579 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4580 {
4581         struct bonding *bond = netdev_priv(bond_dev);
4582
4583         if (change & IFF_PROMISC)
4584                 bond_set_promiscuity(bond,
4585                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
4586
4587         if (change & IFF_ALLMULTI)
4588                 bond_set_allmulti(bond,
4589                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4590 }
4591
4592 static void bond_set_rx_mode(struct net_device *bond_dev)
4593 {
4594         struct bonding *bond = netdev_priv(bond_dev);
4595         struct list_head *iter;
4596         struct slave *slave;
4597
4598         rcu_read_lock();
4599         if (bond_uses_primary(bond)) {
4600                 slave = rcu_dereference(bond->curr_active_slave);
4601                 if (slave) {
4602                         dev_uc_sync(slave->dev, bond_dev);
4603                         dev_mc_sync(slave->dev, bond_dev);
4604                 }
4605         } else {
4606                 bond_for_each_slave_rcu(bond, slave, iter) {
4607                         dev_uc_sync_multiple(slave->dev, bond_dev);
4608                         dev_mc_sync_multiple(slave->dev, bond_dev);
4609                 }
4610         }
4611         rcu_read_unlock();
4612 }
4613
4614 static int bond_neigh_init(struct neighbour *n)
4615 {
4616         struct bonding *bond = netdev_priv(n->dev);
4617         const struct net_device_ops *slave_ops;
4618         struct neigh_parms parms;
4619         struct slave *slave;
4620         int ret = 0;
4621
4622         rcu_read_lock();
4623         slave = bond_first_slave_rcu(bond);
4624         if (!slave)
4625                 goto out;
4626         slave_ops = slave->dev->netdev_ops;
4627         if (!slave_ops->ndo_neigh_setup)
4628                 goto out;
4629
4630         /* TODO: find another way [1] to implement this.
4631          * Passing a zeroed structure is fragile,
4632          * but at least we do not pass garbage.
4633          *
4634          * [1] One way would be that ndo_neigh_setup() never touch
4635          *     struct neigh_parms, but propagate the new neigh_setup()
4636          *     back to ___neigh_create() / neigh_parms_alloc()
4637          */
4638         memset(&parms, 0, sizeof(parms));
4639         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4640
4641         if (ret)
4642                 goto out;
4643
4644         if (parms.neigh_setup)
4645                 ret = parms.neigh_setup(n);
4646 out:
4647         rcu_read_unlock();
4648         return ret;
4649 }
4650
4651 /* The bonding ndo_neigh_setup is called at init time beofre any
4652  * slave exists. So we must declare proxy setup function which will
4653  * be used at run time to resolve the actual slave neigh param setup.
4654  *
4655  * It's also called by master devices (such as vlans) to setup their
4656  * underlying devices. In that case - do nothing, we're already set up from
4657  * our init.
4658  */
4659 static int bond_neigh_setup(struct net_device *dev,
4660                             struct neigh_parms *parms)
4661 {
4662         /* modify only our neigh_parms */
4663         if (parms->dev == dev)
4664                 parms->neigh_setup = bond_neigh_init;
4665
4666         return 0;
4667 }
4668
4669 /* Change the MTU of all of a master's slaves to match the master */
4670 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4671 {
4672         struct bonding *bond = netdev_priv(bond_dev);
4673         struct slave *slave, *rollback_slave;
4674         struct list_head *iter;
4675         int res = 0;
4676
4677         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4678
4679         bond_for_each_slave(bond, slave, iter) {
4680                 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4681                            slave, slave->dev->netdev_ops->ndo_change_mtu);
4682
4683                 res = dev_set_mtu(slave->dev, new_mtu);
4684
4685                 if (res) {
4686                         /* If we failed to set the slave's mtu to the new value
4687                          * we must abort the operation even in ACTIVE_BACKUP
4688                          * mode, because if we allow the backup slaves to have
4689                          * different mtu values than the active slave we'll
4690                          * need to change their mtu when doing a failover. That
4691                          * means changing their mtu from timer context, which
4692                          * is probably not a good idea.
4693                          */
4694                         slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4695                                   res, new_mtu);
4696                         goto unwind;
4697                 }
4698         }
4699
4700         bond_dev->mtu = new_mtu;
4701
4702         return 0;
4703
4704 unwind:
4705         /* unwind from head to the slave that failed */
4706         bond_for_each_slave(bond, rollback_slave, iter) {
4707                 int tmp_res;
4708
4709                 if (rollback_slave == slave)
4710                         break;
4711
4712                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4713                 if (tmp_res)
4714                         slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4715                                   tmp_res);
4716         }
4717
4718         return res;
4719 }
4720
4721 /* Change HW address
4722  *
4723  * Note that many devices must be down to change the HW address, and
4724  * downing the master releases all slaves.  We can make bonds full of
4725  * bonding devices to test this, however.
4726  */
4727 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4728 {
4729         struct bonding *bond = netdev_priv(bond_dev);
4730         struct slave *slave, *rollback_slave;
4731         struct sockaddr_storage *ss = addr, tmp_ss;
4732         struct list_head *iter;
4733         int res = 0;
4734
4735         if (BOND_MODE(bond) == BOND_MODE_ALB)
4736                 return bond_alb_set_mac_address(bond_dev, addr);
4737
4738
4739         netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4740
4741         /* If fail_over_mac is enabled, do nothing and return success.
4742          * Returning an error causes ifenslave to fail.
4743          */
4744         if (bond->params.fail_over_mac &&
4745             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4746                 return 0;
4747
4748         if (!is_valid_ether_addr(ss->__data))
4749                 return -EADDRNOTAVAIL;
4750
4751         bond_for_each_slave(bond, slave, iter) {
4752                 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4753                           __func__, slave);
4754                 res = dev_set_mac_address(slave->dev, addr, NULL);
4755                 if (res) {
4756                         /* TODO: consider downing the slave
4757                          * and retry ?
4758                          * User should expect communications
4759                          * breakage anyway until ARP finish
4760                          * updating, so...
4761                          */
4762                         slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4763                                   __func__, res);
4764                         goto unwind;
4765                 }
4766         }
4767
4768         /* success */
4769         dev_addr_set(bond_dev, ss->__data);
4770         return 0;
4771
4772 unwind:
4773         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4774         tmp_ss.ss_family = bond_dev->type;
4775
4776         /* unwind from head to the slave that failed */
4777         bond_for_each_slave(bond, rollback_slave, iter) {
4778                 int tmp_res;
4779
4780                 if (rollback_slave == slave)
4781                         break;
4782
4783                 tmp_res = dev_set_mac_address(rollback_slave->dev,
4784                                               (struct sockaddr *)&tmp_ss, NULL);
4785                 if (tmp_res) {
4786                         slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4787                                    __func__, tmp_res);
4788                 }
4789         }
4790
4791         return res;
4792 }
4793
4794 /**
4795  * bond_get_slave_by_id - get xmit slave with slave_id
4796  * @bond: bonding device that is transmitting
4797  * @slave_id: slave id up to slave_cnt-1 through which to transmit
4798  *
4799  * This function tries to get slave with slave_id but in case
4800  * it fails, it tries to find the first available slave for transmission.
4801  */
4802 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4803                                           int slave_id)
4804 {
4805         struct list_head *iter;
4806         struct slave *slave;
4807         int i = slave_id;
4808
4809         /* Here we start from the slave with slave_id */
4810         bond_for_each_slave_rcu(bond, slave, iter) {
4811                 if (--i < 0) {
4812                         if (bond_slave_can_tx(slave))
4813                                 return slave;
4814                 }
4815         }
4816
4817         /* Here we start from the first slave up to slave_id */
4818         i = slave_id;
4819         bond_for_each_slave_rcu(bond, slave, iter) {
4820                 if (--i < 0)
4821                         break;
4822                 if (bond_slave_can_tx(slave))
4823                         return slave;
4824         }
4825         /* no slave that can tx has been found */
4826         return NULL;
4827 }
4828
4829 /**
4830  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4831  * @bond: bonding device to use
4832  *
4833  * Based on the value of the bonding device's packets_per_slave parameter
4834  * this function generates a slave id, which is usually used as the next
4835  * slave to transmit through.
4836  */
4837 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4838 {
4839         u32 slave_id;
4840         struct reciprocal_value reciprocal_packets_per_slave;
4841         int packets_per_slave = bond->params.packets_per_slave;
4842
4843         switch (packets_per_slave) {
4844         case 0:
4845                 slave_id = get_random_u32();
4846                 break;
4847         case 1:
4848                 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4849                 break;
4850         default:
4851                 reciprocal_packets_per_slave =
4852                         bond->params.reciprocal_packets_per_slave;
4853                 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4854                 slave_id = reciprocal_divide(slave_id,
4855                                              reciprocal_packets_per_slave);
4856                 break;
4857         }
4858
4859         return slave_id;
4860 }
4861
4862 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4863                                                     struct sk_buff *skb)
4864 {
4865         struct slave *slave;
4866         int slave_cnt;
4867         u32 slave_id;
4868
4869         /* Start with the curr_active_slave that joined the bond as the
4870          * default for sending IGMP traffic.  For failover purposes one
4871          * needs to maintain some consistency for the interface that will
4872          * send the join/membership reports.  The curr_active_slave found
4873          * will send all of this type of traffic.
4874          */
4875         if (skb->protocol == htons(ETH_P_IP)) {
4876                 int noff = skb_network_offset(skb);
4877                 struct iphdr *iph;
4878
4879                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4880                         goto non_igmp;
4881
4882                 iph = ip_hdr(skb);
4883                 if (iph->protocol == IPPROTO_IGMP) {
4884                         slave = rcu_dereference(bond->curr_active_slave);
4885                         if (slave)
4886                                 return slave;
4887                         return bond_get_slave_by_id(bond, 0);
4888                 }
4889         }
4890
4891 non_igmp:
4892         slave_cnt = READ_ONCE(bond->slave_cnt);
4893         if (likely(slave_cnt)) {
4894                 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4895                 return bond_get_slave_by_id(bond, slave_id);
4896         }
4897         return NULL;
4898 }
4899
4900 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4901                                                         struct xdp_buff *xdp)
4902 {
4903         struct slave *slave;
4904         int slave_cnt;
4905         u32 slave_id;
4906         const struct ethhdr *eth;
4907         void *data = xdp->data;
4908
4909         if (data + sizeof(struct ethhdr) > xdp->data_end)
4910                 goto non_igmp;
4911
4912         eth = (struct ethhdr *)data;
4913         data += sizeof(struct ethhdr);
4914
4915         /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4916         if (eth->h_proto == htons(ETH_P_IP)) {
4917                 const struct iphdr *iph;
4918
4919                 if (data + sizeof(struct iphdr) > xdp->data_end)
4920                         goto non_igmp;
4921
4922                 iph = (struct iphdr *)data;
4923
4924                 if (iph->protocol == IPPROTO_IGMP) {
4925                         slave = rcu_dereference(bond->curr_active_slave);
4926                         if (slave)
4927                                 return slave;
4928                         return bond_get_slave_by_id(bond, 0);
4929                 }
4930         }
4931
4932 non_igmp:
4933         slave_cnt = READ_ONCE(bond->slave_cnt);
4934         if (likely(slave_cnt)) {
4935                 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4936                 return bond_get_slave_by_id(bond, slave_id);
4937         }
4938         return NULL;
4939 }
4940
4941 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4942                                         struct net_device *bond_dev)
4943 {
4944         struct bonding *bond = netdev_priv(bond_dev);
4945         struct slave *slave;
4946
4947         slave = bond_xmit_roundrobin_slave_get(bond, skb);
4948         if (likely(slave))
4949                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4950
4951         return bond_tx_drop(bond_dev, skb);
4952 }
4953
4954 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4955 {
4956         return rcu_dereference(bond->curr_active_slave);
4957 }
4958
4959 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4960  * the bond has a usable interface.
4961  */
4962 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4963                                           struct net_device *bond_dev)
4964 {
4965         struct bonding *bond = netdev_priv(bond_dev);
4966         struct slave *slave;
4967
4968         slave = bond_xmit_activebackup_slave_get(bond);
4969         if (slave)
4970                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4971
4972         return bond_tx_drop(bond_dev, skb);
4973 }
4974
4975 /* Use this to update slave_array when (a) it's not appropriate to update
4976  * slave_array right away (note that update_slave_array() may sleep)
4977  * and / or (b) RTNL is not held.
4978  */
4979 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4980 {
4981         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4982 }
4983
4984 /* Slave array work handler. Holds only RTNL */
4985 static void bond_slave_arr_handler(struct work_struct *work)
4986 {
4987         struct bonding *bond = container_of(work, struct bonding,
4988                                             slave_arr_work.work);
4989         int ret;
4990
4991         if (!rtnl_trylock())
4992                 goto err;
4993
4994         ret = bond_update_slave_arr(bond, NULL);
4995         rtnl_unlock();
4996         if (ret) {
4997                 pr_warn_ratelimited("Failed to update slave array from WT\n");
4998                 goto err;
4999         }
5000         return;
5001
5002 err:
5003         bond_slave_arr_work_rearm(bond, 1);
5004 }
5005
5006 static void bond_skip_slave(struct bond_up_slave *slaves,
5007                             struct slave *skipslave)
5008 {
5009         int idx;
5010
5011         /* Rare situation where caller has asked to skip a specific
5012          * slave but allocation failed (most likely!). BTW this is
5013          * only possible when the call is initiated from
5014          * __bond_release_one(). In this situation; overwrite the
5015          * skipslave entry in the array with the last entry from the
5016          * array to avoid a situation where the xmit path may choose
5017          * this to-be-skipped slave to send a packet out.
5018          */
5019         for (idx = 0; slaves && idx < slaves->count; idx++) {
5020                 if (skipslave == slaves->arr[idx]) {
5021                         slaves->arr[idx] =
5022                                 slaves->arr[slaves->count - 1];
5023                         slaves->count--;
5024                         break;
5025                 }
5026         }
5027 }
5028
5029 static void bond_set_slave_arr(struct bonding *bond,
5030                                struct bond_up_slave *usable_slaves,
5031                                struct bond_up_slave *all_slaves)
5032 {
5033         struct bond_up_slave *usable, *all;
5034
5035         usable = rtnl_dereference(bond->usable_slaves);
5036         rcu_assign_pointer(bond->usable_slaves, usable_slaves);
5037         kfree_rcu(usable, rcu);
5038
5039         all = rtnl_dereference(bond->all_slaves);
5040         rcu_assign_pointer(bond->all_slaves, all_slaves);
5041         kfree_rcu(all, rcu);
5042 }
5043
5044 static void bond_reset_slave_arr(struct bonding *bond)
5045 {
5046         struct bond_up_slave *usable, *all;
5047
5048         usable = rtnl_dereference(bond->usable_slaves);
5049         if (usable) {
5050                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5051                 kfree_rcu(usable, rcu);
5052         }
5053
5054         all = rtnl_dereference(bond->all_slaves);
5055         if (all) {
5056                 RCU_INIT_POINTER(bond->all_slaves, NULL);
5057                 kfree_rcu(all, rcu);
5058         }
5059 }
5060
5061 /* Build the usable slaves array in control path for modes that use xmit-hash
5062  * to determine the slave interface -
5063  * (a) BOND_MODE_8023AD
5064  * (b) BOND_MODE_XOR
5065  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5066  *
5067  * The caller is expected to hold RTNL only and NO other lock!
5068  */
5069 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5070 {
5071         struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5072         struct slave *slave;
5073         struct list_head *iter;
5074         int agg_id = 0;
5075         int ret = 0;
5076
5077         might_sleep();
5078
5079         usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5080                                             bond->slave_cnt), GFP_KERNEL);
5081         all_slaves = kzalloc(struct_size(all_slaves, arr,
5082                                          bond->slave_cnt), GFP_KERNEL);
5083         if (!usable_slaves || !all_slaves) {
5084                 ret = -ENOMEM;
5085                 goto out;
5086         }
5087         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5088                 struct ad_info ad_info;
5089
5090                 spin_lock_bh(&bond->mode_lock);
5091                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
5092                         spin_unlock_bh(&bond->mode_lock);
5093                         pr_debug("bond_3ad_get_active_agg_info failed\n");
5094                         /* No active aggragator means it's not safe to use
5095                          * the previous array.
5096                          */
5097                         bond_reset_slave_arr(bond);
5098                         goto out;
5099                 }
5100                 spin_unlock_bh(&bond->mode_lock);
5101                 agg_id = ad_info.aggregator_id;
5102         }
5103         bond_for_each_slave(bond, slave, iter) {
5104                 if (skipslave == slave)
5105                         continue;
5106
5107                 all_slaves->arr[all_slaves->count++] = slave;
5108                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5109                         struct aggregator *agg;
5110
5111                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
5112                         if (!agg || agg->aggregator_identifier != agg_id)
5113                                 continue;
5114                 }
5115                 if (!bond_slave_can_tx(slave))
5116                         continue;
5117
5118                 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5119                           usable_slaves->count);
5120
5121                 usable_slaves->arr[usable_slaves->count++] = slave;
5122         }
5123
5124         bond_set_slave_arr(bond, usable_slaves, all_slaves);
5125         return ret;
5126 out:
5127         if (ret != 0 && skipslave) {
5128                 bond_skip_slave(rtnl_dereference(bond->all_slaves),
5129                                 skipslave);
5130                 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5131                                 skipslave);
5132         }
5133         kfree_rcu(all_slaves, rcu);
5134         kfree_rcu(usable_slaves, rcu);
5135
5136         return ret;
5137 }
5138
5139 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5140                                                  struct sk_buff *skb,
5141                                                  struct bond_up_slave *slaves)
5142 {
5143         struct slave *slave;
5144         unsigned int count;
5145         u32 hash;
5146
5147         hash = bond_xmit_hash(bond, skb);
5148         count = slaves ? READ_ONCE(slaves->count) : 0;
5149         if (unlikely(!count))
5150                 return NULL;
5151
5152         slave = slaves->arr[hash % count];
5153         return slave;
5154 }
5155
5156 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5157                                                      struct xdp_buff *xdp)
5158 {
5159         struct bond_up_slave *slaves;
5160         unsigned int count;
5161         u32 hash;
5162
5163         hash = bond_xmit_hash_xdp(bond, xdp);
5164         slaves = rcu_dereference(bond->usable_slaves);
5165         count = slaves ? READ_ONCE(slaves->count) : 0;
5166         if (unlikely(!count))
5167                 return NULL;
5168
5169         return slaves->arr[hash % count];
5170 }
5171
5172 /* Use this Xmit function for 3AD as well as XOR modes. The current
5173  * usable slave array is formed in the control path. The xmit function
5174  * just calculates hash and sends the packet out.
5175  */
5176 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5177                                      struct net_device *dev)
5178 {
5179         struct bonding *bond = netdev_priv(dev);
5180         struct bond_up_slave *slaves;
5181         struct slave *slave;
5182
5183         slaves = rcu_dereference(bond->usable_slaves);
5184         slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5185         if (likely(slave))
5186                 return bond_dev_queue_xmit(bond, skb, slave->dev);
5187
5188         return bond_tx_drop(dev, skb);
5189 }
5190
5191 /* in broadcast mode, we send everything to all usable interfaces. */
5192 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5193                                        struct net_device *bond_dev)
5194 {
5195         struct bonding *bond = netdev_priv(bond_dev);
5196         struct slave *slave = NULL;
5197         struct list_head *iter;
5198         bool xmit_suc = false;
5199         bool skb_used = false;
5200
5201         bond_for_each_slave_rcu(bond, slave, iter) {
5202                 struct sk_buff *skb2;
5203
5204                 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5205                         continue;
5206
5207                 if (bond_is_last_slave(bond, slave)) {
5208                         skb2 = skb;
5209                         skb_used = true;
5210                 } else {
5211                         skb2 = skb_clone(skb, GFP_ATOMIC);
5212                         if (!skb2) {
5213                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5214                                                     bond_dev->name, __func__);
5215                                 continue;
5216                         }
5217                 }
5218
5219                 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5220                         xmit_suc = true;
5221         }
5222
5223         if (!skb_used)
5224                 dev_kfree_skb_any(skb);
5225
5226         if (xmit_suc)
5227                 return NETDEV_TX_OK;
5228
5229         dev_core_stats_tx_dropped_inc(bond_dev);
5230         return NET_XMIT_DROP;
5231 }
5232
5233 /*------------------------- Device initialization ---------------------------*/
5234
5235 /* Lookup the slave that corresponds to a qid */
5236 static inline int bond_slave_override(struct bonding *bond,
5237                                       struct sk_buff *skb)
5238 {
5239         struct slave *slave = NULL;
5240         struct list_head *iter;
5241
5242         if (!skb_rx_queue_recorded(skb))
5243                 return 1;
5244
5245         /* Find out if any slaves have the same mapping as this skb. */
5246         bond_for_each_slave_rcu(bond, slave, iter) {
5247                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
5248                         if (bond_slave_is_up(slave) &&
5249                             slave->link == BOND_LINK_UP) {
5250                                 bond_dev_queue_xmit(bond, skb, slave->dev);
5251                                 return 0;
5252                         }
5253                         /* If the slave isn't UP, use default transmit policy. */
5254                         break;
5255                 }
5256         }
5257
5258         return 1;
5259 }
5260
5261
5262 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5263                              struct net_device *sb_dev)
5264 {
5265         /* This helper function exists to help dev_pick_tx get the correct
5266          * destination queue.  Using a helper function skips a call to
5267          * skb_tx_hash and will put the skbs in the queue we expect on their
5268          * way down to the bonding driver.
5269          */
5270         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5271
5272         /* Save the original txq to restore before passing to the driver */
5273         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5274
5275         if (unlikely(txq >= dev->real_num_tx_queues)) {
5276                 do {
5277                         txq -= dev->real_num_tx_queues;
5278                 } while (txq >= dev->real_num_tx_queues);
5279         }
5280         return txq;
5281 }
5282
5283 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5284                                               struct sk_buff *skb,
5285                                               bool all_slaves)
5286 {
5287         struct bonding *bond = netdev_priv(master_dev);
5288         struct bond_up_slave *slaves;
5289         struct slave *slave = NULL;
5290
5291         switch (BOND_MODE(bond)) {
5292         case BOND_MODE_ROUNDROBIN:
5293                 slave = bond_xmit_roundrobin_slave_get(bond, skb);
5294                 break;
5295         case BOND_MODE_ACTIVEBACKUP:
5296                 slave = bond_xmit_activebackup_slave_get(bond);
5297                 break;
5298         case BOND_MODE_8023AD:
5299         case BOND_MODE_XOR:
5300                 if (all_slaves)
5301                         slaves = rcu_dereference(bond->all_slaves);
5302                 else
5303                         slaves = rcu_dereference(bond->usable_slaves);
5304                 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5305                 break;
5306         case BOND_MODE_BROADCAST:
5307                 break;
5308         case BOND_MODE_ALB:
5309                 slave = bond_xmit_alb_slave_get(bond, skb);
5310                 break;
5311         case BOND_MODE_TLB:
5312                 slave = bond_xmit_tlb_slave_get(bond, skb);
5313                 break;
5314         default:
5315                 /* Should never happen, mode already checked */
5316                 WARN_ONCE(true, "Unknown bonding mode");
5317                 break;
5318         }
5319
5320         if (slave)
5321                 return slave->dev;
5322         return NULL;
5323 }
5324
5325 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5326 {
5327         switch (sk->sk_family) {
5328 #if IS_ENABLED(CONFIG_IPV6)
5329         case AF_INET6:
5330                 if (ipv6_only_sock(sk) ||
5331                     ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5332                         flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5333                         flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5334                         flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5335                         break;
5336                 }
5337                 fallthrough;
5338 #endif
5339         default: /* AF_INET */
5340                 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5341                 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5342                 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5343                 break;
5344         }
5345
5346         flow->ports.src = inet_sk(sk)->inet_sport;
5347         flow->ports.dst = inet_sk(sk)->inet_dport;
5348 }
5349
5350 /**
5351  * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5352  * @sk: socket to use for headers
5353  *
5354  * This function will extract the necessary field from the socket and use
5355  * them to generate a hash based on the LAYER34 xmit_policy.
5356  * Assumes that sk is a TCP or UDP socket.
5357  */
5358 static u32 bond_sk_hash_l34(struct sock *sk)
5359 {
5360         struct flow_keys flow;
5361         u32 hash;
5362
5363         bond_sk_to_flow(sk, &flow);
5364
5365         /* L4 */
5366         memcpy(&hash, &flow.ports.ports, sizeof(hash));
5367         /* L3 */
5368         return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5369 }
5370
5371 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5372                                                   struct sock *sk)
5373 {
5374         struct bond_up_slave *slaves;
5375         struct slave *slave;
5376         unsigned int count;
5377         u32 hash;
5378
5379         slaves = rcu_dereference(bond->usable_slaves);
5380         count = slaves ? READ_ONCE(slaves->count) : 0;
5381         if (unlikely(!count))
5382                 return NULL;
5383
5384         hash = bond_sk_hash_l34(sk);
5385         slave = slaves->arr[hash % count];
5386
5387         return slave->dev;
5388 }
5389
5390 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5391                                                 struct sock *sk)
5392 {
5393         struct bonding *bond = netdev_priv(dev);
5394         struct net_device *lower = NULL;
5395
5396         rcu_read_lock();
5397         if (bond_sk_check(bond))
5398                 lower = __bond_sk_get_lower_dev(bond, sk);
5399         rcu_read_unlock();
5400
5401         return lower;
5402 }
5403
5404 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5405 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5406                                         struct net_device *dev)
5407 {
5408         struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev);
5409
5410         /* tls_netdev might become NULL, even if tls_is_sk_tx_device_offloaded
5411          * was true, if tls_device_down is running in parallel, but it's OK,
5412          * because bond_get_slave_by_dev has a NULL check.
5413          */
5414         if (likely(bond_get_slave_by_dev(bond, tls_netdev)))
5415                 return bond_dev_queue_xmit(bond, skb, tls_netdev);
5416         return bond_tx_drop(dev, skb);
5417 }
5418 #endif
5419
5420 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5421 {
5422         struct bonding *bond = netdev_priv(dev);
5423
5424         if (bond_should_override_tx_queue(bond) &&
5425             !bond_slave_override(bond, skb))
5426                 return NETDEV_TX_OK;
5427
5428 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5429         if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
5430                 return bond_tls_device_xmit(bond, skb, dev);
5431 #endif
5432
5433         switch (BOND_MODE(bond)) {
5434         case BOND_MODE_ROUNDROBIN:
5435                 return bond_xmit_roundrobin(skb, dev);
5436         case BOND_MODE_ACTIVEBACKUP:
5437                 return bond_xmit_activebackup(skb, dev);
5438         case BOND_MODE_8023AD:
5439         case BOND_MODE_XOR:
5440                 return bond_3ad_xor_xmit(skb, dev);
5441         case BOND_MODE_BROADCAST:
5442                 return bond_xmit_broadcast(skb, dev);
5443         case BOND_MODE_ALB:
5444                 return bond_alb_xmit(skb, dev);
5445         case BOND_MODE_TLB:
5446                 return bond_tlb_xmit(skb, dev);
5447         default:
5448                 /* Should never happen, mode already checked */
5449                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5450                 WARN_ON_ONCE(1);
5451                 return bond_tx_drop(dev, skb);
5452         }
5453 }
5454
5455 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5456 {
5457         struct bonding *bond = netdev_priv(dev);
5458         netdev_tx_t ret = NETDEV_TX_OK;
5459
5460         /* If we risk deadlock from transmitting this in the
5461          * netpoll path, tell netpoll to queue the frame for later tx
5462          */
5463         if (unlikely(is_netpoll_tx_blocked(dev)))
5464                 return NETDEV_TX_BUSY;
5465
5466         rcu_read_lock();
5467         if (bond_has_slaves(bond))
5468                 ret = __bond_start_xmit(skb, dev);
5469         else
5470                 ret = bond_tx_drop(dev, skb);
5471         rcu_read_unlock();
5472
5473         return ret;
5474 }
5475
5476 static struct net_device *
5477 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5478 {
5479         struct bonding *bond = netdev_priv(bond_dev);
5480         struct slave *slave;
5481
5482         /* Caller needs to hold rcu_read_lock() */
5483
5484         switch (BOND_MODE(bond)) {
5485         case BOND_MODE_ROUNDROBIN:
5486                 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5487                 break;
5488
5489         case BOND_MODE_ACTIVEBACKUP:
5490                 slave = bond_xmit_activebackup_slave_get(bond);
5491                 break;
5492
5493         case BOND_MODE_8023AD:
5494         case BOND_MODE_XOR:
5495                 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5496                 break;
5497
5498         default:
5499                 /* Should never happen. Mode guarded by bond_xdp_check() */
5500                 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5501                 WARN_ON_ONCE(1);
5502                 return NULL;
5503         }
5504
5505         if (slave)
5506                 return slave->dev;
5507
5508         return NULL;
5509 }
5510
5511 static int bond_xdp_xmit(struct net_device *bond_dev,
5512                          int n, struct xdp_frame **frames, u32 flags)
5513 {
5514         int nxmit, err = -ENXIO;
5515
5516         rcu_read_lock();
5517
5518         for (nxmit = 0; nxmit < n; nxmit++) {
5519                 struct xdp_frame *frame = frames[nxmit];
5520                 struct xdp_frame *frames1[] = {frame};
5521                 struct net_device *slave_dev;
5522                 struct xdp_buff xdp;
5523
5524                 xdp_convert_frame_to_buff(frame, &xdp);
5525
5526                 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5527                 if (!slave_dev) {
5528                         err = -ENXIO;
5529                         break;
5530                 }
5531
5532                 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5533                 if (err < 1)
5534                         break;
5535         }
5536
5537         rcu_read_unlock();
5538
5539         /* If error happened on the first frame then we can pass the error up, otherwise
5540          * report the number of frames that were xmitted.
5541          */
5542         if (err < 0)
5543                 return (nxmit == 0 ? err : nxmit);
5544
5545         return nxmit;
5546 }
5547
5548 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5549                         struct netlink_ext_ack *extack)
5550 {
5551         struct bonding *bond = netdev_priv(dev);
5552         struct list_head *iter;
5553         struct slave *slave, *rollback_slave;
5554         struct bpf_prog *old_prog;
5555         struct netdev_bpf xdp = {
5556                 .command = XDP_SETUP_PROG,
5557                 .flags   = 0,
5558                 .prog    = prog,
5559                 .extack  = extack,
5560         };
5561         int err;
5562
5563         ASSERT_RTNL();
5564
5565         if (!bond_xdp_check(bond))
5566                 return -EOPNOTSUPP;
5567
5568         old_prog = bond->xdp_prog;
5569         bond->xdp_prog = prog;
5570
5571         bond_for_each_slave(bond, slave, iter) {
5572                 struct net_device *slave_dev = slave->dev;
5573
5574                 if (!slave_dev->netdev_ops->ndo_bpf ||
5575                     !slave_dev->netdev_ops->ndo_xdp_xmit) {
5576                         SLAVE_NL_ERR(dev, slave_dev, extack,
5577                                      "Slave device does not support XDP");
5578                         err = -EOPNOTSUPP;
5579                         goto err;
5580                 }
5581
5582                 if (dev_xdp_prog_count(slave_dev) > 0) {
5583                         SLAVE_NL_ERR(dev, slave_dev, extack,
5584                                      "Slave has XDP program loaded, please unload before enslaving");
5585                         err = -EOPNOTSUPP;
5586                         goto err;
5587                 }
5588
5589                 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5590                 if (err < 0) {
5591                         /* ndo_bpf() sets extack error message */
5592                         slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5593                         goto err;
5594                 }
5595                 if (prog)
5596                         bpf_prog_inc(prog);
5597         }
5598
5599         if (prog) {
5600                 static_branch_inc(&bpf_master_redirect_enabled_key);
5601         } else if (old_prog) {
5602                 bpf_prog_put(old_prog);
5603                 static_branch_dec(&bpf_master_redirect_enabled_key);
5604         }
5605
5606         return 0;
5607
5608 err:
5609         /* unwind the program changes */
5610         bond->xdp_prog = old_prog;
5611         xdp.prog = old_prog;
5612         xdp.extack = NULL; /* do not overwrite original error */
5613
5614         bond_for_each_slave(bond, rollback_slave, iter) {
5615                 struct net_device *slave_dev = rollback_slave->dev;
5616                 int err_unwind;
5617
5618                 if (slave == rollback_slave)
5619                         break;
5620
5621                 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5622                 if (err_unwind < 0)
5623                         slave_err(dev, slave_dev,
5624                                   "Error %d when unwinding XDP program change\n", err_unwind);
5625                 else if (xdp.prog)
5626                         bpf_prog_inc(xdp.prog);
5627         }
5628         return err;
5629 }
5630
5631 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5632 {
5633         switch (xdp->command) {
5634         case XDP_SETUP_PROG:
5635                 return bond_xdp_set(dev, xdp->prog, xdp->extack);
5636         default:
5637                 return -EINVAL;
5638         }
5639 }
5640
5641 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5642 {
5643         if (speed == 0 || speed == SPEED_UNKNOWN)
5644                 speed = slave->speed;
5645         else
5646                 speed = min(speed, slave->speed);
5647
5648         return speed;
5649 }
5650
5651 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5652                                            struct ethtool_link_ksettings *cmd)
5653 {
5654         struct bonding *bond = netdev_priv(bond_dev);
5655         struct list_head *iter;
5656         struct slave *slave;
5657         u32 speed = 0;
5658
5659         cmd->base.duplex = DUPLEX_UNKNOWN;
5660         cmd->base.port = PORT_OTHER;
5661
5662         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5663          * do not need to check mode.  Though link speed might not represent
5664          * the true receive or transmit bandwidth (not all modes are symmetric)
5665          * this is an accurate maximum.
5666          */
5667         bond_for_each_slave(bond, slave, iter) {
5668                 if (bond_slave_can_tx(slave)) {
5669                         if (slave->speed != SPEED_UNKNOWN) {
5670                                 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5671                                         speed = bond_mode_bcast_speed(slave,
5672                                                                       speed);
5673                                 else
5674                                         speed += slave->speed;
5675                         }
5676                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5677                             slave->duplex != DUPLEX_UNKNOWN)
5678                                 cmd->base.duplex = slave->duplex;
5679                 }
5680         }
5681         cmd->base.speed = speed ? : SPEED_UNKNOWN;
5682
5683         return 0;
5684 }
5685
5686 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5687                                      struct ethtool_drvinfo *drvinfo)
5688 {
5689         strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5690         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5691                  BOND_ABI_VERSION);
5692 }
5693
5694 static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5695                                     struct ethtool_ts_info *info)
5696 {
5697         struct bonding *bond = netdev_priv(bond_dev);
5698         const struct ethtool_ops *ops;
5699         struct net_device *real_dev;
5700         struct phy_device *phydev;
5701         int ret = 0;
5702
5703         rcu_read_lock();
5704         real_dev = bond_option_active_slave_get_rcu(bond);
5705         dev_hold(real_dev);
5706         rcu_read_unlock();
5707
5708         if (real_dev) {
5709                 ops = real_dev->ethtool_ops;
5710                 phydev = real_dev->phydev;
5711
5712                 if (phy_has_tsinfo(phydev)) {
5713                         ret = phy_ts_info(phydev, info);
5714                         goto out;
5715                 } else if (ops->get_ts_info) {
5716                         ret = ops->get_ts_info(real_dev, info);
5717                         goto out;
5718                 }
5719         }
5720
5721         info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
5722                                 SOF_TIMESTAMPING_SOFTWARE;
5723         info->phc_index = -1;
5724
5725 out:
5726         dev_put(real_dev);
5727         return ret;
5728 }
5729
5730 static const struct ethtool_ops bond_ethtool_ops = {
5731         .get_drvinfo            = bond_ethtool_get_drvinfo,
5732         .get_link               = ethtool_op_get_link,
5733         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
5734         .get_ts_info            = bond_ethtool_get_ts_info,
5735 };
5736
5737 static const struct net_device_ops bond_netdev_ops = {
5738         .ndo_init               = bond_init,
5739         .ndo_uninit             = bond_uninit,
5740         .ndo_open               = bond_open,
5741         .ndo_stop               = bond_close,
5742         .ndo_start_xmit         = bond_start_xmit,
5743         .ndo_select_queue       = bond_select_queue,
5744         .ndo_get_stats64        = bond_get_stats,
5745         .ndo_eth_ioctl          = bond_eth_ioctl,
5746         .ndo_siocbond           = bond_do_ioctl,
5747         .ndo_siocdevprivate     = bond_siocdevprivate,
5748         .ndo_change_rx_flags    = bond_change_rx_flags,
5749         .ndo_set_rx_mode        = bond_set_rx_mode,
5750         .ndo_change_mtu         = bond_change_mtu,
5751         .ndo_set_mac_address    = bond_set_mac_address,
5752         .ndo_neigh_setup        = bond_neigh_setup,
5753         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
5754         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
5755 #ifdef CONFIG_NET_POLL_CONTROLLER
5756         .ndo_netpoll_setup      = bond_netpoll_setup,
5757         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
5758         .ndo_poll_controller    = bond_poll_controller,
5759 #endif
5760         .ndo_add_slave          = bond_enslave,
5761         .ndo_del_slave          = bond_release,
5762         .ndo_fix_features       = bond_fix_features,
5763         .ndo_features_check     = passthru_features_check,
5764         .ndo_get_xmit_slave     = bond_xmit_get_slave,
5765         .ndo_sk_get_lower_dev   = bond_sk_get_lower_dev,
5766         .ndo_bpf                = bond_xdp,
5767         .ndo_xdp_xmit           = bond_xdp_xmit,
5768         .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5769 };
5770
5771 static const struct device_type bond_type = {
5772         .name = "bond",
5773 };
5774
5775 static void bond_destructor(struct net_device *bond_dev)
5776 {
5777         struct bonding *bond = netdev_priv(bond_dev);
5778
5779         if (bond->wq)
5780                 destroy_workqueue(bond->wq);
5781
5782         if (bond->rr_tx_counter)
5783                 free_percpu(bond->rr_tx_counter);
5784 }
5785
5786 void bond_setup(struct net_device *bond_dev)
5787 {
5788         struct bonding *bond = netdev_priv(bond_dev);
5789
5790         spin_lock_init(&bond->mode_lock);
5791         bond->params = bonding_defaults;
5792
5793         /* Initialize pointers */
5794         bond->dev = bond_dev;
5795
5796         /* Initialize the device entry points */
5797         ether_setup(bond_dev);
5798         bond_dev->max_mtu = ETH_MAX_MTU;
5799         bond_dev->netdev_ops = &bond_netdev_ops;
5800         bond_dev->ethtool_ops = &bond_ethtool_ops;
5801
5802         bond_dev->needs_free_netdev = true;
5803         bond_dev->priv_destructor = bond_destructor;
5804
5805         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5806
5807         /* Initialize the device options */
5808         bond_dev->flags |= IFF_MASTER;
5809         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5810         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5811
5812 #ifdef CONFIG_XFRM_OFFLOAD
5813         /* set up xfrm device ops (only supported in active-backup right now) */
5814         bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5815         INIT_LIST_HEAD(&bond->ipsec_list);
5816         spin_lock_init(&bond->ipsec_lock);
5817 #endif /* CONFIG_XFRM_OFFLOAD */
5818
5819         /* don't acquire bond device's netif_tx_lock when transmitting */
5820         bond_dev->features |= NETIF_F_LLTX;
5821
5822         /* By default, we declare the bond to be fully
5823          * VLAN hardware accelerated capable. Special
5824          * care is taken in the various xmit functions
5825          * when there are slaves that are not hw accel
5826          * capable
5827          */
5828
5829         /* Don't allow bond devices to change network namespaces. */
5830         bond_dev->features |= NETIF_F_NETNS_LOCAL;
5831
5832         bond_dev->hw_features = BOND_VLAN_FEATURES |
5833                                 NETIF_F_HW_VLAN_CTAG_RX |
5834                                 NETIF_F_HW_VLAN_CTAG_FILTER;
5835
5836         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5837         bond_dev->features |= bond_dev->hw_features;
5838         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5839 #ifdef CONFIG_XFRM_OFFLOAD
5840         bond_dev->hw_features |= BOND_XFRM_FEATURES;
5841         /* Only enable XFRM features if this is an active-backup config */
5842         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5843                 bond_dev->features |= BOND_XFRM_FEATURES;
5844 #endif /* CONFIG_XFRM_OFFLOAD */
5845 }
5846
5847 /* Destroy a bonding device.
5848  * Must be under rtnl_lock when this function is called.
5849  */
5850 static void bond_uninit(struct net_device *bond_dev)
5851 {
5852         struct bonding *bond = netdev_priv(bond_dev);
5853         struct bond_up_slave *usable, *all;
5854         struct list_head *iter;
5855         struct slave *slave;
5856
5857         bond_netpoll_cleanup(bond_dev);
5858
5859         /* Release the bonded slaves */
5860         bond_for_each_slave(bond, slave, iter)
5861                 __bond_release_one(bond_dev, slave->dev, true, true);
5862         netdev_info(bond_dev, "Released all slaves\n");
5863
5864         usable = rtnl_dereference(bond->usable_slaves);
5865         if (usable) {
5866                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5867                 kfree_rcu(usable, rcu);
5868         }
5869
5870         all = rtnl_dereference(bond->all_slaves);
5871         if (all) {
5872                 RCU_INIT_POINTER(bond->all_slaves, NULL);
5873                 kfree_rcu(all, rcu);
5874         }
5875
5876         list_del(&bond->bond_list);
5877
5878         bond_debug_unregister(bond);
5879 }
5880
5881 /*------------------------- Module initialization ---------------------------*/
5882
5883 static int bond_check_params(struct bond_params *params)
5884 {
5885         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5886         struct bond_opt_value newval;
5887         const struct bond_opt_value *valptr;
5888         int arp_all_targets_value = 0;
5889         u16 ad_actor_sys_prio = 0;
5890         u16 ad_user_port_key = 0;
5891         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5892         int arp_ip_count;
5893         int bond_mode   = BOND_MODE_ROUNDROBIN;
5894         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5895         int lacp_fast = 0;
5896         int tlb_dynamic_lb;
5897
5898         /* Convert string parameters. */
5899         if (mode) {
5900                 bond_opt_initstr(&newval, mode);
5901                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5902                 if (!valptr) {
5903                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5904                         return -EINVAL;
5905                 }
5906                 bond_mode = valptr->value;
5907         }
5908
5909         if (xmit_hash_policy) {
5910                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5911                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
5912                     bond_mode == BOND_MODE_BROADCAST) {
5913                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5914                                 bond_mode_name(bond_mode));
5915                 } else {
5916                         bond_opt_initstr(&newval, xmit_hash_policy);
5917                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5918                                                 &newval);
5919                         if (!valptr) {
5920                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5921                                        xmit_hash_policy);
5922                                 return -EINVAL;
5923                         }
5924                         xmit_hashtype = valptr->value;
5925                 }
5926         }
5927
5928         if (lacp_rate) {
5929                 if (bond_mode != BOND_MODE_8023AD) {
5930                         pr_info("lacp_rate param is irrelevant in mode %s\n",
5931                                 bond_mode_name(bond_mode));
5932                 } else {
5933                         bond_opt_initstr(&newval, lacp_rate);
5934                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5935                                                 &newval);
5936                         if (!valptr) {
5937                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
5938                                        lacp_rate);
5939                                 return -EINVAL;
5940                         }
5941                         lacp_fast = valptr->value;
5942                 }
5943         }
5944
5945         if (ad_select) {
5946                 bond_opt_initstr(&newval, ad_select);
5947                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
5948                                         &newval);
5949                 if (!valptr) {
5950                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
5951                         return -EINVAL;
5952                 }
5953                 params->ad_select = valptr->value;
5954                 if (bond_mode != BOND_MODE_8023AD)
5955                         pr_warn("ad_select param only affects 802.3ad mode\n");
5956         } else {
5957                 params->ad_select = BOND_AD_STABLE;
5958         }
5959
5960         if (max_bonds < 0) {
5961                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
5962                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
5963                 max_bonds = BOND_DEFAULT_MAX_BONDS;
5964         }
5965
5966         if (miimon < 0) {
5967                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5968                         miimon, INT_MAX);
5969                 miimon = 0;
5970         }
5971
5972         if (updelay < 0) {
5973                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5974                         updelay, INT_MAX);
5975                 updelay = 0;
5976         }
5977
5978         if (downdelay < 0) {
5979                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5980                         downdelay, INT_MAX);
5981                 downdelay = 0;
5982         }
5983
5984         if ((use_carrier != 0) && (use_carrier != 1)) {
5985                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5986                         use_carrier);
5987                 use_carrier = 1;
5988         }
5989
5990         if (num_peer_notif < 0 || num_peer_notif > 255) {
5991                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5992                         num_peer_notif);
5993                 num_peer_notif = 1;
5994         }
5995
5996         /* reset values for 802.3ad/TLB/ALB */
5997         if (!bond_mode_uses_arp(bond_mode)) {
5998                 if (!miimon) {
5999                         pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
6000                         pr_warn("Forcing miimon to 100msec\n");
6001                         miimon = BOND_DEFAULT_MIIMON;
6002                 }
6003         }
6004
6005         if (tx_queues < 1 || tx_queues > 255) {
6006                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
6007                         tx_queues, BOND_DEFAULT_TX_QUEUES);
6008                 tx_queues = BOND_DEFAULT_TX_QUEUES;
6009         }
6010
6011         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
6012                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
6013                         all_slaves_active);
6014                 all_slaves_active = 0;
6015         }
6016
6017         if (resend_igmp < 0 || resend_igmp > 255) {
6018                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
6019                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
6020                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
6021         }
6022
6023         bond_opt_initval(&newval, packets_per_slave);
6024         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
6025                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
6026                         packets_per_slave, USHRT_MAX);
6027                 packets_per_slave = 1;
6028         }
6029
6030         if (bond_mode == BOND_MODE_ALB) {
6031                 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
6032                           updelay);
6033         }
6034
6035         if (!miimon) {
6036                 if (updelay || downdelay) {
6037                         /* just warn the user the up/down delay will have
6038                          * no effect since miimon is zero...
6039                          */
6040                         pr_warn("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
6041                                 updelay, downdelay);
6042                 }
6043         } else {
6044                 /* don't allow arp monitoring */
6045                 if (arp_interval) {
6046                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
6047                                 miimon, arp_interval);
6048                         arp_interval = 0;
6049                 }
6050
6051                 if ((updelay % miimon) != 0) {
6052                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
6053                                 updelay, miimon, (updelay / miimon) * miimon);
6054                 }
6055
6056                 updelay /= miimon;
6057
6058                 if ((downdelay % miimon) != 0) {
6059                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
6060                                 downdelay, miimon,
6061                                 (downdelay / miimon) * miimon);
6062                 }
6063
6064                 downdelay /= miimon;
6065         }
6066
6067         if (arp_interval < 0) {
6068                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6069                         arp_interval, INT_MAX);
6070                 arp_interval = 0;
6071         }
6072
6073         for (arp_ip_count = 0, i = 0;
6074              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6075                 __be32 ip;
6076
6077                 /* not a complete check, but good enough to catch mistakes */
6078                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
6079                     !bond_is_ip_target_ok(ip)) {
6080                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6081                                 arp_ip_target[i]);
6082                         arp_interval = 0;
6083                 } else {
6084                         if (bond_get_targets_ip(arp_target, ip) == -1)
6085                                 arp_target[arp_ip_count++] = ip;
6086                         else
6087                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6088                                         &ip);
6089                 }
6090         }
6091
6092         if (arp_interval && !arp_ip_count) {
6093                 /* don't allow arping if no arp_ip_target given... */
6094                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6095                         arp_interval);
6096                 arp_interval = 0;
6097         }
6098
6099         if (arp_validate) {
6100                 if (!arp_interval) {
6101                         pr_err("arp_validate requires arp_interval\n");
6102                         return -EINVAL;
6103                 }
6104
6105                 bond_opt_initstr(&newval, arp_validate);
6106                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6107                                         &newval);
6108                 if (!valptr) {
6109                         pr_err("Error: invalid arp_validate \"%s\"\n",
6110                                arp_validate);
6111                         return -EINVAL;
6112                 }
6113                 arp_validate_value = valptr->value;
6114         } else {
6115                 arp_validate_value = 0;
6116         }
6117
6118         if (arp_all_targets) {
6119                 bond_opt_initstr(&newval, arp_all_targets);
6120                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6121                                         &newval);
6122                 if (!valptr) {
6123                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6124                                arp_all_targets);
6125                         arp_all_targets_value = 0;
6126                 } else {
6127                         arp_all_targets_value = valptr->value;
6128                 }
6129         }
6130
6131         if (miimon) {
6132                 pr_info("MII link monitoring set to %d ms\n", miimon);
6133         } else if (arp_interval) {
6134                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6135                                           arp_validate_value);
6136                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6137                         arp_interval, valptr->string, arp_ip_count);
6138
6139                 for (i = 0; i < arp_ip_count; i++)
6140                         pr_cont(" %s", arp_ip_target[i]);
6141
6142                 pr_cont("\n");
6143
6144         } else if (max_bonds) {
6145                 /* miimon and arp_interval not set, we need one so things
6146                  * work as expected, see bonding.txt for details
6147                  */
6148                 pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details\n");
6149         }
6150
6151         if (primary && !bond_mode_uses_primary(bond_mode)) {
6152                 /* currently, using a primary only makes sense
6153                  * in active backup, TLB or ALB modes
6154                  */
6155                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6156                         primary, bond_mode_name(bond_mode));
6157                 primary = NULL;
6158         }
6159
6160         if (primary && primary_reselect) {
6161                 bond_opt_initstr(&newval, primary_reselect);
6162                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6163                                         &newval);
6164                 if (!valptr) {
6165                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
6166                                primary_reselect);
6167                         return -EINVAL;
6168                 }
6169                 primary_reselect_value = valptr->value;
6170         } else {
6171                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6172         }
6173
6174         if (fail_over_mac) {
6175                 bond_opt_initstr(&newval, fail_over_mac);
6176                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6177                                         &newval);
6178                 if (!valptr) {
6179                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
6180                                fail_over_mac);
6181                         return -EINVAL;
6182                 }
6183                 fail_over_mac_value = valptr->value;
6184                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6185                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6186         } else {
6187                 fail_over_mac_value = BOND_FOM_NONE;
6188         }
6189
6190         bond_opt_initstr(&newval, "default");
6191         valptr = bond_opt_parse(
6192                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6193                                      &newval);
6194         if (!valptr) {
6195                 pr_err("Error: No ad_actor_sys_prio default value");
6196                 return -EINVAL;
6197         }
6198         ad_actor_sys_prio = valptr->value;
6199
6200         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6201                                 &newval);
6202         if (!valptr) {
6203                 pr_err("Error: No ad_user_port_key default value");
6204                 return -EINVAL;
6205         }
6206         ad_user_port_key = valptr->value;
6207
6208         bond_opt_initstr(&newval, "default");
6209         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6210         if (!valptr) {
6211                 pr_err("Error: No tlb_dynamic_lb default value");
6212                 return -EINVAL;
6213         }
6214         tlb_dynamic_lb = valptr->value;
6215
6216         if (lp_interval == 0) {
6217                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6218                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6219                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6220         }
6221
6222         /* fill params struct with the proper values */
6223         params->mode = bond_mode;
6224         params->xmit_policy = xmit_hashtype;
6225         params->miimon = miimon;
6226         params->num_peer_notif = num_peer_notif;
6227         params->arp_interval = arp_interval;
6228         params->arp_validate = arp_validate_value;
6229         params->arp_all_targets = arp_all_targets_value;
6230         params->missed_max = 2;
6231         params->updelay = updelay;
6232         params->downdelay = downdelay;
6233         params->peer_notif_delay = 0;
6234         params->use_carrier = use_carrier;
6235         params->lacp_active = 1;
6236         params->lacp_fast = lacp_fast;
6237         params->primary[0] = 0;
6238         params->primary_reselect = primary_reselect_value;
6239         params->fail_over_mac = fail_over_mac_value;
6240         params->tx_queues = tx_queues;
6241         params->all_slaves_active = all_slaves_active;
6242         params->resend_igmp = resend_igmp;
6243         params->min_links = min_links;
6244         params->lp_interval = lp_interval;
6245         params->packets_per_slave = packets_per_slave;
6246         params->tlb_dynamic_lb = tlb_dynamic_lb;
6247         params->ad_actor_sys_prio = ad_actor_sys_prio;
6248         eth_zero_addr(params->ad_actor_system);
6249         params->ad_user_port_key = ad_user_port_key;
6250         if (packets_per_slave > 0) {
6251                 params->reciprocal_packets_per_slave =
6252                         reciprocal_value(packets_per_slave);
6253         } else {
6254                 /* reciprocal_packets_per_slave is unused if
6255                  * packets_per_slave is 0 or 1, just initialize it
6256                  */
6257                 params->reciprocal_packets_per_slave =
6258                         (struct reciprocal_value) { 0 };
6259         }
6260
6261         if (primary)
6262                 strscpy_pad(params->primary, primary, sizeof(params->primary));
6263
6264         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6265 #if IS_ENABLED(CONFIG_IPV6)
6266         memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6267 #endif
6268
6269         return 0;
6270 }
6271
6272 /* Called from registration process */
6273 static int bond_init(struct net_device *bond_dev)
6274 {
6275         struct bonding *bond = netdev_priv(bond_dev);
6276         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6277
6278         netdev_dbg(bond_dev, "Begin bond_init\n");
6279
6280         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
6281         if (!bond->wq)
6282                 return -ENOMEM;
6283
6284         spin_lock_init(&bond->stats_lock);
6285         netdev_lockdep_set_classes(bond_dev);
6286
6287         list_add_tail(&bond->bond_list, &bn->dev_list);
6288
6289         bond_prepare_sysfs_group(bond);
6290
6291         bond_debug_register(bond);
6292
6293         /* Ensure valid dev_addr */
6294         if (is_zero_ether_addr(bond_dev->dev_addr) &&
6295             bond_dev->addr_assign_type == NET_ADDR_PERM)
6296                 eth_hw_addr_random(bond_dev);
6297
6298         return 0;
6299 }
6300
6301 unsigned int bond_get_num_tx_queues(void)
6302 {
6303         return tx_queues;
6304 }
6305
6306 /* Create a new bond based on the specified name and bonding parameters.
6307  * If name is NULL, obtain a suitable "bond%d" name for us.
6308  * Caller must NOT hold rtnl_lock; we need to release it here before we
6309  * set up our sysfs entries.
6310  */
6311 int bond_create(struct net *net, const char *name)
6312 {
6313         struct net_device *bond_dev;
6314         struct bonding *bond;
6315         int res = -ENOMEM;
6316
6317         rtnl_lock();
6318
6319         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6320                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
6321                                    bond_setup, tx_queues);
6322         if (!bond_dev)
6323                 goto out;
6324
6325         bond = netdev_priv(bond_dev);
6326         dev_net_set(bond_dev, net);
6327         bond_dev->rtnl_link_ops = &bond_link_ops;
6328
6329         res = register_netdevice(bond_dev);
6330         if (res < 0) {
6331                 free_netdev(bond_dev);
6332                 goto out;
6333         }
6334
6335         netif_carrier_off(bond_dev);
6336
6337         bond_work_init_all(bond);
6338
6339 out:
6340         rtnl_unlock();
6341         return res;
6342 }
6343
6344 static int __net_init bond_net_init(struct net *net)
6345 {
6346         struct bond_net *bn = net_generic(net, bond_net_id);
6347
6348         bn->net = net;
6349         INIT_LIST_HEAD(&bn->dev_list);
6350
6351         bond_create_proc_dir(bn);
6352         bond_create_sysfs(bn);
6353
6354         return 0;
6355 }
6356
6357 static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6358 {
6359         struct bond_net *bn;
6360         struct net *net;
6361         LIST_HEAD(list);
6362
6363         list_for_each_entry(net, net_list, exit_list) {
6364                 bn = net_generic(net, bond_net_id);
6365                 bond_destroy_sysfs(bn);
6366         }
6367
6368         /* Kill off any bonds created after unregistering bond rtnl ops */
6369         rtnl_lock();
6370         list_for_each_entry(net, net_list, exit_list) {
6371                 struct bonding *bond, *tmp_bond;
6372
6373                 bn = net_generic(net, bond_net_id);
6374                 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6375                         unregister_netdevice_queue(bond->dev, &list);
6376         }
6377         unregister_netdevice_many(&list);
6378         rtnl_unlock();
6379
6380         list_for_each_entry(net, net_list, exit_list) {
6381                 bn = net_generic(net, bond_net_id);
6382                 bond_destroy_proc_dir(bn);
6383         }
6384 }
6385
6386 static struct pernet_operations bond_net_ops = {
6387         .init = bond_net_init,
6388         .exit_batch = bond_net_exit_batch,
6389         .id   = &bond_net_id,
6390         .size = sizeof(struct bond_net),
6391 };
6392
6393 static int __init bonding_init(void)
6394 {
6395         int i;
6396         int res;
6397
6398         res = bond_check_params(&bonding_defaults);
6399         if (res)
6400                 goto out;
6401
6402         res = register_pernet_subsys(&bond_net_ops);
6403         if (res)
6404                 goto out;
6405
6406         res = bond_netlink_init();
6407         if (res)
6408                 goto err_link;
6409
6410         bond_create_debugfs();
6411
6412         for (i = 0; i < max_bonds; i++) {
6413                 res = bond_create(&init_net, NULL);
6414                 if (res)
6415                         goto err;
6416         }
6417
6418         skb_flow_dissector_init(&flow_keys_bonding,
6419                                 flow_keys_bonding_keys,
6420                                 ARRAY_SIZE(flow_keys_bonding_keys));
6421
6422         register_netdevice_notifier(&bond_netdev_notifier);
6423 out:
6424         return res;
6425 err:
6426         bond_destroy_debugfs();
6427         bond_netlink_fini();
6428 err_link:
6429         unregister_pernet_subsys(&bond_net_ops);
6430         goto out;
6431
6432 }
6433
6434 static void __exit bonding_exit(void)
6435 {
6436         unregister_netdevice_notifier(&bond_netdev_notifier);
6437
6438         bond_destroy_debugfs();
6439
6440         bond_netlink_fini();
6441         unregister_pernet_subsys(&bond_net_ops);
6442
6443 #ifdef CONFIG_NET_POLL_CONTROLLER
6444         /* Make sure we don't have an imbalance on our netpoll blocking */
6445         WARN_ON(atomic_read(&netpoll_block_tx));
6446 #endif
6447 }
6448
6449 module_init(bonding_init);
6450 module_exit(bonding_exit);
6451 MODULE_LICENSE("GPL");
6452 MODULE_DESCRIPTION(DRV_DESCRIPTION);
6453 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");