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 and IFF_UP
1781  * if they were set
1782  */
1783 static void bond_ether_setup(struct net_device *bond_dev)
1784 {
1785         unsigned int flags = bond_dev->flags & (IFF_SLAVE | IFF_UP);
1786
1787         ether_setup(bond_dev);
1788         bond_dev->flags |= IFF_MASTER | flags;
1789         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1790 }
1791
1792 /* enslave device <slave> to bond device <master> */
1793 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1794                  struct netlink_ext_ack *extack)
1795 {
1796         struct bonding *bond = netdev_priv(bond_dev);
1797         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1798         struct slave *new_slave = NULL, *prev_slave;
1799         struct sockaddr_storage ss;
1800         int link_reporting;
1801         int res = 0, i;
1802
1803         if (slave_dev->flags & IFF_MASTER &&
1804             !netif_is_bond_master(slave_dev)) {
1805                 BOND_NL_ERR(bond_dev, extack,
1806                             "Device type (master device) cannot be enslaved");
1807                 return -EPERM;
1808         }
1809
1810         if (!bond->params.use_carrier &&
1811             slave_dev->ethtool_ops->get_link == NULL &&
1812             slave_ops->ndo_eth_ioctl == NULL) {
1813                 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1814         }
1815
1816         /* already in-use? */
1817         if (netdev_is_rx_handler_busy(slave_dev)) {
1818                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1819                              "Device is in use and cannot be enslaved");
1820                 return -EBUSY;
1821         }
1822
1823         if (bond_dev == slave_dev) {
1824                 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1825                 return -EPERM;
1826         }
1827
1828         /* vlan challenged mutual exclusion */
1829         /* no need to lock since we're protected by rtnl_lock */
1830         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1831                 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1832                 if (vlan_uses_dev(bond_dev)) {
1833                         SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1834                                      "Can not enslave VLAN challenged device to VLAN enabled bond");
1835                         return -EPERM;
1836                 } else {
1837                         slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1838                 }
1839         } else {
1840                 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1841         }
1842
1843         if (slave_dev->features & NETIF_F_HW_ESP)
1844                 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1845
1846         /* Old ifenslave binaries are no longer supported.  These can
1847          * be identified with moderate accuracy by the state of the slave:
1848          * the current ifenslave will set the interface down prior to
1849          * enslaving it; the old ifenslave will not.
1850          */
1851         if (slave_dev->flags & IFF_UP) {
1852                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1853                              "Device can not be enslaved while up");
1854                 return -EPERM;
1855         }
1856
1857         /* set bonding device ether type by slave - bonding netdevices are
1858          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1859          * there is a need to override some of the type dependent attribs/funcs.
1860          *
1861          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1862          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1863          */
1864         if (!bond_has_slaves(bond)) {
1865                 if (bond_dev->type != slave_dev->type) {
1866                         slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1867                                   bond_dev->type, slave_dev->type);
1868
1869                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1870                                                        bond_dev);
1871                         res = notifier_to_errno(res);
1872                         if (res) {
1873                                 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1874                                 return -EBUSY;
1875                         }
1876
1877                         /* Flush unicast and multicast addresses */
1878                         dev_uc_flush(bond_dev);
1879                         dev_mc_flush(bond_dev);
1880
1881                         if (slave_dev->type != ARPHRD_ETHER)
1882                                 bond_setup_by_slave(bond_dev, slave_dev);
1883                         else
1884                                 bond_ether_setup(bond_dev);
1885
1886                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1887                                                  bond_dev);
1888                 }
1889         } else if (bond_dev->type != slave_dev->type) {
1890                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1891                              "Device type is different from other slaves");
1892                 return -EINVAL;
1893         }
1894
1895         if (slave_dev->type == ARPHRD_INFINIBAND &&
1896             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1897                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1898                              "Only active-backup mode is supported for infiniband slaves");
1899                 res = -EOPNOTSUPP;
1900                 goto err_undo_flags;
1901         }
1902
1903         if (!slave_ops->ndo_set_mac_address ||
1904             slave_dev->type == ARPHRD_INFINIBAND) {
1905                 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1906                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1907                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1908                         if (!bond_has_slaves(bond)) {
1909                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1910                                 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1911                         } else {
1912                                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1913                                              "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1914                                 res = -EOPNOTSUPP;
1915                                 goto err_undo_flags;
1916                         }
1917                 }
1918         }
1919
1920         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1921
1922         /* If this is the first slave, then we need to set the master's hardware
1923          * address to be the same as the slave's.
1924          */
1925         if (!bond_has_slaves(bond) &&
1926             bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1927                 res = bond_set_dev_addr(bond->dev, slave_dev);
1928                 if (res)
1929                         goto err_undo_flags;
1930         }
1931
1932         new_slave = bond_alloc_slave(bond, slave_dev);
1933         if (!new_slave) {
1934                 res = -ENOMEM;
1935                 goto err_undo_flags;
1936         }
1937
1938         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1939          * is set via sysfs or module option if desired.
1940          */
1941         new_slave->queue_id = 0;
1942
1943         /* Save slave's original mtu and then set it to match the bond */
1944         new_slave->original_mtu = slave_dev->mtu;
1945         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1946         if (res) {
1947                 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1948                 goto err_free;
1949         }
1950
1951         /* Save slave's original ("permanent") mac address for modes
1952          * that need it, and for restoring it upon release, and then
1953          * set it to the master's address
1954          */
1955         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1956                           slave_dev->addr_len);
1957
1958         if (!bond->params.fail_over_mac ||
1959             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1960                 /* Set slave to master's mac address.  The application already
1961                  * set the master's mac address to that of the first slave
1962                  */
1963                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1964                 ss.ss_family = slave_dev->type;
1965                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1966                                           extack);
1967                 if (res) {
1968                         slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1969                         goto err_restore_mtu;
1970                 }
1971         }
1972
1973         /* set no_addrconf flag before open to prevent IPv6 addrconf */
1974         slave_dev->priv_flags |= IFF_NO_ADDRCONF;
1975
1976         /* open the slave since the application closed it */
1977         res = dev_open(slave_dev, extack);
1978         if (res) {
1979                 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1980                 goto err_restore_mac;
1981         }
1982
1983         slave_dev->priv_flags |= IFF_BONDING;
1984         /* initialize slave stats */
1985         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1986
1987         if (bond_is_lb(bond)) {
1988                 /* bond_alb_init_slave() must be called before all other stages since
1989                  * it might fail and we do not want to have to undo everything
1990                  */
1991                 res = bond_alb_init_slave(bond, new_slave);
1992                 if (res)
1993                         goto err_close;
1994         }
1995
1996         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1997         if (res) {
1998                 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1999                 goto err_close;
2000         }
2001
2002         prev_slave = bond_last_slave(bond);
2003
2004         new_slave->delay = 0;
2005         new_slave->link_failure_count = 0;
2006
2007         if (bond_update_speed_duplex(new_slave) &&
2008             bond_needs_speed_duplex(bond))
2009                 new_slave->link = BOND_LINK_DOWN;
2010
2011         new_slave->last_rx = jiffies -
2012                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
2013         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
2014                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
2015
2016         new_slave->last_tx = new_slave->last_rx;
2017
2018         if (bond->params.miimon && !bond->params.use_carrier) {
2019                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
2020
2021                 if ((link_reporting == -1) && !bond->params.arp_interval) {
2022                         /* miimon is set but a bonded network driver
2023                          * does not support ETHTOOL/MII and
2024                          * arp_interval is not set.  Note: if
2025                          * use_carrier is enabled, we will never go
2026                          * here (because netif_carrier is always
2027                          * supported); thus, we don't need to change
2028                          * the messages for netif_carrier.
2029                          */
2030                         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");
2031                 } else if (link_reporting == -1) {
2032                         /* unable get link status using mii/ethtool */
2033                         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");
2034                 }
2035         }
2036
2037         /* check for initial state */
2038         new_slave->link = BOND_LINK_NOCHANGE;
2039         if (bond->params.miimon) {
2040                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2041                         if (bond->params.updelay) {
2042                                 bond_set_slave_link_state(new_slave,
2043                                                           BOND_LINK_BACK,
2044                                                           BOND_SLAVE_NOTIFY_NOW);
2045                                 new_slave->delay = bond->params.updelay;
2046                         } else {
2047                                 bond_set_slave_link_state(new_slave,
2048                                                           BOND_LINK_UP,
2049                                                           BOND_SLAVE_NOTIFY_NOW);
2050                         }
2051                 } else {
2052                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2053                                                   BOND_SLAVE_NOTIFY_NOW);
2054                 }
2055         } else if (bond->params.arp_interval) {
2056                 bond_set_slave_link_state(new_slave,
2057                                           (netif_carrier_ok(slave_dev) ?
2058                                           BOND_LINK_UP : BOND_LINK_DOWN),
2059                                           BOND_SLAVE_NOTIFY_NOW);
2060         } else {
2061                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2062                                           BOND_SLAVE_NOTIFY_NOW);
2063         }
2064
2065         if (new_slave->link != BOND_LINK_DOWN)
2066                 new_slave->last_link_up = jiffies;
2067         slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2068                   new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2069                   (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2070
2071         if (bond_uses_primary(bond) && bond->params.primary[0]) {
2072                 /* if there is a primary slave, remember it */
2073                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2074                         rcu_assign_pointer(bond->primary_slave, new_slave);
2075                         bond->force_primary = true;
2076                 }
2077         }
2078
2079         switch (BOND_MODE(bond)) {
2080         case BOND_MODE_ACTIVEBACKUP:
2081                 bond_set_slave_inactive_flags(new_slave,
2082                                               BOND_SLAVE_NOTIFY_NOW);
2083                 break;
2084         case BOND_MODE_8023AD:
2085                 /* in 802.3ad mode, the internal mechanism
2086                  * will activate the slaves in the selected
2087                  * aggregator
2088                  */
2089                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2090                 /* if this is the first slave */
2091                 if (!prev_slave) {
2092                         SLAVE_AD_INFO(new_slave)->id = 1;
2093                         /* Initialize AD with the number of times that the AD timer is called in 1 second
2094                          * can be called only after the mac address of the bond is set
2095                          */
2096                         bond_3ad_initialize(bond);
2097                 } else {
2098                         SLAVE_AD_INFO(new_slave)->id =
2099                                 SLAVE_AD_INFO(prev_slave)->id + 1;
2100                 }
2101
2102                 bond_3ad_bind_slave(new_slave);
2103                 break;
2104         case BOND_MODE_TLB:
2105         case BOND_MODE_ALB:
2106                 bond_set_active_slave(new_slave);
2107                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2108                 break;
2109         default:
2110                 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2111
2112                 /* always active in trunk mode */
2113                 bond_set_active_slave(new_slave);
2114
2115                 /* In trunking mode there is little meaning to curr_active_slave
2116                  * anyway (it holds no special properties of the bond device),
2117                  * so we can change it without calling change_active_interface()
2118                  */
2119                 if (!rcu_access_pointer(bond->curr_active_slave) &&
2120                     new_slave->link == BOND_LINK_UP)
2121                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
2122
2123                 break;
2124         } /* switch(bond_mode) */
2125
2126 #ifdef CONFIG_NET_POLL_CONTROLLER
2127         if (bond->dev->npinfo) {
2128                 if (slave_enable_netpoll(new_slave)) {
2129                         slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2130                         res = -EBUSY;
2131                         goto err_detach;
2132                 }
2133         }
2134 #endif
2135
2136         if (!(bond_dev->features & NETIF_F_LRO))
2137                 dev_disable_lro(slave_dev);
2138
2139         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2140                                          new_slave);
2141         if (res) {
2142                 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2143                 goto err_detach;
2144         }
2145
2146         res = bond_master_upper_dev_link(bond, new_slave, extack);
2147         if (res) {
2148                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2149                 goto err_unregister;
2150         }
2151
2152         bond_lower_state_changed(new_slave);
2153
2154         res = bond_sysfs_slave_add(new_slave);
2155         if (res) {
2156                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2157                 goto err_upper_unlink;
2158         }
2159
2160         /* If the mode uses primary, then the following is handled by
2161          * bond_change_active_slave().
2162          */
2163         if (!bond_uses_primary(bond)) {
2164                 /* set promiscuity level to new slave */
2165                 if (bond_dev->flags & IFF_PROMISC) {
2166                         res = dev_set_promiscuity(slave_dev, 1);
2167                         if (res)
2168                                 goto err_sysfs_del;
2169                 }
2170
2171                 /* set allmulti level to new slave */
2172                 if (bond_dev->flags & IFF_ALLMULTI) {
2173                         res = dev_set_allmulti(slave_dev, 1);
2174                         if (res) {
2175                                 if (bond_dev->flags & IFF_PROMISC)
2176                                         dev_set_promiscuity(slave_dev, -1);
2177                                 goto err_sysfs_del;
2178                         }
2179                 }
2180
2181                 if (bond_dev->flags & IFF_UP) {
2182                         netif_addr_lock_bh(bond_dev);
2183                         dev_mc_sync_multiple(slave_dev, bond_dev);
2184                         dev_uc_sync_multiple(slave_dev, bond_dev);
2185                         netif_addr_unlock_bh(bond_dev);
2186
2187                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2188                                 dev_mc_add(slave_dev, lacpdu_mcast_addr);
2189                 }
2190         }
2191
2192         bond->slave_cnt++;
2193         bond_compute_features(bond);
2194         bond_set_carrier(bond);
2195
2196         if (bond_uses_primary(bond)) {
2197                 block_netpoll_tx();
2198                 bond_select_active_slave(bond);
2199                 unblock_netpoll_tx();
2200         }
2201
2202         if (bond_mode_can_use_xmit_hash(bond))
2203                 bond_update_slave_arr(bond, NULL);
2204
2205
2206         if (!slave_dev->netdev_ops->ndo_bpf ||
2207             !slave_dev->netdev_ops->ndo_xdp_xmit) {
2208                 if (bond->xdp_prog) {
2209                         SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2210                                      "Slave does not support XDP");
2211                         res = -EOPNOTSUPP;
2212                         goto err_sysfs_del;
2213                 }
2214         } else if (bond->xdp_prog) {
2215                 struct netdev_bpf xdp = {
2216                         .command = XDP_SETUP_PROG,
2217                         .flags   = 0,
2218                         .prog    = bond->xdp_prog,
2219                         .extack  = extack,
2220                 };
2221
2222                 if (dev_xdp_prog_count(slave_dev) > 0) {
2223                         SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2224                                      "Slave has XDP program loaded, please unload before enslaving");
2225                         res = -EOPNOTSUPP;
2226                         goto err_sysfs_del;
2227                 }
2228
2229                 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2230                 if (res < 0) {
2231                         /* ndo_bpf() sets extack error message */
2232                         slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2233                         goto err_sysfs_del;
2234                 }
2235                 if (bond->xdp_prog)
2236                         bpf_prog_inc(bond->xdp_prog);
2237         }
2238
2239         slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2240                    bond_is_active_slave(new_slave) ? "an active" : "a backup",
2241                    new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2242
2243         /* enslave is successful */
2244         bond_queue_slave_event(new_slave);
2245         return 0;
2246
2247 /* Undo stages on error */
2248 err_sysfs_del:
2249         bond_sysfs_slave_del(new_slave);
2250
2251 err_upper_unlink:
2252         bond_upper_dev_unlink(bond, new_slave);
2253
2254 err_unregister:
2255         netdev_rx_handler_unregister(slave_dev);
2256
2257 err_detach:
2258         vlan_vids_del_by_dev(slave_dev, bond_dev);
2259         if (rcu_access_pointer(bond->primary_slave) == new_slave)
2260                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2261         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2262                 block_netpoll_tx();
2263                 bond_change_active_slave(bond, NULL);
2264                 bond_select_active_slave(bond);
2265                 unblock_netpoll_tx();
2266         }
2267         /* either primary_slave or curr_active_slave might've changed */
2268         synchronize_rcu();
2269         slave_disable_netpoll(new_slave);
2270
2271 err_close:
2272         if (!netif_is_bond_master(slave_dev))
2273                 slave_dev->priv_flags &= ~IFF_BONDING;
2274         dev_close(slave_dev);
2275
2276 err_restore_mac:
2277         slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2278         if (!bond->params.fail_over_mac ||
2279             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2280                 /* XXX TODO - fom follow mode needs to change master's
2281                  * MAC if this slave's MAC is in use by the bond, or at
2282                  * least print a warning.
2283                  */
2284                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2285                                   new_slave->dev->addr_len);
2286                 ss.ss_family = slave_dev->type;
2287                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2288         }
2289
2290 err_restore_mtu:
2291         dev_set_mtu(slave_dev, new_slave->original_mtu);
2292
2293 err_free:
2294         kobject_put(&new_slave->kobj);
2295
2296 err_undo_flags:
2297         /* Enslave of first slave has failed and we need to fix master's mac */
2298         if (!bond_has_slaves(bond)) {
2299                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2300                                             slave_dev->dev_addr))
2301                         eth_hw_addr_random(bond_dev);
2302                 if (bond_dev->type != ARPHRD_ETHER) {
2303                         dev_close(bond_dev);
2304                         bond_ether_setup(bond_dev);
2305                 }
2306         }
2307
2308         return res;
2309 }
2310
2311 /* Try to release the slave device <slave> from the bond device <master>
2312  * It is legal to access curr_active_slave without a lock because all the function
2313  * is RTNL-locked. If "all" is true it means that the function is being called
2314  * while destroying a bond interface and all slaves are being released.
2315  *
2316  * The rules for slave state should be:
2317  *   for Active/Backup:
2318  *     Active stays on all backups go down
2319  *   for Bonded connections:
2320  *     The first up interface should be left on and all others downed.
2321  */
2322 static int __bond_release_one(struct net_device *bond_dev,
2323                               struct net_device *slave_dev,
2324                               bool all, bool unregister)
2325 {
2326         struct bonding *bond = netdev_priv(bond_dev);
2327         struct slave *slave, *oldcurrent;
2328         struct sockaddr_storage ss;
2329         int old_flags = bond_dev->flags;
2330         netdev_features_t old_features = bond_dev->features;
2331
2332         /* slave is not a slave or master is not master of this slave */
2333         if (!(slave_dev->flags & IFF_SLAVE) ||
2334             !netdev_has_upper_dev(slave_dev, bond_dev)) {
2335                 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2336                 return -EINVAL;
2337         }
2338
2339         block_netpoll_tx();
2340
2341         slave = bond_get_slave_by_dev(bond, slave_dev);
2342         if (!slave) {
2343                 /* not a slave of this bond */
2344                 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2345                 unblock_netpoll_tx();
2346                 return -EINVAL;
2347         }
2348
2349         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2350
2351         bond_sysfs_slave_del(slave);
2352
2353         /* recompute stats just before removing the slave */
2354         bond_get_stats(bond->dev, &bond->bond_stats);
2355
2356         if (bond->xdp_prog) {
2357                 struct netdev_bpf xdp = {
2358                         .command = XDP_SETUP_PROG,
2359                         .flags   = 0,
2360                         .prog    = NULL,
2361                         .extack  = NULL,
2362                 };
2363                 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2364                         slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2365         }
2366
2367         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2368          * for this slave anymore.
2369          */
2370         netdev_rx_handler_unregister(slave_dev);
2371
2372         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2373                 bond_3ad_unbind_slave(slave);
2374
2375         bond_upper_dev_unlink(bond, slave);
2376
2377         if (bond_mode_can_use_xmit_hash(bond))
2378                 bond_update_slave_arr(bond, slave);
2379
2380         slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2381                     bond_is_active_slave(slave) ? "active" : "backup");
2382
2383         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2384
2385         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2386
2387         if (!all && (!bond->params.fail_over_mac ||
2388                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2389                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2390                     bond_has_slaves(bond))
2391                         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",
2392                                    slave->perm_hwaddr);
2393         }
2394
2395         if (rtnl_dereference(bond->primary_slave) == slave)
2396                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2397
2398         if (oldcurrent == slave)
2399                 bond_change_active_slave(bond, NULL);
2400
2401         if (bond_is_lb(bond)) {
2402                 /* Must be called only after the slave has been
2403                  * detached from the list and the curr_active_slave
2404                  * has been cleared (if our_slave == old_current),
2405                  * but before a new active slave is selected.
2406                  */
2407                 bond_alb_deinit_slave(bond, slave);
2408         }
2409
2410         if (all) {
2411                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2412         } else if (oldcurrent == slave) {
2413                 /* Note that we hold RTNL over this sequence, so there
2414                  * is no concern that another slave add/remove event
2415                  * will interfere.
2416                  */
2417                 bond_select_active_slave(bond);
2418         }
2419
2420         bond_set_carrier(bond);
2421         if (!bond_has_slaves(bond))
2422                 eth_hw_addr_random(bond_dev);
2423
2424         unblock_netpoll_tx();
2425         synchronize_rcu();
2426         bond->slave_cnt--;
2427
2428         if (!bond_has_slaves(bond)) {
2429                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2430                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2431         }
2432
2433         bond_compute_features(bond);
2434         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2435             (old_features & NETIF_F_VLAN_CHALLENGED))
2436                 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2437
2438         vlan_vids_del_by_dev(slave_dev, bond_dev);
2439
2440         /* If the mode uses primary, then this case was handled above by
2441          * bond_change_active_slave(..., NULL)
2442          */
2443         if (!bond_uses_primary(bond)) {
2444                 /* unset promiscuity level from slave
2445                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
2446                  * of the IFF_PROMISC flag in the bond_dev, but we need the
2447                  * value of that flag before that change, as that was the value
2448                  * when this slave was attached, so we cache at the start of the
2449                  * function and use it here. Same goes for ALLMULTI below
2450                  */
2451                 if (old_flags & IFF_PROMISC)
2452                         dev_set_promiscuity(slave_dev, -1);
2453
2454                 /* unset allmulti level from slave */
2455                 if (old_flags & IFF_ALLMULTI)
2456                         dev_set_allmulti(slave_dev, -1);
2457
2458                 if (old_flags & IFF_UP)
2459                         bond_hw_addr_flush(bond_dev, slave_dev);
2460         }
2461
2462         slave_disable_netpoll(slave);
2463
2464         /* close slave before restoring its mac address */
2465         dev_close(slave_dev);
2466
2467         slave_dev->priv_flags &= ~IFF_NO_ADDRCONF;
2468
2469         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2470             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2471                 /* restore original ("permanent") mac address */
2472                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2473                                   slave->dev->addr_len);
2474                 ss.ss_family = slave_dev->type;
2475                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2476         }
2477
2478         if (unregister)
2479                 __dev_set_mtu(slave_dev, slave->original_mtu);
2480         else
2481                 dev_set_mtu(slave_dev, slave->original_mtu);
2482
2483         if (!netif_is_bond_master(slave_dev))
2484                 slave_dev->priv_flags &= ~IFF_BONDING;
2485
2486         kobject_put(&slave->kobj);
2487
2488         return 0;
2489 }
2490
2491 /* A wrapper used because of ndo_del_link */
2492 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2493 {
2494         return __bond_release_one(bond_dev, slave_dev, false, false);
2495 }
2496
2497 /* First release a slave and then destroy the bond if no more slaves are left.
2498  * Must be under rtnl_lock when this function is called.
2499  */
2500 static int bond_release_and_destroy(struct net_device *bond_dev,
2501                                     struct net_device *slave_dev)
2502 {
2503         struct bonding *bond = netdev_priv(bond_dev);
2504         int ret;
2505
2506         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2507         if (ret == 0 && !bond_has_slaves(bond) &&
2508             bond_dev->reg_state != NETREG_UNREGISTERING) {
2509                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2510                 netdev_info(bond_dev, "Destroying bond\n");
2511                 bond_remove_proc_entry(bond);
2512                 unregister_netdevice(bond_dev);
2513         }
2514         return ret;
2515 }
2516
2517 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2518 {
2519         struct bonding *bond = netdev_priv(bond_dev);
2520
2521         bond_fill_ifbond(bond, info);
2522 }
2523
2524 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2525 {
2526         struct bonding *bond = netdev_priv(bond_dev);
2527         struct list_head *iter;
2528         int i = 0, res = -ENODEV;
2529         struct slave *slave;
2530
2531         bond_for_each_slave(bond, slave, iter) {
2532                 if (i++ == (int)info->slave_id) {
2533                         res = 0;
2534                         bond_fill_ifslave(slave, info);
2535                         break;
2536                 }
2537         }
2538
2539         return res;
2540 }
2541
2542 /*-------------------------------- Monitoring -------------------------------*/
2543
2544 /* called with rcu_read_lock() */
2545 static int bond_miimon_inspect(struct bonding *bond)
2546 {
2547         bool ignore_updelay = false;
2548         int link_state, commit = 0;
2549         struct list_head *iter;
2550         struct slave *slave;
2551
2552         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
2553                 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2554         } else {
2555                 struct bond_up_slave *usable_slaves;
2556
2557                 usable_slaves = rcu_dereference(bond->usable_slaves);
2558
2559                 if (usable_slaves && usable_slaves->count == 0)
2560                         ignore_updelay = true;
2561         }
2562
2563         bond_for_each_slave_rcu(bond, slave, iter) {
2564                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2565
2566                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2567
2568                 switch (slave->link) {
2569                 case BOND_LINK_UP:
2570                         if (link_state)
2571                                 continue;
2572
2573                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2574                         commit++;
2575                         slave->delay = bond->params.downdelay;
2576                         if (slave->delay) {
2577                                 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2578                                            (BOND_MODE(bond) ==
2579                                             BOND_MODE_ACTIVEBACKUP) ?
2580                                             (bond_is_active_slave(slave) ?
2581                                              "active " : "backup ") : "",
2582                                            bond->params.downdelay * bond->params.miimon);
2583                         }
2584                         fallthrough;
2585                 case BOND_LINK_FAIL:
2586                         if (link_state) {
2587                                 /* recovered before downdelay expired */
2588                                 bond_propose_link_state(slave, BOND_LINK_UP);
2589                                 slave->last_link_up = jiffies;
2590                                 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2591                                            (bond->params.downdelay - slave->delay) *
2592                                            bond->params.miimon);
2593                                 commit++;
2594                                 continue;
2595                         }
2596
2597                         if (slave->delay <= 0) {
2598                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2599                                 commit++;
2600                                 continue;
2601                         }
2602
2603                         slave->delay--;
2604                         break;
2605
2606                 case BOND_LINK_DOWN:
2607                         if (!link_state)
2608                                 continue;
2609
2610                         bond_propose_link_state(slave, BOND_LINK_BACK);
2611                         commit++;
2612                         slave->delay = bond->params.updelay;
2613
2614                         if (slave->delay) {
2615                                 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2616                                            ignore_updelay ? 0 :
2617                                            bond->params.updelay *
2618                                            bond->params.miimon);
2619                         }
2620                         fallthrough;
2621                 case BOND_LINK_BACK:
2622                         if (!link_state) {
2623                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2624                                 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2625                                            (bond->params.updelay - slave->delay) *
2626                                            bond->params.miimon);
2627                                 commit++;
2628                                 continue;
2629                         }
2630
2631                         if (ignore_updelay)
2632                                 slave->delay = 0;
2633
2634                         if (slave->delay <= 0) {
2635                                 bond_propose_link_state(slave, BOND_LINK_UP);
2636                                 commit++;
2637                                 ignore_updelay = false;
2638                                 continue;
2639                         }
2640
2641                         slave->delay--;
2642                         break;
2643                 }
2644         }
2645
2646         return commit;
2647 }
2648
2649 static void bond_miimon_link_change(struct bonding *bond,
2650                                     struct slave *slave,
2651                                     char link)
2652 {
2653         switch (BOND_MODE(bond)) {
2654         case BOND_MODE_8023AD:
2655                 bond_3ad_handle_link_change(slave, link);
2656                 break;
2657         case BOND_MODE_TLB:
2658         case BOND_MODE_ALB:
2659                 bond_alb_handle_link_change(bond, slave, link);
2660                 break;
2661         case BOND_MODE_XOR:
2662                 bond_update_slave_arr(bond, NULL);
2663                 break;
2664         }
2665 }
2666
2667 static void bond_miimon_commit(struct bonding *bond)
2668 {
2669         struct slave *slave, *primary, *active;
2670         bool do_failover = false;
2671         struct list_head *iter;
2672
2673         ASSERT_RTNL();
2674
2675         bond_for_each_slave(bond, slave, iter) {
2676                 switch (slave->link_new_state) {
2677                 case BOND_LINK_NOCHANGE:
2678                         /* For 802.3ad mode, check current slave speed and
2679                          * duplex again in case its port was disabled after
2680                          * invalid speed/duplex reporting but recovered before
2681                          * link monitoring could make a decision on the actual
2682                          * link status
2683                          */
2684                         if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2685                             slave->link == BOND_LINK_UP)
2686                                 bond_3ad_adapter_speed_duplex_changed(slave);
2687                         continue;
2688
2689                 case BOND_LINK_UP:
2690                         if (bond_update_speed_duplex(slave) &&
2691                             bond_needs_speed_duplex(bond)) {
2692                                 slave->link = BOND_LINK_DOWN;
2693                                 if (net_ratelimit())
2694                                         slave_warn(bond->dev, slave->dev,
2695                                                    "failed to get link speed/duplex\n");
2696                                 continue;
2697                         }
2698                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2699                                                   BOND_SLAVE_NOTIFY_NOW);
2700                         slave->last_link_up = jiffies;
2701
2702                         primary = rtnl_dereference(bond->primary_slave);
2703                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2704                                 /* prevent it from being the active one */
2705                                 bond_set_backup_slave(slave);
2706                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2707                                 /* make it immediately active */
2708                                 bond_set_active_slave(slave);
2709                         }
2710
2711                         slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2712                                    slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2713                                    slave->duplex ? "full" : "half");
2714
2715                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2716
2717                         active = rtnl_dereference(bond->curr_active_slave);
2718                         if (!active || slave == primary || slave->prio > active->prio)
2719                                 do_failover = true;
2720
2721                         continue;
2722
2723                 case BOND_LINK_DOWN:
2724                         if (slave->link_failure_count < UINT_MAX)
2725                                 slave->link_failure_count++;
2726
2727                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2728                                                   BOND_SLAVE_NOTIFY_NOW);
2729
2730                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2731                             BOND_MODE(bond) == BOND_MODE_8023AD)
2732                                 bond_set_slave_inactive_flags(slave,
2733                                                               BOND_SLAVE_NOTIFY_NOW);
2734
2735                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2736
2737                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2738
2739                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2740                                 do_failover = true;
2741
2742                         continue;
2743
2744                 default:
2745                         slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2746                                   slave->link_new_state);
2747                         bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2748
2749                         continue;
2750                 }
2751         }
2752
2753         if (do_failover) {
2754                 block_netpoll_tx();
2755                 bond_select_active_slave(bond);
2756                 unblock_netpoll_tx();
2757         }
2758
2759         bond_set_carrier(bond);
2760 }
2761
2762 /* bond_mii_monitor
2763  *
2764  * Really a wrapper that splits the mii monitor into two phases: an
2765  * inspection, then (if inspection indicates something needs to be done)
2766  * an acquisition of appropriate locks followed by a commit phase to
2767  * implement whatever link state changes are indicated.
2768  */
2769 static void bond_mii_monitor(struct work_struct *work)
2770 {
2771         struct bonding *bond = container_of(work, struct bonding,
2772                                             mii_work.work);
2773         bool should_notify_peers = false;
2774         bool commit;
2775         unsigned long delay;
2776         struct slave *slave;
2777         struct list_head *iter;
2778
2779         delay = msecs_to_jiffies(bond->params.miimon);
2780
2781         if (!bond_has_slaves(bond))
2782                 goto re_arm;
2783
2784         rcu_read_lock();
2785         should_notify_peers = bond_should_notify_peers(bond);
2786         commit = !!bond_miimon_inspect(bond);
2787         if (bond->send_peer_notif) {
2788                 rcu_read_unlock();
2789                 if (rtnl_trylock()) {
2790                         bond->send_peer_notif--;
2791                         rtnl_unlock();
2792                 }
2793         } else {
2794                 rcu_read_unlock();
2795         }
2796
2797         if (commit) {
2798                 /* Race avoidance with bond_close cancel of workqueue */
2799                 if (!rtnl_trylock()) {
2800                         delay = 1;
2801                         should_notify_peers = false;
2802                         goto re_arm;
2803                 }
2804
2805                 bond_for_each_slave(bond, slave, iter) {
2806                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2807                 }
2808                 bond_miimon_commit(bond);
2809
2810                 rtnl_unlock();  /* might sleep, hold no other locks */
2811         }
2812
2813 re_arm:
2814         if (bond->params.miimon)
2815                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2816
2817         if (should_notify_peers) {
2818                 if (!rtnl_trylock())
2819                         return;
2820                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2821                 rtnl_unlock();
2822         }
2823 }
2824
2825 static int bond_upper_dev_walk(struct net_device *upper,
2826                                struct netdev_nested_priv *priv)
2827 {
2828         __be32 ip = *(__be32 *)priv->data;
2829
2830         return ip == bond_confirm_addr(upper, 0, ip);
2831 }
2832
2833 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2834 {
2835         struct netdev_nested_priv priv = {
2836                 .data = (void *)&ip,
2837         };
2838         bool ret = false;
2839
2840         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2841                 return true;
2842
2843         rcu_read_lock();
2844         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2845                 ret = true;
2846         rcu_read_unlock();
2847
2848         return ret;
2849 }
2850
2851 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2852                              struct sk_buff *skb)
2853 {
2854         struct net_device *bond_dev = slave->bond->dev;
2855         struct net_device *slave_dev = slave->dev;
2856         struct bond_vlan_tag *outer_tag = tags;
2857
2858         if (!tags || tags->vlan_proto == VLAN_N_VID)
2859                 return true;
2860
2861         tags++;
2862
2863         /* Go through all the tags backwards and add them to the packet */
2864         while (tags->vlan_proto != VLAN_N_VID) {
2865                 if (!tags->vlan_id) {
2866                         tags++;
2867                         continue;
2868                 }
2869
2870                 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2871                           ntohs(outer_tag->vlan_proto), tags->vlan_id);
2872                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2873                                                 tags->vlan_id);
2874                 if (!skb) {
2875                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2876                         return false;
2877                 }
2878
2879                 tags++;
2880         }
2881         /* Set the outer tag */
2882         if (outer_tag->vlan_id) {
2883                 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2884                           ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2885                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2886                                        outer_tag->vlan_id);
2887         }
2888
2889         return true;
2890 }
2891
2892 /* We go to the (large) trouble of VLAN tagging ARP frames because
2893  * switches in VLAN mode (especially if ports are configured as
2894  * "native" to a VLAN) might not pass non-tagged frames.
2895  */
2896 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2897                           __be32 src_ip, struct bond_vlan_tag *tags)
2898 {
2899         struct net_device *bond_dev = slave->bond->dev;
2900         struct net_device *slave_dev = slave->dev;
2901         struct sk_buff *skb;
2902
2903         slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2904                   arp_op, &dest_ip, &src_ip);
2905
2906         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2907                          NULL, slave_dev->dev_addr, NULL);
2908
2909         if (!skb) {
2910                 net_err_ratelimited("ARP packet allocation failed\n");
2911                 return;
2912         }
2913
2914         if (bond_handle_vlan(slave, tags, skb)) {
2915                 slave_update_last_tx(slave);
2916                 arp_xmit(skb);
2917         }
2918
2919         return;
2920 }
2921
2922 /* Validate the device path between the @start_dev and the @end_dev.
2923  * The path is valid if the @end_dev is reachable through device
2924  * stacking.
2925  * When the path is validated, collect any vlan information in the
2926  * path.
2927  */
2928 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2929                                               struct net_device *end_dev,
2930                                               int level)
2931 {
2932         struct bond_vlan_tag *tags;
2933         struct net_device *upper;
2934         struct list_head  *iter;
2935
2936         if (start_dev == end_dev) {
2937                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2938                 if (!tags)
2939                         return ERR_PTR(-ENOMEM);
2940                 tags[level].vlan_proto = VLAN_N_VID;
2941                 return tags;
2942         }
2943
2944         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2945                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2946                 if (IS_ERR_OR_NULL(tags)) {
2947                         if (IS_ERR(tags))
2948                                 return tags;
2949                         continue;
2950                 }
2951                 if (is_vlan_dev(upper)) {
2952                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2953                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2954                 }
2955
2956                 return tags;
2957         }
2958
2959         return NULL;
2960 }
2961
2962 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2963 {
2964         struct rtable *rt;
2965         struct bond_vlan_tag *tags;
2966         __be32 *targets = bond->params.arp_targets, addr;
2967         int i;
2968
2969         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2970                 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2971                           __func__, &targets[i]);
2972                 tags = NULL;
2973
2974                 /* Find out through which dev should the packet go */
2975                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2976                                      RTO_ONLINK, 0);
2977                 if (IS_ERR(rt)) {
2978                         /* there's no route to target - try to send arp
2979                          * probe to generate any traffic (arp_validate=0)
2980                          */
2981                         if (bond->params.arp_validate)
2982                                 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2983                                              bond->dev->name,
2984                                              &targets[i]);
2985                         bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2986                                       0, tags);
2987                         continue;
2988                 }
2989
2990                 /* bond device itself */
2991                 if (rt->dst.dev == bond->dev)
2992                         goto found;
2993
2994                 rcu_read_lock();
2995                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2996                 rcu_read_unlock();
2997
2998                 if (!IS_ERR_OR_NULL(tags))
2999                         goto found;
3000
3001                 /* Not our device - skip */
3002                 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
3003                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
3004
3005                 ip_rt_put(rt);
3006                 continue;
3007
3008 found:
3009                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
3010                 ip_rt_put(rt);
3011                 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
3012                 kfree(tags);
3013         }
3014 }
3015
3016 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
3017 {
3018         int i;
3019
3020         if (!sip || !bond_has_this_ip(bond, tip)) {
3021                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
3022                            __func__, &sip, &tip);
3023                 return;
3024         }
3025
3026         i = bond_get_targets_ip(bond->params.arp_targets, sip);
3027         if (i == -1) {
3028                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
3029                            __func__, &sip);
3030                 return;
3031         }
3032         slave->last_rx = jiffies;
3033         slave->target_last_arp_rx[i] = jiffies;
3034 }
3035
3036 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
3037                         struct slave *slave)
3038 {
3039         struct arphdr *arp = (struct arphdr *)skb->data;
3040         struct slave *curr_active_slave, *curr_arp_slave;
3041         unsigned char *arp_ptr;
3042         __be32 sip, tip;
3043         unsigned int alen;
3044
3045         alen = arp_hdr_len(bond->dev);
3046
3047         if (alen > skb_headlen(skb)) {
3048                 arp = kmalloc(alen, GFP_ATOMIC);
3049                 if (!arp)
3050                         goto out_unlock;
3051                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
3052                         goto out_unlock;
3053         }
3054
3055         if (arp->ar_hln != bond->dev->addr_len ||
3056             skb->pkt_type == PACKET_OTHERHOST ||
3057             skb->pkt_type == PACKET_LOOPBACK ||
3058             arp->ar_hrd != htons(ARPHRD_ETHER) ||
3059             arp->ar_pro != htons(ETH_P_IP) ||
3060             arp->ar_pln != 4)
3061                 goto out_unlock;
3062
3063         arp_ptr = (unsigned char *)(arp + 1);
3064         arp_ptr += bond->dev->addr_len;
3065         memcpy(&sip, arp_ptr, 4);
3066         arp_ptr += 4 + bond->dev->addr_len;
3067         memcpy(&tip, arp_ptr, 4);
3068
3069         slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3070                   __func__, slave->dev->name, bond_slave_state(slave),
3071                   bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3072                   &sip, &tip);
3073
3074         curr_active_slave = rcu_dereference(bond->curr_active_slave);
3075         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3076
3077         /* We 'trust' the received ARP enough to validate it if:
3078          *
3079          * (a) the slave receiving the ARP is active (which includes the
3080          * current ARP slave, if any), or
3081          *
3082          * (b) the receiving slave isn't active, but there is a currently
3083          * active slave and it received valid arp reply(s) after it became
3084          * the currently active slave, or
3085          *
3086          * (c) there is an ARP slave that sent an ARP during the prior ARP
3087          * interval, and we receive an ARP reply on any slave.  We accept
3088          * these because switch FDB update delays may deliver the ARP
3089          * reply to a slave other than the sender of the ARP request.
3090          *
3091          * Note: for (b), backup slaves are receiving the broadcast ARP
3092          * request, not a reply.  This request passes from the sending
3093          * slave through the L2 switch(es) to the receiving slave.  Since
3094          * this is checking the request, sip/tip are swapped for
3095          * validation.
3096          *
3097          * This is done to avoid endless looping when we can't reach the
3098          * arp_ip_target and fool ourselves with our own arp requests.
3099          */
3100         if (bond_is_active_slave(slave))
3101                 bond_validate_arp(bond, slave, sip, tip);
3102         else if (curr_active_slave &&
3103                  time_after(slave_last_rx(bond, curr_active_slave),
3104                             curr_active_slave->last_link_up))
3105                 bond_validate_arp(bond, slave, tip, sip);
3106         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3107                  bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3108                 bond_validate_arp(bond, slave, sip, tip);
3109
3110 out_unlock:
3111         if (arp != (struct arphdr *)skb->data)
3112                 kfree(arp);
3113         return RX_HANDLER_ANOTHER;
3114 }
3115
3116 #if IS_ENABLED(CONFIG_IPV6)
3117 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3118                          const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3119 {
3120         struct net_device *bond_dev = slave->bond->dev;
3121         struct net_device *slave_dev = slave->dev;
3122         struct in6_addr mcaddr;
3123         struct sk_buff *skb;
3124
3125         slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3126                   daddr, saddr);
3127
3128         skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3129         if (!skb) {
3130                 net_err_ratelimited("NS packet allocation failed\n");
3131                 return;
3132         }
3133
3134         addrconf_addr_solict_mult(daddr, &mcaddr);
3135         if (bond_handle_vlan(slave, tags, skb)) {
3136                 slave_update_last_tx(slave);
3137                 ndisc_send_skb(skb, &mcaddr, saddr);
3138         }
3139 }
3140
3141 static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3142 {
3143         struct in6_addr *targets = bond->params.ns_targets;
3144         struct bond_vlan_tag *tags;
3145         struct dst_entry *dst;
3146         struct in6_addr saddr;
3147         struct flowi6 fl6;
3148         int i;
3149
3150         for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3151                 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3152                           __func__, &targets[i]);
3153                 tags = NULL;
3154
3155                 /* Find out through which dev should the packet go */
3156                 memset(&fl6, 0, sizeof(struct flowi6));
3157                 fl6.daddr = targets[i];
3158                 fl6.flowi6_oif = bond->dev->ifindex;
3159
3160                 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3161                 if (dst->error) {
3162                         dst_release(dst);
3163                         /* there's no route to target - try to send arp
3164                          * probe to generate any traffic (arp_validate=0)
3165                          */
3166                         if (bond->params.arp_validate)
3167                                 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3168                                              bond->dev->name,
3169                                              &targets[i]);
3170                         bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3171                         continue;
3172                 }
3173
3174                 /* bond device itself */
3175                 if (dst->dev == bond->dev)
3176                         goto found;
3177
3178                 rcu_read_lock();
3179                 tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3180                 rcu_read_unlock();
3181
3182                 if (!IS_ERR_OR_NULL(tags))
3183                         goto found;
3184
3185                 /* Not our device - skip */
3186                 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3187                           &targets[i], dst->dev ? dst->dev->name : "NULL");
3188
3189                 dst_release(dst);
3190                 continue;
3191
3192 found:
3193                 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3194                         bond_ns_send(slave, &targets[i], &saddr, tags);
3195                 else
3196                         bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3197
3198                 dst_release(dst);
3199                 kfree(tags);
3200         }
3201 }
3202
3203 static int bond_confirm_addr6(struct net_device *dev,
3204                               struct netdev_nested_priv *priv)
3205 {
3206         struct in6_addr *addr = (struct in6_addr *)priv->data;
3207
3208         return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3209 }
3210
3211 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3212 {
3213         struct netdev_nested_priv priv = {
3214                 .data = addr,
3215         };
3216         int ret = false;
3217
3218         if (bond_confirm_addr6(bond->dev, &priv))
3219                 return true;
3220
3221         rcu_read_lock();
3222         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3223                 ret = true;
3224         rcu_read_unlock();
3225
3226         return ret;
3227 }
3228
3229 static void bond_validate_na(struct bonding *bond, struct slave *slave,
3230                              struct in6_addr *saddr, struct in6_addr *daddr)
3231 {
3232         int i;
3233
3234         /* Ignore NAs that:
3235          * 1. Source address is unspecified address.
3236          * 2. Dest address is neither all-nodes multicast address nor
3237          *    exist on bond interface.
3238          */
3239         if (ipv6_addr_any(saddr) ||
3240             (!ipv6_addr_equal(daddr, &in6addr_linklocal_allnodes) &&
3241              !bond_has_this_ip6(bond, daddr))) {
3242                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3243                           __func__, saddr, daddr);
3244                 return;
3245         }
3246
3247         i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3248         if (i == -1) {
3249                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3250                           __func__, saddr);
3251                 return;
3252         }
3253         slave->last_rx = jiffies;
3254         slave->target_last_arp_rx[i] = jiffies;
3255 }
3256
3257 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3258                        struct slave *slave)
3259 {
3260         struct slave *curr_active_slave, *curr_arp_slave;
3261         struct in6_addr *saddr, *daddr;
3262         struct {
3263                 struct ipv6hdr ip6;
3264                 struct icmp6hdr icmp6;
3265         } *combined, _combined;
3266
3267         if (skb->pkt_type == PACKET_OTHERHOST ||
3268             skb->pkt_type == PACKET_LOOPBACK)
3269                 goto out;
3270
3271         combined = skb_header_pointer(skb, 0, sizeof(_combined), &_combined);
3272         if (!combined || combined->ip6.nexthdr != NEXTHDR_ICMP ||
3273             (combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_SOLICITATION &&
3274              combined->icmp6.icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT))
3275                 goto out;
3276
3277         saddr = &combined->ip6.saddr;
3278         daddr = &combined->ip6.daddr;
3279
3280         slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3281                   __func__, slave->dev->name, bond_slave_state(slave),
3282                   bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3283                   saddr, daddr);
3284
3285         curr_active_slave = rcu_dereference(bond->curr_active_slave);
3286         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3287
3288         /* We 'trust' the received ARP enough to validate it if:
3289          * see bond_arp_rcv().
3290          */
3291         if (bond_is_active_slave(slave))
3292                 bond_validate_na(bond, slave, saddr, daddr);
3293         else if (curr_active_slave &&
3294                  time_after(slave_last_rx(bond, curr_active_slave),
3295                             curr_active_slave->last_link_up))
3296                 bond_validate_na(bond, slave, daddr, saddr);
3297         else if (curr_arp_slave &&
3298                  bond_time_in_interval(bond, slave_last_tx(curr_arp_slave), 1))
3299                 bond_validate_na(bond, slave, saddr, daddr);
3300
3301 out:
3302         return RX_HANDLER_ANOTHER;
3303 }
3304 #endif
3305
3306 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3307                       struct slave *slave)
3308 {
3309 #if IS_ENABLED(CONFIG_IPV6)
3310         bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3311 #endif
3312         bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3313
3314         slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3315                   __func__, skb->dev->name);
3316
3317         /* Use arp validate logic for both ARP and NS */
3318         if (!slave_do_arp_validate(bond, slave)) {
3319                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
3320 #if IS_ENABLED(CONFIG_IPV6)
3321                     (slave_do_arp_validate_only(bond) && is_ipv6) ||
3322 #endif
3323                     !slave_do_arp_validate_only(bond))
3324                         slave->last_rx = jiffies;
3325                 return RX_HANDLER_ANOTHER;
3326         } else if (is_arp) {
3327                 return bond_arp_rcv(skb, bond, slave);
3328 #if IS_ENABLED(CONFIG_IPV6)
3329         } else if (is_ipv6) {
3330                 return bond_na_rcv(skb, bond, slave);
3331 #endif
3332         } else {
3333                 return RX_HANDLER_ANOTHER;
3334         }
3335 }
3336
3337 static void bond_send_validate(struct bonding *bond, struct slave *slave)
3338 {
3339         bond_arp_send_all(bond, slave);
3340 #if IS_ENABLED(CONFIG_IPV6)
3341         bond_ns_send_all(bond, slave);
3342 #endif
3343 }
3344
3345 /* function to verify if we're in the arp_interval timeslice, returns true if
3346  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3347  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3348  */
3349 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3350                                   int mod)
3351 {
3352         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3353
3354         return time_in_range(jiffies,
3355                              last_act - delta_in_ticks,
3356                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
3357 }
3358
3359 /* This function is called regularly to monitor each slave's link
3360  * ensuring that traffic is being sent and received when arp monitoring
3361  * is used in load-balancing mode. if the adapter has been dormant, then an
3362  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3363  * arp monitoring in active backup mode.
3364  */
3365 static void bond_loadbalance_arp_mon(struct bonding *bond)
3366 {
3367         struct slave *slave, *oldcurrent;
3368         struct list_head *iter;
3369         int do_failover = 0, slave_state_changed = 0;
3370
3371         if (!bond_has_slaves(bond))
3372                 goto re_arm;
3373
3374         rcu_read_lock();
3375
3376         oldcurrent = rcu_dereference(bond->curr_active_slave);
3377         /* see if any of the previous devices are up now (i.e. they have
3378          * xmt and rcv traffic). the curr_active_slave does not come into
3379          * the picture unless it is null. also, slave->last_link_up is not
3380          * needed here because we send an arp on each slave and give a slave
3381          * as long as it needs to get the tx/rx within the delta.
3382          * TODO: what about up/down delay in arp mode? it wasn't here before
3383          *       so it can wait
3384          */
3385         bond_for_each_slave_rcu(bond, slave, iter) {
3386                 unsigned long last_tx = slave_last_tx(slave);
3387
3388                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3389
3390                 if (slave->link != BOND_LINK_UP) {
3391                         if (bond_time_in_interval(bond, last_tx, 1) &&
3392                             bond_time_in_interval(bond, slave->last_rx, 1)) {
3393
3394                                 bond_propose_link_state(slave, BOND_LINK_UP);
3395                                 slave_state_changed = 1;
3396
3397                                 /* primary_slave has no meaning in round-robin
3398                                  * mode. the window of a slave being up and
3399                                  * curr_active_slave being null after enslaving
3400                                  * is closed.
3401                                  */
3402                                 if (!oldcurrent) {
3403                                         slave_info(bond->dev, slave->dev, "link status definitely up\n");
3404                                         do_failover = 1;
3405                                 } else {
3406                                         slave_info(bond->dev, slave->dev, "interface is now up\n");
3407                                 }
3408                         }
3409                 } else {
3410                         /* slave->link == BOND_LINK_UP */
3411
3412                         /* not all switches will respond to an arp request
3413                          * when the source ip is 0, so don't take the link down
3414                          * if we don't know our ip yet
3415                          */
3416                         if (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3417                             !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3418
3419                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
3420                                 slave_state_changed = 1;
3421
3422                                 if (slave->link_failure_count < UINT_MAX)
3423                                         slave->link_failure_count++;
3424
3425                                 slave_info(bond->dev, slave->dev, "interface is now down\n");
3426
3427                                 if (slave == oldcurrent)
3428                                         do_failover = 1;
3429                         }
3430                 }
3431
3432                 /* note: if switch is in round-robin mode, all links
3433                  * must tx arp to ensure all links rx an arp - otherwise
3434                  * links may oscillate or not come up at all; if switch is
3435                  * in something like xor mode, there is nothing we can
3436                  * do - all replies will be rx'ed on same link causing slaves
3437                  * to be unstable during low/no traffic periods
3438                  */
3439                 if (bond_slave_is_up(slave))
3440                         bond_send_validate(bond, slave);
3441         }
3442
3443         rcu_read_unlock();
3444
3445         if (do_failover || slave_state_changed) {
3446                 if (!rtnl_trylock())
3447                         goto re_arm;
3448
3449                 bond_for_each_slave(bond, slave, iter) {
3450                         if (slave->link_new_state != BOND_LINK_NOCHANGE)
3451                                 slave->link = slave->link_new_state;
3452                 }
3453
3454                 if (slave_state_changed) {
3455                         bond_slave_state_change(bond);
3456                         if (BOND_MODE(bond) == BOND_MODE_XOR)
3457                                 bond_update_slave_arr(bond, NULL);
3458                 }
3459                 if (do_failover) {
3460                         block_netpoll_tx();
3461                         bond_select_active_slave(bond);
3462                         unblock_netpoll_tx();
3463                 }
3464                 rtnl_unlock();
3465         }
3466
3467 re_arm:
3468         if (bond->params.arp_interval)
3469                 queue_delayed_work(bond->wq, &bond->arp_work,
3470                                    msecs_to_jiffies(bond->params.arp_interval));
3471 }
3472
3473 /* Called to inspect slaves for active-backup mode ARP monitor link state
3474  * changes.  Sets proposed link state in slaves to specify what action
3475  * should take place for the slave.  Returns 0 if no changes are found, >0
3476  * if changes to link states must be committed.
3477  *
3478  * Called with rcu_read_lock held.
3479  */
3480 static int bond_ab_arp_inspect(struct bonding *bond)
3481 {
3482         unsigned long last_tx, last_rx;
3483         struct list_head *iter;
3484         struct slave *slave;
3485         int commit = 0;
3486
3487         bond_for_each_slave_rcu(bond, slave, iter) {
3488                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3489                 last_rx = slave_last_rx(bond, slave);
3490
3491                 if (slave->link != BOND_LINK_UP) {
3492                         if (bond_time_in_interval(bond, last_rx, 1)) {
3493                                 bond_propose_link_state(slave, BOND_LINK_UP);
3494                                 commit++;
3495                         } else if (slave->link == BOND_LINK_BACK) {
3496                                 bond_propose_link_state(slave, BOND_LINK_FAIL);
3497                                 commit++;
3498                         }
3499                         continue;
3500                 }
3501
3502                 /* Give slaves 2*delta after being enslaved or made
3503                  * active.  This avoids bouncing, as the last receive
3504                  * times need a full ARP monitor cycle to be updated.
3505                  */
3506                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3507                         continue;
3508
3509                 /* Backup slave is down if:
3510                  * - No current_arp_slave AND
3511                  * - more than (missed_max+1)*delta since last receive AND
3512                  * - the bond has an IP address
3513                  *
3514                  * Note: a non-null current_arp_slave indicates
3515                  * the curr_active_slave went down and we are
3516                  * searching for a new one; under this condition
3517                  * we only take the curr_active_slave down - this
3518                  * gives each slave a chance to tx/rx traffic
3519                  * before being taken out
3520                  */
3521                 if (!bond_is_active_slave(slave) &&
3522                     !rcu_access_pointer(bond->current_arp_slave) &&
3523                     !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3524                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3525                         commit++;
3526                 }
3527
3528                 /* Active slave is down if:
3529                  * - more than missed_max*delta since transmitting OR
3530                  * - (more than missed_max*delta since receive AND
3531                  *    the bond has an IP address)
3532                  */
3533                 last_tx = slave_last_tx(slave);
3534                 if (bond_is_active_slave(slave) &&
3535                     (!bond_time_in_interval(bond, last_tx, bond->params.missed_max) ||
3536                      !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3537                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3538                         commit++;
3539                 }
3540         }
3541
3542         return commit;
3543 }
3544
3545 /* Called to commit link state changes noted by inspection step of
3546  * active-backup mode ARP monitor.
3547  *
3548  * Called with RTNL hold.
3549  */
3550 static void bond_ab_arp_commit(struct bonding *bond)
3551 {
3552         bool do_failover = false;
3553         struct list_head *iter;
3554         unsigned long last_tx;
3555         struct slave *slave;
3556
3557         bond_for_each_slave(bond, slave, iter) {
3558                 switch (slave->link_new_state) {
3559                 case BOND_LINK_NOCHANGE:
3560                         continue;
3561
3562                 case BOND_LINK_UP:
3563                         last_tx = slave_last_tx(slave);
3564                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
3565                             (!rtnl_dereference(bond->curr_active_slave) &&
3566                              bond_time_in_interval(bond, last_tx, 1))) {
3567                                 struct slave *current_arp_slave;
3568
3569                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3570                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
3571                                                           BOND_SLAVE_NOTIFY_NOW);
3572                                 if (current_arp_slave) {
3573                                         bond_set_slave_inactive_flags(
3574                                                 current_arp_slave,
3575                                                 BOND_SLAVE_NOTIFY_NOW);
3576                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3577                                 }
3578
3579                                 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3580
3581                                 if (!rtnl_dereference(bond->curr_active_slave) ||
3582                                     slave == rtnl_dereference(bond->primary_slave) ||
3583                                     slave->prio > rtnl_dereference(bond->curr_active_slave)->prio)
3584                                         do_failover = true;
3585
3586                         }
3587
3588                         continue;
3589
3590                 case BOND_LINK_DOWN:
3591                         if (slave->link_failure_count < UINT_MAX)
3592                                 slave->link_failure_count++;
3593
3594                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3595                                                   BOND_SLAVE_NOTIFY_NOW);
3596                         bond_set_slave_inactive_flags(slave,
3597                                                       BOND_SLAVE_NOTIFY_NOW);
3598
3599                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3600
3601                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
3602                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3603                                 do_failover = true;
3604                         }
3605
3606                         continue;
3607
3608                 case BOND_LINK_FAIL:
3609                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3610                                                   BOND_SLAVE_NOTIFY_NOW);
3611                         bond_set_slave_inactive_flags(slave,
3612                                                       BOND_SLAVE_NOTIFY_NOW);
3613
3614                         /* A slave has just been enslaved and has become
3615                          * the current active slave.
3616                          */
3617                         if (rtnl_dereference(bond->curr_active_slave))
3618                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3619                         continue;
3620
3621                 default:
3622                         slave_err(bond->dev, slave->dev,
3623                                   "impossible: link_new_state %d on slave\n",
3624                                   slave->link_new_state);
3625                         continue;
3626                 }
3627         }
3628
3629         if (do_failover) {
3630                 block_netpoll_tx();
3631                 bond_select_active_slave(bond);
3632                 unblock_netpoll_tx();
3633         }
3634
3635         bond_set_carrier(bond);
3636 }
3637
3638 /* Send ARP probes for active-backup mode ARP monitor.
3639  *
3640  * Called with rcu_read_lock held.
3641  */
3642 static bool bond_ab_arp_probe(struct bonding *bond)
3643 {
3644         struct slave *slave, *before = NULL, *new_slave = NULL,
3645                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3646                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3647         struct list_head *iter;
3648         bool found = false;
3649         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3650
3651         if (curr_arp_slave && curr_active_slave)
3652                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3653                             curr_arp_slave->dev->name,
3654                             curr_active_slave->dev->name);
3655
3656         if (curr_active_slave) {
3657                 bond_send_validate(bond, curr_active_slave);
3658                 return should_notify_rtnl;
3659         }
3660
3661         /* if we don't have a curr_active_slave, search for the next available
3662          * backup slave from the current_arp_slave and make it the candidate
3663          * for becoming the curr_active_slave
3664          */
3665
3666         if (!curr_arp_slave) {
3667                 curr_arp_slave = bond_first_slave_rcu(bond);
3668                 if (!curr_arp_slave)
3669                         return should_notify_rtnl;
3670         }
3671
3672         bond_for_each_slave_rcu(bond, slave, iter) {
3673                 if (!found && !before && bond_slave_is_up(slave))
3674                         before = slave;
3675
3676                 if (found && !new_slave && bond_slave_is_up(slave))
3677                         new_slave = slave;
3678                 /* if the link state is up at this point, we
3679                  * mark it down - this can happen if we have
3680                  * simultaneous link failures and
3681                  * reselect_active_interface doesn't make this
3682                  * one the current slave so it is still marked
3683                  * up when it is actually down
3684                  */
3685                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3686                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3687                                                   BOND_SLAVE_NOTIFY_LATER);
3688                         if (slave->link_failure_count < UINT_MAX)
3689                                 slave->link_failure_count++;
3690
3691                         bond_set_slave_inactive_flags(slave,
3692                                                       BOND_SLAVE_NOTIFY_LATER);
3693
3694                         slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3695                 }
3696                 if (slave == curr_arp_slave)
3697                         found = true;
3698         }
3699
3700         if (!new_slave && before)
3701                 new_slave = before;
3702
3703         if (!new_slave)
3704                 goto check_state;
3705
3706         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3707                                   BOND_SLAVE_NOTIFY_LATER);
3708         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3709         bond_send_validate(bond, new_slave);
3710         new_slave->last_link_up = jiffies;
3711         rcu_assign_pointer(bond->current_arp_slave, new_slave);
3712
3713 check_state:
3714         bond_for_each_slave_rcu(bond, slave, iter) {
3715                 if (slave->should_notify || slave->should_notify_link) {
3716                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3717                         break;
3718                 }
3719         }
3720         return should_notify_rtnl;
3721 }
3722
3723 static void bond_activebackup_arp_mon(struct bonding *bond)
3724 {
3725         bool should_notify_peers = false;
3726         bool should_notify_rtnl = false;
3727         int delta_in_ticks;
3728
3729         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3730
3731         if (!bond_has_slaves(bond))
3732                 goto re_arm;
3733
3734         rcu_read_lock();
3735
3736         should_notify_peers = bond_should_notify_peers(bond);
3737
3738         if (bond_ab_arp_inspect(bond)) {
3739                 rcu_read_unlock();
3740
3741                 /* Race avoidance with bond_close flush of workqueue */
3742                 if (!rtnl_trylock()) {
3743                         delta_in_ticks = 1;
3744                         should_notify_peers = false;
3745                         goto re_arm;
3746                 }
3747
3748                 bond_ab_arp_commit(bond);
3749
3750                 rtnl_unlock();
3751                 rcu_read_lock();
3752         }
3753
3754         should_notify_rtnl = bond_ab_arp_probe(bond);
3755         rcu_read_unlock();
3756
3757 re_arm:
3758         if (bond->params.arp_interval)
3759                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3760
3761         if (should_notify_peers || should_notify_rtnl) {
3762                 if (!rtnl_trylock())
3763                         return;
3764
3765                 if (should_notify_peers) {
3766                         bond->send_peer_notif--;
3767                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3768                                                  bond->dev);
3769                 }
3770                 if (should_notify_rtnl) {
3771                         bond_slave_state_notify(bond);
3772                         bond_slave_link_notify(bond);
3773                 }
3774
3775                 rtnl_unlock();
3776         }
3777 }
3778
3779 static void bond_arp_monitor(struct work_struct *work)
3780 {
3781         struct bonding *bond = container_of(work, struct bonding,
3782                                             arp_work.work);
3783
3784         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3785                 bond_activebackup_arp_mon(bond);
3786         else
3787                 bond_loadbalance_arp_mon(bond);
3788 }
3789
3790 /*-------------------------- netdev event handling --------------------------*/
3791
3792 /* Change device name */
3793 static int bond_event_changename(struct bonding *bond)
3794 {
3795         bond_remove_proc_entry(bond);
3796         bond_create_proc_entry(bond);
3797
3798         bond_debug_reregister(bond);
3799
3800         return NOTIFY_DONE;
3801 }
3802
3803 static int bond_master_netdev_event(unsigned long event,
3804                                     struct net_device *bond_dev)
3805 {
3806         struct bonding *event_bond = netdev_priv(bond_dev);
3807
3808         netdev_dbg(bond_dev, "%s called\n", __func__);
3809
3810         switch (event) {
3811         case NETDEV_CHANGENAME:
3812                 return bond_event_changename(event_bond);
3813         case NETDEV_UNREGISTER:
3814                 bond_remove_proc_entry(event_bond);
3815 #ifdef CONFIG_XFRM_OFFLOAD
3816                 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3817 #endif /* CONFIG_XFRM_OFFLOAD */
3818                 break;
3819         case NETDEV_REGISTER:
3820                 bond_create_proc_entry(event_bond);
3821                 break;
3822         default:
3823                 break;
3824         }
3825
3826         return NOTIFY_DONE;
3827 }
3828
3829 static int bond_slave_netdev_event(unsigned long event,
3830                                    struct net_device *slave_dev)
3831 {
3832         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3833         struct bonding *bond;
3834         struct net_device *bond_dev;
3835
3836         /* A netdev event can be generated while enslaving a device
3837          * before netdev_rx_handler_register is called in which case
3838          * slave will be NULL
3839          */
3840         if (!slave) {
3841                 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3842                 return NOTIFY_DONE;
3843         }
3844
3845         bond_dev = slave->bond->dev;
3846         bond = slave->bond;
3847         primary = rtnl_dereference(bond->primary_slave);
3848
3849         slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3850
3851         switch (event) {
3852         case NETDEV_UNREGISTER:
3853                 if (bond_dev->type != ARPHRD_ETHER)
3854                         bond_release_and_destroy(bond_dev, slave_dev);
3855                 else
3856                         __bond_release_one(bond_dev, slave_dev, false, true);
3857                 break;
3858         case NETDEV_UP:
3859         case NETDEV_CHANGE:
3860                 /* For 802.3ad mode only:
3861                  * Getting invalid Speed/Duplex values here will put slave
3862                  * in weird state. Mark it as link-fail if the link was
3863                  * previously up or link-down if it hasn't yet come up, and
3864                  * let link-monitoring (miimon) set it right when correct
3865                  * speeds/duplex are available.
3866                  */
3867                 if (bond_update_speed_duplex(slave) &&
3868                     BOND_MODE(bond) == BOND_MODE_8023AD) {
3869                         if (slave->last_link_up)
3870                                 slave->link = BOND_LINK_FAIL;
3871                         else
3872                                 slave->link = BOND_LINK_DOWN;
3873                 }
3874
3875                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3876                         bond_3ad_adapter_speed_duplex_changed(slave);
3877                 fallthrough;
3878         case NETDEV_DOWN:
3879                 /* Refresh slave-array if applicable!
3880                  * If the setup does not use miimon or arpmon (mode-specific!),
3881                  * then these events will not cause the slave-array to be
3882                  * refreshed. This will cause xmit to use a slave that is not
3883                  * usable. Avoid such situation by refeshing the array at these
3884                  * events. If these (miimon/arpmon) parameters are configured
3885                  * then array gets refreshed twice and that should be fine!
3886                  */
3887                 if (bond_mode_can_use_xmit_hash(bond))
3888                         bond_update_slave_arr(bond, NULL);
3889                 break;
3890         case NETDEV_CHANGEMTU:
3891                 /* TODO: Should slaves be allowed to
3892                  * independently alter their MTU?  For
3893                  * an active-backup bond, slaves need
3894                  * not be the same type of device, so
3895                  * MTUs may vary.  For other modes,
3896                  * slaves arguably should have the
3897                  * same MTUs. To do this, we'd need to
3898                  * take over the slave's change_mtu
3899                  * function for the duration of their
3900                  * servitude.
3901                  */
3902                 break;
3903         case NETDEV_CHANGENAME:
3904                 /* we don't care if we don't have primary set */
3905                 if (!bond_uses_primary(bond) ||
3906                     !bond->params.primary[0])
3907                         break;
3908
3909                 if (slave == primary) {
3910                         /* slave's name changed - he's no longer primary */
3911                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3912                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3913                         /* we have a new primary slave */
3914                         rcu_assign_pointer(bond->primary_slave, slave);
3915                 } else { /* we didn't change primary - exit */
3916                         break;
3917                 }
3918
3919                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3920                             primary ? slave_dev->name : "none");
3921
3922                 block_netpoll_tx();
3923                 bond_select_active_slave(bond);
3924                 unblock_netpoll_tx();
3925                 break;
3926         case NETDEV_FEAT_CHANGE:
3927                 bond_compute_features(bond);
3928                 break;
3929         case NETDEV_RESEND_IGMP:
3930                 /* Propagate to master device */
3931                 call_netdevice_notifiers(event, slave->bond->dev);
3932                 break;
3933         default:
3934                 break;
3935         }
3936
3937         return NOTIFY_DONE;
3938 }
3939
3940 /* bond_netdev_event: handle netdev notifier chain events.
3941  *
3942  * This function receives events for the netdev chain.  The caller (an
3943  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3944  * locks for us to safely manipulate the slave devices (RTNL lock,
3945  * dev_probe_lock).
3946  */
3947 static int bond_netdev_event(struct notifier_block *this,
3948                              unsigned long event, void *ptr)
3949 {
3950         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3951
3952         netdev_dbg(event_dev, "%s received %s\n",
3953                    __func__, netdev_cmd_to_name(event));
3954
3955         if (!(event_dev->priv_flags & IFF_BONDING))
3956                 return NOTIFY_DONE;
3957
3958         if (event_dev->flags & IFF_MASTER) {
3959                 int ret;
3960
3961                 ret = bond_master_netdev_event(event, event_dev);
3962                 if (ret != NOTIFY_DONE)
3963                         return ret;
3964         }
3965
3966         if (event_dev->flags & IFF_SLAVE)
3967                 return bond_slave_netdev_event(event, event_dev);
3968
3969         return NOTIFY_DONE;
3970 }
3971
3972 static struct notifier_block bond_netdev_notifier = {
3973         .notifier_call = bond_netdev_event,
3974 };
3975
3976 /*---------------------------- Hashing Policies -----------------------------*/
3977
3978 /* Helper to access data in a packet, with or without a backing skb.
3979  * If skb is given the data is linearized if necessary via pskb_may_pull.
3980  */
3981 static inline const void *bond_pull_data(struct sk_buff *skb,
3982                                          const void *data, int hlen, int n)
3983 {
3984         if (likely(n <= hlen))
3985                 return data;
3986         else if (skb && likely(pskb_may_pull(skb, n)))
3987                 return skb->head;
3988
3989         return NULL;
3990 }
3991
3992 /* L2 hash helper */
3993 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
3994 {
3995         struct ethhdr *ep;
3996
3997         data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
3998         if (!data)
3999                 return 0;
4000
4001         ep = (struct ethhdr *)(data + mhoff);
4002         return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
4003 }
4004
4005 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
4006                          int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
4007 {
4008         const struct ipv6hdr *iph6;
4009         const struct iphdr *iph;
4010
4011         if (l2_proto == htons(ETH_P_IP)) {
4012                 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
4013                 if (!data)
4014                         return false;
4015
4016                 iph = (const struct iphdr *)(data + *nhoff);
4017                 iph_to_flow_copy_v4addrs(fk, iph);
4018                 *nhoff += iph->ihl << 2;
4019                 if (!ip_is_fragment(iph))
4020                         *ip_proto = iph->protocol;
4021         } else if (l2_proto == htons(ETH_P_IPV6)) {
4022                 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
4023                 if (!data)
4024                         return false;
4025
4026                 iph6 = (const struct ipv6hdr *)(data + *nhoff);
4027                 iph_to_flow_copy_v6addrs(fk, iph6);
4028                 *nhoff += sizeof(*iph6);
4029                 *ip_proto = iph6->nexthdr;
4030         } else {
4031                 return false;
4032         }
4033
4034         if (l34 && *ip_proto >= 0)
4035                 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
4036
4037         return true;
4038 }
4039
4040 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
4041 {
4042         u32 srcmac_vendor = 0, srcmac_dev = 0;
4043         struct ethhdr *mac_hdr;
4044         u16 vlan = 0;
4045         int i;
4046
4047         data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
4048         if (!data)
4049                 return 0;
4050         mac_hdr = (struct ethhdr *)(data + mhoff);
4051
4052         for (i = 0; i < 3; i++)
4053                 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
4054
4055         for (i = 3; i < ETH_ALEN; i++)
4056                 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
4057
4058         if (skb && skb_vlan_tag_present(skb))
4059                 vlan = skb_vlan_tag_get(skb);
4060
4061         return vlan ^ srcmac_vendor ^ srcmac_dev;
4062 }
4063
4064 /* Extract the appropriate headers based on bond's xmit policy */
4065 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
4066                               __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
4067 {
4068         bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
4069         int ip_proto = -1;
4070
4071         switch (bond->params.xmit_policy) {
4072         case BOND_XMIT_POLICY_ENCAP23:
4073         case BOND_XMIT_POLICY_ENCAP34:
4074                 memset(fk, 0, sizeof(*fk));
4075                 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
4076                                           fk, data, l2_proto, nhoff, hlen, 0);
4077         default:
4078                 break;
4079         }
4080
4081         fk->ports.ports = 0;
4082         memset(&fk->icmp, 0, sizeof(fk->icmp));
4083         if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4084                 return false;
4085
4086         /* ICMP error packets contains at least 8 bytes of the header
4087          * of the packet which generated the error. Use this information
4088          * to correlate ICMP error packets within the same flow which
4089          * generated the error.
4090          */
4091         if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4092                 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4093                 if (ip_proto == IPPROTO_ICMP) {
4094                         if (!icmp_is_err(fk->icmp.type))
4095                                 return true;
4096
4097                         nhoff += sizeof(struct icmphdr);
4098                 } else if (ip_proto == IPPROTO_ICMPV6) {
4099                         if (!icmpv6_is_err(fk->icmp.type))
4100                                 return true;
4101
4102                         nhoff += sizeof(struct icmp6hdr);
4103                 }
4104                 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4105         }
4106
4107         return true;
4108 }
4109
4110 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4111 {
4112         hash ^= (__force u32)flow_get_u32_dst(flow) ^
4113                 (__force u32)flow_get_u32_src(flow);
4114         hash ^= (hash >> 16);
4115         hash ^= (hash >> 8);
4116
4117         /* discard lowest hash bit to deal with the common even ports pattern */
4118         if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4119                 xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4120                 return hash >> 1;
4121
4122         return hash;
4123 }
4124
4125 /* Generate hash based on xmit policy. If @skb is given it is used to linearize
4126  * the data as required, but this function can be used without it if the data is
4127  * known to be linear (e.g. with xdp_buff).
4128  */
4129 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4130                             __be16 l2_proto, int mhoff, int nhoff, int hlen)
4131 {
4132         struct flow_keys flow;
4133         u32 hash;
4134
4135         if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4136                 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4137
4138         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4139             !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4140                 return bond_eth_hash(skb, data, mhoff, hlen);
4141
4142         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4143             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4144                 hash = bond_eth_hash(skb, data, mhoff, hlen);
4145         } else {
4146                 if (flow.icmp.id)
4147                         memcpy(&hash, &flow.icmp, sizeof(hash));
4148                 else
4149                         memcpy(&hash, &flow.ports.ports, sizeof(hash));
4150         }
4151
4152         return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4153 }
4154
4155 /**
4156  * bond_xmit_hash - generate a hash value based on the xmit policy
4157  * @bond: bonding device
4158  * @skb: buffer to use for headers
4159  *
4160  * This function will extract the necessary headers from the skb buffer and use
4161  * them to generate a hash based on the xmit_policy set in the bonding device
4162  */
4163 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4164 {
4165         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4166             skb->l4_hash)
4167                 return skb->hash;
4168
4169         return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4170                                 skb_mac_offset(skb), skb_network_offset(skb),
4171                                 skb_headlen(skb));
4172 }
4173
4174 /**
4175  * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4176  * @bond: bonding device
4177  * @xdp: buffer to use for headers
4178  *
4179  * The XDP variant of bond_xmit_hash.
4180  */
4181 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4182 {
4183         struct ethhdr *eth;
4184
4185         if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4186                 return 0;
4187
4188         eth = (struct ethhdr *)xdp->data;
4189
4190         return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4191                                 sizeof(struct ethhdr), xdp->data_end - xdp->data);
4192 }
4193
4194 /*-------------------------- Device entry points ----------------------------*/
4195
4196 void bond_work_init_all(struct bonding *bond)
4197 {
4198         INIT_DELAYED_WORK(&bond->mcast_work,
4199                           bond_resend_igmp_join_requests_delayed);
4200         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4201         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4202         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4203         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4204         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4205 }
4206
4207 static void bond_work_cancel_all(struct bonding *bond)
4208 {
4209         cancel_delayed_work_sync(&bond->mii_work);
4210         cancel_delayed_work_sync(&bond->arp_work);
4211         cancel_delayed_work_sync(&bond->alb_work);
4212         cancel_delayed_work_sync(&bond->ad_work);
4213         cancel_delayed_work_sync(&bond->mcast_work);
4214         cancel_delayed_work_sync(&bond->slave_arr_work);
4215 }
4216
4217 static int bond_open(struct net_device *bond_dev)
4218 {
4219         struct bonding *bond = netdev_priv(bond_dev);
4220         struct list_head *iter;
4221         struct slave *slave;
4222
4223         if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN && !bond->rr_tx_counter) {
4224                 bond->rr_tx_counter = alloc_percpu(u32);
4225                 if (!bond->rr_tx_counter)
4226                         return -ENOMEM;
4227         }
4228
4229         /* reset slave->backup and slave->inactive */
4230         if (bond_has_slaves(bond)) {
4231                 bond_for_each_slave(bond, slave, iter) {
4232                         if (bond_uses_primary(bond) &&
4233                             slave != rcu_access_pointer(bond->curr_active_slave)) {
4234                                 bond_set_slave_inactive_flags(slave,
4235                                                               BOND_SLAVE_NOTIFY_NOW);
4236                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4237                                 bond_set_slave_active_flags(slave,
4238                                                             BOND_SLAVE_NOTIFY_NOW);
4239                         }
4240                 }
4241         }
4242
4243         if (bond_is_lb(bond)) {
4244                 /* bond_alb_initialize must be called before the timer
4245                  * is started.
4246                  */
4247                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4248                         return -ENOMEM;
4249                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4250                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
4251         }
4252
4253         if (bond->params.miimon)  /* link check interval, in milliseconds. */
4254                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
4255
4256         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
4257                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
4258                 bond->recv_probe = bond_rcv_validate;
4259         }
4260
4261         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4262                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
4263                 /* register to receive LACPDUs */
4264                 bond->recv_probe = bond_3ad_lacpdu_recv;
4265                 bond_3ad_initiate_agg_selection(bond, 1);
4266
4267                 bond_for_each_slave(bond, slave, iter)
4268                         dev_mc_add(slave->dev, lacpdu_mcast_addr);
4269         }
4270
4271         if (bond_mode_can_use_xmit_hash(bond))
4272                 bond_update_slave_arr(bond, NULL);
4273
4274         return 0;
4275 }
4276
4277 static int bond_close(struct net_device *bond_dev)
4278 {
4279         struct bonding *bond = netdev_priv(bond_dev);
4280         struct slave *slave;
4281
4282         bond_work_cancel_all(bond);
4283         bond->send_peer_notif = 0;
4284         if (bond_is_lb(bond))
4285                 bond_alb_deinitialize(bond);
4286         bond->recv_probe = NULL;
4287
4288         if (bond_uses_primary(bond)) {
4289                 rcu_read_lock();
4290                 slave = rcu_dereference(bond->curr_active_slave);
4291                 if (slave)
4292                         bond_hw_addr_flush(bond_dev, slave->dev);
4293                 rcu_read_unlock();
4294         } else {
4295                 struct list_head *iter;
4296
4297                 bond_for_each_slave(bond, slave, iter)
4298                         bond_hw_addr_flush(bond_dev, slave->dev);
4299         }
4300
4301         return 0;
4302 }
4303
4304 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4305  * that some drivers can provide 32bit values only.
4306  */
4307 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4308                             const struct rtnl_link_stats64 *_new,
4309                             const struct rtnl_link_stats64 *_old)
4310 {
4311         const u64 *new = (const u64 *)_new;
4312         const u64 *old = (const u64 *)_old;
4313         u64 *res = (u64 *)_res;
4314         int i;
4315
4316         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4317                 u64 nv = new[i];
4318                 u64 ov = old[i];
4319                 s64 delta = nv - ov;
4320
4321                 /* detects if this particular field is 32bit only */
4322                 if (((nv | ov) >> 32) == 0)
4323                         delta = (s64)(s32)((u32)nv - (u32)ov);
4324
4325                 /* filter anomalies, some drivers reset their stats
4326                  * at down/up events.
4327                  */
4328                 if (delta > 0)
4329                         res[i] += delta;
4330         }
4331 }
4332
4333 #ifdef CONFIG_LOCKDEP
4334 static int bond_get_lowest_level_rcu(struct net_device *dev)
4335 {
4336         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4337         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4338         int cur = 0, max = 0;
4339
4340         now = dev;
4341         iter = &dev->adj_list.lower;
4342
4343         while (1) {
4344                 next = NULL;
4345                 while (1) {
4346                         ldev = netdev_next_lower_dev_rcu(now, &iter);
4347                         if (!ldev)
4348                                 break;
4349
4350                         next = ldev;
4351                         niter = &ldev->adj_list.lower;
4352                         dev_stack[cur] = now;
4353                         iter_stack[cur++] = iter;
4354                         if (max <= cur)
4355                                 max = cur;
4356                         break;
4357                 }
4358
4359                 if (!next) {
4360                         if (!cur)
4361                                 return max;
4362                         next = dev_stack[--cur];
4363                         niter = iter_stack[cur];
4364                 }
4365
4366                 now = next;
4367                 iter = niter;
4368         }
4369
4370         return max;
4371 }
4372 #endif
4373
4374 static void bond_get_stats(struct net_device *bond_dev,
4375                            struct rtnl_link_stats64 *stats)
4376 {
4377         struct bonding *bond = netdev_priv(bond_dev);
4378         struct rtnl_link_stats64 temp;
4379         struct list_head *iter;
4380         struct slave *slave;
4381         int nest_level = 0;
4382
4383
4384         rcu_read_lock();
4385 #ifdef CONFIG_LOCKDEP
4386         nest_level = bond_get_lowest_level_rcu(bond_dev);
4387 #endif
4388
4389         spin_lock_nested(&bond->stats_lock, nest_level);
4390         memcpy(stats, &bond->bond_stats, sizeof(*stats));
4391
4392         bond_for_each_slave_rcu(bond, slave, iter) {
4393                 const struct rtnl_link_stats64 *new =
4394                         dev_get_stats(slave->dev, &temp);
4395
4396                 bond_fold_stats(stats, new, &slave->slave_stats);
4397
4398                 /* save off the slave stats for the next run */
4399                 memcpy(&slave->slave_stats, new, sizeof(*new));
4400         }
4401
4402         memcpy(&bond->bond_stats, stats, sizeof(*stats));
4403         spin_unlock(&bond->stats_lock);
4404         rcu_read_unlock();
4405 }
4406
4407 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4408 {
4409         struct bonding *bond = netdev_priv(bond_dev);
4410         struct mii_ioctl_data *mii = NULL;
4411         const struct net_device_ops *ops;
4412         struct net_device *real_dev;
4413         struct hwtstamp_config cfg;
4414         struct ifreq ifrr;
4415         int res = 0;
4416
4417         netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4418
4419         switch (cmd) {
4420         case SIOCGMIIPHY:
4421                 mii = if_mii(ifr);
4422                 if (!mii)
4423                         return -EINVAL;
4424
4425                 mii->phy_id = 0;
4426                 fallthrough;
4427         case SIOCGMIIREG:
4428                 /* We do this again just in case we were called by SIOCGMIIREG
4429                  * instead of SIOCGMIIPHY.
4430                  */
4431                 mii = if_mii(ifr);
4432                 if (!mii)
4433                         return -EINVAL;
4434
4435                 if (mii->reg_num == 1) {
4436                         mii->val_out = 0;
4437                         if (netif_carrier_ok(bond->dev))
4438                                 mii->val_out = BMSR_LSTATUS;
4439                 }
4440
4441                 break;
4442         case SIOCSHWTSTAMP:
4443                 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4444                         return -EFAULT;
4445
4446                 if (!(cfg.flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
4447                         return -EOPNOTSUPP;
4448
4449                 fallthrough;
4450         case SIOCGHWTSTAMP:
4451                 real_dev = bond_option_active_slave_get_rcu(bond);
4452                 if (!real_dev)
4453                         return -EOPNOTSUPP;
4454
4455                 strscpy_pad(ifrr.ifr_name, real_dev->name, IFNAMSIZ);
4456                 ifrr.ifr_ifru = ifr->ifr_ifru;
4457
4458                 ops = real_dev->netdev_ops;
4459                 if (netif_device_present(real_dev) && ops->ndo_eth_ioctl) {
4460                         res = ops->ndo_eth_ioctl(real_dev, &ifrr, cmd);
4461                         if (res)
4462                                 return res;
4463
4464                         ifr->ifr_ifru = ifrr.ifr_ifru;
4465                         if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4466                                 return -EFAULT;
4467
4468                         /* Set the BOND_PHC_INDEX flag to notify user space */
4469                         cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
4470
4471                         return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ?
4472                                 -EFAULT : 0;
4473                 }
4474                 fallthrough;
4475         default:
4476                 res = -EOPNOTSUPP;
4477         }
4478
4479         return res;
4480 }
4481
4482 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4483 {
4484         struct bonding *bond = netdev_priv(bond_dev);
4485         struct net_device *slave_dev = NULL;
4486         struct ifbond k_binfo;
4487         struct ifbond __user *u_binfo = NULL;
4488         struct ifslave k_sinfo;
4489         struct ifslave __user *u_sinfo = NULL;
4490         struct bond_opt_value newval;
4491         struct net *net;
4492         int res = 0;
4493
4494         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4495
4496         switch (cmd) {
4497         case SIOCBONDINFOQUERY:
4498                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4499
4500                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4501                         return -EFAULT;
4502
4503                 bond_info_query(bond_dev, &k_binfo);
4504                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4505                         return -EFAULT;
4506
4507                 return 0;
4508         case SIOCBONDSLAVEINFOQUERY:
4509                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4510
4511                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4512                         return -EFAULT;
4513
4514                 res = bond_slave_info_query(bond_dev, &k_sinfo);
4515                 if (res == 0 &&
4516                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4517                         return -EFAULT;
4518
4519                 return res;
4520         default:
4521                 break;
4522         }
4523
4524         net = dev_net(bond_dev);
4525
4526         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4527                 return -EPERM;
4528
4529         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4530
4531         slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4532
4533         if (!slave_dev)
4534                 return -ENODEV;
4535
4536         switch (cmd) {
4537         case SIOCBONDENSLAVE:
4538                 res = bond_enslave(bond_dev, slave_dev, NULL);
4539                 break;
4540         case SIOCBONDRELEASE:
4541                 res = bond_release(bond_dev, slave_dev);
4542                 break;
4543         case SIOCBONDSETHWADDR:
4544                 res = bond_set_dev_addr(bond_dev, slave_dev);
4545                 break;
4546         case SIOCBONDCHANGEACTIVE:
4547                 bond_opt_initstr(&newval, slave_dev->name);
4548                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4549                                             &newval);
4550                 break;
4551         default:
4552                 res = -EOPNOTSUPP;
4553         }
4554
4555         return res;
4556 }
4557
4558 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4559                                void __user *data, int cmd)
4560 {
4561         struct ifreq ifrdata = { .ifr_data = data };
4562
4563         switch (cmd) {
4564         case BOND_INFO_QUERY_OLD:
4565                 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4566         case BOND_SLAVE_INFO_QUERY_OLD:
4567                 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4568         case BOND_ENSLAVE_OLD:
4569                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4570         case BOND_RELEASE_OLD:
4571                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4572         case BOND_SETHWADDR_OLD:
4573                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4574         case BOND_CHANGE_ACTIVE_OLD:
4575                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4576         }
4577
4578         return -EOPNOTSUPP;
4579 }
4580
4581 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4582 {
4583         struct bonding *bond = netdev_priv(bond_dev);
4584
4585         if (change & IFF_PROMISC)
4586                 bond_set_promiscuity(bond,
4587                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
4588
4589         if (change & IFF_ALLMULTI)
4590                 bond_set_allmulti(bond,
4591                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4592 }
4593
4594 static void bond_set_rx_mode(struct net_device *bond_dev)
4595 {
4596         struct bonding *bond = netdev_priv(bond_dev);
4597         struct list_head *iter;
4598         struct slave *slave;
4599
4600         rcu_read_lock();
4601         if (bond_uses_primary(bond)) {
4602                 slave = rcu_dereference(bond->curr_active_slave);
4603                 if (slave) {
4604                         dev_uc_sync(slave->dev, bond_dev);
4605                         dev_mc_sync(slave->dev, bond_dev);
4606                 }
4607         } else {
4608                 bond_for_each_slave_rcu(bond, slave, iter) {
4609                         dev_uc_sync_multiple(slave->dev, bond_dev);
4610                         dev_mc_sync_multiple(slave->dev, bond_dev);
4611                 }
4612         }
4613         rcu_read_unlock();
4614 }
4615
4616 static int bond_neigh_init(struct neighbour *n)
4617 {
4618         struct bonding *bond = netdev_priv(n->dev);
4619         const struct net_device_ops *slave_ops;
4620         struct neigh_parms parms;
4621         struct slave *slave;
4622         int ret = 0;
4623
4624         rcu_read_lock();
4625         slave = bond_first_slave_rcu(bond);
4626         if (!slave)
4627                 goto out;
4628         slave_ops = slave->dev->netdev_ops;
4629         if (!slave_ops->ndo_neigh_setup)
4630                 goto out;
4631
4632         /* TODO: find another way [1] to implement this.
4633          * Passing a zeroed structure is fragile,
4634          * but at least we do not pass garbage.
4635          *
4636          * [1] One way would be that ndo_neigh_setup() never touch
4637          *     struct neigh_parms, but propagate the new neigh_setup()
4638          *     back to ___neigh_create() / neigh_parms_alloc()
4639          */
4640         memset(&parms, 0, sizeof(parms));
4641         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4642
4643         if (ret)
4644                 goto out;
4645
4646         if (parms.neigh_setup)
4647                 ret = parms.neigh_setup(n);
4648 out:
4649         rcu_read_unlock();
4650         return ret;
4651 }
4652
4653 /* The bonding ndo_neigh_setup is called at init time beofre any
4654  * slave exists. So we must declare proxy setup function which will
4655  * be used at run time to resolve the actual slave neigh param setup.
4656  *
4657  * It's also called by master devices (such as vlans) to setup their
4658  * underlying devices. In that case - do nothing, we're already set up from
4659  * our init.
4660  */
4661 static int bond_neigh_setup(struct net_device *dev,
4662                             struct neigh_parms *parms)
4663 {
4664         /* modify only our neigh_parms */
4665         if (parms->dev == dev)
4666                 parms->neigh_setup = bond_neigh_init;
4667
4668         return 0;
4669 }
4670
4671 /* Change the MTU of all of a master's slaves to match the master */
4672 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4673 {
4674         struct bonding *bond = netdev_priv(bond_dev);
4675         struct slave *slave, *rollback_slave;
4676         struct list_head *iter;
4677         int res = 0;
4678
4679         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4680
4681         bond_for_each_slave(bond, slave, iter) {
4682                 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4683                            slave, slave->dev->netdev_ops->ndo_change_mtu);
4684
4685                 res = dev_set_mtu(slave->dev, new_mtu);
4686
4687                 if (res) {
4688                         /* If we failed to set the slave's mtu to the new value
4689                          * we must abort the operation even in ACTIVE_BACKUP
4690                          * mode, because if we allow the backup slaves to have
4691                          * different mtu values than the active slave we'll
4692                          * need to change their mtu when doing a failover. That
4693                          * means changing their mtu from timer context, which
4694                          * is probably not a good idea.
4695                          */
4696                         slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4697                                   res, new_mtu);
4698                         goto unwind;
4699                 }
4700         }
4701
4702         bond_dev->mtu = new_mtu;
4703
4704         return 0;
4705
4706 unwind:
4707         /* unwind from head to the slave that failed */
4708         bond_for_each_slave(bond, rollback_slave, iter) {
4709                 int tmp_res;
4710
4711                 if (rollback_slave == slave)
4712                         break;
4713
4714                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4715                 if (tmp_res)
4716                         slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4717                                   tmp_res);
4718         }
4719
4720         return res;
4721 }
4722
4723 /* Change HW address
4724  *
4725  * Note that many devices must be down to change the HW address, and
4726  * downing the master releases all slaves.  We can make bonds full of
4727  * bonding devices to test this, however.
4728  */
4729 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4730 {
4731         struct bonding *bond = netdev_priv(bond_dev);
4732         struct slave *slave, *rollback_slave;
4733         struct sockaddr_storage *ss = addr, tmp_ss;
4734         struct list_head *iter;
4735         int res = 0;
4736
4737         if (BOND_MODE(bond) == BOND_MODE_ALB)
4738                 return bond_alb_set_mac_address(bond_dev, addr);
4739
4740
4741         netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4742
4743         /* If fail_over_mac is enabled, do nothing and return success.
4744          * Returning an error causes ifenslave to fail.
4745          */
4746         if (bond->params.fail_over_mac &&
4747             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4748                 return 0;
4749
4750         if (!is_valid_ether_addr(ss->__data))
4751                 return -EADDRNOTAVAIL;
4752
4753         bond_for_each_slave(bond, slave, iter) {
4754                 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4755                           __func__, slave);
4756                 res = dev_set_mac_address(slave->dev, addr, NULL);
4757                 if (res) {
4758                         /* TODO: consider downing the slave
4759                          * and retry ?
4760                          * User should expect communications
4761                          * breakage anyway until ARP finish
4762                          * updating, so...
4763                          */
4764                         slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4765                                   __func__, res);
4766                         goto unwind;
4767                 }
4768         }
4769
4770         /* success */
4771         dev_addr_set(bond_dev, ss->__data);
4772         return 0;
4773
4774 unwind:
4775         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4776         tmp_ss.ss_family = bond_dev->type;
4777
4778         /* unwind from head to the slave that failed */
4779         bond_for_each_slave(bond, rollback_slave, iter) {
4780                 int tmp_res;
4781
4782                 if (rollback_slave == slave)
4783                         break;
4784
4785                 tmp_res = dev_set_mac_address(rollback_slave->dev,
4786                                               (struct sockaddr *)&tmp_ss, NULL);
4787                 if (tmp_res) {
4788                         slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4789                                    __func__, tmp_res);
4790                 }
4791         }
4792
4793         return res;
4794 }
4795
4796 /**
4797  * bond_get_slave_by_id - get xmit slave with slave_id
4798  * @bond: bonding device that is transmitting
4799  * @slave_id: slave id up to slave_cnt-1 through which to transmit
4800  *
4801  * This function tries to get slave with slave_id but in case
4802  * it fails, it tries to find the first available slave for transmission.
4803  */
4804 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4805                                           int slave_id)
4806 {
4807         struct list_head *iter;
4808         struct slave *slave;
4809         int i = slave_id;
4810
4811         /* Here we start from the slave with slave_id */
4812         bond_for_each_slave_rcu(bond, slave, iter) {
4813                 if (--i < 0) {
4814                         if (bond_slave_can_tx(slave))
4815                                 return slave;
4816                 }
4817         }
4818
4819         /* Here we start from the first slave up to slave_id */
4820         i = slave_id;
4821         bond_for_each_slave_rcu(bond, slave, iter) {
4822                 if (--i < 0)
4823                         break;
4824                 if (bond_slave_can_tx(slave))
4825                         return slave;
4826         }
4827         /* no slave that can tx has been found */
4828         return NULL;
4829 }
4830
4831 /**
4832  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4833  * @bond: bonding device to use
4834  *
4835  * Based on the value of the bonding device's packets_per_slave parameter
4836  * this function generates a slave id, which is usually used as the next
4837  * slave to transmit through.
4838  */
4839 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4840 {
4841         u32 slave_id;
4842         struct reciprocal_value reciprocal_packets_per_slave;
4843         int packets_per_slave = bond->params.packets_per_slave;
4844
4845         switch (packets_per_slave) {
4846         case 0:
4847                 slave_id = get_random_u32();
4848                 break;
4849         case 1:
4850                 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4851                 break;
4852         default:
4853                 reciprocal_packets_per_slave =
4854                         bond->params.reciprocal_packets_per_slave;
4855                 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4856                 slave_id = reciprocal_divide(slave_id,
4857                                              reciprocal_packets_per_slave);
4858                 break;
4859         }
4860
4861         return slave_id;
4862 }
4863
4864 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4865                                                     struct sk_buff *skb)
4866 {
4867         struct slave *slave;
4868         int slave_cnt;
4869         u32 slave_id;
4870
4871         /* Start with the curr_active_slave that joined the bond as the
4872          * default for sending IGMP traffic.  For failover purposes one
4873          * needs to maintain some consistency for the interface that will
4874          * send the join/membership reports.  The curr_active_slave found
4875          * will send all of this type of traffic.
4876          */
4877         if (skb->protocol == htons(ETH_P_IP)) {
4878                 int noff = skb_network_offset(skb);
4879                 struct iphdr *iph;
4880
4881                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4882                         goto non_igmp;
4883
4884                 iph = ip_hdr(skb);
4885                 if (iph->protocol == IPPROTO_IGMP) {
4886                         slave = rcu_dereference(bond->curr_active_slave);
4887                         if (slave)
4888                                 return slave;
4889                         return bond_get_slave_by_id(bond, 0);
4890                 }
4891         }
4892
4893 non_igmp:
4894         slave_cnt = READ_ONCE(bond->slave_cnt);
4895         if (likely(slave_cnt)) {
4896                 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4897                 return bond_get_slave_by_id(bond, slave_id);
4898         }
4899         return NULL;
4900 }
4901
4902 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4903                                                         struct xdp_buff *xdp)
4904 {
4905         struct slave *slave;
4906         int slave_cnt;
4907         u32 slave_id;
4908         const struct ethhdr *eth;
4909         void *data = xdp->data;
4910
4911         if (data + sizeof(struct ethhdr) > xdp->data_end)
4912                 goto non_igmp;
4913
4914         eth = (struct ethhdr *)data;
4915         data += sizeof(struct ethhdr);
4916
4917         /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4918         if (eth->h_proto == htons(ETH_P_IP)) {
4919                 const struct iphdr *iph;
4920
4921                 if (data + sizeof(struct iphdr) > xdp->data_end)
4922                         goto non_igmp;
4923
4924                 iph = (struct iphdr *)data;
4925
4926                 if (iph->protocol == IPPROTO_IGMP) {
4927                         slave = rcu_dereference(bond->curr_active_slave);
4928                         if (slave)
4929                                 return slave;
4930                         return bond_get_slave_by_id(bond, 0);
4931                 }
4932         }
4933
4934 non_igmp:
4935         slave_cnt = READ_ONCE(bond->slave_cnt);
4936         if (likely(slave_cnt)) {
4937                 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4938                 return bond_get_slave_by_id(bond, slave_id);
4939         }
4940         return NULL;
4941 }
4942
4943 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4944                                         struct net_device *bond_dev)
4945 {
4946         struct bonding *bond = netdev_priv(bond_dev);
4947         struct slave *slave;
4948
4949         slave = bond_xmit_roundrobin_slave_get(bond, skb);
4950         if (likely(slave))
4951                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4952
4953         return bond_tx_drop(bond_dev, skb);
4954 }
4955
4956 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4957 {
4958         return rcu_dereference(bond->curr_active_slave);
4959 }
4960
4961 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4962  * the bond has a usable interface.
4963  */
4964 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4965                                           struct net_device *bond_dev)
4966 {
4967         struct bonding *bond = netdev_priv(bond_dev);
4968         struct slave *slave;
4969
4970         slave = bond_xmit_activebackup_slave_get(bond);
4971         if (slave)
4972                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4973
4974         return bond_tx_drop(bond_dev, skb);
4975 }
4976
4977 /* Use this to update slave_array when (a) it's not appropriate to update
4978  * slave_array right away (note that update_slave_array() may sleep)
4979  * and / or (b) RTNL is not held.
4980  */
4981 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4982 {
4983         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4984 }
4985
4986 /* Slave array work handler. Holds only RTNL */
4987 static void bond_slave_arr_handler(struct work_struct *work)
4988 {
4989         struct bonding *bond = container_of(work, struct bonding,
4990                                             slave_arr_work.work);
4991         int ret;
4992
4993         if (!rtnl_trylock())
4994                 goto err;
4995
4996         ret = bond_update_slave_arr(bond, NULL);
4997         rtnl_unlock();
4998         if (ret) {
4999                 pr_warn_ratelimited("Failed to update slave array from WT\n");
5000                 goto err;
5001         }
5002         return;
5003
5004 err:
5005         bond_slave_arr_work_rearm(bond, 1);
5006 }
5007
5008 static void bond_skip_slave(struct bond_up_slave *slaves,
5009                             struct slave *skipslave)
5010 {
5011         int idx;
5012
5013         /* Rare situation where caller has asked to skip a specific
5014          * slave but allocation failed (most likely!). BTW this is
5015          * only possible when the call is initiated from
5016          * __bond_release_one(). In this situation; overwrite the
5017          * skipslave entry in the array with the last entry from the
5018          * array to avoid a situation where the xmit path may choose
5019          * this to-be-skipped slave to send a packet out.
5020          */
5021         for (idx = 0; slaves && idx < slaves->count; idx++) {
5022                 if (skipslave == slaves->arr[idx]) {
5023                         slaves->arr[idx] =
5024                                 slaves->arr[slaves->count - 1];
5025                         slaves->count--;
5026                         break;
5027                 }
5028         }
5029 }
5030
5031 static void bond_set_slave_arr(struct bonding *bond,
5032                                struct bond_up_slave *usable_slaves,
5033                                struct bond_up_slave *all_slaves)
5034 {
5035         struct bond_up_slave *usable, *all;
5036
5037         usable = rtnl_dereference(bond->usable_slaves);
5038         rcu_assign_pointer(bond->usable_slaves, usable_slaves);
5039         kfree_rcu(usable, rcu);
5040
5041         all = rtnl_dereference(bond->all_slaves);
5042         rcu_assign_pointer(bond->all_slaves, all_slaves);
5043         kfree_rcu(all, rcu);
5044 }
5045
5046 static void bond_reset_slave_arr(struct bonding *bond)
5047 {
5048         struct bond_up_slave *usable, *all;
5049
5050         usable = rtnl_dereference(bond->usable_slaves);
5051         if (usable) {
5052                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5053                 kfree_rcu(usable, rcu);
5054         }
5055
5056         all = rtnl_dereference(bond->all_slaves);
5057         if (all) {
5058                 RCU_INIT_POINTER(bond->all_slaves, NULL);
5059                 kfree_rcu(all, rcu);
5060         }
5061 }
5062
5063 /* Build the usable slaves array in control path for modes that use xmit-hash
5064  * to determine the slave interface -
5065  * (a) BOND_MODE_8023AD
5066  * (b) BOND_MODE_XOR
5067  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
5068  *
5069  * The caller is expected to hold RTNL only and NO other lock!
5070  */
5071 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
5072 {
5073         struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
5074         struct slave *slave;
5075         struct list_head *iter;
5076         int agg_id = 0;
5077         int ret = 0;
5078
5079         might_sleep();
5080
5081         usable_slaves = kzalloc(struct_size(usable_slaves, arr,
5082                                             bond->slave_cnt), GFP_KERNEL);
5083         all_slaves = kzalloc(struct_size(all_slaves, arr,
5084                                          bond->slave_cnt), GFP_KERNEL);
5085         if (!usable_slaves || !all_slaves) {
5086                 ret = -ENOMEM;
5087                 goto out;
5088         }
5089         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5090                 struct ad_info ad_info;
5091
5092                 spin_lock_bh(&bond->mode_lock);
5093                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
5094                         spin_unlock_bh(&bond->mode_lock);
5095                         pr_debug("bond_3ad_get_active_agg_info failed\n");
5096                         /* No active aggragator means it's not safe to use
5097                          * the previous array.
5098                          */
5099                         bond_reset_slave_arr(bond);
5100                         goto out;
5101                 }
5102                 spin_unlock_bh(&bond->mode_lock);
5103                 agg_id = ad_info.aggregator_id;
5104         }
5105         bond_for_each_slave(bond, slave, iter) {
5106                 if (skipslave == slave)
5107                         continue;
5108
5109                 all_slaves->arr[all_slaves->count++] = slave;
5110                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5111                         struct aggregator *agg;
5112
5113                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
5114                         if (!agg || agg->aggregator_identifier != agg_id)
5115                                 continue;
5116                 }
5117                 if (!bond_slave_can_tx(slave))
5118                         continue;
5119
5120                 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5121                           usable_slaves->count);
5122
5123                 usable_slaves->arr[usable_slaves->count++] = slave;
5124         }
5125
5126         bond_set_slave_arr(bond, usable_slaves, all_slaves);
5127         return ret;
5128 out:
5129         if (ret != 0 && skipslave) {
5130                 bond_skip_slave(rtnl_dereference(bond->all_slaves),
5131                                 skipslave);
5132                 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5133                                 skipslave);
5134         }
5135         kfree_rcu(all_slaves, rcu);
5136         kfree_rcu(usable_slaves, rcu);
5137
5138         return ret;
5139 }
5140
5141 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5142                                                  struct sk_buff *skb,
5143                                                  struct bond_up_slave *slaves)
5144 {
5145         struct slave *slave;
5146         unsigned int count;
5147         u32 hash;
5148
5149         hash = bond_xmit_hash(bond, skb);
5150         count = slaves ? READ_ONCE(slaves->count) : 0;
5151         if (unlikely(!count))
5152                 return NULL;
5153
5154         slave = slaves->arr[hash % count];
5155         return slave;
5156 }
5157
5158 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5159                                                      struct xdp_buff *xdp)
5160 {
5161         struct bond_up_slave *slaves;
5162         unsigned int count;
5163         u32 hash;
5164
5165         hash = bond_xmit_hash_xdp(bond, xdp);
5166         slaves = rcu_dereference(bond->usable_slaves);
5167         count = slaves ? READ_ONCE(slaves->count) : 0;
5168         if (unlikely(!count))
5169                 return NULL;
5170
5171         return slaves->arr[hash % count];
5172 }
5173
5174 /* Use this Xmit function for 3AD as well as XOR modes. The current
5175  * usable slave array is formed in the control path. The xmit function
5176  * just calculates hash and sends the packet out.
5177  */
5178 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5179                                      struct net_device *dev)
5180 {
5181         struct bonding *bond = netdev_priv(dev);
5182         struct bond_up_slave *slaves;
5183         struct slave *slave;
5184
5185         slaves = rcu_dereference(bond->usable_slaves);
5186         slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5187         if (likely(slave))
5188                 return bond_dev_queue_xmit(bond, skb, slave->dev);
5189
5190         return bond_tx_drop(dev, skb);
5191 }
5192
5193 /* in broadcast mode, we send everything to all usable interfaces. */
5194 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5195                                        struct net_device *bond_dev)
5196 {
5197         struct bonding *bond = netdev_priv(bond_dev);
5198         struct slave *slave = NULL;
5199         struct list_head *iter;
5200         bool xmit_suc = false;
5201         bool skb_used = false;
5202
5203         bond_for_each_slave_rcu(bond, slave, iter) {
5204                 struct sk_buff *skb2;
5205
5206                 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5207                         continue;
5208
5209                 if (bond_is_last_slave(bond, slave)) {
5210                         skb2 = skb;
5211                         skb_used = true;
5212                 } else {
5213                         skb2 = skb_clone(skb, GFP_ATOMIC);
5214                         if (!skb2) {
5215                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5216                                                     bond_dev->name, __func__);
5217                                 continue;
5218                         }
5219                 }
5220
5221                 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5222                         xmit_suc = true;
5223         }
5224
5225         if (!skb_used)
5226                 dev_kfree_skb_any(skb);
5227
5228         if (xmit_suc)
5229                 return NETDEV_TX_OK;
5230
5231         dev_core_stats_tx_dropped_inc(bond_dev);
5232         return NET_XMIT_DROP;
5233 }
5234
5235 /*------------------------- Device initialization ---------------------------*/
5236
5237 /* Lookup the slave that corresponds to a qid */
5238 static inline int bond_slave_override(struct bonding *bond,
5239                                       struct sk_buff *skb)
5240 {
5241         struct slave *slave = NULL;
5242         struct list_head *iter;
5243
5244         if (!skb_rx_queue_recorded(skb))
5245                 return 1;
5246
5247         /* Find out if any slaves have the same mapping as this skb. */
5248         bond_for_each_slave_rcu(bond, slave, iter) {
5249                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
5250                         if (bond_slave_is_up(slave) &&
5251                             slave->link == BOND_LINK_UP) {
5252                                 bond_dev_queue_xmit(bond, skb, slave->dev);
5253                                 return 0;
5254                         }
5255                         /* If the slave isn't UP, use default transmit policy. */
5256                         break;
5257                 }
5258         }
5259
5260         return 1;
5261 }
5262
5263
5264 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5265                              struct net_device *sb_dev)
5266 {
5267         /* This helper function exists to help dev_pick_tx get the correct
5268          * destination queue.  Using a helper function skips a call to
5269          * skb_tx_hash and will put the skbs in the queue we expect on their
5270          * way down to the bonding driver.
5271          */
5272         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5273
5274         /* Save the original txq to restore before passing to the driver */
5275         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5276
5277         if (unlikely(txq >= dev->real_num_tx_queues)) {
5278                 do {
5279                         txq -= dev->real_num_tx_queues;
5280                 } while (txq >= dev->real_num_tx_queues);
5281         }
5282         return txq;
5283 }
5284
5285 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5286                                               struct sk_buff *skb,
5287                                               bool all_slaves)
5288 {
5289         struct bonding *bond = netdev_priv(master_dev);
5290         struct bond_up_slave *slaves;
5291         struct slave *slave = NULL;
5292
5293         switch (BOND_MODE(bond)) {
5294         case BOND_MODE_ROUNDROBIN:
5295                 slave = bond_xmit_roundrobin_slave_get(bond, skb);
5296                 break;
5297         case BOND_MODE_ACTIVEBACKUP:
5298                 slave = bond_xmit_activebackup_slave_get(bond);
5299                 break;
5300         case BOND_MODE_8023AD:
5301         case BOND_MODE_XOR:
5302                 if (all_slaves)
5303                         slaves = rcu_dereference(bond->all_slaves);
5304                 else
5305                         slaves = rcu_dereference(bond->usable_slaves);
5306                 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5307                 break;
5308         case BOND_MODE_BROADCAST:
5309                 break;
5310         case BOND_MODE_ALB:
5311                 slave = bond_xmit_alb_slave_get(bond, skb);
5312                 break;
5313         case BOND_MODE_TLB:
5314                 slave = bond_xmit_tlb_slave_get(bond, skb);
5315                 break;
5316         default:
5317                 /* Should never happen, mode already checked */
5318                 WARN_ONCE(true, "Unknown bonding mode");
5319                 break;
5320         }
5321
5322         if (slave)
5323                 return slave->dev;
5324         return NULL;
5325 }
5326
5327 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5328 {
5329         switch (sk->sk_family) {
5330 #if IS_ENABLED(CONFIG_IPV6)
5331         case AF_INET6:
5332                 if (ipv6_only_sock(sk) ||
5333                     ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5334                         flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5335                         flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5336                         flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5337                         break;
5338                 }
5339                 fallthrough;
5340 #endif
5341         default: /* AF_INET */
5342                 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5343                 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5344                 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5345                 break;
5346         }
5347
5348         flow->ports.src = inet_sk(sk)->inet_sport;
5349         flow->ports.dst = inet_sk(sk)->inet_dport;
5350 }
5351
5352 /**
5353  * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5354  * @sk: socket to use for headers
5355  *
5356  * This function will extract the necessary field from the socket and use
5357  * them to generate a hash based on the LAYER34 xmit_policy.
5358  * Assumes that sk is a TCP or UDP socket.
5359  */
5360 static u32 bond_sk_hash_l34(struct sock *sk)
5361 {
5362         struct flow_keys flow;
5363         u32 hash;
5364
5365         bond_sk_to_flow(sk, &flow);
5366
5367         /* L4 */
5368         memcpy(&hash, &flow.ports.ports, sizeof(hash));
5369         /* L3 */
5370         return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5371 }
5372
5373 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5374                                                   struct sock *sk)
5375 {
5376         struct bond_up_slave *slaves;
5377         struct slave *slave;
5378         unsigned int count;
5379         u32 hash;
5380
5381         slaves = rcu_dereference(bond->usable_slaves);
5382         count = slaves ? READ_ONCE(slaves->count) : 0;
5383         if (unlikely(!count))
5384                 return NULL;
5385
5386         hash = bond_sk_hash_l34(sk);
5387         slave = slaves->arr[hash % count];
5388
5389         return slave->dev;
5390 }
5391
5392 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5393                                                 struct sock *sk)
5394 {
5395         struct bonding *bond = netdev_priv(dev);
5396         struct net_device *lower = NULL;
5397
5398         rcu_read_lock();
5399         if (bond_sk_check(bond))
5400                 lower = __bond_sk_get_lower_dev(bond, sk);
5401         rcu_read_unlock();
5402
5403         return lower;
5404 }
5405
5406 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5407 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5408                                         struct net_device *dev)
5409 {
5410         struct net_device *tls_netdev = rcu_dereference(tls_get_ctx(skb->sk)->netdev);
5411
5412         /* tls_netdev might become NULL, even if tls_is_sk_tx_device_offloaded
5413          * was true, if tls_device_down is running in parallel, but it's OK,
5414          * because bond_get_slave_by_dev has a NULL check.
5415          */
5416         if (likely(bond_get_slave_by_dev(bond, tls_netdev)))
5417                 return bond_dev_queue_xmit(bond, skb, tls_netdev);
5418         return bond_tx_drop(dev, skb);
5419 }
5420 #endif
5421
5422 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5423 {
5424         struct bonding *bond = netdev_priv(dev);
5425
5426         if (bond_should_override_tx_queue(bond) &&
5427             !bond_slave_override(bond, skb))
5428                 return NETDEV_TX_OK;
5429
5430 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5431         if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
5432                 return bond_tls_device_xmit(bond, skb, dev);
5433 #endif
5434
5435         switch (BOND_MODE(bond)) {
5436         case BOND_MODE_ROUNDROBIN:
5437                 return bond_xmit_roundrobin(skb, dev);
5438         case BOND_MODE_ACTIVEBACKUP:
5439                 return bond_xmit_activebackup(skb, dev);
5440         case BOND_MODE_8023AD:
5441         case BOND_MODE_XOR:
5442                 return bond_3ad_xor_xmit(skb, dev);
5443         case BOND_MODE_BROADCAST:
5444                 return bond_xmit_broadcast(skb, dev);
5445         case BOND_MODE_ALB:
5446                 return bond_alb_xmit(skb, dev);
5447         case BOND_MODE_TLB:
5448                 return bond_tlb_xmit(skb, dev);
5449         default:
5450                 /* Should never happen, mode already checked */
5451                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5452                 WARN_ON_ONCE(1);
5453                 return bond_tx_drop(dev, skb);
5454         }
5455 }
5456
5457 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5458 {
5459         struct bonding *bond = netdev_priv(dev);
5460         netdev_tx_t ret = NETDEV_TX_OK;
5461
5462         /* If we risk deadlock from transmitting this in the
5463          * netpoll path, tell netpoll to queue the frame for later tx
5464          */
5465         if (unlikely(is_netpoll_tx_blocked(dev)))
5466                 return NETDEV_TX_BUSY;
5467
5468         rcu_read_lock();
5469         if (bond_has_slaves(bond))
5470                 ret = __bond_start_xmit(skb, dev);
5471         else
5472                 ret = bond_tx_drop(dev, skb);
5473         rcu_read_unlock();
5474
5475         return ret;
5476 }
5477
5478 static struct net_device *
5479 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5480 {
5481         struct bonding *bond = netdev_priv(bond_dev);
5482         struct slave *slave;
5483
5484         /* Caller needs to hold rcu_read_lock() */
5485
5486         switch (BOND_MODE(bond)) {
5487         case BOND_MODE_ROUNDROBIN:
5488                 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5489                 break;
5490
5491         case BOND_MODE_ACTIVEBACKUP:
5492                 slave = bond_xmit_activebackup_slave_get(bond);
5493                 break;
5494
5495         case BOND_MODE_8023AD:
5496         case BOND_MODE_XOR:
5497                 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5498                 break;
5499
5500         default:
5501                 /* Should never happen. Mode guarded by bond_xdp_check() */
5502                 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5503                 WARN_ON_ONCE(1);
5504                 return NULL;
5505         }
5506
5507         if (slave)
5508                 return slave->dev;
5509
5510         return NULL;
5511 }
5512
5513 static int bond_xdp_xmit(struct net_device *bond_dev,
5514                          int n, struct xdp_frame **frames, u32 flags)
5515 {
5516         int nxmit, err = -ENXIO;
5517
5518         rcu_read_lock();
5519
5520         for (nxmit = 0; nxmit < n; nxmit++) {
5521                 struct xdp_frame *frame = frames[nxmit];
5522                 struct xdp_frame *frames1[] = {frame};
5523                 struct net_device *slave_dev;
5524                 struct xdp_buff xdp;
5525
5526                 xdp_convert_frame_to_buff(frame, &xdp);
5527
5528                 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5529                 if (!slave_dev) {
5530                         err = -ENXIO;
5531                         break;
5532                 }
5533
5534                 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5535                 if (err < 1)
5536                         break;
5537         }
5538
5539         rcu_read_unlock();
5540
5541         /* If error happened on the first frame then we can pass the error up, otherwise
5542          * report the number of frames that were xmitted.
5543          */
5544         if (err < 0)
5545                 return (nxmit == 0 ? err : nxmit);
5546
5547         return nxmit;
5548 }
5549
5550 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5551                         struct netlink_ext_ack *extack)
5552 {
5553         struct bonding *bond = netdev_priv(dev);
5554         struct list_head *iter;
5555         struct slave *slave, *rollback_slave;
5556         struct bpf_prog *old_prog;
5557         struct netdev_bpf xdp = {
5558                 .command = XDP_SETUP_PROG,
5559                 .flags   = 0,
5560                 .prog    = prog,
5561                 .extack  = extack,
5562         };
5563         int err;
5564
5565         ASSERT_RTNL();
5566
5567         if (!bond_xdp_check(bond))
5568                 return -EOPNOTSUPP;
5569
5570         old_prog = bond->xdp_prog;
5571         bond->xdp_prog = prog;
5572
5573         bond_for_each_slave(bond, slave, iter) {
5574                 struct net_device *slave_dev = slave->dev;
5575
5576                 if (!slave_dev->netdev_ops->ndo_bpf ||
5577                     !slave_dev->netdev_ops->ndo_xdp_xmit) {
5578                         SLAVE_NL_ERR(dev, slave_dev, extack,
5579                                      "Slave device does not support XDP");
5580                         err = -EOPNOTSUPP;
5581                         goto err;
5582                 }
5583
5584                 if (dev_xdp_prog_count(slave_dev) > 0) {
5585                         SLAVE_NL_ERR(dev, slave_dev, extack,
5586                                      "Slave has XDP program loaded, please unload before enslaving");
5587                         err = -EOPNOTSUPP;
5588                         goto err;
5589                 }
5590
5591                 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5592                 if (err < 0) {
5593                         /* ndo_bpf() sets extack error message */
5594                         slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5595                         goto err;
5596                 }
5597                 if (prog)
5598                         bpf_prog_inc(prog);
5599         }
5600
5601         if (prog) {
5602                 static_branch_inc(&bpf_master_redirect_enabled_key);
5603         } else if (old_prog) {
5604                 bpf_prog_put(old_prog);
5605                 static_branch_dec(&bpf_master_redirect_enabled_key);
5606         }
5607
5608         return 0;
5609
5610 err:
5611         /* unwind the program changes */
5612         bond->xdp_prog = old_prog;
5613         xdp.prog = old_prog;
5614         xdp.extack = NULL; /* do not overwrite original error */
5615
5616         bond_for_each_slave(bond, rollback_slave, iter) {
5617                 struct net_device *slave_dev = rollback_slave->dev;
5618                 int err_unwind;
5619
5620                 if (slave == rollback_slave)
5621                         break;
5622
5623                 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5624                 if (err_unwind < 0)
5625                         slave_err(dev, slave_dev,
5626                                   "Error %d when unwinding XDP program change\n", err_unwind);
5627                 else if (xdp.prog)
5628                         bpf_prog_inc(xdp.prog);
5629         }
5630         return err;
5631 }
5632
5633 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5634 {
5635         switch (xdp->command) {
5636         case XDP_SETUP_PROG:
5637                 return bond_xdp_set(dev, xdp->prog, xdp->extack);
5638         default:
5639                 return -EINVAL;
5640         }
5641 }
5642
5643 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5644 {
5645         if (speed == 0 || speed == SPEED_UNKNOWN)
5646                 speed = slave->speed;
5647         else
5648                 speed = min(speed, slave->speed);
5649
5650         return speed;
5651 }
5652
5653 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5654                                            struct ethtool_link_ksettings *cmd)
5655 {
5656         struct bonding *bond = netdev_priv(bond_dev);
5657         struct list_head *iter;
5658         struct slave *slave;
5659         u32 speed = 0;
5660
5661         cmd->base.duplex = DUPLEX_UNKNOWN;
5662         cmd->base.port = PORT_OTHER;
5663
5664         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5665          * do not need to check mode.  Though link speed might not represent
5666          * the true receive or transmit bandwidth (not all modes are symmetric)
5667          * this is an accurate maximum.
5668          */
5669         bond_for_each_slave(bond, slave, iter) {
5670                 if (bond_slave_can_tx(slave)) {
5671                         if (slave->speed != SPEED_UNKNOWN) {
5672                                 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5673                                         speed = bond_mode_bcast_speed(slave,
5674                                                                       speed);
5675                                 else
5676                                         speed += slave->speed;
5677                         }
5678                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5679                             slave->duplex != DUPLEX_UNKNOWN)
5680                                 cmd->base.duplex = slave->duplex;
5681                 }
5682         }
5683         cmd->base.speed = speed ? : SPEED_UNKNOWN;
5684
5685         return 0;
5686 }
5687
5688 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5689                                      struct ethtool_drvinfo *drvinfo)
5690 {
5691         strscpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5692         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5693                  BOND_ABI_VERSION);
5694 }
5695
5696 static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5697                                     struct ethtool_ts_info *info)
5698 {
5699         struct bonding *bond = netdev_priv(bond_dev);
5700         struct ethtool_ts_info ts_info;
5701         const struct ethtool_ops *ops;
5702         struct net_device *real_dev;
5703         bool sw_tx_support = false;
5704         struct phy_device *phydev;
5705         struct list_head *iter;
5706         struct slave *slave;
5707         int ret = 0;
5708
5709         rcu_read_lock();
5710         real_dev = bond_option_active_slave_get_rcu(bond);
5711         dev_hold(real_dev);
5712         rcu_read_unlock();
5713
5714         if (real_dev) {
5715                 ops = real_dev->ethtool_ops;
5716                 phydev = real_dev->phydev;
5717
5718                 if (phy_has_tsinfo(phydev)) {
5719                         ret = phy_ts_info(phydev, info);
5720                         goto out;
5721                 } else if (ops->get_ts_info) {
5722                         ret = ops->get_ts_info(real_dev, info);
5723                         goto out;
5724                 }
5725         } else {
5726                 /* Check if all slaves support software tx timestamping */
5727                 rcu_read_lock();
5728                 bond_for_each_slave_rcu(bond, slave, iter) {
5729                         ret = -1;
5730                         ops = slave->dev->ethtool_ops;
5731                         phydev = slave->dev->phydev;
5732
5733                         if (phy_has_tsinfo(phydev))
5734                                 ret = phy_ts_info(phydev, &ts_info);
5735                         else if (ops->get_ts_info)
5736                                 ret = ops->get_ts_info(slave->dev, &ts_info);
5737
5738                         if (!ret && (ts_info.so_timestamping & SOF_TIMESTAMPING_TX_SOFTWARE)) {
5739                                 sw_tx_support = true;
5740                                 continue;
5741                         }
5742
5743                         sw_tx_support = false;
5744                         break;
5745                 }
5746                 rcu_read_unlock();
5747         }
5748
5749         ret = 0;
5750         info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
5751                                 SOF_TIMESTAMPING_SOFTWARE;
5752         if (sw_tx_support)
5753                 info->so_timestamping |= SOF_TIMESTAMPING_TX_SOFTWARE;
5754
5755         info->phc_index = -1;
5756
5757 out:
5758         dev_put(real_dev);
5759         return ret;
5760 }
5761
5762 static const struct ethtool_ops bond_ethtool_ops = {
5763         .get_drvinfo            = bond_ethtool_get_drvinfo,
5764         .get_link               = ethtool_op_get_link,
5765         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
5766         .get_ts_info            = bond_ethtool_get_ts_info,
5767 };
5768
5769 static const struct net_device_ops bond_netdev_ops = {
5770         .ndo_init               = bond_init,
5771         .ndo_uninit             = bond_uninit,
5772         .ndo_open               = bond_open,
5773         .ndo_stop               = bond_close,
5774         .ndo_start_xmit         = bond_start_xmit,
5775         .ndo_select_queue       = bond_select_queue,
5776         .ndo_get_stats64        = bond_get_stats,
5777         .ndo_eth_ioctl          = bond_eth_ioctl,
5778         .ndo_siocbond           = bond_do_ioctl,
5779         .ndo_siocdevprivate     = bond_siocdevprivate,
5780         .ndo_change_rx_flags    = bond_change_rx_flags,
5781         .ndo_set_rx_mode        = bond_set_rx_mode,
5782         .ndo_change_mtu         = bond_change_mtu,
5783         .ndo_set_mac_address    = bond_set_mac_address,
5784         .ndo_neigh_setup        = bond_neigh_setup,
5785         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
5786         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
5787 #ifdef CONFIG_NET_POLL_CONTROLLER
5788         .ndo_netpoll_setup      = bond_netpoll_setup,
5789         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
5790         .ndo_poll_controller    = bond_poll_controller,
5791 #endif
5792         .ndo_add_slave          = bond_enslave,
5793         .ndo_del_slave          = bond_release,
5794         .ndo_fix_features       = bond_fix_features,
5795         .ndo_features_check     = passthru_features_check,
5796         .ndo_get_xmit_slave     = bond_xmit_get_slave,
5797         .ndo_sk_get_lower_dev   = bond_sk_get_lower_dev,
5798         .ndo_bpf                = bond_xdp,
5799         .ndo_xdp_xmit           = bond_xdp_xmit,
5800         .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5801 };
5802
5803 static const struct device_type bond_type = {
5804         .name = "bond",
5805 };
5806
5807 static void bond_destructor(struct net_device *bond_dev)
5808 {
5809         struct bonding *bond = netdev_priv(bond_dev);
5810
5811         if (bond->wq)
5812                 destroy_workqueue(bond->wq);
5813
5814         if (bond->rr_tx_counter)
5815                 free_percpu(bond->rr_tx_counter);
5816 }
5817
5818 void bond_setup(struct net_device *bond_dev)
5819 {
5820         struct bonding *bond = netdev_priv(bond_dev);
5821
5822         spin_lock_init(&bond->mode_lock);
5823         bond->params = bonding_defaults;
5824
5825         /* Initialize pointers */
5826         bond->dev = bond_dev;
5827
5828         /* Initialize the device entry points */
5829         ether_setup(bond_dev);
5830         bond_dev->max_mtu = ETH_MAX_MTU;
5831         bond_dev->netdev_ops = &bond_netdev_ops;
5832         bond_dev->ethtool_ops = &bond_ethtool_ops;
5833
5834         bond_dev->needs_free_netdev = true;
5835         bond_dev->priv_destructor = bond_destructor;
5836
5837         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5838
5839         /* Initialize the device options */
5840         bond_dev->flags |= IFF_MASTER;
5841         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5842         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5843
5844 #ifdef CONFIG_XFRM_OFFLOAD
5845         /* set up xfrm device ops (only supported in active-backup right now) */
5846         bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5847         INIT_LIST_HEAD(&bond->ipsec_list);
5848         spin_lock_init(&bond->ipsec_lock);
5849 #endif /* CONFIG_XFRM_OFFLOAD */
5850
5851         /* don't acquire bond device's netif_tx_lock when transmitting */
5852         bond_dev->features |= NETIF_F_LLTX;
5853
5854         /* By default, we declare the bond to be fully
5855          * VLAN hardware accelerated capable. Special
5856          * care is taken in the various xmit functions
5857          * when there are slaves that are not hw accel
5858          * capable
5859          */
5860
5861         /* Don't allow bond devices to change network namespaces. */
5862         bond_dev->features |= NETIF_F_NETNS_LOCAL;
5863
5864         bond_dev->hw_features = BOND_VLAN_FEATURES |
5865                                 NETIF_F_HW_VLAN_CTAG_RX |
5866                                 NETIF_F_HW_VLAN_CTAG_FILTER;
5867
5868         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5869         bond_dev->features |= bond_dev->hw_features;
5870         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5871 #ifdef CONFIG_XFRM_OFFLOAD
5872         bond_dev->hw_features |= BOND_XFRM_FEATURES;
5873         /* Only enable XFRM features if this is an active-backup config */
5874         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5875                 bond_dev->features |= BOND_XFRM_FEATURES;
5876 #endif /* CONFIG_XFRM_OFFLOAD */
5877 }
5878
5879 /* Destroy a bonding device.
5880  * Must be under rtnl_lock when this function is called.
5881  */
5882 static void bond_uninit(struct net_device *bond_dev)
5883 {
5884         struct bonding *bond = netdev_priv(bond_dev);
5885         struct bond_up_slave *usable, *all;
5886         struct list_head *iter;
5887         struct slave *slave;
5888
5889         bond_netpoll_cleanup(bond_dev);
5890
5891         /* Release the bonded slaves */
5892         bond_for_each_slave(bond, slave, iter)
5893                 __bond_release_one(bond_dev, slave->dev, true, true);
5894         netdev_info(bond_dev, "Released all slaves\n");
5895
5896         usable = rtnl_dereference(bond->usable_slaves);
5897         if (usable) {
5898                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5899                 kfree_rcu(usable, rcu);
5900         }
5901
5902         all = rtnl_dereference(bond->all_slaves);
5903         if (all) {
5904                 RCU_INIT_POINTER(bond->all_slaves, NULL);
5905                 kfree_rcu(all, rcu);
5906         }
5907
5908         list_del(&bond->bond_list);
5909
5910         bond_debug_unregister(bond);
5911 }
5912
5913 /*------------------------- Module initialization ---------------------------*/
5914
5915 static int bond_check_params(struct bond_params *params)
5916 {
5917         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5918         struct bond_opt_value newval;
5919         const struct bond_opt_value *valptr;
5920         int arp_all_targets_value = 0;
5921         u16 ad_actor_sys_prio = 0;
5922         u16 ad_user_port_key = 0;
5923         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5924         int arp_ip_count;
5925         int bond_mode   = BOND_MODE_ROUNDROBIN;
5926         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5927         int lacp_fast = 0;
5928         int tlb_dynamic_lb;
5929
5930         /* Convert string parameters. */
5931         if (mode) {
5932                 bond_opt_initstr(&newval, mode);
5933                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5934                 if (!valptr) {
5935                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5936                         return -EINVAL;
5937                 }
5938                 bond_mode = valptr->value;
5939         }
5940
5941         if (xmit_hash_policy) {
5942                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5943                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
5944                     bond_mode == BOND_MODE_BROADCAST) {
5945                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5946                                 bond_mode_name(bond_mode));
5947                 } else {
5948                         bond_opt_initstr(&newval, xmit_hash_policy);
5949                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5950                                                 &newval);
5951                         if (!valptr) {
5952                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5953                                        xmit_hash_policy);
5954                                 return -EINVAL;
5955                         }
5956                         xmit_hashtype = valptr->value;
5957                 }
5958         }
5959
5960         if (lacp_rate) {
5961                 if (bond_mode != BOND_MODE_8023AD) {
5962                         pr_info("lacp_rate param is irrelevant in mode %s\n",
5963                                 bond_mode_name(bond_mode));
5964                 } else {
5965                         bond_opt_initstr(&newval, lacp_rate);
5966                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5967                                                 &newval);
5968                         if (!valptr) {
5969                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
5970                                        lacp_rate);
5971                                 return -EINVAL;
5972                         }
5973                         lacp_fast = valptr->value;
5974                 }
5975         }
5976
5977         if (ad_select) {
5978                 bond_opt_initstr(&newval, ad_select);
5979                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
5980                                         &newval);
5981                 if (!valptr) {
5982                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
5983                         return -EINVAL;
5984                 }
5985                 params->ad_select = valptr->value;
5986                 if (bond_mode != BOND_MODE_8023AD)
5987                         pr_warn("ad_select param only affects 802.3ad mode\n");
5988         } else {
5989                 params->ad_select = BOND_AD_STABLE;
5990         }
5991
5992         if (max_bonds < 0) {
5993                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
5994                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
5995                 max_bonds = BOND_DEFAULT_MAX_BONDS;
5996         }
5997
5998         if (miimon < 0) {
5999                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6000                         miimon, INT_MAX);
6001                 miimon = 0;
6002         }
6003
6004         if (updelay < 0) {
6005                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6006                         updelay, INT_MAX);
6007                 updelay = 0;
6008         }
6009
6010         if (downdelay < 0) {
6011                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6012                         downdelay, INT_MAX);
6013                 downdelay = 0;
6014         }
6015
6016         if ((use_carrier != 0) && (use_carrier != 1)) {
6017                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
6018                         use_carrier);
6019                 use_carrier = 1;
6020         }
6021
6022         if (num_peer_notif < 0 || num_peer_notif > 255) {
6023                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
6024                         num_peer_notif);
6025                 num_peer_notif = 1;
6026         }
6027
6028         /* reset values for 802.3ad/TLB/ALB */
6029         if (!bond_mode_uses_arp(bond_mode)) {
6030                 if (!miimon) {
6031                         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");
6032                         pr_warn("Forcing miimon to 100msec\n");
6033                         miimon = BOND_DEFAULT_MIIMON;
6034                 }
6035         }
6036
6037         if (tx_queues < 1 || tx_queues > 255) {
6038                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
6039                         tx_queues, BOND_DEFAULT_TX_QUEUES);
6040                 tx_queues = BOND_DEFAULT_TX_QUEUES;
6041         }
6042
6043         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
6044                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
6045                         all_slaves_active);
6046                 all_slaves_active = 0;
6047         }
6048
6049         if (resend_igmp < 0 || resend_igmp > 255) {
6050                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
6051                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
6052                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
6053         }
6054
6055         bond_opt_initval(&newval, packets_per_slave);
6056         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
6057                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
6058                         packets_per_slave, USHRT_MAX);
6059                 packets_per_slave = 1;
6060         }
6061
6062         if (bond_mode == BOND_MODE_ALB) {
6063                 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",
6064                           updelay);
6065         }
6066
6067         if (!miimon) {
6068                 if (updelay || downdelay) {
6069                         /* just warn the user the up/down delay will have
6070                          * no effect since miimon is zero...
6071                          */
6072                         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",
6073                                 updelay, downdelay);
6074                 }
6075         } else {
6076                 /* don't allow arp monitoring */
6077                 if (arp_interval) {
6078                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
6079                                 miimon, arp_interval);
6080                         arp_interval = 0;
6081                 }
6082
6083                 if ((updelay % miimon) != 0) {
6084                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
6085                                 updelay, miimon, (updelay / miimon) * miimon);
6086                 }
6087
6088                 updelay /= miimon;
6089
6090                 if ((downdelay % miimon) != 0) {
6091                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
6092                                 downdelay, miimon,
6093                                 (downdelay / miimon) * miimon);
6094                 }
6095
6096                 downdelay /= miimon;
6097         }
6098
6099         if (arp_interval < 0) {
6100                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
6101                         arp_interval, INT_MAX);
6102                 arp_interval = 0;
6103         }
6104
6105         for (arp_ip_count = 0, i = 0;
6106              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
6107                 __be32 ip;
6108
6109                 /* not a complete check, but good enough to catch mistakes */
6110                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
6111                     !bond_is_ip_target_ok(ip)) {
6112                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
6113                                 arp_ip_target[i]);
6114                         arp_interval = 0;
6115                 } else {
6116                         if (bond_get_targets_ip(arp_target, ip) == -1)
6117                                 arp_target[arp_ip_count++] = ip;
6118                         else
6119                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
6120                                         &ip);
6121                 }
6122         }
6123
6124         if (arp_interval && !arp_ip_count) {
6125                 /* don't allow arping if no arp_ip_target given... */
6126                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
6127                         arp_interval);
6128                 arp_interval = 0;
6129         }
6130
6131         if (arp_validate) {
6132                 if (!arp_interval) {
6133                         pr_err("arp_validate requires arp_interval\n");
6134                         return -EINVAL;
6135                 }
6136
6137                 bond_opt_initstr(&newval, arp_validate);
6138                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6139                                         &newval);
6140                 if (!valptr) {
6141                         pr_err("Error: invalid arp_validate \"%s\"\n",
6142                                arp_validate);
6143                         return -EINVAL;
6144                 }
6145                 arp_validate_value = valptr->value;
6146         } else {
6147                 arp_validate_value = 0;
6148         }
6149
6150         if (arp_all_targets) {
6151                 bond_opt_initstr(&newval, arp_all_targets);
6152                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6153                                         &newval);
6154                 if (!valptr) {
6155                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6156                                arp_all_targets);
6157                         arp_all_targets_value = 0;
6158                 } else {
6159                         arp_all_targets_value = valptr->value;
6160                 }
6161         }
6162
6163         if (miimon) {
6164                 pr_info("MII link monitoring set to %d ms\n", miimon);
6165         } else if (arp_interval) {
6166                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6167                                           arp_validate_value);
6168                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6169                         arp_interval, valptr->string, arp_ip_count);
6170
6171                 for (i = 0; i < arp_ip_count; i++)
6172                         pr_cont(" %s", arp_ip_target[i]);
6173
6174                 pr_cont("\n");
6175
6176         } else if (max_bonds) {
6177                 /* miimon and arp_interval not set, we need one so things
6178                  * work as expected, see bonding.txt for details
6179                  */
6180                 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");
6181         }
6182
6183         if (primary && !bond_mode_uses_primary(bond_mode)) {
6184                 /* currently, using a primary only makes sense
6185                  * in active backup, TLB or ALB modes
6186                  */
6187                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6188                         primary, bond_mode_name(bond_mode));
6189                 primary = NULL;
6190         }
6191
6192         if (primary && primary_reselect) {
6193                 bond_opt_initstr(&newval, primary_reselect);
6194                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6195                                         &newval);
6196                 if (!valptr) {
6197                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
6198                                primary_reselect);
6199                         return -EINVAL;
6200                 }
6201                 primary_reselect_value = valptr->value;
6202         } else {
6203                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6204         }
6205
6206         if (fail_over_mac) {
6207                 bond_opt_initstr(&newval, fail_over_mac);
6208                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6209                                         &newval);
6210                 if (!valptr) {
6211                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
6212                                fail_over_mac);
6213                         return -EINVAL;
6214                 }
6215                 fail_over_mac_value = valptr->value;
6216                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6217                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6218         } else {
6219                 fail_over_mac_value = BOND_FOM_NONE;
6220         }
6221
6222         bond_opt_initstr(&newval, "default");
6223         valptr = bond_opt_parse(
6224                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6225                                      &newval);
6226         if (!valptr) {
6227                 pr_err("Error: No ad_actor_sys_prio default value");
6228                 return -EINVAL;
6229         }
6230         ad_actor_sys_prio = valptr->value;
6231
6232         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6233                                 &newval);
6234         if (!valptr) {
6235                 pr_err("Error: No ad_user_port_key default value");
6236                 return -EINVAL;
6237         }
6238         ad_user_port_key = valptr->value;
6239
6240         bond_opt_initstr(&newval, "default");
6241         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6242         if (!valptr) {
6243                 pr_err("Error: No tlb_dynamic_lb default value");
6244                 return -EINVAL;
6245         }
6246         tlb_dynamic_lb = valptr->value;
6247
6248         if (lp_interval == 0) {
6249                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6250                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6251                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6252         }
6253
6254         /* fill params struct with the proper values */
6255         params->mode = bond_mode;
6256         params->xmit_policy = xmit_hashtype;
6257         params->miimon = miimon;
6258         params->num_peer_notif = num_peer_notif;
6259         params->arp_interval = arp_interval;
6260         params->arp_validate = arp_validate_value;
6261         params->arp_all_targets = arp_all_targets_value;
6262         params->missed_max = 2;
6263         params->updelay = updelay;
6264         params->downdelay = downdelay;
6265         params->peer_notif_delay = 0;
6266         params->use_carrier = use_carrier;
6267         params->lacp_active = 1;
6268         params->lacp_fast = lacp_fast;
6269         params->primary[0] = 0;
6270         params->primary_reselect = primary_reselect_value;
6271         params->fail_over_mac = fail_over_mac_value;
6272         params->tx_queues = tx_queues;
6273         params->all_slaves_active = all_slaves_active;
6274         params->resend_igmp = resend_igmp;
6275         params->min_links = min_links;
6276         params->lp_interval = lp_interval;
6277         params->packets_per_slave = packets_per_slave;
6278         params->tlb_dynamic_lb = tlb_dynamic_lb;
6279         params->ad_actor_sys_prio = ad_actor_sys_prio;
6280         eth_zero_addr(params->ad_actor_system);
6281         params->ad_user_port_key = ad_user_port_key;
6282         if (packets_per_slave > 0) {
6283                 params->reciprocal_packets_per_slave =
6284                         reciprocal_value(packets_per_slave);
6285         } else {
6286                 /* reciprocal_packets_per_slave is unused if
6287                  * packets_per_slave is 0 or 1, just initialize it
6288                  */
6289                 params->reciprocal_packets_per_slave =
6290                         (struct reciprocal_value) { 0 };
6291         }
6292
6293         if (primary)
6294                 strscpy_pad(params->primary, primary, sizeof(params->primary));
6295
6296         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6297 #if IS_ENABLED(CONFIG_IPV6)
6298         memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6299 #endif
6300
6301         return 0;
6302 }
6303
6304 /* Called from registration process */
6305 static int bond_init(struct net_device *bond_dev)
6306 {
6307         struct bonding *bond = netdev_priv(bond_dev);
6308         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6309
6310         netdev_dbg(bond_dev, "Begin bond_init\n");
6311
6312         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
6313         if (!bond->wq)
6314                 return -ENOMEM;
6315
6316         spin_lock_init(&bond->stats_lock);
6317         netdev_lockdep_set_classes(bond_dev);
6318
6319         list_add_tail(&bond->bond_list, &bn->dev_list);
6320
6321         bond_prepare_sysfs_group(bond);
6322
6323         bond_debug_register(bond);
6324
6325         /* Ensure valid dev_addr */
6326         if (is_zero_ether_addr(bond_dev->dev_addr) &&
6327             bond_dev->addr_assign_type == NET_ADDR_PERM)
6328                 eth_hw_addr_random(bond_dev);
6329
6330         return 0;
6331 }
6332
6333 unsigned int bond_get_num_tx_queues(void)
6334 {
6335         return tx_queues;
6336 }
6337
6338 /* Create a new bond based on the specified name and bonding parameters.
6339  * If name is NULL, obtain a suitable "bond%d" name for us.
6340  * Caller must NOT hold rtnl_lock; we need to release it here before we
6341  * set up our sysfs entries.
6342  */
6343 int bond_create(struct net *net, const char *name)
6344 {
6345         struct net_device *bond_dev;
6346         struct bonding *bond;
6347         int res = -ENOMEM;
6348
6349         rtnl_lock();
6350
6351         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6352                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
6353                                    bond_setup, tx_queues);
6354         if (!bond_dev)
6355                 goto out;
6356
6357         bond = netdev_priv(bond_dev);
6358         dev_net_set(bond_dev, net);
6359         bond_dev->rtnl_link_ops = &bond_link_ops;
6360
6361         res = register_netdevice(bond_dev);
6362         if (res < 0) {
6363                 free_netdev(bond_dev);
6364                 goto out;
6365         }
6366
6367         netif_carrier_off(bond_dev);
6368
6369         bond_work_init_all(bond);
6370
6371 out:
6372         rtnl_unlock();
6373         return res;
6374 }
6375
6376 static int __net_init bond_net_init(struct net *net)
6377 {
6378         struct bond_net *bn = net_generic(net, bond_net_id);
6379
6380         bn->net = net;
6381         INIT_LIST_HEAD(&bn->dev_list);
6382
6383         bond_create_proc_dir(bn);
6384         bond_create_sysfs(bn);
6385
6386         return 0;
6387 }
6388
6389 static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6390 {
6391         struct bond_net *bn;
6392         struct net *net;
6393         LIST_HEAD(list);
6394
6395         list_for_each_entry(net, net_list, exit_list) {
6396                 bn = net_generic(net, bond_net_id);
6397                 bond_destroy_sysfs(bn);
6398         }
6399
6400         /* Kill off any bonds created after unregistering bond rtnl ops */
6401         rtnl_lock();
6402         list_for_each_entry(net, net_list, exit_list) {
6403                 struct bonding *bond, *tmp_bond;
6404
6405                 bn = net_generic(net, bond_net_id);
6406                 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6407                         unregister_netdevice_queue(bond->dev, &list);
6408         }
6409         unregister_netdevice_many(&list);
6410         rtnl_unlock();
6411
6412         list_for_each_entry(net, net_list, exit_list) {
6413                 bn = net_generic(net, bond_net_id);
6414                 bond_destroy_proc_dir(bn);
6415         }
6416 }
6417
6418 static struct pernet_operations bond_net_ops = {
6419         .init = bond_net_init,
6420         .exit_batch = bond_net_exit_batch,
6421         .id   = &bond_net_id,
6422         .size = sizeof(struct bond_net),
6423 };
6424
6425 static int __init bonding_init(void)
6426 {
6427         int i;
6428         int res;
6429
6430         res = bond_check_params(&bonding_defaults);
6431         if (res)
6432                 goto out;
6433
6434         res = register_pernet_subsys(&bond_net_ops);
6435         if (res)
6436                 goto out;
6437
6438         res = bond_netlink_init();
6439         if (res)
6440                 goto err_link;
6441
6442         bond_create_debugfs();
6443
6444         for (i = 0; i < max_bonds; i++) {
6445                 res = bond_create(&init_net, NULL);
6446                 if (res)
6447                         goto err;
6448         }
6449
6450         skb_flow_dissector_init(&flow_keys_bonding,
6451                                 flow_keys_bonding_keys,
6452                                 ARRAY_SIZE(flow_keys_bonding_keys));
6453
6454         register_netdevice_notifier(&bond_netdev_notifier);
6455 out:
6456         return res;
6457 err:
6458         bond_destroy_debugfs();
6459         bond_netlink_fini();
6460 err_link:
6461         unregister_pernet_subsys(&bond_net_ops);
6462         goto out;
6463
6464 }
6465
6466 static void __exit bonding_exit(void)
6467 {
6468         unregister_netdevice_notifier(&bond_netdev_notifier);
6469
6470         bond_destroy_debugfs();
6471
6472         bond_netlink_fini();
6473         unregister_pernet_subsys(&bond_net_ops);
6474
6475 #ifdef CONFIG_NET_POLL_CONTROLLER
6476         /* Make sure we don't have an imbalance on our netpoll blocking */
6477         WARN_ON(atomic_read(&netpoll_block_tx));
6478 #endif
6479 }
6480
6481 module_init(bonding_init);
6482 module_exit(bonding_exit);
6483 MODULE_LICENSE("GPL");
6484 MODULE_DESCRIPTION(DRV_DESCRIPTION);
6485 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");