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 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  **/
 423 static int bond_ipsec_add_sa(struct xfrm_state *xs)
 424 {
 425         struct net_device *bond_dev = xs->xso.dev;
 426         struct bond_ipsec *ipsec;
 427         struct bonding *bond;
 428         struct slave *slave;
 429         int err;
 430
 431         if (!bond_dev)
 432                 return -EINVAL;
 433
 434         rcu_read_lock();
 435         bond = netdev_priv(bond_dev);
 436         slave = rcu_dereference(bond->curr_active_slave);
 437         if (!slave) {
 438                 rcu_read_unlock();
 439                 return -ENODEV;
 440         }
 441
 442         if (!slave->dev->xfrmdev_ops ||
 443             !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
 444             netif_is_bond_master(slave->dev)) {
 445                 slave_warn(bond_dev, slave->dev, "Slave does not support ipsec offload\n");
 446                 rcu_read_unlock();
 447                 return -EINVAL;
 448         }
 449
 450         ipsec = kmalloc(sizeof(*ipsec), GFP_ATOMIC);
 451         if (!ipsec) {
 452                 rcu_read_unlock();
 453                 return -ENOMEM;
 454         }
 455         xs->xso.real_dev = slave->dev;
 456
 457         err = slave->dev->xfrmdev_ops->xdo_dev_state_add(xs);
 458         if (!err) {
 459                 ipsec->xs = xs;
 460                 INIT_LIST_HEAD(&ipsec->list);
 461                 spin_lock_bh(&bond->ipsec_lock);
 462                 list_add(&ipsec->list, &bond->ipsec_list);
 463                 spin_unlock_bh(&bond->ipsec_lock);
 464         } else {
 465                 kfree(ipsec);
 466         }
 467         rcu_read_unlock();
 468         return err;
 469 }
 470
 471 static void bond_ipsec_add_sa_all(struct bonding *bond)
 472 {
 473         struct net_device *bond_dev = bond->dev;
 474         struct bond_ipsec *ipsec;
 475         struct slave *slave;
 476
 477         rcu_read_lock();
 478         slave = rcu_dereference(bond->curr_active_slave);
 479         if (!slave)
 480                 goto out;
 481
 482         if (!slave->dev->xfrmdev_ops ||
 483             !slave->dev->xfrmdev_ops->xdo_dev_state_add ||
 484             netif_is_bond_master(slave->dev)) {
 485                 spin_lock_bh(&bond->ipsec_lock);
 486                 if (!list_empty(&bond->ipsec_list))
 487                         slave_warn(bond_dev, slave->dev,
 488                                    "%s: no slave xdo_dev_state_add\n",
 489                                    __func__);
 490                 spin_unlock_bh(&bond->ipsec_lock);
 491                 goto out;
 492         }
 493
 494         spin_lock_bh(&bond->ipsec_lock);
 495         list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 496                 ipsec->xs->xso.real_dev = slave->dev;
 497                 if (slave->dev->xfrmdev_ops->xdo_dev_state_add(ipsec->xs)) {
 498                         slave_warn(bond_dev, slave->dev, "%s: failed to add SA\n", __func__);
 499                         ipsec->xs->xso.real_dev = NULL;
 500                 }
 501         }
 502         spin_unlock_bh(&bond->ipsec_lock);
 503 out:
 504         rcu_read_unlock();
 505 }
 506
 507 /**
 508  * bond_ipsec_del_sa - clear out this specific SA
 509  * @xs: pointer to transformer state struct
 510  **/
 511 static void bond_ipsec_del_sa(struct xfrm_state *xs)
 512 {
 513         struct net_device *bond_dev = xs->xso.dev;
 514         struct bond_ipsec *ipsec;
 515         struct bonding *bond;
 516         struct slave *slave;
 517
 518         if (!bond_dev)
 519                 return;
 520
 521         rcu_read_lock();
 522         bond = netdev_priv(bond_dev);
 523         slave = rcu_dereference(bond->curr_active_slave);
 524
 525         if (!slave)
 526                 goto out;
 527
 528         if (!xs->xso.real_dev)
 529                 goto out;
 530
 531         WARN_ON(xs->xso.real_dev != slave->dev);
 532
 533         if (!slave->dev->xfrmdev_ops ||
 534             !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
 535             netif_is_bond_master(slave->dev)) {
 536                 slave_warn(bond_dev, slave->dev, "%s: no slave xdo_dev_state_delete\n", __func__);
 537                 goto out;
 538         }
 539
 540         slave->dev->xfrmdev_ops->xdo_dev_state_delete(xs);
 541 out:
 542         spin_lock_bh(&bond->ipsec_lock);
 543         list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 544                 if (ipsec->xs == xs) {
 545                         list_del(&ipsec->list);
 546                         kfree(ipsec);
 547                         break;
 548                 }
 549         }
 550         spin_unlock_bh(&bond->ipsec_lock);
 551         rcu_read_unlock();
 552 }
 553
 554 static void bond_ipsec_del_sa_all(struct bonding *bond)
 555 {
 556         struct net_device *bond_dev = bond->dev;
 557         struct bond_ipsec *ipsec;
 558         struct slave *slave;
 559
 560         rcu_read_lock();
 561         slave = rcu_dereference(bond->curr_active_slave);
 562         if (!slave) {
 563                 rcu_read_unlock();
 564                 return;
 565         }
 566
 567         spin_lock_bh(&bond->ipsec_lock);
 568         list_for_each_entry(ipsec, &bond->ipsec_list, list) {
 569                 if (!ipsec->xs->xso.real_dev)
 570                         continue;
 571
 572                 if (!slave->dev->xfrmdev_ops ||
 573                     !slave->dev->xfrmdev_ops->xdo_dev_state_delete ||
 574                     netif_is_bond_master(slave->dev)) {
 575                         slave_warn(bond_dev, slave->dev,
 576                                    "%s: no slave xdo_dev_state_delete\n",
 577                                    __func__);
 578                 } else {
 579                         slave->dev->xfrmdev_ops->xdo_dev_state_delete(ipsec->xs);
 580                 }
 581                 ipsec->xs->xso.real_dev = NULL;
 582         }
 583         spin_unlock_bh(&bond->ipsec_lock);
 584         rcu_read_unlock();
 585 }
 586
 587 /**
 588  * bond_ipsec_offload_ok - can this packet use the xfrm hw offload
 589  * @skb: current data packet
 590  * @xs: pointer to transformer state struct
 591  **/
 592 static bool bond_ipsec_offload_ok(struct sk_buff *skb, struct xfrm_state *xs)
 593 {
 594         struct net_device *bond_dev = xs->xso.dev;
 595         struct net_device *real_dev;
 596         struct slave *curr_active;
 597         struct bonding *bond;
 598         int err;
 599
 600         bond = netdev_priv(bond_dev);
 601         rcu_read_lock();
 602         curr_active = rcu_dereference(bond->curr_active_slave);
 603         real_dev = curr_active->dev;
 604
 605         if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
 606                 err = false;
 607                 goto out;
 608         }
 609
 610         if (!xs->xso.real_dev) {
 611                 err = false;
 612                 goto out;
 613         }
 614
 615         if (!real_dev->xfrmdev_ops ||
 616             !real_dev->xfrmdev_ops->xdo_dev_offload_ok ||
 617             netif_is_bond_master(real_dev)) {
 618                 err = false;
 619                 goto out;
 620         }
 621
 622         err = real_dev->xfrmdev_ops->xdo_dev_offload_ok(skb, xs);
 623 out:
 624         rcu_read_unlock();
 625         return err;
 626 }
 627
 628 static const struct xfrmdev_ops bond_xfrmdev_ops = {
 629         .xdo_dev_state_add = bond_ipsec_add_sa,
 630         .xdo_dev_state_delete = bond_ipsec_del_sa,
 631         .xdo_dev_offload_ok = bond_ipsec_offload_ok,
 632 };
 633 #endif /* CONFIG_XFRM_OFFLOAD */
 634
 635 /*------------------------------- Link status -------------------------------*/
 636
 637 /* Set the carrier state for the master according to the state of its
 638  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
 639  * do special 802.3ad magic.
 640  *
 641  * Returns zero if carrier state does not change, nonzero if it does.
 642  */
 643 int bond_set_carrier(struct bonding *bond)
 644 {
 645         struct list_head *iter;
 646         struct slave *slave;
 647
 648         if (!bond_has_slaves(bond))
 649                 goto down;
 650
 651         if (BOND_MODE(bond) == BOND_MODE_8023AD)
 652                 return bond_3ad_set_carrier(bond);
 653
 654         bond_for_each_slave(bond, slave, iter) {
 655                 if (slave->link == BOND_LINK_UP) {
 656                         if (!netif_carrier_ok(bond->dev)) {
 657                                 netif_carrier_on(bond->dev);
 658                                 return 1;
 659                         }
 660                         return 0;
 661                 }
 662         }
 663
 664 down:
 665         if (netif_carrier_ok(bond->dev)) {
 666                 netif_carrier_off(bond->dev);
 667                 return 1;
 668         }
 669         return 0;
 670 }
 671
 672 /* Get link speed and duplex from the slave's base driver
 673  * using ethtool. If for some reason the call fails or the
 674  * values are invalid, set speed and duplex to -1,
 675  * and return. Return 1 if speed or duplex settings are
 676  * UNKNOWN; 0 otherwise.
 677  */
 678 static int bond_update_speed_duplex(struct slave *slave)
 679 {
 680         struct net_device *slave_dev = slave->dev;
 681         struct ethtool_link_ksettings ecmd;
 682         int res;
 683
 684         slave->speed = SPEED_UNKNOWN;
 685         slave->duplex = DUPLEX_UNKNOWN;
 686
 687         res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
 688         if (res < 0)
 689                 return 1;
 690         if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
 691                 return 1;
 692         switch (ecmd.base.duplex) {
 693         case DUPLEX_FULL:
 694         case DUPLEX_HALF:
 695                 break;
 696         default:
 697                 return 1;
 698         }
 699
 700         slave->speed = ecmd.base.speed;
 701         slave->duplex = ecmd.base.duplex;
 702
 703         return 0;
 704 }
 705
 706 const char *bond_slave_link_status(s8 link)
 707 {
 708         switch (link) {
 709         case BOND_LINK_UP:
 710                 return "up";
 711         case BOND_LINK_FAIL:
 712                 return "going down";
 713         case BOND_LINK_DOWN:
 714                 return "down";
 715         case BOND_LINK_BACK:
 716                 return "going back";
 717         default:
 718                 return "unknown";
 719         }
 720 }
 721
 722 /* if <dev> supports MII link status reporting, check its link status.
 723  *
 724  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
 725  * depending upon the setting of the use_carrier parameter.
 726  *
 727  * Return either BMSR_LSTATUS, meaning that the link is up (or we
 728  * can't tell and just pretend it is), or 0, meaning that the link is
 729  * down.
 730  *
 731  * If reporting is non-zero, instead of faking link up, return -1 if
 732  * both ETHTOOL and MII ioctls fail (meaning the device does not
 733  * support them).  If use_carrier is set, return whatever it says.
 734  * It'd be nice if there was a good way to tell if a driver supports
 735  * netif_carrier, but there really isn't.
 736  */
 737 static int bond_check_dev_link(struct bonding *bond,
 738                                struct net_device *slave_dev, int reporting)
 739 {
 740         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
 741         int (*ioctl)(struct net_device *, struct ifreq *, int);
 742         struct ifreq ifr;
 743         struct mii_ioctl_data *mii;
 744
 745         if (!reporting && !netif_running(slave_dev))
 746                 return 0;
 747
 748         if (bond->params.use_carrier)
 749                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
 750
 751         /* Try to get link status using Ethtool first. */
 752         if (slave_dev->ethtool_ops->get_link)
 753                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
 754                         BMSR_LSTATUS : 0;
 755
 756         /* Ethtool can't be used, fallback to MII ioctls. */
 757         ioctl = slave_ops->ndo_eth_ioctl;
 758         if (ioctl) {
 759                 /* TODO: set pointer to correct ioctl on a per team member
 760                  *       bases to make this more efficient. that is, once
 761                  *       we determine the correct ioctl, we will always
 762                  *       call it and not the others for that team
 763                  *       member.
 764                  */
 765
 766                 /* We cannot assume that SIOCGMIIPHY will also read a
 767                  * register; not all network drivers (e.g., e100)
 768                  * support that.
 769                  */
 770
 771                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
 772                 strscpy_pad(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
 773                 mii = if_mii(&ifr);
 774                 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
 775                         mii->reg_num = MII_BMSR;
 776                         if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
 777                                 return mii->val_out & BMSR_LSTATUS;
 778                 }
 779         }
 780
 781         /* If reporting, report that either there's no ndo_eth_ioctl,
 782          * or both SIOCGMIIREG and get_link failed (meaning that we
 783          * cannot report link status).  If not reporting, pretend
 784          * we're ok.
 785          */
 786         return reporting ? -1 : BMSR_LSTATUS;
 787 }
 788
 789 /*----------------------------- Multicast list ------------------------------*/
 790
 791 /* Push the promiscuity flag down to appropriate slaves */
 792 static int bond_set_promiscuity(struct bonding *bond, int inc)
 793 {
 794         struct list_head *iter;
 795         int err = 0;
 796
 797         if (bond_uses_primary(bond)) {
 798                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 799
 800                 if (curr_active)
 801                         err = dev_set_promiscuity(curr_active->dev, inc);
 802         } else {
 803                 struct slave *slave;
 804
 805                 bond_for_each_slave(bond, slave, iter) {
 806                         err = dev_set_promiscuity(slave->dev, inc);
 807                         if (err)
 808                                 return err;
 809                 }
 810         }
 811         return err;
 812 }
 813
 814 /* Push the allmulti flag down to all slaves */
 815 static int bond_set_allmulti(struct bonding *bond, int inc)
 816 {
 817         struct list_head *iter;
 818         int err = 0;
 819
 820         if (bond_uses_primary(bond)) {
 821                 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
 822
 823                 if (curr_active)
 824                         err = dev_set_allmulti(curr_active->dev, inc);
 825         } else {
 826                 struct slave *slave;
 827
 828                 bond_for_each_slave(bond, slave, iter) {
 829                         err = dev_set_allmulti(slave->dev, inc);
 830                         if (err)
 831                                 return err;
 832                 }
 833         }
 834         return err;
 835 }
 836
 837 /* Retrieve the list of registered multicast addresses for the bonding
 838  * device and retransmit an IGMP JOIN request to the current active
 839  * slave.
 840  */
 841 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
 842 {
 843         struct bonding *bond = container_of(work, struct bonding,
 844                                             mcast_work.work);
 845
 846         if (!rtnl_trylock()) {
 847                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
 848                 return;
 849         }
 850         call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
 851
 852         if (bond->igmp_retrans > 1) {
 853                 bond->igmp_retrans--;
 854                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
 855         }
 856         rtnl_unlock();
 857 }
 858
 859 /* Flush bond's hardware addresses from slave */
 860 static void bond_hw_addr_flush(struct net_device *bond_dev,
 861                                struct net_device *slave_dev)
 862 {
 863         struct bonding *bond = netdev_priv(bond_dev);
 864
 865         dev_uc_unsync(slave_dev, bond_dev);
 866         dev_mc_unsync(slave_dev, bond_dev);
 867
 868         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
 869                 /* del lacpdu mc addr from mc list */
 870                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
 871
 872                 dev_mc_del(slave_dev, lacpdu_multicast);
 873         }
 874 }
 875
 876 /*--------------------------- Active slave change ---------------------------*/
 877
 878 /* Update the hardware address list and promisc/allmulti for the new and
 879  * old active slaves (if any).  Modes that are not using primary keep all
 880  * slaves up date at all times; only the modes that use primary need to call
 881  * this function to swap these settings during a failover.
 882  */
 883 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
 884                               struct slave *old_active)
 885 {
 886         if (old_active) {
 887                 if (bond->dev->flags & IFF_PROMISC)
 888                         dev_set_promiscuity(old_active->dev, -1);
 889
 890                 if (bond->dev->flags & IFF_ALLMULTI)
 891                         dev_set_allmulti(old_active->dev, -1);
 892
 893                 bond_hw_addr_flush(bond->dev, old_active->dev);
 894         }
 895
 896         if (new_active) {
 897                 /* FIXME: Signal errors upstream. */
 898                 if (bond->dev->flags & IFF_PROMISC)
 899                         dev_set_promiscuity(new_active->dev, 1);
 900
 901                 if (bond->dev->flags & IFF_ALLMULTI)
 902                         dev_set_allmulti(new_active->dev, 1);
 903
 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  * bond_set_dev_addr - clone slave's address to bond
 913  * @bond_dev: bond net device
 914  * @slave_dev: slave net device
 915  *
 916  * Should be called with RTNL held.
 917  */
 918 static int bond_set_dev_addr(struct net_device *bond_dev,
 919                              struct net_device *slave_dev)
 920 {
 921         int err;
 922
 923         slave_dbg(bond_dev, slave_dev, "bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
 924                   bond_dev, slave_dev, slave_dev->addr_len);
 925         err = dev_pre_changeaddr_notify(bond_dev, slave_dev->dev_addr, NULL);
 926         if (err)
 927                 return err;
 928
 929         __dev_addr_set(bond_dev, slave_dev->dev_addr, slave_dev->addr_len);
 930         bond_dev->addr_assign_type = NET_ADDR_STOLEN;
 931         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
 932         return 0;
 933 }
 934
 935 static struct slave *bond_get_old_active(struct bonding *bond,
 936                                          struct slave *new_active)
 937 {
 938         struct slave *slave;
 939         struct list_head *iter;
 940
 941         bond_for_each_slave(bond, slave, iter) {
 942                 if (slave == new_active)
 943                         continue;
 944
 945                 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
 946                         return slave;
 947         }
 948
 949         return NULL;
 950 }
 951
 952 /* bond_do_fail_over_mac
 953  *
 954  * Perform special MAC address swapping for fail_over_mac settings
 955  *
 956  * Called with RTNL
 957  */
 958 static void bond_do_fail_over_mac(struct bonding *bond,
 959                                   struct slave *new_active,
 960                                   struct slave *old_active)
 961 {
 962         u8 tmp_mac[MAX_ADDR_LEN];
 963         struct sockaddr_storage ss;
 964         int rv;
 965
 966         switch (bond->params.fail_over_mac) {
 967         case BOND_FOM_ACTIVE:
 968                 if (new_active) {
 969                         rv = bond_set_dev_addr(bond->dev, new_active->dev);
 970                         if (rv)
 971                                 slave_err(bond->dev, new_active->dev, "Error %d setting bond MAC from slave\n",
 972                                           -rv);
 973                 }
 974                 break;
 975         case BOND_FOM_FOLLOW:
 976                 /* if new_active && old_active, swap them
 977                  * if just old_active, do nothing (going to no active slave)
 978                  * if just new_active, set new_active to bond's MAC
 979                  */
 980                 if (!new_active)
 981                         return;
 982
 983                 if (!old_active)
 984                         old_active = bond_get_old_active(bond, new_active);
 985
 986                 if (old_active) {
 987                         bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
 988                                           new_active->dev->addr_len);
 989                         bond_hw_addr_copy(ss.__data,
 990                                           old_active->dev->dev_addr,
 991                                           old_active->dev->addr_len);
 992                         ss.ss_family = new_active->dev->type;
 993                 } else {
 994                         bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
 995                                           bond->dev->addr_len);
 996                         ss.ss_family = bond->dev->type;
 997                 }
 998
 999                 rv = dev_set_mac_address(new_active->dev,
1000                                          (struct sockaddr *)&ss, NULL);
1001                 if (rv) {
1002                         slave_err(bond->dev, new_active->dev, "Error %d setting MAC of new active slave\n",
1003                                   -rv);
1004                         goto out;
1005                 }
1006
1007                 if (!old_active)
1008                         goto out;
1009
1010                 bond_hw_addr_copy(ss.__data, tmp_mac,
1011                                   new_active->dev->addr_len);
1012                 ss.ss_family = old_active->dev->type;
1013
1014                 rv = dev_set_mac_address(old_active->dev,
1015                                          (struct sockaddr *)&ss, NULL);
1016                 if (rv)
1017                         slave_err(bond->dev, old_active->dev, "Error %d setting MAC of old active slave\n",
1018                                   -rv);
1019 out:
1020                 break;
1021         default:
1022                 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
1023                            bond->params.fail_over_mac);
1024                 break;
1025         }
1026
1027 }
1028
1029 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
1030 {
1031         struct slave *prim = rtnl_dereference(bond->primary_slave);
1032         struct slave *curr = rtnl_dereference(bond->curr_active_slave);
1033
1034         if (!prim || prim->link != BOND_LINK_UP) {
1035                 if (!curr || curr->link != BOND_LINK_UP)
1036                         return NULL;
1037                 return curr;
1038         }
1039
1040         if (bond->force_primary) {
1041                 bond->force_primary = false;
1042                 return prim;
1043         }
1044
1045         if (!curr || curr->link != BOND_LINK_UP)
1046                 return prim;
1047
1048         /* At this point, prim and curr are both up */
1049         switch (bond->params.primary_reselect) {
1050         case BOND_PRI_RESELECT_ALWAYS:
1051                 return prim;
1052         case BOND_PRI_RESELECT_BETTER:
1053                 if (prim->speed < curr->speed)
1054                         return curr;
1055                 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
1056                         return curr;
1057                 return prim;
1058         case BOND_PRI_RESELECT_FAILURE:
1059                 return curr;
1060         default:
1061                 netdev_err(bond->dev, "impossible primary_reselect %d\n",
1062                            bond->params.primary_reselect);
1063                 return curr;
1064         }
1065 }
1066
1067 /**
1068  * bond_find_best_slave - select the best available slave to be the active one
1069  * @bond: our bonding struct
1070  */
1071 static struct slave *bond_find_best_slave(struct bonding *bond)
1072 {
1073         struct slave *slave, *bestslave = NULL;
1074         struct list_head *iter;
1075         int mintime = bond->params.updelay;
1076
1077         slave = bond_choose_primary_or_current(bond);
1078         if (slave)
1079                 return slave;
1080
1081         bond_for_each_slave(bond, slave, iter) {
1082                 if (slave->link == BOND_LINK_UP)
1083                         return slave;
1084                 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
1085                     slave->delay < mintime) {
1086                         mintime = slave->delay;
1087                         bestslave = slave;
1088                 }
1089         }
1090
1091         return bestslave;
1092 }
1093
1094 static bool bond_should_notify_peers(struct bonding *bond)
1095 {
1096         struct slave *slave;
1097
1098         rcu_read_lock();
1099         slave = rcu_dereference(bond->curr_active_slave);
1100         rcu_read_unlock();
1101
1102         if (!slave || !bond->send_peer_notif ||
1103             bond->send_peer_notif %
1104             max(1, bond->params.peer_notif_delay) != 0 ||
1105             !netif_carrier_ok(bond->dev) ||
1106             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1107                 return false;
1108
1109         netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
1110                    slave ? slave->dev->name : "NULL");
1111
1112         return true;
1113 }
1114
1115 /**
1116  * bond_change_active_slave - change the active slave into the specified one
1117  * @bond: our bonding struct
1118  * @new_active: the new slave to make the active one
1119  *
1120  * Set the new slave to the bond's settings and unset them on the old
1121  * curr_active_slave.
1122  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1123  *
1124  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1125  * because it is apparently the best available slave we have, even though its
1126  * updelay hasn't timed out yet.
1127  *
1128  * Caller must hold RTNL.
1129  */
1130 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1131 {
1132         struct slave *old_active;
1133
1134         ASSERT_RTNL();
1135
1136         old_active = rtnl_dereference(bond->curr_active_slave);
1137
1138         if (old_active == new_active)
1139                 return;
1140
1141 #ifdef CONFIG_XFRM_OFFLOAD
1142         bond_ipsec_del_sa_all(bond);
1143 #endif /* CONFIG_XFRM_OFFLOAD */
1144
1145         if (new_active) {
1146                 new_active->last_link_up = jiffies;
1147
1148                 if (new_active->link == BOND_LINK_BACK) {
1149                         if (bond_uses_primary(bond)) {
1150                                 slave_info(bond->dev, new_active->dev, "making interface the new active one %d ms earlier\n",
1151                                            (bond->params.updelay - new_active->delay) * bond->params.miimon);
1152                         }
1153
1154                         new_active->delay = 0;
1155                         bond_set_slave_link_state(new_active, BOND_LINK_UP,
1156                                                   BOND_SLAVE_NOTIFY_NOW);
1157
1158                         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1159                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1160
1161                         if (bond_is_lb(bond))
1162                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1163                 } else {
1164                         if (bond_uses_primary(bond))
1165                                 slave_info(bond->dev, new_active->dev, "making interface the new active one\n");
1166                 }
1167         }
1168
1169         if (bond_uses_primary(bond))
1170                 bond_hw_addr_swap(bond, new_active, old_active);
1171
1172         if (bond_is_lb(bond)) {
1173                 bond_alb_handle_active_change(bond, new_active);
1174                 if (old_active)
1175                         bond_set_slave_inactive_flags(old_active,
1176                                                       BOND_SLAVE_NOTIFY_NOW);
1177                 if (new_active)
1178                         bond_set_slave_active_flags(new_active,
1179                                                     BOND_SLAVE_NOTIFY_NOW);
1180         } else {
1181                 rcu_assign_pointer(bond->curr_active_slave, new_active);
1182         }
1183
1184         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
1185                 if (old_active)
1186                         bond_set_slave_inactive_flags(old_active,
1187                                                       BOND_SLAVE_NOTIFY_NOW);
1188
1189                 if (new_active) {
1190                         bool should_notify_peers = false;
1191
1192                         bond_set_slave_active_flags(new_active,
1193                                                     BOND_SLAVE_NOTIFY_NOW);
1194
1195                         if (bond->params.fail_over_mac)
1196                                 bond_do_fail_over_mac(bond, new_active,
1197                                                       old_active);
1198
1199                         if (netif_running(bond->dev)) {
1200                                 bond->send_peer_notif =
1201                                         bond->params.num_peer_notif *
1202                                         max(1, bond->params.peer_notif_delay);
1203                                 should_notify_peers =
1204                                         bond_should_notify_peers(bond);
1205                         }
1206
1207                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1208                         if (should_notify_peers) {
1209                                 bond->send_peer_notif--;
1210                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1211                                                          bond->dev);
1212                         }
1213                 }
1214         }
1215
1216 #ifdef CONFIG_XFRM_OFFLOAD
1217         bond_ipsec_add_sa_all(bond);
1218 #endif /* CONFIG_XFRM_OFFLOAD */
1219
1220         /* resend IGMP joins since active slave has changed or
1221          * all were sent on curr_active_slave.
1222          * resend only if bond is brought up with the affected
1223          * bonding modes and the retransmission is enabled
1224          */
1225         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1226             ((bond_uses_primary(bond) && new_active) ||
1227              BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
1228                 bond->igmp_retrans = bond->params.resend_igmp;
1229                 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
1230         }
1231 }
1232
1233 /**
1234  * bond_select_active_slave - select a new active slave, if needed
1235  * @bond: our bonding struct
1236  *
1237  * This functions should be called when one of the following occurs:
1238  * - The old curr_active_slave has been released or lost its link.
1239  * - The primary_slave has got its link back.
1240  * - A slave has got its link back and there's no old curr_active_slave.
1241  *
1242  * Caller must hold RTNL.
1243  */
1244 void bond_select_active_slave(struct bonding *bond)
1245 {
1246         struct slave *best_slave;
1247         int rv;
1248
1249         ASSERT_RTNL();
1250
1251         best_slave = bond_find_best_slave(bond);
1252         if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
1253                 bond_change_active_slave(bond, best_slave);
1254                 rv = bond_set_carrier(bond);
1255                 if (!rv)
1256                         return;
1257
1258                 if (netif_carrier_ok(bond->dev))
1259                         netdev_info(bond->dev, "active interface up!\n");
1260                 else
1261                         netdev_info(bond->dev, "now running without any active interface!\n");
1262         }
1263 }
1264
1265 #ifdef CONFIG_NET_POLL_CONTROLLER
1266 static inline int slave_enable_netpoll(struct slave *slave)
1267 {
1268         struct netpoll *np;
1269         int err = 0;
1270
1271         np = kzalloc(sizeof(*np), GFP_KERNEL);
1272         err = -ENOMEM;
1273         if (!np)
1274                 goto out;
1275
1276         err = __netpoll_setup(np, slave->dev);
1277         if (err) {
1278                 kfree(np);
1279                 goto out;
1280         }
1281         slave->np = np;
1282 out:
1283         return err;
1284 }
1285 static inline void slave_disable_netpoll(struct slave *slave)
1286 {
1287         struct netpoll *np = slave->np;
1288
1289         if (!np)
1290                 return;
1291
1292         slave->np = NULL;
1293
1294         __netpoll_free(np);
1295 }
1296
1297 static void bond_poll_controller(struct net_device *bond_dev)
1298 {
1299         struct bonding *bond = netdev_priv(bond_dev);
1300         struct slave *slave = NULL;
1301         struct list_head *iter;
1302         struct ad_info ad_info;
1303
1304         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1305                 if (bond_3ad_get_active_agg_info(bond, &ad_info))
1306                         return;
1307
1308         bond_for_each_slave_rcu(bond, slave, iter) {
1309                 if (!bond_slave_is_up(slave))
1310                         continue;
1311
1312                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1313                         struct aggregator *agg =
1314                             SLAVE_AD_INFO(slave)->port.aggregator;
1315
1316                         if (agg &&
1317                             agg->aggregator_identifier != ad_info.aggregator_id)
1318                                 continue;
1319                 }
1320
1321                 netpoll_poll_dev(slave->dev);
1322         }
1323 }
1324
1325 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1326 {
1327         struct bonding *bond = netdev_priv(bond_dev);
1328         struct list_head *iter;
1329         struct slave *slave;
1330
1331         bond_for_each_slave(bond, slave, iter)
1332                 if (bond_slave_is_up(slave))
1333                         slave_disable_netpoll(slave);
1334 }
1335
1336 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1337 {
1338         struct bonding *bond = netdev_priv(dev);
1339         struct list_head *iter;
1340         struct slave *slave;
1341         int err = 0;
1342
1343         bond_for_each_slave(bond, slave, iter) {
1344                 err = slave_enable_netpoll(slave);
1345                 if (err) {
1346                         bond_netpoll_cleanup(dev);
1347                         break;
1348                 }
1349         }
1350         return err;
1351 }
1352 #else
1353 static inline int slave_enable_netpoll(struct slave *slave)
1354 {
1355         return 0;
1356 }
1357 static inline void slave_disable_netpoll(struct slave *slave)
1358 {
1359 }
1360 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1361 {
1362 }
1363 #endif
1364
1365 /*---------------------------------- IOCTL ----------------------------------*/
1366
1367 static netdev_features_t bond_fix_features(struct net_device *dev,
1368                                            netdev_features_t features)
1369 {
1370         struct bonding *bond = netdev_priv(dev);
1371         struct list_head *iter;
1372         netdev_features_t mask;
1373         struct slave *slave;
1374
1375 #if IS_ENABLED(CONFIG_TLS_DEVICE)
1376         if (bond_sk_check(bond))
1377                 features |= BOND_TLS_FEATURES;
1378         else
1379                 features &= ~BOND_TLS_FEATURES;
1380 #endif
1381
1382         mask = features;
1383
1384         features &= ~NETIF_F_ONE_FOR_ALL;
1385         features |= NETIF_F_ALL_FOR_ALL;
1386
1387         bond_for_each_slave(bond, slave, iter) {
1388                 features = netdev_increment_features(features,
1389                                                      slave->dev->features,
1390                                                      mask);
1391         }
1392         features = netdev_add_tso_features(features, mask);
1393
1394         return features;
1395 }
1396
1397 #define BOND_VLAN_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1398                                  NETIF_F_FRAGLIST | NETIF_F_GSO_SOFTWARE | \
1399                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1400
1401 #define BOND_ENC_FEATURES       (NETIF_F_HW_CSUM | NETIF_F_SG | \
1402                                  NETIF_F_RXCSUM | NETIF_F_GSO_SOFTWARE)
1403
1404 #define BOND_MPLS_FEATURES      (NETIF_F_HW_CSUM | NETIF_F_SG | \
1405                                  NETIF_F_GSO_SOFTWARE)
1406
1407
1408 static void bond_compute_features(struct bonding *bond)
1409 {
1410         unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1411                                         IFF_XMIT_DST_RELEASE_PERM;
1412         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1413         netdev_features_t enc_features  = BOND_ENC_FEATURES;
1414 #ifdef CONFIG_XFRM_OFFLOAD
1415         netdev_features_t xfrm_features  = BOND_XFRM_FEATURES;
1416 #endif /* CONFIG_XFRM_OFFLOAD */
1417         netdev_features_t mpls_features  = BOND_MPLS_FEATURES;
1418         struct net_device *bond_dev = bond->dev;
1419         struct list_head *iter;
1420         struct slave *slave;
1421         unsigned short max_hard_header_len = ETH_HLEN;
1422         unsigned int tso_max_size = TSO_MAX_SIZE;
1423         u16 tso_max_segs = TSO_MAX_SEGS;
1424
1425         if (!bond_has_slaves(bond))
1426                 goto done;
1427         vlan_features &= NETIF_F_ALL_FOR_ALL;
1428         mpls_features &= NETIF_F_ALL_FOR_ALL;
1429
1430         bond_for_each_slave(bond, slave, iter) {
1431                 vlan_features = netdev_increment_features(vlan_features,
1432                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1433
1434                 enc_features = netdev_increment_features(enc_features,
1435                                                          slave->dev->hw_enc_features,
1436                                                          BOND_ENC_FEATURES);
1437
1438 #ifdef CONFIG_XFRM_OFFLOAD
1439                 xfrm_features = netdev_increment_features(xfrm_features,
1440                                                           slave->dev->hw_enc_features,
1441                                                           BOND_XFRM_FEATURES);
1442 #endif /* CONFIG_XFRM_OFFLOAD */
1443
1444                 mpls_features = netdev_increment_features(mpls_features,
1445                                                           slave->dev->mpls_features,
1446                                                           BOND_MPLS_FEATURES);
1447
1448                 dst_release_flag &= slave->dev->priv_flags;
1449                 if (slave->dev->hard_header_len > max_hard_header_len)
1450                         max_hard_header_len = slave->dev->hard_header_len;
1451
1452                 tso_max_size = min(tso_max_size, slave->dev->tso_max_size);
1453                 tso_max_segs = min(tso_max_segs, slave->dev->tso_max_segs);
1454         }
1455         bond_dev->hard_header_len = max_hard_header_len;
1456
1457 done:
1458         bond_dev->vlan_features = vlan_features;
1459         bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1460                                     NETIF_F_HW_VLAN_CTAG_TX |
1461                                     NETIF_F_HW_VLAN_STAG_TX;
1462 #ifdef CONFIG_XFRM_OFFLOAD
1463         bond_dev->hw_enc_features |= xfrm_features;
1464 #endif /* CONFIG_XFRM_OFFLOAD */
1465         bond_dev->mpls_features = mpls_features;
1466         netif_set_tso_max_segs(bond_dev, tso_max_segs);
1467         netif_set_tso_max_size(bond_dev, tso_max_size);
1468
1469         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1470         if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1471             dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1472                 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1473
1474         netdev_change_features(bond_dev);
1475 }
1476
1477 static void bond_setup_by_slave(struct net_device *bond_dev,
1478                                 struct net_device *slave_dev)
1479 {
1480         bond_dev->header_ops        = slave_dev->header_ops;
1481
1482         bond_dev->type              = slave_dev->type;
1483         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1484         bond_dev->needed_headroom   = slave_dev->needed_headroom;
1485         bond_dev->addr_len          = slave_dev->addr_len;
1486
1487         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1488                 slave_dev->addr_len);
1489 }
1490
1491 /* On bonding slaves other than the currently active slave, suppress
1492  * duplicates except for alb non-mcast/bcast.
1493  */
1494 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1495                                             struct slave *slave,
1496                                             struct bonding *bond)
1497 {
1498         if (bond_is_slave_inactive(slave)) {
1499                 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1500                     skb->pkt_type != PACKET_BROADCAST &&
1501                     skb->pkt_type != PACKET_MULTICAST)
1502                         return false;
1503                 return true;
1504         }
1505         return false;
1506 }
1507
1508 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1509 {
1510         struct sk_buff *skb = *pskb;
1511         struct slave *slave;
1512         struct bonding *bond;
1513         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1514                           struct slave *);
1515         int ret = RX_HANDLER_ANOTHER;
1516
1517         skb = skb_share_check(skb, GFP_ATOMIC);
1518         if (unlikely(!skb))
1519                 return RX_HANDLER_CONSUMED;
1520
1521         *pskb = skb;
1522
1523         slave = bond_slave_get_rcu(skb->dev);
1524         bond = slave->bond;
1525
1526         recv_probe = READ_ONCE(bond->recv_probe);
1527         if (recv_probe) {
1528                 ret = recv_probe(skb, bond, slave);
1529                 if (ret == RX_HANDLER_CONSUMED) {
1530                         consume_skb(skb);
1531                         return ret;
1532                 }
1533         }
1534
1535         /*
1536          * For packets determined by bond_should_deliver_exact_match() call to
1537          * be suppressed we want to make an exception for link-local packets.
1538          * This is necessary for e.g. LLDP daemons to be able to monitor
1539          * inactive slave links without being forced to bind to them
1540          * explicitly.
1541          *
1542          * At the same time, packets that are passed to the bonding master
1543          * (including link-local ones) can have their originating interface
1544          * determined via PACKET_ORIGDEV socket option.
1545          */
1546         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1547                 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1548                         return RX_HANDLER_PASS;
1549                 return RX_HANDLER_EXACT;
1550         }
1551
1552         skb->dev = bond->dev;
1553
1554         if (BOND_MODE(bond) == BOND_MODE_ALB &&
1555             netif_is_bridge_port(bond->dev) &&
1556             skb->pkt_type == PACKET_HOST) {
1557
1558                 if (unlikely(skb_cow_head(skb,
1559                                           skb->data - skb_mac_header(skb)))) {
1560                         kfree_skb(skb);
1561                         return RX_HANDLER_CONSUMED;
1562                 }
1563                 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1564                                   bond->dev->addr_len);
1565         }
1566
1567         return ret;
1568 }
1569
1570 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1571 {
1572         switch (BOND_MODE(bond)) {
1573         case BOND_MODE_ROUNDROBIN:
1574                 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1575         case BOND_MODE_ACTIVEBACKUP:
1576                 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1577         case BOND_MODE_BROADCAST:
1578                 return NETDEV_LAG_TX_TYPE_BROADCAST;
1579         case BOND_MODE_XOR:
1580         case BOND_MODE_8023AD:
1581                 return NETDEV_LAG_TX_TYPE_HASH;
1582         default:
1583                 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1584         }
1585 }
1586
1587 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1588                                                enum netdev_lag_tx_type type)
1589 {
1590         if (type != NETDEV_LAG_TX_TYPE_HASH)
1591                 return NETDEV_LAG_HASH_NONE;
1592
1593         switch (bond->params.xmit_policy) {
1594         case BOND_XMIT_POLICY_LAYER2:
1595                 return NETDEV_LAG_HASH_L2;
1596         case BOND_XMIT_POLICY_LAYER34:
1597                 return NETDEV_LAG_HASH_L34;
1598         case BOND_XMIT_POLICY_LAYER23:
1599                 return NETDEV_LAG_HASH_L23;
1600         case BOND_XMIT_POLICY_ENCAP23:
1601                 return NETDEV_LAG_HASH_E23;
1602         case BOND_XMIT_POLICY_ENCAP34:
1603                 return NETDEV_LAG_HASH_E34;
1604         case BOND_XMIT_POLICY_VLAN_SRCMAC:
1605                 return NETDEV_LAG_HASH_VLAN_SRCMAC;
1606         default:
1607                 return NETDEV_LAG_HASH_UNKNOWN;
1608         }
1609 }
1610
1611 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1612                                       struct netlink_ext_ack *extack)
1613 {
1614         struct netdev_lag_upper_info lag_upper_info;
1615         enum netdev_lag_tx_type type;
1616
1617         type = bond_lag_tx_type(bond);
1618         lag_upper_info.tx_type = type;
1619         lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1620
1621         return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1622                                             &lag_upper_info, extack);
1623 }
1624
1625 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1626 {
1627         netdev_upper_dev_unlink(slave->dev, bond->dev);
1628         slave->dev->flags &= ~IFF_SLAVE;
1629 }
1630
1631 static void slave_kobj_release(struct kobject *kobj)
1632 {
1633         struct slave *slave = to_slave(kobj);
1634         struct bonding *bond = bond_get_bond_by_slave(slave);
1635
1636         cancel_delayed_work_sync(&slave->notify_work);
1637         if (BOND_MODE(bond) == BOND_MODE_8023AD)
1638                 kfree(SLAVE_AD_INFO(slave));
1639
1640         kfree(slave);
1641 }
1642
1643 static struct kobj_type slave_ktype = {
1644         .release = slave_kobj_release,
1645 #ifdef CONFIG_SYSFS
1646         .sysfs_ops = &slave_sysfs_ops,
1647 #endif
1648 };
1649
1650 static int bond_kobj_init(struct slave *slave)
1651 {
1652         int err;
1653
1654         err = kobject_init_and_add(&slave->kobj, &slave_ktype,
1655                                    &(slave->dev->dev.kobj), "bonding_slave");
1656         if (err)
1657                 kobject_put(&slave->kobj);
1658
1659         return err;
1660 }
1661
1662 static struct slave *bond_alloc_slave(struct bonding *bond,
1663                                       struct net_device *slave_dev)
1664 {
1665         struct slave *slave = NULL;
1666
1667         slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1668         if (!slave)
1669                 return NULL;
1670
1671         slave->bond = bond;
1672         slave->dev = slave_dev;
1673         INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1674
1675         if (bond_kobj_init(slave))
1676                 return NULL;
1677
1678         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1679                 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1680                                                GFP_KERNEL);
1681                 if (!SLAVE_AD_INFO(slave)) {
1682                         kobject_put(&slave->kobj);
1683                         return NULL;
1684                 }
1685         }
1686
1687         return slave;
1688 }
1689
1690 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1691 {
1692         info->bond_mode = BOND_MODE(bond);
1693         info->miimon = bond->params.miimon;
1694         info->num_slaves = bond->slave_cnt;
1695 }
1696
1697 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1698 {
1699         strcpy(info->slave_name, slave->dev->name);
1700         info->link = slave->link;
1701         info->state = bond_slave_state(slave);
1702         info->link_failure_count = slave->link_failure_count;
1703 }
1704
1705 static void bond_netdev_notify_work(struct work_struct *_work)
1706 {
1707         struct slave *slave = container_of(_work, struct slave,
1708                                            notify_work.work);
1709
1710         if (rtnl_trylock()) {
1711                 struct netdev_bonding_info binfo;
1712
1713                 bond_fill_ifslave(slave, &binfo.slave);
1714                 bond_fill_ifbond(slave->bond, &binfo.master);
1715                 netdev_bonding_info_change(slave->dev, &binfo);
1716                 rtnl_unlock();
1717         } else {
1718                 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1719         }
1720 }
1721
1722 void bond_queue_slave_event(struct slave *slave)
1723 {
1724         queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1725 }
1726
1727 void bond_lower_state_changed(struct slave *slave)
1728 {
1729         struct netdev_lag_lower_state_info info;
1730
1731         info.link_up = slave->link == BOND_LINK_UP ||
1732                        slave->link == BOND_LINK_FAIL;
1733         info.tx_enabled = bond_is_active_slave(slave);
1734         netdev_lower_state_changed(slave->dev, &info);
1735 }
1736
1737 #define BOND_NL_ERR(bond_dev, extack, errmsg) do {              \
1738         if (extack)                                             \
1739                 NL_SET_ERR_MSG(extack, errmsg);                 \
1740         else                                                    \
1741                 netdev_err(bond_dev, "Error: %s\n", errmsg);    \
1742 } while (0)
1743
1744 #define SLAVE_NL_ERR(bond_dev, slave_dev, extack, errmsg) do {          \
1745         if (extack)                                                     \
1746                 NL_SET_ERR_MSG(extack, errmsg);                         \
1747         else                                                            \
1748                 slave_err(bond_dev, slave_dev, "Error: %s\n", errmsg);  \
1749 } while (0)
1750
1751 /* enslave device <slave> to bond device <master> */
1752 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1753                  struct netlink_ext_ack *extack)
1754 {
1755         struct bonding *bond = netdev_priv(bond_dev);
1756         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1757         struct slave *new_slave = NULL, *prev_slave;
1758         struct sockaddr_storage ss;
1759         int link_reporting;
1760         int res = 0, i;
1761
1762         if (slave_dev->flags & IFF_MASTER &&
1763             !netif_is_bond_master(slave_dev)) {
1764                 BOND_NL_ERR(bond_dev, extack,
1765                             "Device type (master device) cannot be enslaved");
1766                 return -EPERM;
1767         }
1768
1769         if (!bond->params.use_carrier &&
1770             slave_dev->ethtool_ops->get_link == NULL &&
1771             slave_ops->ndo_eth_ioctl == NULL) {
1772                 slave_warn(bond_dev, slave_dev, "no link monitoring support\n");
1773         }
1774
1775         /* already in-use? */
1776         if (netdev_is_rx_handler_busy(slave_dev)) {
1777                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1778                              "Device is in use and cannot be enslaved");
1779                 return -EBUSY;
1780         }
1781
1782         if (bond_dev == slave_dev) {
1783                 BOND_NL_ERR(bond_dev, extack, "Cannot enslave bond to itself.");
1784                 return -EPERM;
1785         }
1786
1787         /* vlan challenged mutual exclusion */
1788         /* no need to lock since we're protected by rtnl_lock */
1789         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1790                 slave_dbg(bond_dev, slave_dev, "is NETIF_F_VLAN_CHALLENGED\n");
1791                 if (vlan_uses_dev(bond_dev)) {
1792                         SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1793                                      "Can not enslave VLAN challenged device to VLAN enabled bond");
1794                         return -EPERM;
1795                 } else {
1796                         slave_warn(bond_dev, slave_dev, "enslaved VLAN challenged slave. Adding VLANs will be blocked as long as it is part of bond.\n");
1797                 }
1798         } else {
1799                 slave_dbg(bond_dev, slave_dev, "is !NETIF_F_VLAN_CHALLENGED\n");
1800         }
1801
1802         if (slave_dev->features & NETIF_F_HW_ESP)
1803                 slave_dbg(bond_dev, slave_dev, "is esp-hw-offload capable\n");
1804
1805         /* Old ifenslave binaries are no longer supported.  These can
1806          * be identified with moderate accuracy by the state of the slave:
1807          * the current ifenslave will set the interface down prior to
1808          * enslaving it; the old ifenslave will not.
1809          */
1810         if (slave_dev->flags & IFF_UP) {
1811                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1812                              "Device can not be enslaved while up");
1813                 return -EPERM;
1814         }
1815
1816         /* set bonding device ether type by slave - bonding netdevices are
1817          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1818          * there is a need to override some of the type dependent attribs/funcs.
1819          *
1820          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1821          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1822          */
1823         if (!bond_has_slaves(bond)) {
1824                 if (bond_dev->type != slave_dev->type) {
1825                         slave_dbg(bond_dev, slave_dev, "change device type from %d to %d\n",
1826                                   bond_dev->type, slave_dev->type);
1827
1828                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1829                                                        bond_dev);
1830                         res = notifier_to_errno(res);
1831                         if (res) {
1832                                 slave_err(bond_dev, slave_dev, "refused to change device type\n");
1833                                 return -EBUSY;
1834                         }
1835
1836                         /* Flush unicast and multicast addresses */
1837                         dev_uc_flush(bond_dev);
1838                         dev_mc_flush(bond_dev);
1839
1840                         if (slave_dev->type != ARPHRD_ETHER)
1841                                 bond_setup_by_slave(bond_dev, slave_dev);
1842                         else {
1843                                 ether_setup(bond_dev);
1844                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1845                         }
1846
1847                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1848                                                  bond_dev);
1849                 }
1850         } else if (bond_dev->type != slave_dev->type) {
1851                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1852                              "Device type is different from other slaves");
1853                 return -EINVAL;
1854         }
1855
1856         if (slave_dev->type == ARPHRD_INFINIBAND &&
1857             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1858                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1859                              "Only active-backup mode is supported for infiniband slaves");
1860                 res = -EOPNOTSUPP;
1861                 goto err_undo_flags;
1862         }
1863
1864         if (!slave_ops->ndo_set_mac_address ||
1865             slave_dev->type == ARPHRD_INFINIBAND) {
1866                 slave_warn(bond_dev, slave_dev, "The slave device specified does not support setting the MAC address\n");
1867                 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1868                     bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1869                         if (!bond_has_slaves(bond)) {
1870                                 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1871                                 slave_warn(bond_dev, slave_dev, "Setting fail_over_mac to active for active-backup mode\n");
1872                         } else {
1873                                 SLAVE_NL_ERR(bond_dev, slave_dev, extack,
1874                                              "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1875                                 res = -EOPNOTSUPP;
1876                                 goto err_undo_flags;
1877                         }
1878                 }
1879         }
1880
1881         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1882
1883         /* If this is the first slave, then we need to set the master's hardware
1884          * address to be the same as the slave's.
1885          */
1886         if (!bond_has_slaves(bond) &&
1887             bond->dev->addr_assign_type == NET_ADDR_RANDOM) {
1888                 res = bond_set_dev_addr(bond->dev, slave_dev);
1889                 if (res)
1890                         goto err_undo_flags;
1891         }
1892
1893         new_slave = bond_alloc_slave(bond, slave_dev);
1894         if (!new_slave) {
1895                 res = -ENOMEM;
1896                 goto err_undo_flags;
1897         }
1898
1899         /* Set the new_slave's queue_id to be zero.  Queue ID mapping
1900          * is set via sysfs or module option if desired.
1901          */
1902         new_slave->queue_id = 0;
1903
1904         /* Save slave's original mtu and then set it to match the bond */
1905         new_slave->original_mtu = slave_dev->mtu;
1906         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1907         if (res) {
1908                 slave_err(bond_dev, slave_dev, "Error %d calling dev_set_mtu\n", res);
1909                 goto err_free;
1910         }
1911
1912         /* Save slave's original ("permanent") mac address for modes
1913          * that need it, and for restoring it upon release, and then
1914          * set it to the master's address
1915          */
1916         bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1917                           slave_dev->addr_len);
1918
1919         if (!bond->params.fail_over_mac ||
1920             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1921                 /* Set slave to master's mac address.  The application already
1922                  * set the master's mac address to that of the first slave
1923                  */
1924                 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1925                 ss.ss_family = slave_dev->type;
1926                 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss,
1927                                           extack);
1928                 if (res) {
1929                         slave_err(bond_dev, slave_dev, "Error %d calling set_mac_address\n", res);
1930                         goto err_restore_mtu;
1931                 }
1932         }
1933
1934         /* set slave flag before open to prevent IPv6 addrconf */
1935         slave_dev->flags |= IFF_SLAVE;
1936
1937         /* open the slave since the application closed it */
1938         res = dev_open(slave_dev, extack);
1939         if (res) {
1940                 slave_err(bond_dev, slave_dev, "Opening slave failed\n");
1941                 goto err_restore_mac;
1942         }
1943
1944         slave_dev->priv_flags |= IFF_BONDING;
1945         /* initialize slave stats */
1946         dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1947
1948         if (bond_is_lb(bond)) {
1949                 /* bond_alb_init_slave() must be called before all other stages since
1950                  * it might fail and we do not want to have to undo everything
1951                  */
1952                 res = bond_alb_init_slave(bond, new_slave);
1953                 if (res)
1954                         goto err_close;
1955         }
1956
1957         res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1958         if (res) {
1959                 slave_err(bond_dev, slave_dev, "Couldn't add bond vlan ids\n");
1960                 goto err_close;
1961         }
1962
1963         prev_slave = bond_last_slave(bond);
1964
1965         new_slave->delay = 0;
1966         new_slave->link_failure_count = 0;
1967
1968         if (bond_update_speed_duplex(new_slave) &&
1969             bond_needs_speed_duplex(bond))
1970                 new_slave->link = BOND_LINK_DOWN;
1971
1972         new_slave->last_rx = jiffies -
1973                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1974         for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1975                 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1976
1977         if (bond->params.miimon && !bond->params.use_carrier) {
1978                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1979
1980                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1981                         /* miimon is set but a bonded network driver
1982                          * does not support ETHTOOL/MII and
1983                          * arp_interval is not set.  Note: if
1984                          * use_carrier is enabled, we will never go
1985                          * here (because netif_carrier is always
1986                          * supported); thus, we don't need to change
1987                          * the messages for netif_carrier.
1988                          */
1989                         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");
1990                 } else if (link_reporting == -1) {
1991                         /* unable get link status using mii/ethtool */
1992                         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");
1993                 }
1994         }
1995
1996         /* check for initial state */
1997         new_slave->link = BOND_LINK_NOCHANGE;
1998         if (bond->params.miimon) {
1999                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
2000                         if (bond->params.updelay) {
2001                                 bond_set_slave_link_state(new_slave,
2002                                                           BOND_LINK_BACK,
2003                                                           BOND_SLAVE_NOTIFY_NOW);
2004                                 new_slave->delay = bond->params.updelay;
2005                         } else {
2006                                 bond_set_slave_link_state(new_slave,
2007                                                           BOND_LINK_UP,
2008                                                           BOND_SLAVE_NOTIFY_NOW);
2009                         }
2010                 } else {
2011                         bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
2012                                                   BOND_SLAVE_NOTIFY_NOW);
2013                 }
2014         } else if (bond->params.arp_interval) {
2015                 bond_set_slave_link_state(new_slave,
2016                                           (netif_carrier_ok(slave_dev) ?
2017                                           BOND_LINK_UP : BOND_LINK_DOWN),
2018                                           BOND_SLAVE_NOTIFY_NOW);
2019         } else {
2020                 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
2021                                           BOND_SLAVE_NOTIFY_NOW);
2022         }
2023
2024         if (new_slave->link != BOND_LINK_DOWN)
2025                 new_slave->last_link_up = jiffies;
2026         slave_dbg(bond_dev, slave_dev, "Initial state of slave is BOND_LINK_%s\n",
2027                   new_slave->link == BOND_LINK_DOWN ? "DOWN" :
2028                   (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
2029
2030         if (bond_uses_primary(bond) && bond->params.primary[0]) {
2031                 /* if there is a primary slave, remember it */
2032                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
2033                         rcu_assign_pointer(bond->primary_slave, new_slave);
2034                         bond->force_primary = true;
2035                 }
2036         }
2037
2038         switch (BOND_MODE(bond)) {
2039         case BOND_MODE_ACTIVEBACKUP:
2040                 bond_set_slave_inactive_flags(new_slave,
2041                                               BOND_SLAVE_NOTIFY_NOW);
2042                 break;
2043         case BOND_MODE_8023AD:
2044                 /* in 802.3ad mode, the internal mechanism
2045                  * will activate the slaves in the selected
2046                  * aggregator
2047                  */
2048                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2049                 /* if this is the first slave */
2050                 if (!prev_slave) {
2051                         SLAVE_AD_INFO(new_slave)->id = 1;
2052                         /* Initialize AD with the number of times that the AD timer is called in 1 second
2053                          * can be called only after the mac address of the bond is set
2054                          */
2055                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
2056                 } else {
2057                         SLAVE_AD_INFO(new_slave)->id =
2058                                 SLAVE_AD_INFO(prev_slave)->id + 1;
2059                 }
2060
2061                 bond_3ad_bind_slave(new_slave);
2062                 break;
2063         case BOND_MODE_TLB:
2064         case BOND_MODE_ALB:
2065                 bond_set_active_slave(new_slave);
2066                 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
2067                 break;
2068         default:
2069                 slave_dbg(bond_dev, slave_dev, "This slave is always active in trunk mode\n");
2070
2071                 /* always active in trunk mode */
2072                 bond_set_active_slave(new_slave);
2073
2074                 /* In trunking mode there is little meaning to curr_active_slave
2075                  * anyway (it holds no special properties of the bond device),
2076                  * so we can change it without calling change_active_interface()
2077                  */
2078                 if (!rcu_access_pointer(bond->curr_active_slave) &&
2079                     new_slave->link == BOND_LINK_UP)
2080                         rcu_assign_pointer(bond->curr_active_slave, new_slave);
2081
2082                 break;
2083         } /* switch(bond_mode) */
2084
2085 #ifdef CONFIG_NET_POLL_CONTROLLER
2086         if (bond->dev->npinfo) {
2087                 if (slave_enable_netpoll(new_slave)) {
2088                         slave_info(bond_dev, slave_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
2089                         res = -EBUSY;
2090                         goto err_detach;
2091                 }
2092         }
2093 #endif
2094
2095         if (!(bond_dev->features & NETIF_F_LRO))
2096                 dev_disable_lro(slave_dev);
2097
2098         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
2099                                          new_slave);
2100         if (res) {
2101                 slave_dbg(bond_dev, slave_dev, "Error %d calling netdev_rx_handler_register\n", res);
2102                 goto err_detach;
2103         }
2104
2105         res = bond_master_upper_dev_link(bond, new_slave, extack);
2106         if (res) {
2107                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_master_upper_dev_link\n", res);
2108                 goto err_unregister;
2109         }
2110
2111         bond_lower_state_changed(new_slave);
2112
2113         res = bond_sysfs_slave_add(new_slave);
2114         if (res) {
2115                 slave_dbg(bond_dev, slave_dev, "Error %d calling bond_sysfs_slave_add\n", res);
2116                 goto err_upper_unlink;
2117         }
2118
2119         /* If the mode uses primary, then the following is handled by
2120          * bond_change_active_slave().
2121          */
2122         if (!bond_uses_primary(bond)) {
2123                 /* set promiscuity level to new slave */
2124                 if (bond_dev->flags & IFF_PROMISC) {
2125                         res = dev_set_promiscuity(slave_dev, 1);
2126                         if (res)
2127                                 goto err_sysfs_del;
2128                 }
2129
2130                 /* set allmulti level to new slave */
2131                 if (bond_dev->flags & IFF_ALLMULTI) {
2132                         res = dev_set_allmulti(slave_dev, 1);
2133                         if (res) {
2134                                 if (bond_dev->flags & IFF_PROMISC)
2135                                         dev_set_promiscuity(slave_dev, -1);
2136                                 goto err_sysfs_del;
2137                         }
2138                 }
2139
2140                 netif_addr_lock_bh(bond_dev);
2141                 dev_mc_sync_multiple(slave_dev, bond_dev);
2142                 dev_uc_sync_multiple(slave_dev, bond_dev);
2143                 netif_addr_unlock_bh(bond_dev);
2144
2145                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2146                         /* add lacpdu mc addr to mc list */
2147                         u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
2148
2149                         dev_mc_add(slave_dev, lacpdu_multicast);
2150                 }
2151         }
2152
2153         bond->slave_cnt++;
2154         bond_compute_features(bond);
2155         bond_set_carrier(bond);
2156
2157         if (bond_uses_primary(bond)) {
2158                 block_netpoll_tx();
2159                 bond_select_active_slave(bond);
2160                 unblock_netpoll_tx();
2161         }
2162
2163         if (bond_mode_can_use_xmit_hash(bond))
2164                 bond_update_slave_arr(bond, NULL);
2165
2166
2167         if (!slave_dev->netdev_ops->ndo_bpf ||
2168             !slave_dev->netdev_ops->ndo_xdp_xmit) {
2169                 if (bond->xdp_prog) {
2170                         SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2171                                      "Slave does not support XDP");
2172                         res = -EOPNOTSUPP;
2173                         goto err_sysfs_del;
2174                 }
2175         } else if (bond->xdp_prog) {
2176                 struct netdev_bpf xdp = {
2177                         .command = XDP_SETUP_PROG,
2178                         .flags   = 0,
2179                         .prog    = bond->xdp_prog,
2180                         .extack  = extack,
2181                 };
2182
2183                 if (dev_xdp_prog_count(slave_dev) > 0) {
2184                         SLAVE_NL_ERR(bond_dev, slave_dev, extack,
2185                                      "Slave has XDP program loaded, please unload before enslaving");
2186                         res = -EOPNOTSUPP;
2187                         goto err_sysfs_del;
2188                 }
2189
2190                 res = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
2191                 if (res < 0) {
2192                         /* ndo_bpf() sets extack error message */
2193                         slave_dbg(bond_dev, slave_dev, "Error %d calling ndo_bpf\n", res);
2194                         goto err_sysfs_del;
2195                 }
2196                 if (bond->xdp_prog)
2197                         bpf_prog_inc(bond->xdp_prog);
2198         }
2199
2200         slave_info(bond_dev, slave_dev, "Enslaving as %s interface with %s link\n",
2201                    bond_is_active_slave(new_slave) ? "an active" : "a backup",
2202                    new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
2203
2204         /* enslave is successful */
2205         bond_queue_slave_event(new_slave);
2206         return 0;
2207
2208 /* Undo stages on error */
2209 err_sysfs_del:
2210         bond_sysfs_slave_del(new_slave);
2211
2212 err_upper_unlink:
2213         bond_upper_dev_unlink(bond, new_slave);
2214
2215 err_unregister:
2216         netdev_rx_handler_unregister(slave_dev);
2217
2218 err_detach:
2219         vlan_vids_del_by_dev(slave_dev, bond_dev);
2220         if (rcu_access_pointer(bond->primary_slave) == new_slave)
2221                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2222         if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
2223                 block_netpoll_tx();
2224                 bond_change_active_slave(bond, NULL);
2225                 bond_select_active_slave(bond);
2226                 unblock_netpoll_tx();
2227         }
2228         /* either primary_slave or curr_active_slave might've changed */
2229         synchronize_rcu();
2230         slave_disable_netpoll(new_slave);
2231
2232 err_close:
2233         if (!netif_is_bond_master(slave_dev))
2234                 slave_dev->priv_flags &= ~IFF_BONDING;
2235         dev_close(slave_dev);
2236
2237 err_restore_mac:
2238         slave_dev->flags &= ~IFF_SLAVE;
2239         if (!bond->params.fail_over_mac ||
2240             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2241                 /* XXX TODO - fom follow mode needs to change master's
2242                  * MAC if this slave's MAC is in use by the bond, or at
2243                  * least print a warning.
2244                  */
2245                 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
2246                                   new_slave->dev->addr_len);
2247                 ss.ss_family = slave_dev->type;
2248                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2249         }
2250
2251 err_restore_mtu:
2252         dev_set_mtu(slave_dev, new_slave->original_mtu);
2253
2254 err_free:
2255         kobject_put(&new_slave->kobj);
2256
2257 err_undo_flags:
2258         /* Enslave of first slave has failed and we need to fix master's mac */
2259         if (!bond_has_slaves(bond)) {
2260                 if (ether_addr_equal_64bits(bond_dev->dev_addr,
2261                                             slave_dev->dev_addr))
2262                         eth_hw_addr_random(bond_dev);
2263                 if (bond_dev->type != ARPHRD_ETHER) {
2264                         dev_close(bond_dev);
2265                         ether_setup(bond_dev);
2266                         bond_dev->flags |= IFF_MASTER;
2267                         bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
2268                 }
2269         }
2270
2271         return res;
2272 }
2273
2274 /* Try to release the slave device <slave> from the bond device <master>
2275  * It is legal to access curr_active_slave without a lock because all the function
2276  * is RTNL-locked. If "all" is true it means that the function is being called
2277  * while destroying a bond interface and all slaves are being released.
2278  *
2279  * The rules for slave state should be:
2280  *   for Active/Backup:
2281  *     Active stays on all backups go down
2282  *   for Bonded connections:
2283  *     The first up interface should be left on and all others downed.
2284  */
2285 static int __bond_release_one(struct net_device *bond_dev,
2286                               struct net_device *slave_dev,
2287                               bool all, bool unregister)
2288 {
2289         struct bonding *bond = netdev_priv(bond_dev);
2290         struct slave *slave, *oldcurrent;
2291         struct sockaddr_storage ss;
2292         int old_flags = bond_dev->flags;
2293         netdev_features_t old_features = bond_dev->features;
2294
2295         /* slave is not a slave or master is not master of this slave */
2296         if (!(slave_dev->flags & IFF_SLAVE) ||
2297             !netdev_has_upper_dev(slave_dev, bond_dev)) {
2298                 slave_dbg(bond_dev, slave_dev, "cannot release slave\n");
2299                 return -EINVAL;
2300         }
2301
2302         block_netpoll_tx();
2303
2304         slave = bond_get_slave_by_dev(bond, slave_dev);
2305         if (!slave) {
2306                 /* not a slave of this bond */
2307                 slave_info(bond_dev, slave_dev, "interface not enslaved\n");
2308                 unblock_netpoll_tx();
2309                 return -EINVAL;
2310         }
2311
2312         bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
2313
2314         bond_sysfs_slave_del(slave);
2315
2316         /* recompute stats just before removing the slave */
2317         bond_get_stats(bond->dev, &bond->bond_stats);
2318
2319         if (bond->xdp_prog) {
2320                 struct netdev_bpf xdp = {
2321                         .command = XDP_SETUP_PROG,
2322                         .flags   = 0,
2323                         .prog    = NULL,
2324                         .extack  = NULL,
2325                 };
2326                 if (slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp))
2327                         slave_warn(bond_dev, slave_dev, "failed to unload XDP program\n");
2328         }
2329
2330         /* unregister rx_handler early so bond_handle_frame wouldn't be called
2331          * for this slave anymore.
2332          */
2333         netdev_rx_handler_unregister(slave_dev);
2334
2335         if (BOND_MODE(bond) == BOND_MODE_8023AD)
2336                 bond_3ad_unbind_slave(slave);
2337
2338         bond_upper_dev_unlink(bond, slave);
2339
2340         if (bond_mode_can_use_xmit_hash(bond))
2341                 bond_update_slave_arr(bond, slave);
2342
2343         slave_info(bond_dev, slave_dev, "Releasing %s interface\n",
2344                     bond_is_active_slave(slave) ? "active" : "backup");
2345
2346         oldcurrent = rcu_access_pointer(bond->curr_active_slave);
2347
2348         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2349
2350         if (!all && (!bond->params.fail_over_mac ||
2351                      BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
2352                 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
2353                     bond_has_slaves(bond))
2354                         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",
2355                                    slave->perm_hwaddr);
2356         }
2357
2358         if (rtnl_dereference(bond->primary_slave) == slave)
2359                 RCU_INIT_POINTER(bond->primary_slave, NULL);
2360
2361         if (oldcurrent == slave)
2362                 bond_change_active_slave(bond, NULL);
2363
2364         if (bond_is_lb(bond)) {
2365                 /* Must be called only after the slave has been
2366                  * detached from the list and the curr_active_slave
2367                  * has been cleared (if our_slave == old_current),
2368                  * but before a new active slave is selected.
2369                  */
2370                 bond_alb_deinit_slave(bond, slave);
2371         }
2372
2373         if (all) {
2374                 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
2375         } else if (oldcurrent == slave) {
2376                 /* Note that we hold RTNL over this sequence, so there
2377                  * is no concern that another slave add/remove event
2378                  * will interfere.
2379                  */
2380                 bond_select_active_slave(bond);
2381         }
2382
2383         bond_set_carrier(bond);
2384         if (!bond_has_slaves(bond))
2385                 eth_hw_addr_random(bond_dev);
2386
2387         unblock_netpoll_tx();
2388         synchronize_rcu();
2389         bond->slave_cnt--;
2390
2391         if (!bond_has_slaves(bond)) {
2392                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2393                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2394         }
2395
2396         bond_compute_features(bond);
2397         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2398             (old_features & NETIF_F_VLAN_CHALLENGED))
2399                 slave_info(bond_dev, slave_dev, "last VLAN challenged slave left bond - VLAN blocking is removed\n");
2400
2401         vlan_vids_del_by_dev(slave_dev, bond_dev);
2402
2403         /* If the mode uses primary, then this case was handled above by
2404          * bond_change_active_slave(..., NULL)
2405          */
2406         if (!bond_uses_primary(bond)) {
2407                 /* unset promiscuity level from slave
2408                  * NOTE: The NETDEV_CHANGEADDR call above may change the value
2409                  * of the IFF_PROMISC flag in the bond_dev, but we need the
2410                  * value of that flag before that change, as that was the value
2411                  * when this slave was attached, so we cache at the start of the
2412                  * function and use it here. Same goes for ALLMULTI below
2413                  */
2414                 if (old_flags & IFF_PROMISC)
2415                         dev_set_promiscuity(slave_dev, -1);
2416
2417                 /* unset allmulti level from slave */
2418                 if (old_flags & IFF_ALLMULTI)
2419                         dev_set_allmulti(slave_dev, -1);
2420
2421                 bond_hw_addr_flush(bond_dev, slave_dev);
2422         }
2423
2424         slave_disable_netpoll(slave);
2425
2426         /* close slave before restoring its mac address */
2427         dev_close(slave_dev);
2428
2429         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
2430             BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2431                 /* restore original ("permanent") mac address */
2432                 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
2433                                   slave->dev->addr_len);
2434                 ss.ss_family = slave_dev->type;
2435                 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss, NULL);
2436         }
2437
2438         if (unregister)
2439                 __dev_set_mtu(slave_dev, slave->original_mtu);
2440         else
2441                 dev_set_mtu(slave_dev, slave->original_mtu);
2442
2443         if (!netif_is_bond_master(slave_dev))
2444                 slave_dev->priv_flags &= ~IFF_BONDING;
2445
2446         kobject_put(&slave->kobj);
2447
2448         return 0;
2449 }
2450
2451 /* A wrapper used because of ndo_del_link */
2452 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2453 {
2454         return __bond_release_one(bond_dev, slave_dev, false, false);
2455 }
2456
2457 /* First release a slave and then destroy the bond if no more slaves are left.
2458  * Must be under rtnl_lock when this function is called.
2459  */
2460 static int bond_release_and_destroy(struct net_device *bond_dev,
2461                                     struct net_device *slave_dev)
2462 {
2463         struct bonding *bond = netdev_priv(bond_dev);
2464         int ret;
2465
2466         ret = __bond_release_one(bond_dev, slave_dev, false, true);
2467         if (ret == 0 && !bond_has_slaves(bond) &&
2468             bond_dev->reg_state != NETREG_UNREGISTERING) {
2469                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2470                 netdev_info(bond_dev, "Destroying bond\n");
2471                 bond_remove_proc_entry(bond);
2472                 unregister_netdevice(bond_dev);
2473         }
2474         return ret;
2475 }
2476
2477 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2478 {
2479         struct bonding *bond = netdev_priv(bond_dev);
2480
2481         bond_fill_ifbond(bond, info);
2482 }
2483
2484 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2485 {
2486         struct bonding *bond = netdev_priv(bond_dev);
2487         struct list_head *iter;
2488         int i = 0, res = -ENODEV;
2489         struct slave *slave;
2490
2491         bond_for_each_slave(bond, slave, iter) {
2492                 if (i++ == (int)info->slave_id) {
2493                         res = 0;
2494                         bond_fill_ifslave(slave, info);
2495                         break;
2496                 }
2497         }
2498
2499         return res;
2500 }
2501
2502 /*-------------------------------- Monitoring -------------------------------*/
2503
2504 /* called with rcu_read_lock() */
2505 static int bond_miimon_inspect(struct bonding *bond)
2506 {
2507         int link_state, commit = 0;
2508         struct list_head *iter;
2509         struct slave *slave;
2510         bool ignore_updelay;
2511
2512         ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2513
2514         bond_for_each_slave_rcu(bond, slave, iter) {
2515                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2516
2517                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2518
2519                 switch (slave->link) {
2520                 case BOND_LINK_UP:
2521                         if (link_state)
2522                                 continue;
2523
2524                         bond_propose_link_state(slave, BOND_LINK_FAIL);
2525                         commit++;
2526                         slave->delay = bond->params.downdelay;
2527                         if (slave->delay) {
2528                                 slave_info(bond->dev, slave->dev, "link status down for %sinterface, disabling it in %d ms\n",
2529                                            (BOND_MODE(bond) ==
2530                                             BOND_MODE_ACTIVEBACKUP) ?
2531                                             (bond_is_active_slave(slave) ?
2532                                              "active " : "backup ") : "",
2533                                            bond->params.downdelay * bond->params.miimon);
2534                         }
2535                         fallthrough;
2536                 case BOND_LINK_FAIL:
2537                         if (link_state) {
2538                                 /* recovered before downdelay expired */
2539                                 bond_propose_link_state(slave, BOND_LINK_UP);
2540                                 slave->last_link_up = jiffies;
2541                                 slave_info(bond->dev, slave->dev, "link status up again after %d ms\n",
2542                                            (bond->params.downdelay - slave->delay) *
2543                                            bond->params.miimon);
2544                                 commit++;
2545                                 continue;
2546                         }
2547
2548                         if (slave->delay <= 0) {
2549                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2550                                 commit++;
2551                                 continue;
2552                         }
2553
2554                         slave->delay--;
2555                         break;
2556
2557                 case BOND_LINK_DOWN:
2558                         if (!link_state)
2559                                 continue;
2560
2561                         bond_propose_link_state(slave, BOND_LINK_BACK);
2562                         commit++;
2563                         slave->delay = bond->params.updelay;
2564
2565                         if (slave->delay) {
2566                                 slave_info(bond->dev, slave->dev, "link status up, enabling it in %d ms\n",
2567                                            ignore_updelay ? 0 :
2568                                            bond->params.updelay *
2569                                            bond->params.miimon);
2570                         }
2571                         fallthrough;
2572                 case BOND_LINK_BACK:
2573                         if (!link_state) {
2574                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
2575                                 slave_info(bond->dev, slave->dev, "link status down again after %d ms\n",
2576                                            (bond->params.updelay - slave->delay) *
2577                                            bond->params.miimon);
2578                                 commit++;
2579                                 continue;
2580                         }
2581
2582                         if (ignore_updelay)
2583                                 slave->delay = 0;
2584
2585                         if (slave->delay <= 0) {
2586                                 bond_propose_link_state(slave, BOND_LINK_UP);
2587                                 commit++;
2588                                 ignore_updelay = false;
2589                                 continue;
2590                         }
2591
2592                         slave->delay--;
2593                         break;
2594                 }
2595         }
2596
2597         return commit;
2598 }
2599
2600 static void bond_miimon_link_change(struct bonding *bond,
2601                                     struct slave *slave,
2602                                     char link)
2603 {
2604         switch (BOND_MODE(bond)) {
2605         case BOND_MODE_8023AD:
2606                 bond_3ad_handle_link_change(slave, link);
2607                 break;
2608         case BOND_MODE_TLB:
2609         case BOND_MODE_ALB:
2610                 bond_alb_handle_link_change(bond, slave, link);
2611                 break;
2612         case BOND_MODE_XOR:
2613                 bond_update_slave_arr(bond, NULL);
2614                 break;
2615         }
2616 }
2617
2618 static void bond_miimon_commit(struct bonding *bond)
2619 {
2620         struct list_head *iter;
2621         struct slave *slave, *primary;
2622
2623         bond_for_each_slave(bond, slave, iter) {
2624                 switch (slave->link_new_state) {
2625                 case BOND_LINK_NOCHANGE:
2626                         /* For 802.3ad mode, check current slave speed and
2627                          * duplex again in case its port was disabled after
2628                          * invalid speed/duplex reporting but recovered before
2629                          * link monitoring could make a decision on the actual
2630                          * link status
2631                          */
2632                         if (BOND_MODE(bond) == BOND_MODE_8023AD &&
2633                             slave->link == BOND_LINK_UP)
2634                                 bond_3ad_adapter_speed_duplex_changed(slave);
2635                         continue;
2636
2637                 case BOND_LINK_UP:
2638                         if (bond_update_speed_duplex(slave) &&
2639                             bond_needs_speed_duplex(bond)) {
2640                                 slave->link = BOND_LINK_DOWN;
2641                                 if (net_ratelimit())
2642                                         slave_warn(bond->dev, slave->dev,
2643                                                    "failed to get link speed/duplex\n");
2644                                 continue;
2645                         }
2646                         bond_set_slave_link_state(slave, BOND_LINK_UP,
2647                                                   BOND_SLAVE_NOTIFY_NOW);
2648                         slave->last_link_up = jiffies;
2649
2650                         primary = rtnl_dereference(bond->primary_slave);
2651                         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2652                                 /* prevent it from being the active one */
2653                                 bond_set_backup_slave(slave);
2654                         } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2655                                 /* make it immediately active */
2656                                 bond_set_active_slave(slave);
2657                         }
2658
2659                         slave_info(bond->dev, slave->dev, "link status definitely up, %u Mbps %s duplex\n",
2660                                    slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2661                                    slave->duplex ? "full" : "half");
2662
2663                         bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2664
2665                         if (!bond->curr_active_slave || slave == primary)
2666                                 goto do_failover;
2667
2668                         continue;
2669
2670                 case BOND_LINK_DOWN:
2671                         if (slave->link_failure_count < UINT_MAX)
2672                                 slave->link_failure_count++;
2673
2674                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2675                                                   BOND_SLAVE_NOTIFY_NOW);
2676
2677                         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2678                             BOND_MODE(bond) == BOND_MODE_8023AD)
2679                                 bond_set_slave_inactive_flags(slave,
2680                                                               BOND_SLAVE_NOTIFY_NOW);
2681
2682                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
2683
2684                         bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2685
2686                         if (slave == rcu_access_pointer(bond->curr_active_slave))
2687                                 goto do_failover;
2688
2689                         continue;
2690
2691                 default:
2692                         slave_err(bond->dev, slave->dev, "invalid new link %d on slave\n",
2693                                   slave->link_new_state);
2694                         bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
2695
2696                         continue;
2697                 }
2698
2699 do_failover:
2700                 block_netpoll_tx();
2701                 bond_select_active_slave(bond);
2702                 unblock_netpoll_tx();
2703         }
2704
2705         bond_set_carrier(bond);
2706 }
2707
2708 /* bond_mii_monitor
2709  *
2710  * Really a wrapper that splits the mii monitor into two phases: an
2711  * inspection, then (if inspection indicates something needs to be done)
2712  * an acquisition of appropriate locks followed by a commit phase to
2713  * implement whatever link state changes are indicated.
2714  */
2715 static void bond_mii_monitor(struct work_struct *work)
2716 {
2717         struct bonding *bond = container_of(work, struct bonding,
2718                                             mii_work.work);
2719         bool should_notify_peers = false;
2720         bool commit;
2721         unsigned long delay;
2722         struct slave *slave;
2723         struct list_head *iter;
2724
2725         delay = msecs_to_jiffies(bond->params.miimon);
2726
2727         if (!bond_has_slaves(bond))
2728                 goto re_arm;
2729
2730         rcu_read_lock();
2731         should_notify_peers = bond_should_notify_peers(bond);
2732         commit = !!bond_miimon_inspect(bond);
2733         if (bond->send_peer_notif) {
2734                 rcu_read_unlock();
2735                 if (rtnl_trylock()) {
2736                         bond->send_peer_notif--;
2737                         rtnl_unlock();
2738                 }
2739         } else {
2740                 rcu_read_unlock();
2741         }
2742
2743         if (commit) {
2744                 /* Race avoidance with bond_close cancel of workqueue */
2745                 if (!rtnl_trylock()) {
2746                         delay = 1;
2747                         should_notify_peers = false;
2748                         goto re_arm;
2749                 }
2750
2751                 bond_for_each_slave(bond, slave, iter) {
2752                         bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2753                 }
2754                 bond_miimon_commit(bond);
2755
2756                 rtnl_unlock();  /* might sleep, hold no other locks */
2757         }
2758
2759 re_arm:
2760         if (bond->params.miimon)
2761                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2762
2763         if (should_notify_peers) {
2764                 if (!rtnl_trylock())
2765                         return;
2766                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2767                 rtnl_unlock();
2768         }
2769 }
2770
2771 static int bond_upper_dev_walk(struct net_device *upper,
2772                                struct netdev_nested_priv *priv)
2773 {
2774         __be32 ip = *(__be32 *)priv->data;
2775
2776         return ip == bond_confirm_addr(upper, 0, ip);
2777 }
2778
2779 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2780 {
2781         struct netdev_nested_priv priv = {
2782                 .data = (void *)&ip,
2783         };
2784         bool ret = false;
2785
2786         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2787                 return true;
2788
2789         rcu_read_lock();
2790         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &priv))
2791                 ret = true;
2792         rcu_read_unlock();
2793
2794         return ret;
2795 }
2796
2797 static bool bond_handle_vlan(struct slave *slave, struct bond_vlan_tag *tags,
2798                              struct sk_buff *skb)
2799 {
2800         struct net_device *bond_dev = slave->bond->dev;
2801         struct net_device *slave_dev = slave->dev;
2802         struct bond_vlan_tag *outer_tag = tags;
2803
2804         if (!tags || tags->vlan_proto == VLAN_N_VID)
2805                 return true;
2806
2807         tags++;
2808
2809         /* Go through all the tags backwards and add them to the packet */
2810         while (tags->vlan_proto != VLAN_N_VID) {
2811                 if (!tags->vlan_id) {
2812                         tags++;
2813                         continue;
2814                 }
2815
2816                 slave_dbg(bond_dev, slave_dev, "inner tag: proto %X vid %X\n",
2817                           ntohs(outer_tag->vlan_proto), tags->vlan_id);
2818                 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2819                                                 tags->vlan_id);
2820                 if (!skb) {
2821                         net_err_ratelimited("failed to insert inner VLAN tag\n");
2822                         return false;
2823                 }
2824
2825                 tags++;
2826         }
2827         /* Set the outer tag */
2828         if (outer_tag->vlan_id) {
2829                 slave_dbg(bond_dev, slave_dev, "outer tag: proto %X vid %X\n",
2830                           ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2831                 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2832                                        outer_tag->vlan_id);
2833         }
2834
2835         return true;
2836 }
2837
2838 /* We go to the (large) trouble of VLAN tagging ARP frames because
2839  * switches in VLAN mode (especially if ports are configured as
2840  * "native" to a VLAN) might not pass non-tagged frames.
2841  */
2842 static void bond_arp_send(struct slave *slave, int arp_op, __be32 dest_ip,
2843                           __be32 src_ip, struct bond_vlan_tag *tags)
2844 {
2845         struct net_device *bond_dev = slave->bond->dev;
2846         struct net_device *slave_dev = slave->dev;
2847         struct sk_buff *skb;
2848
2849         slave_dbg(bond_dev, slave_dev, "arp %d on slave: dst %pI4 src %pI4\n",
2850                   arp_op, &dest_ip, &src_ip);
2851
2852         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2853                          NULL, slave_dev->dev_addr, NULL);
2854
2855         if (!skb) {
2856                 net_err_ratelimited("ARP packet allocation failed\n");
2857                 return;
2858         }
2859
2860         if (bond_handle_vlan(slave, tags, skb))
2861                 arp_xmit(skb);
2862         return;
2863 }
2864
2865 /* Validate the device path between the @start_dev and the @end_dev.
2866  * The path is valid if the @end_dev is reachable through device
2867  * stacking.
2868  * When the path is validated, collect any vlan information in the
2869  * path.
2870  */
2871 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2872                                               struct net_device *end_dev,
2873                                               int level)
2874 {
2875         struct bond_vlan_tag *tags;
2876         struct net_device *upper;
2877         struct list_head  *iter;
2878
2879         if (start_dev == end_dev) {
2880                 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2881                 if (!tags)
2882                         return ERR_PTR(-ENOMEM);
2883                 tags[level].vlan_proto = VLAN_N_VID;
2884                 return tags;
2885         }
2886
2887         netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2888                 tags = bond_verify_device_path(upper, end_dev, level + 1);
2889                 if (IS_ERR_OR_NULL(tags)) {
2890                         if (IS_ERR(tags))
2891                                 return tags;
2892                         continue;
2893                 }
2894                 if (is_vlan_dev(upper)) {
2895                         tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2896                         tags[level].vlan_id = vlan_dev_vlan_id(upper);
2897                 }
2898
2899                 return tags;
2900         }
2901
2902         return NULL;
2903 }
2904
2905 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2906 {
2907         struct rtable *rt;
2908         struct bond_vlan_tag *tags;
2909         __be32 *targets = bond->params.arp_targets, addr;
2910         int i;
2911
2912         for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2913                 slave_dbg(bond->dev, slave->dev, "%s: target %pI4\n",
2914                           __func__, &targets[i]);
2915                 tags = NULL;
2916
2917                 /* Find out through which dev should the packet go */
2918                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2919                                      RTO_ONLINK, 0);
2920                 if (IS_ERR(rt)) {
2921                         /* there's no route to target - try to send arp
2922                          * probe to generate any traffic (arp_validate=0)
2923                          */
2924                         if (bond->params.arp_validate)
2925                                 pr_warn_once("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2926                                              bond->dev->name,
2927                                              &targets[i]);
2928                         bond_arp_send(slave, ARPOP_REQUEST, targets[i],
2929                                       0, tags);
2930                         continue;
2931                 }
2932
2933                 /* bond device itself */
2934                 if (rt->dst.dev == bond->dev)
2935                         goto found;
2936
2937                 rcu_read_lock();
2938                 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2939                 rcu_read_unlock();
2940
2941                 if (!IS_ERR_OR_NULL(tags))
2942                         goto found;
2943
2944                 /* Not our device - skip */
2945                 slave_dbg(bond->dev, slave->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2946                            &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2947
2948                 ip_rt_put(rt);
2949                 continue;
2950
2951 found:
2952                 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2953                 ip_rt_put(rt);
2954                 bond_arp_send(slave, ARPOP_REQUEST, targets[i], addr, tags);
2955                 kfree(tags);
2956         }
2957 }
2958
2959 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2960 {
2961         int i;
2962
2963         if (!sip || !bond_has_this_ip(bond, tip)) {
2964                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 tip %pI4 not found\n",
2965                            __func__, &sip, &tip);
2966                 return;
2967         }
2968
2969         i = bond_get_targets_ip(bond->params.arp_targets, sip);
2970         if (i == -1) {
2971                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI4 not found in targets\n",
2972                            __func__, &sip);
2973                 return;
2974         }
2975         slave->last_rx = jiffies;
2976         slave->target_last_arp_rx[i] = jiffies;
2977 }
2978
2979 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2980                         struct slave *slave)
2981 {
2982         struct arphdr *arp = (struct arphdr *)skb->data;
2983         struct slave *curr_active_slave, *curr_arp_slave;
2984         unsigned char *arp_ptr;
2985         __be32 sip, tip;
2986         unsigned int alen;
2987
2988         alen = arp_hdr_len(bond->dev);
2989
2990         if (alen > skb_headlen(skb)) {
2991                 arp = kmalloc(alen, GFP_ATOMIC);
2992                 if (!arp)
2993                         goto out_unlock;
2994                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2995                         goto out_unlock;
2996         }
2997
2998         if (arp->ar_hln != bond->dev->addr_len ||
2999             skb->pkt_type == PACKET_OTHERHOST ||
3000             skb->pkt_type == PACKET_LOOPBACK ||
3001             arp->ar_hrd != htons(ARPHRD_ETHER) ||
3002             arp->ar_pro != htons(ETH_P_IP) ||
3003             arp->ar_pln != 4)
3004                 goto out_unlock;
3005
3006         arp_ptr = (unsigned char *)(arp + 1);
3007         arp_ptr += bond->dev->addr_len;
3008         memcpy(&sip, arp_ptr, 4);
3009         arp_ptr += 4 + bond->dev->addr_len;
3010         memcpy(&tip, arp_ptr, 4);
3011
3012         slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
3013                   __func__, slave->dev->name, bond_slave_state(slave),
3014                   bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3015                   &sip, &tip);
3016
3017         curr_active_slave = rcu_dereference(bond->curr_active_slave);
3018         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3019
3020         /* We 'trust' the received ARP enough to validate it if:
3021          *
3022          * (a) the slave receiving the ARP is active (which includes the
3023          * current ARP slave, if any), or
3024          *
3025          * (b) the receiving slave isn't active, but there is a currently
3026          * active slave and it received valid arp reply(s) after it became
3027          * the currently active slave, or
3028          *
3029          * (c) there is an ARP slave that sent an ARP during the prior ARP
3030          * interval, and we receive an ARP reply on any slave.  We accept
3031          * these because switch FDB update delays may deliver the ARP
3032          * reply to a slave other than the sender of the ARP request.
3033          *
3034          * Note: for (b), backup slaves are receiving the broadcast ARP
3035          * request, not a reply.  This request passes from the sending
3036          * slave through the L2 switch(es) to the receiving slave.  Since
3037          * this is checking the request, sip/tip are swapped for
3038          * validation.
3039          *
3040          * This is done to avoid endless looping when we can't reach the
3041          * arp_ip_target and fool ourselves with our own arp requests.
3042          */
3043         if (bond_is_active_slave(slave))
3044                 bond_validate_arp(bond, slave, sip, tip);
3045         else if (curr_active_slave &&
3046                  time_after(slave_last_rx(bond, curr_active_slave),
3047                             curr_active_slave->last_link_up))
3048                 bond_validate_arp(bond, slave, tip, sip);
3049         else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
3050                  bond_time_in_interval(bond,
3051                                        dev_trans_start(curr_arp_slave->dev), 1))
3052                 bond_validate_arp(bond, slave, sip, tip);
3053
3054 out_unlock:
3055         if (arp != (struct arphdr *)skb->data)
3056                 kfree(arp);
3057         return RX_HANDLER_ANOTHER;
3058 }
3059
3060 #if IS_ENABLED(CONFIG_IPV6)
3061 static void bond_ns_send(struct slave *slave, const struct in6_addr *daddr,
3062                          const struct in6_addr *saddr, struct bond_vlan_tag *tags)
3063 {
3064         struct net_device *bond_dev = slave->bond->dev;
3065         struct net_device *slave_dev = slave->dev;
3066         struct in6_addr mcaddr;
3067         struct sk_buff *skb;
3068
3069         slave_dbg(bond_dev, slave_dev, "NS on slave: dst %pI6c src %pI6c\n",
3070                   daddr, saddr);
3071
3072         skb = ndisc_ns_create(slave_dev, daddr, saddr, 0);
3073         if (!skb) {
3074                 net_err_ratelimited("NS packet allocation failed\n");
3075                 return;
3076         }
3077
3078         addrconf_addr_solict_mult(daddr, &mcaddr);
3079         if (bond_handle_vlan(slave, tags, skb))
3080                 ndisc_send_skb(skb, &mcaddr, saddr);
3081 }
3082
3083 static void bond_ns_send_all(struct bonding *bond, struct slave *slave)
3084 {
3085         struct in6_addr *targets = bond->params.ns_targets;
3086         struct bond_vlan_tag *tags;
3087         struct dst_entry *dst;
3088         struct in6_addr saddr;
3089         struct flowi6 fl6;
3090         int i;
3091
3092         for (i = 0; i < BOND_MAX_NS_TARGETS && !ipv6_addr_any(&targets[i]); i++) {
3093                 slave_dbg(bond->dev, slave->dev, "%s: target %pI6c\n",
3094                           __func__, &targets[i]);
3095                 tags = NULL;
3096
3097                 /* Find out through which dev should the packet go */
3098                 memset(&fl6, 0, sizeof(struct flowi6));
3099                 fl6.daddr = targets[i];
3100                 fl6.flowi6_oif = bond->dev->ifindex;
3101
3102                 dst = ip6_route_output(dev_net(bond->dev), NULL, &fl6);
3103                 if (dst->error) {
3104                         dst_release(dst);
3105                         /* there's no route to target - try to send arp
3106                          * probe to generate any traffic (arp_validate=0)
3107                          */
3108                         if (bond->params.arp_validate)
3109                                 pr_warn_once("%s: no route to ns_ip6_target %pI6c and arp_validate is set\n",
3110                                              bond->dev->name,
3111                                              &targets[i]);
3112                         bond_ns_send(slave, &targets[i], &in6addr_any, tags);
3113                         continue;
3114                 }
3115
3116                 /* bond device itself */
3117                 if (dst->dev == bond->dev)
3118                         goto found;
3119
3120                 rcu_read_lock();
3121                 tags = bond_verify_device_path(bond->dev, dst->dev, 0);
3122                 rcu_read_unlock();
3123
3124                 if (!IS_ERR_OR_NULL(tags))
3125                         goto found;
3126
3127                 /* Not our device - skip */
3128                 slave_dbg(bond->dev, slave->dev, "no path to ns_ip6_target %pI6c via dst->dev %s\n",
3129                           &targets[i], dst->dev ? dst->dev->name : "NULL");
3130
3131                 dst_release(dst);
3132                 continue;
3133
3134 found:
3135                 if (!ipv6_dev_get_saddr(dev_net(dst->dev), dst->dev, &targets[i], 0, &saddr))
3136                         bond_ns_send(slave, &targets[i], &saddr, tags);
3137                 dst_release(dst);
3138                 kfree(tags);
3139         }
3140 }
3141
3142 static int bond_confirm_addr6(struct net_device *dev,
3143                               struct netdev_nested_priv *priv)
3144 {
3145         struct in6_addr *addr = (struct in6_addr *)priv->data;
3146
3147         return ipv6_chk_addr(dev_net(dev), addr, dev, 0);
3148 }
3149
3150 static bool bond_has_this_ip6(struct bonding *bond, struct in6_addr *addr)
3151 {
3152         struct netdev_nested_priv priv = {
3153                 .data = addr,
3154         };
3155         int ret = false;
3156
3157         if (bond_confirm_addr6(bond->dev, &priv))
3158                 return true;
3159
3160         rcu_read_lock();
3161         if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_confirm_addr6, &priv))
3162                 ret = true;
3163         rcu_read_unlock();
3164
3165         return ret;
3166 }
3167
3168 static void bond_validate_ns(struct bonding *bond, struct slave *slave,
3169                              struct in6_addr *saddr, struct in6_addr *daddr)
3170 {
3171         int i;
3172
3173         if (ipv6_addr_any(saddr) || !bond_has_this_ip6(bond, daddr)) {
3174                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c tip %pI6c not found\n",
3175                           __func__, saddr, daddr);
3176                 return;
3177         }
3178
3179         i = bond_get_targets_ip6(bond->params.ns_targets, saddr);
3180         if (i == -1) {
3181                 slave_dbg(bond->dev, slave->dev, "%s: sip %pI6c not found in targets\n",
3182                           __func__, saddr);
3183                 return;
3184         }
3185         slave->last_rx = jiffies;
3186         slave->target_last_arp_rx[i] = jiffies;
3187 }
3188
3189 static int bond_na_rcv(const struct sk_buff *skb, struct bonding *bond,
3190                        struct slave *slave)
3191 {
3192         struct slave *curr_active_slave, *curr_arp_slave;
3193         struct icmp6hdr *hdr = icmp6_hdr(skb);
3194         struct in6_addr *saddr, *daddr;
3195
3196         if (skb->pkt_type == PACKET_OTHERHOST ||
3197             skb->pkt_type == PACKET_LOOPBACK ||
3198             hdr->icmp6_type != NDISC_NEIGHBOUR_ADVERTISEMENT)
3199                 goto out;
3200
3201         saddr = &ipv6_hdr(skb)->saddr;
3202         daddr = &ipv6_hdr(skb)->daddr;
3203
3204         slave_dbg(bond->dev, slave->dev, "%s: %s/%d av %d sv %d sip %pI6c tip %pI6c\n",
3205                   __func__, slave->dev->name, bond_slave_state(slave),
3206                   bond->params.arp_validate, slave_do_arp_validate(bond, slave),
3207                   saddr, daddr);
3208
3209         curr_active_slave = rcu_dereference(bond->curr_active_slave);
3210         curr_arp_slave = rcu_dereference(bond->current_arp_slave);
3211
3212         /* We 'trust' the received ARP enough to validate it if:
3213          * see bond_arp_rcv().
3214          */
3215         if (bond_is_active_slave(slave))
3216                 bond_validate_ns(bond, slave, saddr, daddr);
3217         else if (curr_active_slave &&
3218                  time_after(slave_last_rx(bond, curr_active_slave),
3219                             curr_active_slave->last_link_up))
3220                 bond_validate_ns(bond, slave, saddr, daddr);
3221         else if (curr_arp_slave &&
3222                  bond_time_in_interval(bond,
3223                                        dev_trans_start(curr_arp_slave->dev), 1))
3224                 bond_validate_ns(bond, slave, saddr, daddr);
3225
3226 out:
3227         return RX_HANDLER_ANOTHER;
3228 }
3229 #endif
3230
3231 int bond_rcv_validate(const struct sk_buff *skb, struct bonding *bond,
3232                       struct slave *slave)
3233 {
3234 #if IS_ENABLED(CONFIG_IPV6)
3235         bool is_ipv6 = skb->protocol == __cpu_to_be16(ETH_P_IPV6);
3236 #endif
3237         bool is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
3238
3239         slave_dbg(bond->dev, slave->dev, "%s: skb->dev %s\n",
3240                   __func__, skb->dev->name);
3241
3242         /* Use arp validate logic for both ARP and NS */
3243         if (!slave_do_arp_validate(bond, slave)) {
3244                 if ((slave_do_arp_validate_only(bond) && is_arp) ||
3245 #if IS_ENABLED(CONFIG_IPV6)
3246                     (slave_do_arp_validate_only(bond) && is_ipv6) ||
3247 #endif
3248                     !slave_do_arp_validate_only(bond))
3249                         slave->last_rx = jiffies;
3250                 return RX_HANDLER_ANOTHER;
3251         } else if (is_arp) {
3252                 return bond_arp_rcv(skb, bond, slave);
3253 #if IS_ENABLED(CONFIG_IPV6)
3254         } else if (is_ipv6) {
3255                 return bond_na_rcv(skb, bond, slave);
3256 #endif
3257         } else {
3258                 return RX_HANDLER_ANOTHER;
3259         }
3260 }
3261
3262 static void bond_send_validate(struct bonding *bond, struct slave *slave)
3263 {
3264         bond_arp_send_all(bond, slave);
3265 #if IS_ENABLED(CONFIG_IPV6)
3266         bond_ns_send_all(bond, slave);
3267 #endif
3268 }
3269
3270 /* function to verify if we're in the arp_interval timeslice, returns true if
3271  * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
3272  * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
3273  */
3274 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
3275                                   int mod)
3276 {
3277         int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3278
3279         return time_in_range(jiffies,
3280                              last_act - delta_in_ticks,
3281                              last_act + mod * delta_in_ticks + delta_in_ticks/2);
3282 }
3283
3284 /* This function is called regularly to monitor each slave's link
3285  * ensuring that traffic is being sent and received when arp monitoring
3286  * is used in load-balancing mode. if the adapter has been dormant, then an
3287  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
3288  * arp monitoring in active backup mode.
3289  */
3290 static void bond_loadbalance_arp_mon(struct bonding *bond)
3291 {
3292         struct slave *slave, *oldcurrent;
3293         struct list_head *iter;
3294         int do_failover = 0, slave_state_changed = 0;
3295
3296         if (!bond_has_slaves(bond))
3297                 goto re_arm;
3298
3299         rcu_read_lock();
3300
3301         oldcurrent = rcu_dereference(bond->curr_active_slave);
3302         /* see if any of the previous devices are up now (i.e. they have
3303          * xmt and rcv traffic). the curr_active_slave does not come into
3304          * the picture unless it is null. also, slave->last_link_up is not
3305          * needed here because we send an arp on each slave and give a slave
3306          * as long as it needs to get the tx/rx within the delta.
3307          * TODO: what about up/down delay in arp mode? it wasn't here before
3308          *       so it can wait
3309          */
3310         bond_for_each_slave_rcu(bond, slave, iter) {
3311                 unsigned long trans_start = dev_trans_start(slave->dev);
3312
3313                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3314
3315                 if (slave->link != BOND_LINK_UP) {
3316                         if (bond_time_in_interval(bond, trans_start, 1) &&
3317                             bond_time_in_interval(bond, slave->last_rx, 1)) {
3318
3319                                 bond_propose_link_state(slave, BOND_LINK_UP);
3320                                 slave_state_changed = 1;
3321
3322                                 /* primary_slave has no meaning in round-robin
3323                                  * mode. the window of a slave being up and
3324                                  * curr_active_slave being null after enslaving
3325                                  * is closed.
3326                                  */
3327                                 if (!oldcurrent) {
3328                                         slave_info(bond->dev, slave->dev, "link status definitely up\n");
3329                                         do_failover = 1;
3330                                 } else {
3331                                         slave_info(bond->dev, slave->dev, "interface is now up\n");
3332                                 }
3333                         }
3334                 } else {
3335                         /* slave->link == BOND_LINK_UP */
3336
3337                         /* not all switches will respond to an arp request
3338                          * when the source ip is 0, so don't take the link down
3339                          * if we don't know our ip yet
3340                          */
3341                         if (!bond_time_in_interval(bond, trans_start, bond->params.missed_max) ||
3342                             !bond_time_in_interval(bond, slave->last_rx, bond->params.missed_max)) {
3343
3344                                 bond_propose_link_state(slave, BOND_LINK_DOWN);
3345                                 slave_state_changed = 1;
3346
3347                                 if (slave->link_failure_count < UINT_MAX)
3348                                         slave->link_failure_count++;
3349
3350                                 slave_info(bond->dev, slave->dev, "interface is now down\n");
3351
3352                                 if (slave == oldcurrent)
3353                                         do_failover = 1;
3354                         }
3355                 }
3356
3357                 /* note: if switch is in round-robin mode, all links
3358                  * must tx arp to ensure all links rx an arp - otherwise
3359                  * links may oscillate or not come up at all; if switch is
3360                  * in something like xor mode, there is nothing we can
3361                  * do - all replies will be rx'ed on same link causing slaves
3362                  * to be unstable during low/no traffic periods
3363                  */
3364                 if (bond_slave_is_up(slave))
3365                         bond_send_validate(bond, slave);
3366         }
3367
3368         rcu_read_unlock();
3369
3370         if (do_failover || slave_state_changed) {
3371                 if (!rtnl_trylock())
3372                         goto re_arm;
3373
3374                 bond_for_each_slave(bond, slave, iter) {
3375                         if (slave->link_new_state != BOND_LINK_NOCHANGE)
3376                                 slave->link = slave->link_new_state;
3377                 }
3378
3379                 if (slave_state_changed) {
3380                         bond_slave_state_change(bond);
3381                         if (BOND_MODE(bond) == BOND_MODE_XOR)
3382                                 bond_update_slave_arr(bond, NULL);
3383                 }
3384                 if (do_failover) {
3385                         block_netpoll_tx();
3386                         bond_select_active_slave(bond);
3387                         unblock_netpoll_tx();
3388                 }
3389                 rtnl_unlock();
3390         }
3391
3392 re_arm:
3393         if (bond->params.arp_interval)
3394                 queue_delayed_work(bond->wq, &bond->arp_work,
3395                                    msecs_to_jiffies(bond->params.arp_interval));
3396 }
3397
3398 /* Called to inspect slaves for active-backup mode ARP monitor link state
3399  * changes.  Sets proposed link state in slaves to specify what action
3400  * should take place for the slave.  Returns 0 if no changes are found, >0
3401  * if changes to link states must be committed.
3402  *
3403  * Called with rcu_read_lock held.
3404  */
3405 static int bond_ab_arp_inspect(struct bonding *bond)
3406 {
3407         unsigned long trans_start, last_rx;
3408         struct list_head *iter;
3409         struct slave *slave;
3410         int commit = 0;
3411
3412         bond_for_each_slave_rcu(bond, slave, iter) {
3413                 bond_propose_link_state(slave, BOND_LINK_NOCHANGE);
3414                 last_rx = slave_last_rx(bond, slave);
3415
3416                 if (slave->link != BOND_LINK_UP) {
3417                         if (bond_time_in_interval(bond, last_rx, 1)) {
3418                                 bond_propose_link_state(slave, BOND_LINK_UP);
3419                                 commit++;
3420                         } else if (slave->link == BOND_LINK_BACK) {
3421                                 bond_propose_link_state(slave, BOND_LINK_FAIL);
3422                                 commit++;
3423                         }
3424                         continue;
3425                 }
3426
3427                 /* Give slaves 2*delta after being enslaved or made
3428                  * active.  This avoids bouncing, as the last receive
3429                  * times need a full ARP monitor cycle to be updated.
3430                  */
3431                 if (bond_time_in_interval(bond, slave->last_link_up, 2))
3432                         continue;
3433
3434                 /* Backup slave is down if:
3435                  * - No current_arp_slave AND
3436                  * - more than (missed_max+1)*delta since last receive AND
3437                  * - the bond has an IP address
3438                  *
3439                  * Note: a non-null current_arp_slave indicates
3440                  * the curr_active_slave went down and we are
3441                  * searching for a new one; under this condition
3442                  * we only take the curr_active_slave down - this
3443                  * gives each slave a chance to tx/rx traffic
3444                  * before being taken out
3445                  */
3446                 if (!bond_is_active_slave(slave) &&
3447                     !rcu_access_pointer(bond->current_arp_slave) &&
3448                     !bond_time_in_interval(bond, last_rx, bond->params.missed_max + 1)) {
3449                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3450                         commit++;
3451                 }
3452
3453                 /* Active slave is down if:
3454                  * - more than missed_max*delta since transmitting OR
3455                  * - (more than missed_max*delta since receive AND
3456                  *    the bond has an IP address)
3457                  */
3458                 trans_start = dev_trans_start(slave->dev);
3459                 if (bond_is_active_slave(slave) &&
3460                     (!bond_time_in_interval(bond, trans_start, bond->params.missed_max) ||
3461                      !bond_time_in_interval(bond, last_rx, bond->params.missed_max))) {
3462                         bond_propose_link_state(slave, BOND_LINK_DOWN);
3463                         commit++;
3464                 }
3465         }
3466
3467         return commit;
3468 }
3469
3470 /* Called to commit link state changes noted by inspection step of
3471  * active-backup mode ARP monitor.
3472  *
3473  * Called with RTNL hold.
3474  */
3475 static void bond_ab_arp_commit(struct bonding *bond)
3476 {
3477         unsigned long trans_start;
3478         struct list_head *iter;
3479         struct slave *slave;
3480
3481         bond_for_each_slave(bond, slave, iter) {
3482                 switch (slave->link_new_state) {
3483                 case BOND_LINK_NOCHANGE:
3484                         continue;
3485
3486                 case BOND_LINK_UP:
3487                         trans_start = dev_trans_start(slave->dev);
3488                         if (rtnl_dereference(bond->curr_active_slave) != slave ||
3489                             (!rtnl_dereference(bond->curr_active_slave) &&
3490                              bond_time_in_interval(bond, trans_start, 1))) {
3491                                 struct slave *current_arp_slave;
3492
3493                                 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
3494                                 bond_set_slave_link_state(slave, BOND_LINK_UP,
3495                                                           BOND_SLAVE_NOTIFY_NOW);
3496                                 if (current_arp_slave) {
3497                                         bond_set_slave_inactive_flags(
3498                                                 current_arp_slave,
3499                                                 BOND_SLAVE_NOTIFY_NOW);
3500                                         RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3501                                 }
3502
3503                                 slave_info(bond->dev, slave->dev, "link status definitely up\n");
3504
3505                                 if (!rtnl_dereference(bond->curr_active_slave) ||
3506                                     slave == rtnl_dereference(bond->primary_slave))
3507                                         goto do_failover;
3508
3509                         }
3510
3511                         continue;
3512
3513                 case BOND_LINK_DOWN:
3514                         if (slave->link_failure_count < UINT_MAX)
3515                                 slave->link_failure_count++;
3516
3517                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3518                                                   BOND_SLAVE_NOTIFY_NOW);
3519                         bond_set_slave_inactive_flags(slave,
3520                                                       BOND_SLAVE_NOTIFY_NOW);
3521
3522                         slave_info(bond->dev, slave->dev, "link status definitely down, disabling slave\n");
3523
3524                         if (slave == rtnl_dereference(bond->curr_active_slave)) {
3525                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3526                                 goto do_failover;
3527                         }
3528
3529                         continue;
3530
3531                 case BOND_LINK_FAIL:
3532                         bond_set_slave_link_state(slave, BOND_LINK_FAIL,
3533                                                   BOND_SLAVE_NOTIFY_NOW);
3534                         bond_set_slave_inactive_flags(slave,
3535                                                       BOND_SLAVE_NOTIFY_NOW);
3536
3537                         /* A slave has just been enslaved and has become
3538                          * the current active slave.
3539                          */
3540                         if (rtnl_dereference(bond->curr_active_slave))
3541                                 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
3542                         continue;
3543
3544                 default:
3545                         slave_err(bond->dev, slave->dev,
3546                                   "impossible: link_new_state %d on slave\n",
3547                                   slave->link_new_state);
3548                         continue;
3549                 }
3550
3551 do_failover:
3552                 block_netpoll_tx();
3553                 bond_select_active_slave(bond);
3554                 unblock_netpoll_tx();
3555         }
3556
3557         bond_set_carrier(bond);
3558 }
3559
3560 /* Send ARP probes for active-backup mode ARP monitor.
3561  *
3562  * Called with rcu_read_lock held.
3563  */
3564 static bool bond_ab_arp_probe(struct bonding *bond)
3565 {
3566         struct slave *slave, *before = NULL, *new_slave = NULL,
3567                      *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
3568                      *curr_active_slave = rcu_dereference(bond->curr_active_slave);
3569         struct list_head *iter;
3570         bool found = false;
3571         bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
3572
3573         if (curr_arp_slave && curr_active_slave)
3574                 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
3575                             curr_arp_slave->dev->name,
3576                             curr_active_slave->dev->name);
3577
3578         if (curr_active_slave) {
3579                 bond_send_validate(bond, curr_active_slave);
3580                 return should_notify_rtnl;
3581         }
3582
3583         /* if we don't have a curr_active_slave, search for the next available
3584          * backup slave from the current_arp_slave and make it the candidate
3585          * for becoming the curr_active_slave
3586          */
3587
3588         if (!curr_arp_slave) {
3589                 curr_arp_slave = bond_first_slave_rcu(bond);
3590                 if (!curr_arp_slave)
3591                         return should_notify_rtnl;
3592         }
3593
3594         bond_for_each_slave_rcu(bond, slave, iter) {
3595                 if (!found && !before && bond_slave_is_up(slave))
3596                         before = slave;
3597
3598                 if (found && !new_slave && bond_slave_is_up(slave))
3599                         new_slave = slave;
3600                 /* if the link state is up at this point, we
3601                  * mark it down - this can happen if we have
3602                  * simultaneous link failures and
3603                  * reselect_active_interface doesn't make this
3604                  * one the current slave so it is still marked
3605                  * up when it is actually down
3606                  */
3607                 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
3608                         bond_set_slave_link_state(slave, BOND_LINK_DOWN,
3609                                                   BOND_SLAVE_NOTIFY_LATER);
3610                         if (slave->link_failure_count < UINT_MAX)
3611                                 slave->link_failure_count++;
3612
3613                         bond_set_slave_inactive_flags(slave,
3614                                                       BOND_SLAVE_NOTIFY_LATER);
3615
3616                         slave_info(bond->dev, slave->dev, "backup interface is now down\n");
3617                 }
3618                 if (slave == curr_arp_slave)
3619                         found = true;
3620         }
3621
3622         if (!new_slave && before)
3623                 new_slave = before;
3624
3625         if (!new_slave)
3626                 goto check_state;
3627
3628         bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
3629                                   BOND_SLAVE_NOTIFY_LATER);
3630         bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
3631         bond_send_validate(bond, new_slave);
3632         new_slave->last_link_up = jiffies;
3633         rcu_assign_pointer(bond->current_arp_slave, new_slave);
3634
3635 check_state:
3636         bond_for_each_slave_rcu(bond, slave, iter) {
3637                 if (slave->should_notify || slave->should_notify_link) {
3638                         should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
3639                         break;
3640                 }
3641         }
3642         return should_notify_rtnl;
3643 }
3644
3645 static void bond_activebackup_arp_mon(struct bonding *bond)
3646 {
3647         bool should_notify_peers = false;
3648         bool should_notify_rtnl = false;
3649         int delta_in_ticks;
3650
3651         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3652
3653         if (!bond_has_slaves(bond))
3654                 goto re_arm;
3655
3656         rcu_read_lock();
3657
3658         should_notify_peers = bond_should_notify_peers(bond);
3659
3660         if (bond_ab_arp_inspect(bond)) {
3661                 rcu_read_unlock();
3662
3663                 /* Race avoidance with bond_close flush of workqueue */
3664                 if (!rtnl_trylock()) {
3665                         delta_in_ticks = 1;
3666                         should_notify_peers = false;
3667                         goto re_arm;
3668                 }
3669
3670                 bond_ab_arp_commit(bond);
3671
3672                 rtnl_unlock();
3673                 rcu_read_lock();
3674         }
3675
3676         should_notify_rtnl = bond_ab_arp_probe(bond);
3677         rcu_read_unlock();
3678
3679 re_arm:
3680         if (bond->params.arp_interval)
3681                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3682
3683         if (should_notify_peers || should_notify_rtnl) {
3684                 if (!rtnl_trylock())
3685                         return;
3686
3687                 if (should_notify_peers)
3688                         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3689                                                  bond->dev);
3690                 if (should_notify_rtnl) {
3691                         bond_slave_state_notify(bond);
3692                         bond_slave_link_notify(bond);
3693                 }
3694
3695                 rtnl_unlock();
3696         }
3697 }
3698
3699 static void bond_arp_monitor(struct work_struct *work)
3700 {
3701         struct bonding *bond = container_of(work, struct bonding,
3702                                             arp_work.work);
3703
3704         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3705                 bond_activebackup_arp_mon(bond);
3706         else
3707                 bond_loadbalance_arp_mon(bond);
3708 }
3709
3710 /*-------------------------- netdev event handling --------------------------*/
3711
3712 /* Change device name */
3713 static int bond_event_changename(struct bonding *bond)
3714 {
3715         bond_remove_proc_entry(bond);
3716         bond_create_proc_entry(bond);
3717
3718         bond_debug_reregister(bond);
3719
3720         return NOTIFY_DONE;
3721 }
3722
3723 static int bond_master_netdev_event(unsigned long event,
3724                                     struct net_device *bond_dev)
3725 {
3726         struct bonding *event_bond = netdev_priv(bond_dev);
3727
3728         netdev_dbg(bond_dev, "%s called\n", __func__);
3729
3730         switch (event) {
3731         case NETDEV_CHANGENAME:
3732                 return bond_event_changename(event_bond);
3733         case NETDEV_UNREGISTER:
3734                 bond_remove_proc_entry(event_bond);
3735 #ifdef CONFIG_XFRM_OFFLOAD
3736                 xfrm_dev_state_flush(dev_net(bond_dev), bond_dev, true);
3737 #endif /* CONFIG_XFRM_OFFLOAD */
3738                 break;
3739         case NETDEV_REGISTER:
3740                 bond_create_proc_entry(event_bond);
3741                 break;
3742         default:
3743                 break;
3744         }
3745
3746         return NOTIFY_DONE;
3747 }
3748
3749 static int bond_slave_netdev_event(unsigned long event,
3750                                    struct net_device *slave_dev)
3751 {
3752         struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3753         struct bonding *bond;
3754         struct net_device *bond_dev;
3755
3756         /* A netdev event can be generated while enslaving a device
3757          * before netdev_rx_handler_register is called in which case
3758          * slave will be NULL
3759          */
3760         if (!slave) {
3761                 netdev_dbg(slave_dev, "%s called on NULL slave\n", __func__);
3762                 return NOTIFY_DONE;
3763         }
3764
3765         bond_dev = slave->bond->dev;
3766         bond = slave->bond;
3767         primary = rtnl_dereference(bond->primary_slave);
3768
3769         slave_dbg(bond_dev, slave_dev, "%s called\n", __func__);
3770
3771         switch (event) {
3772         case NETDEV_UNREGISTER:
3773                 if (bond_dev->type != ARPHRD_ETHER)
3774                         bond_release_and_destroy(bond_dev, slave_dev);
3775                 else
3776                         __bond_release_one(bond_dev, slave_dev, false, true);
3777                 break;
3778         case NETDEV_UP:
3779         case NETDEV_CHANGE:
3780                 /* For 802.3ad mode only:
3781                  * Getting invalid Speed/Duplex values here will put slave
3782                  * in weird state. Mark it as link-fail if the link was
3783                  * previously up or link-down if it hasn't yet come up, and
3784                  * let link-monitoring (miimon) set it right when correct
3785                  * speeds/duplex are available.
3786                  */
3787                 if (bond_update_speed_duplex(slave) &&
3788                     BOND_MODE(bond) == BOND_MODE_8023AD) {
3789                         if (slave->last_link_up)
3790                                 slave->link = BOND_LINK_FAIL;
3791                         else
3792                                 slave->link = BOND_LINK_DOWN;
3793                 }
3794
3795                 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3796                         bond_3ad_adapter_speed_duplex_changed(slave);
3797                 fallthrough;
3798         case NETDEV_DOWN:
3799                 /* Refresh slave-array if applicable!
3800                  * If the setup does not use miimon or arpmon (mode-specific!),
3801                  * then these events will not cause the slave-array to be
3802                  * refreshed. This will cause xmit to use a slave that is not
3803                  * usable. Avoid such situation by refeshing the array at these
3804                  * events. If these (miimon/arpmon) parameters are configured
3805                  * then array gets refreshed twice and that should be fine!
3806                  */
3807                 if (bond_mode_can_use_xmit_hash(bond))
3808                         bond_update_slave_arr(bond, NULL);
3809                 break;
3810         case NETDEV_CHANGEMTU:
3811                 /* TODO: Should slaves be allowed to
3812                  * independently alter their MTU?  For
3813                  * an active-backup bond, slaves need
3814                  * not be the same type of device, so
3815                  * MTUs may vary.  For other modes,
3816                  * slaves arguably should have the
3817                  * same MTUs. To do this, we'd need to
3818                  * take over the slave's change_mtu
3819                  * function for the duration of their
3820                  * servitude.
3821                  */
3822                 break;
3823         case NETDEV_CHANGENAME:
3824                 /* we don't care if we don't have primary set */
3825                 if (!bond_uses_primary(bond) ||
3826                     !bond->params.primary[0])
3827                         break;
3828
3829                 if (slave == primary) {
3830                         /* slave's name changed - he's no longer primary */
3831                         RCU_INIT_POINTER(bond->primary_slave, NULL);
3832                 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3833                         /* we have a new primary slave */
3834                         rcu_assign_pointer(bond->primary_slave, slave);
3835                 } else { /* we didn't change primary - exit */
3836                         break;
3837                 }
3838
3839                 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3840                             primary ? slave_dev->name : "none");
3841
3842                 block_netpoll_tx();
3843                 bond_select_active_slave(bond);
3844                 unblock_netpoll_tx();
3845                 break;
3846         case NETDEV_FEAT_CHANGE:
3847                 bond_compute_features(bond);
3848                 break;
3849         case NETDEV_RESEND_IGMP:
3850                 /* Propagate to master device */
3851                 call_netdevice_notifiers(event, slave->bond->dev);
3852                 break;
3853         default:
3854                 break;
3855         }
3856
3857         return NOTIFY_DONE;
3858 }
3859
3860 /* bond_netdev_event: handle netdev notifier chain events.
3861  *
3862  * This function receives events for the netdev chain.  The caller (an
3863  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3864  * locks for us to safely manipulate the slave devices (RTNL lock,
3865  * dev_probe_lock).
3866  */
3867 static int bond_netdev_event(struct notifier_block *this,
3868                              unsigned long event, void *ptr)
3869 {
3870         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3871
3872         netdev_dbg(event_dev, "%s received %s\n",
3873                    __func__, netdev_cmd_to_name(event));
3874
3875         if (!(event_dev->priv_flags & IFF_BONDING))
3876                 return NOTIFY_DONE;
3877
3878         if (event_dev->flags & IFF_MASTER) {
3879                 int ret;
3880
3881                 ret = bond_master_netdev_event(event, event_dev);
3882                 if (ret != NOTIFY_DONE)
3883                         return ret;
3884         }
3885
3886         if (event_dev->flags & IFF_SLAVE)
3887                 return bond_slave_netdev_event(event, event_dev);
3888
3889         return NOTIFY_DONE;
3890 }
3891
3892 static struct notifier_block bond_netdev_notifier = {
3893         .notifier_call = bond_netdev_event,
3894 };
3895
3896 /*---------------------------- Hashing Policies -----------------------------*/
3897
3898 /* Helper to access data in a packet, with or without a backing skb.
3899  * If skb is given the data is linearized if necessary via pskb_may_pull.
3900  */
3901 static inline const void *bond_pull_data(struct sk_buff *skb,
3902                                          const void *data, int hlen, int n)
3903 {
3904         if (likely(n <= hlen))
3905                 return data;
3906         else if (skb && likely(pskb_may_pull(skb, n)))
3907                 return skb->head;
3908
3909         return NULL;
3910 }
3911
3912 /* L2 hash helper */
3913 static inline u32 bond_eth_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
3914 {
3915         struct ethhdr *ep;
3916
3917         data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
3918         if (!data)
3919                 return 0;
3920
3921         ep = (struct ethhdr *)(data + mhoff);
3922         return ep->h_dest[5] ^ ep->h_source[5] ^ be16_to_cpu(ep->h_proto);
3923 }
3924
3925 static bool bond_flow_ip(struct sk_buff *skb, struct flow_keys *fk, const void *data,
3926                          int hlen, __be16 l2_proto, int *nhoff, int *ip_proto, bool l34)
3927 {
3928         const struct ipv6hdr *iph6;
3929         const struct iphdr *iph;
3930
3931         if (l2_proto == htons(ETH_P_IP)) {
3932                 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph));
3933                 if (!data)
3934                         return false;
3935
3936                 iph = (const struct iphdr *)(data + *nhoff);
3937                 iph_to_flow_copy_v4addrs(fk, iph);
3938                 *nhoff += iph->ihl << 2;
3939                 if (!ip_is_fragment(iph))
3940                         *ip_proto = iph->protocol;
3941         } else if (l2_proto == htons(ETH_P_IPV6)) {
3942                 data = bond_pull_data(skb, data, hlen, *nhoff + sizeof(*iph6));
3943                 if (!data)
3944                         return false;
3945
3946                 iph6 = (const struct ipv6hdr *)(data + *nhoff);
3947                 iph_to_flow_copy_v6addrs(fk, iph6);
3948                 *nhoff += sizeof(*iph6);
3949                 *ip_proto = iph6->nexthdr;
3950         } else {
3951                 return false;
3952         }
3953
3954         if (l34 && *ip_proto >= 0)
3955                 fk->ports.ports = __skb_flow_get_ports(skb, *nhoff, *ip_proto, data, hlen);
3956
3957         return true;
3958 }
3959
3960 static u32 bond_vlan_srcmac_hash(struct sk_buff *skb, const void *data, int mhoff, int hlen)
3961 {
3962         u32 srcmac_vendor = 0, srcmac_dev = 0;
3963         struct ethhdr *mac_hdr;
3964         u16 vlan = 0;
3965         int i;
3966
3967         data = bond_pull_data(skb, data, hlen, mhoff + sizeof(struct ethhdr));
3968         if (!data)
3969                 return 0;
3970         mac_hdr = (struct ethhdr *)(data + mhoff);
3971
3972         for (i = 0; i < 3; i++)
3973                 srcmac_vendor = (srcmac_vendor << 8) | mac_hdr->h_source[i];
3974
3975         for (i = 3; i < ETH_ALEN; i++)
3976                 srcmac_dev = (srcmac_dev << 8) | mac_hdr->h_source[i];
3977
3978         if (skb && skb_vlan_tag_present(skb))
3979                 vlan = skb_vlan_tag_get(skb);
3980
3981         return vlan ^ srcmac_vendor ^ srcmac_dev;
3982 }
3983
3984 /* Extract the appropriate headers based on bond's xmit policy */
3985 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb, const void *data,
3986                               __be16 l2_proto, int nhoff, int hlen, struct flow_keys *fk)
3987 {
3988         bool l34 = bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34;
3989         int ip_proto = -1;
3990
3991         switch (bond->params.xmit_policy) {
3992         case BOND_XMIT_POLICY_ENCAP23:
3993         case BOND_XMIT_POLICY_ENCAP34:
3994                 memset(fk, 0, sizeof(*fk));
3995                 return __skb_flow_dissect(NULL, skb, &flow_keys_bonding,
3996                                           fk, data, l2_proto, nhoff, hlen, 0);
3997         default:
3998                 break;
3999         }
4000
4001         fk->ports.ports = 0;
4002         memset(&fk->icmp, 0, sizeof(fk->icmp));
4003         if (!bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34))
4004                 return false;
4005
4006         /* ICMP error packets contains at least 8 bytes of the header
4007          * of the packet which generated the error. Use this information
4008          * to correlate ICMP error packets within the same flow which
4009          * generated the error.
4010          */
4011         if (ip_proto == IPPROTO_ICMP || ip_proto == IPPROTO_ICMPV6) {
4012                 skb_flow_get_icmp_tci(skb, &fk->icmp, data, nhoff, hlen);
4013                 if (ip_proto == IPPROTO_ICMP) {
4014                         if (!icmp_is_err(fk->icmp.type))
4015                                 return true;
4016
4017                         nhoff += sizeof(struct icmphdr);
4018                 } else if (ip_proto == IPPROTO_ICMPV6) {
4019                         if (!icmpv6_is_err(fk->icmp.type))
4020                                 return true;
4021
4022                         nhoff += sizeof(struct icmp6hdr);
4023                 }
4024                 return bond_flow_ip(skb, fk, data, hlen, l2_proto, &nhoff, &ip_proto, l34);
4025         }
4026
4027         return true;
4028 }
4029
4030 static u32 bond_ip_hash(u32 hash, struct flow_keys *flow, int xmit_policy)
4031 {
4032         hash ^= (__force u32)flow_get_u32_dst(flow) ^
4033                 (__force u32)flow_get_u32_src(flow);
4034         hash ^= (hash >> 16);
4035         hash ^= (hash >> 8);
4036
4037         /* discard lowest hash bit to deal with the common even ports pattern */
4038         if (xmit_policy == BOND_XMIT_POLICY_LAYER34 ||
4039                 xmit_policy == BOND_XMIT_POLICY_ENCAP34)
4040                 return hash >> 1;
4041
4042         return hash;
4043 }
4044
4045 /* Generate hash based on xmit policy. If @skb is given it is used to linearize
4046  * the data as required, but this function can be used without it if the data is
4047  * known to be linear (e.g. with xdp_buff).
4048  */
4049 static u32 __bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, const void *data,
4050                             __be16 l2_proto, int mhoff, int nhoff, int hlen)
4051 {
4052         struct flow_keys flow;
4053         u32 hash;
4054
4055         if (bond->params.xmit_policy == BOND_XMIT_POLICY_VLAN_SRCMAC)
4056                 return bond_vlan_srcmac_hash(skb, data, mhoff, hlen);
4057
4058         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
4059             !bond_flow_dissect(bond, skb, data, l2_proto, nhoff, hlen, &flow))
4060                 return bond_eth_hash(skb, data, mhoff, hlen);
4061
4062         if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
4063             bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23) {
4064                 hash = bond_eth_hash(skb, data, mhoff, hlen);
4065         } else {
4066                 if (flow.icmp.id)
4067                         memcpy(&hash, &flow.icmp, sizeof(hash));
4068                 else
4069                         memcpy(&hash, &flow.ports.ports, sizeof(hash));
4070         }
4071
4072         return bond_ip_hash(hash, &flow, bond->params.xmit_policy);
4073 }
4074
4075 /**
4076  * bond_xmit_hash - generate a hash value based on the xmit policy
4077  * @bond: bonding device
4078  * @skb: buffer to use for headers
4079  *
4080  * This function will extract the necessary headers from the skb buffer and use
4081  * them to generate a hash based on the xmit_policy set in the bonding device
4082  */
4083 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
4084 {
4085         if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
4086             skb->l4_hash)
4087                 return skb->hash;
4088
4089         return __bond_xmit_hash(bond, skb, skb->data, skb->protocol,
4090                                 skb_mac_offset(skb), skb_network_offset(skb),
4091                                 skb_headlen(skb));
4092 }
4093
4094 /**
4095  * bond_xmit_hash_xdp - generate a hash value based on the xmit policy
4096  * @bond: bonding device
4097  * @xdp: buffer to use for headers
4098  *
4099  * The XDP variant of bond_xmit_hash.
4100  */
4101 static u32 bond_xmit_hash_xdp(struct bonding *bond, struct xdp_buff *xdp)
4102 {
4103         struct ethhdr *eth;
4104
4105         if (xdp->data + sizeof(struct ethhdr) > xdp->data_end)
4106                 return 0;
4107
4108         eth = (struct ethhdr *)xdp->data;
4109
4110         return __bond_xmit_hash(bond, NULL, xdp->data, eth->h_proto, 0,
4111                                 sizeof(struct ethhdr), xdp->data_end - xdp->data);
4112 }
4113
4114 /*-------------------------- Device entry points ----------------------------*/
4115
4116 void bond_work_init_all(struct bonding *bond)
4117 {
4118         INIT_DELAYED_WORK(&bond->mcast_work,
4119                           bond_resend_igmp_join_requests_delayed);
4120         INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
4121         INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
4122         INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
4123         INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
4124         INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
4125 }
4126
4127 static void bond_work_cancel_all(struct bonding *bond)
4128 {
4129         cancel_delayed_work_sync(&bond->mii_work);
4130         cancel_delayed_work_sync(&bond->arp_work);
4131         cancel_delayed_work_sync(&bond->alb_work);
4132         cancel_delayed_work_sync(&bond->ad_work);
4133         cancel_delayed_work_sync(&bond->mcast_work);
4134         cancel_delayed_work_sync(&bond->slave_arr_work);
4135 }
4136
4137 static int bond_open(struct net_device *bond_dev)
4138 {
4139         struct bonding *bond = netdev_priv(bond_dev);
4140         struct list_head *iter;
4141         struct slave *slave;
4142
4143         /* reset slave->backup and slave->inactive */
4144         if (bond_has_slaves(bond)) {
4145                 bond_for_each_slave(bond, slave, iter) {
4146                         if (bond_uses_primary(bond) &&
4147                             slave != rcu_access_pointer(bond->curr_active_slave)) {
4148                                 bond_set_slave_inactive_flags(slave,
4149                                                               BOND_SLAVE_NOTIFY_NOW);
4150                         } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
4151                                 bond_set_slave_active_flags(slave,
4152                                                             BOND_SLAVE_NOTIFY_NOW);
4153                         }
4154                 }
4155         }
4156
4157         if (bond_is_lb(bond)) {
4158                 /* bond_alb_initialize must be called before the timer
4159                  * is started.
4160                  */
4161                 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
4162                         return -ENOMEM;
4163                 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
4164                         queue_delayed_work(bond->wq, &bond->alb_work, 0);
4165         }
4166
4167         if (bond->params.miimon)  /* link check interval, in milliseconds. */
4168                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
4169
4170         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
4171                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
4172                 bond->recv_probe = bond_rcv_validate;
4173         }
4174
4175         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4176                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
4177                 /* register to receive LACPDUs */
4178                 bond->recv_probe = bond_3ad_lacpdu_recv;
4179                 bond_3ad_initiate_agg_selection(bond, 1);
4180         }
4181
4182         if (bond_mode_can_use_xmit_hash(bond))
4183                 bond_update_slave_arr(bond, NULL);
4184
4185         return 0;
4186 }
4187
4188 static int bond_close(struct net_device *bond_dev)
4189 {
4190         struct bonding *bond = netdev_priv(bond_dev);
4191
4192         bond_work_cancel_all(bond);
4193         bond->send_peer_notif = 0;
4194         if (bond_is_lb(bond))
4195                 bond_alb_deinitialize(bond);
4196         bond->recv_probe = NULL;
4197
4198         return 0;
4199 }
4200
4201 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
4202  * that some drivers can provide 32bit values only.
4203  */
4204 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
4205                             const struct rtnl_link_stats64 *_new,
4206                             const struct rtnl_link_stats64 *_old)
4207 {
4208         const u64 *new = (const u64 *)_new;
4209         const u64 *old = (const u64 *)_old;
4210         u64 *res = (u64 *)_res;
4211         int i;
4212
4213         for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
4214                 u64 nv = new[i];
4215                 u64 ov = old[i];
4216                 s64 delta = nv - ov;
4217
4218                 /* detects if this particular field is 32bit only */
4219                 if (((nv | ov) >> 32) == 0)
4220                         delta = (s64)(s32)((u32)nv - (u32)ov);
4221
4222                 /* filter anomalies, some drivers reset their stats
4223                  * at down/up events.
4224                  */
4225                 if (delta > 0)
4226                         res[i] += delta;
4227         }
4228 }
4229
4230 #ifdef CONFIG_LOCKDEP
4231 static int bond_get_lowest_level_rcu(struct net_device *dev)
4232 {
4233         struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
4234         struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
4235         int cur = 0, max = 0;
4236
4237         now = dev;
4238         iter = &dev->adj_list.lower;
4239
4240         while (1) {
4241                 next = NULL;
4242                 while (1) {
4243                         ldev = netdev_next_lower_dev_rcu(now, &iter);
4244                         if (!ldev)
4245                                 break;
4246
4247                         next = ldev;
4248                         niter = &ldev->adj_list.lower;
4249                         dev_stack[cur] = now;
4250                         iter_stack[cur++] = iter;
4251                         if (max <= cur)
4252                                 max = cur;
4253                         break;
4254                 }
4255
4256                 if (!next) {
4257                         if (!cur)
4258                                 return max;
4259                         next = dev_stack[--cur];
4260                         niter = iter_stack[cur];
4261                 }
4262
4263                 now = next;
4264                 iter = niter;
4265         }
4266
4267         return max;
4268 }
4269 #endif
4270
4271 static void bond_get_stats(struct net_device *bond_dev,
4272                            struct rtnl_link_stats64 *stats)
4273 {
4274         struct bonding *bond = netdev_priv(bond_dev);
4275         struct rtnl_link_stats64 temp;
4276         struct list_head *iter;
4277         struct slave *slave;
4278         int nest_level = 0;
4279
4280
4281         rcu_read_lock();
4282 #ifdef CONFIG_LOCKDEP
4283         nest_level = bond_get_lowest_level_rcu(bond_dev);
4284 #endif
4285
4286         spin_lock_nested(&bond->stats_lock, nest_level);
4287         memcpy(stats, &bond->bond_stats, sizeof(*stats));
4288
4289         bond_for_each_slave_rcu(bond, slave, iter) {
4290                 const struct rtnl_link_stats64 *new =
4291                         dev_get_stats(slave->dev, &temp);
4292
4293                 bond_fold_stats(stats, new, &slave->slave_stats);
4294
4295                 /* save off the slave stats for the next run */
4296                 memcpy(&slave->slave_stats, new, sizeof(*new));
4297         }
4298
4299         memcpy(&bond->bond_stats, stats, sizeof(*stats));
4300         spin_unlock(&bond->stats_lock);
4301         rcu_read_unlock();
4302 }
4303
4304 static int bond_eth_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4305 {
4306         struct bonding *bond = netdev_priv(bond_dev);
4307         struct mii_ioctl_data *mii = NULL;
4308         const struct net_device_ops *ops;
4309         struct net_device *real_dev;
4310         struct hwtstamp_config cfg;
4311         struct ifreq ifrr;
4312         int res = 0;
4313
4314         netdev_dbg(bond_dev, "bond_eth_ioctl: cmd=%d\n", cmd);
4315
4316         switch (cmd) {
4317         case SIOCGMIIPHY:
4318                 mii = if_mii(ifr);
4319                 if (!mii)
4320                         return -EINVAL;
4321
4322                 mii->phy_id = 0;
4323                 fallthrough;
4324         case SIOCGMIIREG:
4325                 /* We do this again just in case we were called by SIOCGMIIREG
4326                  * instead of SIOCGMIIPHY.
4327                  */
4328                 mii = if_mii(ifr);
4329                 if (!mii)
4330                         return -EINVAL;
4331
4332                 if (mii->reg_num == 1) {
4333                         mii->val_out = 0;
4334                         if (netif_carrier_ok(bond->dev))
4335                                 mii->val_out = BMSR_LSTATUS;
4336                 }
4337
4338                 break;
4339         case SIOCSHWTSTAMP:
4340                 if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4341                         return -EFAULT;
4342
4343                 if (!(cfg.flags & HWTSTAMP_FLAG_BONDED_PHC_INDEX))
4344                         return -EOPNOTSUPP;
4345
4346                 fallthrough;
4347         case SIOCGHWTSTAMP:
4348                 real_dev = bond_option_active_slave_get_rcu(bond);
4349                 if (!real_dev)
4350                         return -EOPNOTSUPP;
4351
4352                 strscpy_pad(ifrr.ifr_name, real_dev->name, IFNAMSIZ);
4353                 ifrr.ifr_ifru = ifr->ifr_ifru;
4354
4355                 ops = real_dev->netdev_ops;
4356                 if (netif_device_present(real_dev) && ops->ndo_eth_ioctl) {
4357                         res = ops->ndo_eth_ioctl(real_dev, &ifrr, cmd);
4358                         if (res)
4359                                 return res;
4360
4361                         ifr->ifr_ifru = ifrr.ifr_ifru;
4362                         if (copy_from_user(&cfg, ifr->ifr_data, sizeof(cfg)))
4363                                 return -EFAULT;
4364
4365                         /* Set the BOND_PHC_INDEX flag to notify user space */
4366                         cfg.flags |= HWTSTAMP_FLAG_BONDED_PHC_INDEX;
4367
4368                         return copy_to_user(ifr->ifr_data, &cfg, sizeof(cfg)) ?
4369                                 -EFAULT : 0;
4370                 }
4371                 fallthrough;
4372         default:
4373                 res = -EOPNOTSUPP;
4374         }
4375
4376         return res;
4377 }
4378
4379 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
4380 {
4381         struct bonding *bond = netdev_priv(bond_dev);
4382         struct net_device *slave_dev = NULL;
4383         struct ifbond k_binfo;
4384         struct ifbond __user *u_binfo = NULL;
4385         struct ifslave k_sinfo;
4386         struct ifslave __user *u_sinfo = NULL;
4387         struct bond_opt_value newval;
4388         struct net *net;
4389         int res = 0;
4390
4391         netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
4392
4393         switch (cmd) {
4394         case SIOCBONDINFOQUERY:
4395                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
4396
4397                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
4398                         return -EFAULT;
4399
4400                 bond_info_query(bond_dev, &k_binfo);
4401                 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4402                         return -EFAULT;
4403
4404                 return 0;
4405         case SIOCBONDSLAVEINFOQUERY:
4406                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4407
4408                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4409                         return -EFAULT;
4410
4411                 res = bond_slave_info_query(bond_dev, &k_sinfo);
4412                 if (res == 0 &&
4413                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4414                         return -EFAULT;
4415
4416                 return res;
4417         default:
4418                 break;
4419         }
4420
4421         net = dev_net(bond_dev);
4422
4423         if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
4424                 return -EPERM;
4425
4426         slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
4427
4428         slave_dbg(bond_dev, slave_dev, "slave_dev=%p:\n", slave_dev);
4429
4430         if (!slave_dev)
4431                 return -ENODEV;
4432
4433         switch (cmd) {
4434         case SIOCBONDENSLAVE:
4435                 res = bond_enslave(bond_dev, slave_dev, NULL);
4436                 break;
4437         case SIOCBONDRELEASE:
4438                 res = bond_release(bond_dev, slave_dev);
4439                 break;
4440         case SIOCBONDSETHWADDR:
4441                 res = bond_set_dev_addr(bond_dev, slave_dev);
4442                 break;
4443         case SIOCBONDCHANGEACTIVE:
4444                 bond_opt_initstr(&newval, slave_dev->name);
4445                 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
4446                                             &newval);
4447                 break;
4448         default:
4449                 res = -EOPNOTSUPP;
4450         }
4451
4452         return res;
4453 }
4454
4455 static int bond_siocdevprivate(struct net_device *bond_dev, struct ifreq *ifr,
4456                                void __user *data, int cmd)
4457 {
4458         struct ifreq ifrdata = { .ifr_data = data };
4459
4460         switch (cmd) {
4461         case BOND_INFO_QUERY_OLD:
4462                 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDINFOQUERY);
4463         case BOND_SLAVE_INFO_QUERY_OLD:
4464                 return bond_do_ioctl(bond_dev, &ifrdata, SIOCBONDSLAVEINFOQUERY);
4465         case BOND_ENSLAVE_OLD:
4466                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDENSLAVE);
4467         case BOND_RELEASE_OLD:
4468                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDRELEASE);
4469         case BOND_SETHWADDR_OLD:
4470                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDSETHWADDR);
4471         case BOND_CHANGE_ACTIVE_OLD:
4472                 return bond_do_ioctl(bond_dev, ifr, SIOCBONDCHANGEACTIVE);
4473         }
4474
4475         return -EOPNOTSUPP;
4476 }
4477
4478 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
4479 {
4480         struct bonding *bond = netdev_priv(bond_dev);
4481
4482         if (change & IFF_PROMISC)
4483                 bond_set_promiscuity(bond,
4484                                      bond_dev->flags & IFF_PROMISC ? 1 : -1);
4485
4486         if (change & IFF_ALLMULTI)
4487                 bond_set_allmulti(bond,
4488                                   bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
4489 }
4490
4491 static void bond_set_rx_mode(struct net_device *bond_dev)
4492 {
4493         struct bonding *bond = netdev_priv(bond_dev);
4494         struct list_head *iter;
4495         struct slave *slave;
4496
4497         rcu_read_lock();
4498         if (bond_uses_primary(bond)) {
4499                 slave = rcu_dereference(bond->curr_active_slave);
4500                 if (slave) {
4501                         dev_uc_sync(slave->dev, bond_dev);
4502                         dev_mc_sync(slave->dev, bond_dev);
4503                 }
4504         } else {
4505                 bond_for_each_slave_rcu(bond, slave, iter) {
4506                         dev_uc_sync_multiple(slave->dev, bond_dev);
4507                         dev_mc_sync_multiple(slave->dev, bond_dev);
4508                 }
4509         }
4510         rcu_read_unlock();
4511 }
4512
4513 static int bond_neigh_init(struct neighbour *n)
4514 {
4515         struct bonding *bond = netdev_priv(n->dev);
4516         const struct net_device_ops *slave_ops;
4517         struct neigh_parms parms;
4518         struct slave *slave;
4519         int ret = 0;
4520
4521         rcu_read_lock();
4522         slave = bond_first_slave_rcu(bond);
4523         if (!slave)
4524                 goto out;
4525         slave_ops = slave->dev->netdev_ops;
4526         if (!slave_ops->ndo_neigh_setup)
4527                 goto out;
4528
4529         /* TODO: find another way [1] to implement this.
4530          * Passing a zeroed structure is fragile,
4531          * but at least we do not pass garbage.
4532          *
4533          * [1] One way would be that ndo_neigh_setup() never touch
4534          *     struct neigh_parms, but propagate the new neigh_setup()
4535          *     back to ___neigh_create() / neigh_parms_alloc()
4536          */
4537         memset(&parms, 0, sizeof(parms));
4538         ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
4539
4540         if (ret)
4541                 goto out;
4542
4543         if (parms.neigh_setup)
4544                 ret = parms.neigh_setup(n);
4545 out:
4546         rcu_read_unlock();
4547         return ret;
4548 }
4549
4550 /* The bonding ndo_neigh_setup is called at init time beofre any
4551  * slave exists. So we must declare proxy setup function which will
4552  * be used at run time to resolve the actual slave neigh param setup.
4553  *
4554  * It's also called by master devices (such as vlans) to setup their
4555  * underlying devices. In that case - do nothing, we're already set up from
4556  * our init.
4557  */
4558 static int bond_neigh_setup(struct net_device *dev,
4559                             struct neigh_parms *parms)
4560 {
4561         /* modify only our neigh_parms */
4562         if (parms->dev == dev)
4563                 parms->neigh_setup = bond_neigh_init;
4564
4565         return 0;
4566 }
4567
4568 /* Change the MTU of all of a master's slaves to match the master */
4569 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4570 {
4571         struct bonding *bond = netdev_priv(bond_dev);
4572         struct slave *slave, *rollback_slave;
4573         struct list_head *iter;
4574         int res = 0;
4575
4576         netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
4577
4578         bond_for_each_slave(bond, slave, iter) {
4579                 slave_dbg(bond_dev, slave->dev, "s %p c_m %p\n",
4580                            slave, slave->dev->netdev_ops->ndo_change_mtu);
4581
4582                 res = dev_set_mtu(slave->dev, new_mtu);
4583
4584                 if (res) {
4585                         /* If we failed to set the slave's mtu to the new value
4586                          * we must abort the operation even in ACTIVE_BACKUP
4587                          * mode, because if we allow the backup slaves to have
4588                          * different mtu values than the active slave we'll
4589                          * need to change their mtu when doing a failover. That
4590                          * means changing their mtu from timer context, which
4591                          * is probably not a good idea.
4592                          */
4593                         slave_dbg(bond_dev, slave->dev, "err %d setting mtu to %d\n",
4594                                   res, new_mtu);
4595                         goto unwind;
4596                 }
4597         }
4598
4599         bond_dev->mtu = new_mtu;
4600
4601         return 0;
4602
4603 unwind:
4604         /* unwind from head to the slave that failed */
4605         bond_for_each_slave(bond, rollback_slave, iter) {
4606                 int tmp_res;
4607
4608                 if (rollback_slave == slave)
4609                         break;
4610
4611                 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
4612                 if (tmp_res)
4613                         slave_dbg(bond_dev, rollback_slave->dev, "unwind err %d\n",
4614                                   tmp_res);
4615         }
4616
4617         return res;
4618 }
4619
4620 /* Change HW address
4621  *
4622  * Note that many devices must be down to change the HW address, and
4623  * downing the master releases all slaves.  We can make bonds full of
4624  * bonding devices to test this, however.
4625  */
4626 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4627 {
4628         struct bonding *bond = netdev_priv(bond_dev);
4629         struct slave *slave, *rollback_slave;
4630         struct sockaddr_storage *ss = addr, tmp_ss;
4631         struct list_head *iter;
4632         int res = 0;
4633
4634         if (BOND_MODE(bond) == BOND_MODE_ALB)
4635                 return bond_alb_set_mac_address(bond_dev, addr);
4636
4637
4638         netdev_dbg(bond_dev, "%s: bond=%p\n", __func__, bond);
4639
4640         /* If fail_over_mac is enabled, do nothing and return success.
4641          * Returning an error causes ifenslave to fail.
4642          */
4643         if (bond->params.fail_over_mac &&
4644             BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
4645                 return 0;
4646
4647         if (!is_valid_ether_addr(ss->__data))
4648                 return -EADDRNOTAVAIL;
4649
4650         bond_for_each_slave(bond, slave, iter) {
4651                 slave_dbg(bond_dev, slave->dev, "%s: slave=%p\n",
4652                           __func__, slave);
4653                 res = dev_set_mac_address(slave->dev, addr, NULL);
4654                 if (res) {
4655                         /* TODO: consider downing the slave
4656                          * and retry ?
4657                          * User should expect communications
4658                          * breakage anyway until ARP finish
4659                          * updating, so...
4660                          */
4661                         slave_dbg(bond_dev, slave->dev, "%s: err %d\n",
4662                                   __func__, res);
4663                         goto unwind;
4664                 }
4665         }
4666
4667         /* success */
4668         dev_addr_set(bond_dev, ss->__data);
4669         return 0;
4670
4671 unwind:
4672         memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
4673         tmp_ss.ss_family = bond_dev->type;
4674
4675         /* unwind from head to the slave that failed */
4676         bond_for_each_slave(bond, rollback_slave, iter) {
4677                 int tmp_res;
4678
4679                 if (rollback_slave == slave)
4680                         break;
4681
4682                 tmp_res = dev_set_mac_address(rollback_slave->dev,
4683                                               (struct sockaddr *)&tmp_ss, NULL);
4684                 if (tmp_res) {
4685                         slave_dbg(bond_dev, rollback_slave->dev, "%s: unwind err %d\n",
4686                                    __func__, tmp_res);
4687                 }
4688         }
4689
4690         return res;
4691 }
4692
4693 /**
4694  * bond_get_slave_by_id - get xmit slave with slave_id
4695  * @bond: bonding device that is transmitting
4696  * @slave_id: slave id up to slave_cnt-1 through which to transmit
4697  *
4698  * This function tries to get slave with slave_id but in case
4699  * it fails, it tries to find the first available slave for transmission.
4700  */
4701 static struct slave *bond_get_slave_by_id(struct bonding *bond,
4702                                           int slave_id)
4703 {
4704         struct list_head *iter;
4705         struct slave *slave;
4706         int i = slave_id;
4707
4708         /* Here we start from the slave with slave_id */
4709         bond_for_each_slave_rcu(bond, slave, iter) {
4710                 if (--i < 0) {
4711                         if (bond_slave_can_tx(slave))
4712                                 return slave;
4713                 }
4714         }
4715
4716         /* Here we start from the first slave up to slave_id */
4717         i = slave_id;
4718         bond_for_each_slave_rcu(bond, slave, iter) {
4719                 if (--i < 0)
4720                         break;
4721                 if (bond_slave_can_tx(slave))
4722                         return slave;
4723         }
4724         /* no slave that can tx has been found */
4725         return NULL;
4726 }
4727
4728 /**
4729  * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
4730  * @bond: bonding device to use
4731  *
4732  * Based on the value of the bonding device's packets_per_slave parameter
4733  * this function generates a slave id, which is usually used as the next
4734  * slave to transmit through.
4735  */
4736 static u32 bond_rr_gen_slave_id(struct bonding *bond)
4737 {
4738         u32 slave_id;
4739         struct reciprocal_value reciprocal_packets_per_slave;
4740         int packets_per_slave = bond->params.packets_per_slave;
4741
4742         switch (packets_per_slave) {
4743         case 0:
4744                 slave_id = prandom_u32();
4745                 break;
4746         case 1:
4747                 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4748                 break;
4749         default:
4750                 reciprocal_packets_per_slave =
4751                         bond->params.reciprocal_packets_per_slave;
4752                 slave_id = this_cpu_inc_return(*bond->rr_tx_counter);
4753                 slave_id = reciprocal_divide(slave_id,
4754                                              reciprocal_packets_per_slave);
4755                 break;
4756         }
4757
4758         return slave_id;
4759 }
4760
4761 static struct slave *bond_xmit_roundrobin_slave_get(struct bonding *bond,
4762                                                     struct sk_buff *skb)
4763 {
4764         struct slave *slave;
4765         int slave_cnt;
4766         u32 slave_id;
4767
4768         /* Start with the curr_active_slave that joined the bond as the
4769          * default for sending IGMP traffic.  For failover purposes one
4770          * needs to maintain some consistency for the interface that will
4771          * send the join/membership reports.  The curr_active_slave found
4772          * will send all of this type of traffic.
4773          */
4774         if (skb->protocol == htons(ETH_P_IP)) {
4775                 int noff = skb_network_offset(skb);
4776                 struct iphdr *iph;
4777
4778                 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
4779                         goto non_igmp;
4780
4781                 iph = ip_hdr(skb);
4782                 if (iph->protocol == IPPROTO_IGMP) {
4783                         slave = rcu_dereference(bond->curr_active_slave);
4784                         if (slave)
4785                                 return slave;
4786                         return bond_get_slave_by_id(bond, 0);
4787                 }
4788         }
4789
4790 non_igmp:
4791         slave_cnt = READ_ONCE(bond->slave_cnt);
4792         if (likely(slave_cnt)) {
4793                 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4794                 return bond_get_slave_by_id(bond, slave_id);
4795         }
4796         return NULL;
4797 }
4798
4799 static struct slave *bond_xdp_xmit_roundrobin_slave_get(struct bonding *bond,
4800                                                         struct xdp_buff *xdp)
4801 {
4802         struct slave *slave;
4803         int slave_cnt;
4804         u32 slave_id;
4805         const struct ethhdr *eth;
4806         void *data = xdp->data;
4807
4808         if (data + sizeof(struct ethhdr) > xdp->data_end)
4809                 goto non_igmp;
4810
4811         eth = (struct ethhdr *)data;
4812         data += sizeof(struct ethhdr);
4813
4814         /* See comment on IGMP in bond_xmit_roundrobin_slave_get() */
4815         if (eth->h_proto == htons(ETH_P_IP)) {
4816                 const struct iphdr *iph;
4817
4818                 if (data + sizeof(struct iphdr) > xdp->data_end)
4819                         goto non_igmp;
4820
4821                 iph = (struct iphdr *)data;
4822
4823                 if (iph->protocol == IPPROTO_IGMP) {
4824                         slave = rcu_dereference(bond->curr_active_slave);
4825                         if (slave)
4826                                 return slave;
4827                         return bond_get_slave_by_id(bond, 0);
4828                 }
4829         }
4830
4831 non_igmp:
4832         slave_cnt = READ_ONCE(bond->slave_cnt);
4833         if (likely(slave_cnt)) {
4834                 slave_id = bond_rr_gen_slave_id(bond) % slave_cnt;
4835                 return bond_get_slave_by_id(bond, slave_id);
4836         }
4837         return NULL;
4838 }
4839
4840 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
4841                                         struct net_device *bond_dev)
4842 {
4843         struct bonding *bond = netdev_priv(bond_dev);
4844         struct slave *slave;
4845
4846         slave = bond_xmit_roundrobin_slave_get(bond, skb);
4847         if (likely(slave))
4848                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4849
4850         return bond_tx_drop(bond_dev, skb);
4851 }
4852
4853 static struct slave *bond_xmit_activebackup_slave_get(struct bonding *bond)
4854 {
4855         return rcu_dereference(bond->curr_active_slave);
4856 }
4857
4858 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
4859  * the bond has a usable interface.
4860  */
4861 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
4862                                           struct net_device *bond_dev)
4863 {
4864         struct bonding *bond = netdev_priv(bond_dev);
4865         struct slave *slave;
4866
4867         slave = bond_xmit_activebackup_slave_get(bond);
4868         if (slave)
4869                 return bond_dev_queue_xmit(bond, skb, slave->dev);
4870
4871         return bond_tx_drop(bond_dev, skb);
4872 }
4873
4874 /* Use this to update slave_array when (a) it's not appropriate to update
4875  * slave_array right away (note that update_slave_array() may sleep)
4876  * and / or (b) RTNL is not held.
4877  */
4878 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
4879 {
4880         queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
4881 }
4882
4883 /* Slave array work handler. Holds only RTNL */
4884 static void bond_slave_arr_handler(struct work_struct *work)
4885 {
4886         struct bonding *bond = container_of(work, struct bonding,
4887                                             slave_arr_work.work);
4888         int ret;
4889
4890         if (!rtnl_trylock())
4891                 goto err;
4892
4893         ret = bond_update_slave_arr(bond, NULL);
4894         rtnl_unlock();
4895         if (ret) {
4896                 pr_warn_ratelimited("Failed to update slave array from WT\n");
4897                 goto err;
4898         }
4899         return;
4900
4901 err:
4902         bond_slave_arr_work_rearm(bond, 1);
4903 }
4904
4905 static void bond_skip_slave(struct bond_up_slave *slaves,
4906                             struct slave *skipslave)
4907 {
4908         int idx;
4909
4910         /* Rare situation where caller has asked to skip a specific
4911          * slave but allocation failed (most likely!). BTW this is
4912          * only possible when the call is initiated from
4913          * __bond_release_one(). In this situation; overwrite the
4914          * skipslave entry in the array with the last entry from the
4915          * array to avoid a situation where the xmit path may choose
4916          * this to-be-skipped slave to send a packet out.
4917          */
4918         for (idx = 0; slaves && idx < slaves->count; idx++) {
4919                 if (skipslave == slaves->arr[idx]) {
4920                         slaves->arr[idx] =
4921                                 slaves->arr[slaves->count - 1];
4922                         slaves->count--;
4923                         break;
4924                 }
4925         }
4926 }
4927
4928 static void bond_set_slave_arr(struct bonding *bond,
4929                                struct bond_up_slave *usable_slaves,
4930                                struct bond_up_slave *all_slaves)
4931 {
4932         struct bond_up_slave *usable, *all;
4933
4934         usable = rtnl_dereference(bond->usable_slaves);
4935         rcu_assign_pointer(bond->usable_slaves, usable_slaves);
4936         kfree_rcu(usable, rcu);
4937
4938         all = rtnl_dereference(bond->all_slaves);
4939         rcu_assign_pointer(bond->all_slaves, all_slaves);
4940         kfree_rcu(all, rcu);
4941 }
4942
4943 static void bond_reset_slave_arr(struct bonding *bond)
4944 {
4945         struct bond_up_slave *usable, *all;
4946
4947         usable = rtnl_dereference(bond->usable_slaves);
4948         if (usable) {
4949                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
4950                 kfree_rcu(usable, rcu);
4951         }
4952
4953         all = rtnl_dereference(bond->all_slaves);
4954         if (all) {
4955                 RCU_INIT_POINTER(bond->all_slaves, NULL);
4956                 kfree_rcu(all, rcu);
4957         }
4958 }
4959
4960 /* Build the usable slaves array in control path for modes that use xmit-hash
4961  * to determine the slave interface -
4962  * (a) BOND_MODE_8023AD
4963  * (b) BOND_MODE_XOR
4964  * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
4965  *
4966  * The caller is expected to hold RTNL only and NO other lock!
4967  */
4968 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
4969 {
4970         struct bond_up_slave *usable_slaves = NULL, *all_slaves = NULL;
4971         struct slave *slave;
4972         struct list_head *iter;
4973         int agg_id = 0;
4974         int ret = 0;
4975
4976         might_sleep();
4977
4978         usable_slaves = kzalloc(struct_size(usable_slaves, arr,
4979                                             bond->slave_cnt), GFP_KERNEL);
4980         all_slaves = kzalloc(struct_size(all_slaves, arr,
4981                                          bond->slave_cnt), GFP_KERNEL);
4982         if (!usable_slaves || !all_slaves) {
4983                 ret = -ENOMEM;
4984                 goto out;
4985         }
4986         if (BOND_MODE(bond) == BOND_MODE_8023AD) {
4987                 struct ad_info ad_info;
4988
4989                 spin_lock_bh(&bond->mode_lock);
4990                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
4991                         spin_unlock_bh(&bond->mode_lock);
4992                         pr_debug("bond_3ad_get_active_agg_info failed\n");
4993                         /* No active aggragator means it's not safe to use
4994                          * the previous array.
4995                          */
4996                         bond_reset_slave_arr(bond);
4997                         goto out;
4998                 }
4999                 spin_unlock_bh(&bond->mode_lock);
5000                 agg_id = ad_info.aggregator_id;
5001         }
5002         bond_for_each_slave(bond, slave, iter) {
5003                 if (skipslave == slave)
5004                         continue;
5005
5006                 all_slaves->arr[all_slaves->count++] = slave;
5007                 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
5008                         struct aggregator *agg;
5009
5010                         agg = SLAVE_AD_INFO(slave)->port.aggregator;
5011                         if (!agg || agg->aggregator_identifier != agg_id)
5012                                 continue;
5013                 }
5014                 if (!bond_slave_can_tx(slave))
5015                         continue;
5016
5017                 slave_dbg(bond->dev, slave->dev, "Adding slave to tx hash array[%d]\n",
5018                           usable_slaves->count);
5019
5020                 usable_slaves->arr[usable_slaves->count++] = slave;
5021         }
5022
5023         bond_set_slave_arr(bond, usable_slaves, all_slaves);
5024         return ret;
5025 out:
5026         if (ret != 0 && skipslave) {
5027                 bond_skip_slave(rtnl_dereference(bond->all_slaves),
5028                                 skipslave);
5029                 bond_skip_slave(rtnl_dereference(bond->usable_slaves),
5030                                 skipslave);
5031         }
5032         kfree_rcu(all_slaves, rcu);
5033         kfree_rcu(usable_slaves, rcu);
5034
5035         return ret;
5036 }
5037
5038 static struct slave *bond_xmit_3ad_xor_slave_get(struct bonding *bond,
5039                                                  struct sk_buff *skb,
5040                                                  struct bond_up_slave *slaves)
5041 {
5042         struct slave *slave;
5043         unsigned int count;
5044         u32 hash;
5045
5046         hash = bond_xmit_hash(bond, skb);
5047         count = slaves ? READ_ONCE(slaves->count) : 0;
5048         if (unlikely(!count))
5049                 return NULL;
5050
5051         slave = slaves->arr[hash % count];
5052         return slave;
5053 }
5054
5055 static struct slave *bond_xdp_xmit_3ad_xor_slave_get(struct bonding *bond,
5056                                                      struct xdp_buff *xdp)
5057 {
5058         struct bond_up_slave *slaves;
5059         unsigned int count;
5060         u32 hash;
5061
5062         hash = bond_xmit_hash_xdp(bond, xdp);
5063         slaves = rcu_dereference(bond->usable_slaves);
5064         count = slaves ? READ_ONCE(slaves->count) : 0;
5065         if (unlikely(!count))
5066                 return NULL;
5067
5068         return slaves->arr[hash % count];
5069 }
5070
5071 /* Use this Xmit function for 3AD as well as XOR modes. The current
5072  * usable slave array is formed in the control path. The xmit function
5073  * just calculates hash and sends the packet out.
5074  */
5075 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
5076                                      struct net_device *dev)
5077 {
5078         struct bonding *bond = netdev_priv(dev);
5079         struct bond_up_slave *slaves;
5080         struct slave *slave;
5081
5082         slaves = rcu_dereference(bond->usable_slaves);
5083         slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5084         if (likely(slave))
5085                 return bond_dev_queue_xmit(bond, skb, slave->dev);
5086
5087         return bond_tx_drop(dev, skb);
5088 }
5089
5090 /* in broadcast mode, we send everything to all usable interfaces. */
5091 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
5092                                        struct net_device *bond_dev)
5093 {
5094         struct bonding *bond = netdev_priv(bond_dev);
5095         struct slave *slave = NULL;
5096         struct list_head *iter;
5097         bool xmit_suc = false;
5098         bool skb_used = false;
5099
5100         bond_for_each_slave_rcu(bond, slave, iter) {
5101                 struct sk_buff *skb2;
5102
5103                 if (!(bond_slave_is_up(slave) && slave->link == BOND_LINK_UP))
5104                         continue;
5105
5106                 if (bond_is_last_slave(bond, slave)) {
5107                         skb2 = skb;
5108                         skb_used = true;
5109                 } else {
5110                         skb2 = skb_clone(skb, GFP_ATOMIC);
5111                         if (!skb2) {
5112                                 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
5113                                                     bond_dev->name, __func__);
5114                                 continue;
5115                         }
5116                 }
5117
5118                 if (bond_dev_queue_xmit(bond, skb2, slave->dev) == NETDEV_TX_OK)
5119                         xmit_suc = true;
5120         }
5121
5122         if (!skb_used)
5123                 dev_kfree_skb_any(skb);
5124
5125         if (xmit_suc)
5126                 return NETDEV_TX_OK;
5127
5128         dev_core_stats_tx_dropped_inc(bond_dev);
5129         return NET_XMIT_DROP;
5130 }
5131
5132 /*------------------------- Device initialization ---------------------------*/
5133
5134 /* Lookup the slave that corresponds to a qid */
5135 static inline int bond_slave_override(struct bonding *bond,
5136                                       struct sk_buff *skb)
5137 {
5138         struct slave *slave = NULL;
5139         struct list_head *iter;
5140
5141         if (!skb_rx_queue_recorded(skb))
5142                 return 1;
5143
5144         /* Find out if any slaves have the same mapping as this skb. */
5145         bond_for_each_slave_rcu(bond, slave, iter) {
5146                 if (slave->queue_id == skb_get_queue_mapping(skb)) {
5147                         if (bond_slave_is_up(slave) &&
5148                             slave->link == BOND_LINK_UP) {
5149                                 bond_dev_queue_xmit(bond, skb, slave->dev);
5150                                 return 0;
5151                         }
5152                         /* If the slave isn't UP, use default transmit policy. */
5153                         break;
5154                 }
5155         }
5156
5157         return 1;
5158 }
5159
5160
5161 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
5162                              struct net_device *sb_dev)
5163 {
5164         /* This helper function exists to help dev_pick_tx get the correct
5165          * destination queue.  Using a helper function skips a call to
5166          * skb_tx_hash and will put the skbs in the queue we expect on their
5167          * way down to the bonding driver.
5168          */
5169         u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
5170
5171         /* Save the original txq to restore before passing to the driver */
5172         qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
5173
5174         if (unlikely(txq >= dev->real_num_tx_queues)) {
5175                 do {
5176                         txq -= dev->real_num_tx_queues;
5177                 } while (txq >= dev->real_num_tx_queues);
5178         }
5179         return txq;
5180 }
5181
5182 static struct net_device *bond_xmit_get_slave(struct net_device *master_dev,
5183                                               struct sk_buff *skb,
5184                                               bool all_slaves)
5185 {
5186         struct bonding *bond = netdev_priv(master_dev);
5187         struct bond_up_slave *slaves;
5188         struct slave *slave = NULL;
5189
5190         switch (BOND_MODE(bond)) {
5191         case BOND_MODE_ROUNDROBIN:
5192                 slave = bond_xmit_roundrobin_slave_get(bond, skb);
5193                 break;
5194         case BOND_MODE_ACTIVEBACKUP:
5195                 slave = bond_xmit_activebackup_slave_get(bond);
5196                 break;
5197         case BOND_MODE_8023AD:
5198         case BOND_MODE_XOR:
5199                 if (all_slaves)
5200                         slaves = rcu_dereference(bond->all_slaves);
5201                 else
5202                         slaves = rcu_dereference(bond->usable_slaves);
5203                 slave = bond_xmit_3ad_xor_slave_get(bond, skb, slaves);
5204                 break;
5205         case BOND_MODE_BROADCAST:
5206                 break;
5207         case BOND_MODE_ALB:
5208                 slave = bond_xmit_alb_slave_get(bond, skb);
5209                 break;
5210         case BOND_MODE_TLB:
5211                 slave = bond_xmit_tlb_slave_get(bond, skb);
5212                 break;
5213         default:
5214                 /* Should never happen, mode already checked */
5215                 WARN_ONCE(true, "Unknown bonding mode");
5216                 break;
5217         }
5218
5219         if (slave)
5220                 return slave->dev;
5221         return NULL;
5222 }
5223
5224 static void bond_sk_to_flow(struct sock *sk, struct flow_keys *flow)
5225 {
5226         switch (sk->sk_family) {
5227 #if IS_ENABLED(CONFIG_IPV6)
5228         case AF_INET6:
5229                 if (ipv6_only_sock(sk) ||
5230                     ipv6_addr_type(&sk->sk_v6_daddr) != IPV6_ADDR_MAPPED) {
5231                         flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV6_ADDRS;
5232                         flow->addrs.v6addrs.src = inet6_sk(sk)->saddr;
5233                         flow->addrs.v6addrs.dst = sk->sk_v6_daddr;
5234                         break;
5235                 }
5236                 fallthrough;
5237 #endif
5238         default: /* AF_INET */
5239                 flow->control.addr_type = FLOW_DISSECTOR_KEY_IPV4_ADDRS;
5240                 flow->addrs.v4addrs.src = inet_sk(sk)->inet_rcv_saddr;
5241                 flow->addrs.v4addrs.dst = inet_sk(sk)->inet_daddr;
5242                 break;
5243         }
5244
5245         flow->ports.src = inet_sk(sk)->inet_sport;
5246         flow->ports.dst = inet_sk(sk)->inet_dport;
5247 }
5248
5249 /**
5250  * bond_sk_hash_l34 - generate a hash value based on the socket's L3 and L4 fields
5251  * @sk: socket to use for headers
5252  *
5253  * This function will extract the necessary field from the socket and use
5254  * them to generate a hash based on the LAYER34 xmit_policy.
5255  * Assumes that sk is a TCP or UDP socket.
5256  */
5257 static u32 bond_sk_hash_l34(struct sock *sk)
5258 {
5259         struct flow_keys flow;
5260         u32 hash;
5261
5262         bond_sk_to_flow(sk, &flow);
5263
5264         /* L4 */
5265         memcpy(&hash, &flow.ports.ports, sizeof(hash));
5266         /* L3 */
5267         return bond_ip_hash(hash, &flow, BOND_XMIT_POLICY_LAYER34);
5268 }
5269
5270 static struct net_device *__bond_sk_get_lower_dev(struct bonding *bond,
5271                                                   struct sock *sk)
5272 {
5273         struct bond_up_slave *slaves;
5274         struct slave *slave;
5275         unsigned int count;
5276         u32 hash;
5277
5278         slaves = rcu_dereference(bond->usable_slaves);
5279         count = slaves ? READ_ONCE(slaves->count) : 0;
5280         if (unlikely(!count))
5281                 return NULL;
5282
5283         hash = bond_sk_hash_l34(sk);
5284         slave = slaves->arr[hash % count];
5285
5286         return slave->dev;
5287 }
5288
5289 static struct net_device *bond_sk_get_lower_dev(struct net_device *dev,
5290                                                 struct sock *sk)
5291 {
5292         struct bonding *bond = netdev_priv(dev);
5293         struct net_device *lower = NULL;
5294
5295         rcu_read_lock();
5296         if (bond_sk_check(bond))
5297                 lower = __bond_sk_get_lower_dev(bond, sk);
5298         rcu_read_unlock();
5299
5300         return lower;
5301 }
5302
5303 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5304 static netdev_tx_t bond_tls_device_xmit(struct bonding *bond, struct sk_buff *skb,
5305                                         struct net_device *dev)
5306 {
5307         if (likely(bond_get_slave_by_dev(bond, tls_get_ctx(skb->sk)->netdev)))
5308                 return bond_dev_queue_xmit(bond, skb, tls_get_ctx(skb->sk)->netdev);
5309         return bond_tx_drop(dev, skb);
5310 }
5311 #endif
5312
5313 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5314 {
5315         struct bonding *bond = netdev_priv(dev);
5316
5317         if (bond_should_override_tx_queue(bond) &&
5318             !bond_slave_override(bond, skb))
5319                 return NETDEV_TX_OK;
5320
5321 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5322         if (skb->sk && tls_is_sk_tx_device_offloaded(skb->sk))
5323                 return bond_tls_device_xmit(bond, skb, dev);
5324 #endif
5325
5326         switch (BOND_MODE(bond)) {
5327         case BOND_MODE_ROUNDROBIN:
5328                 return bond_xmit_roundrobin(skb, dev);
5329         case BOND_MODE_ACTIVEBACKUP:
5330                 return bond_xmit_activebackup(skb, dev);
5331         case BOND_MODE_8023AD:
5332         case BOND_MODE_XOR:
5333                 return bond_3ad_xor_xmit(skb, dev);
5334         case BOND_MODE_BROADCAST:
5335                 return bond_xmit_broadcast(skb, dev);
5336         case BOND_MODE_ALB:
5337                 return bond_alb_xmit(skb, dev);
5338         case BOND_MODE_TLB:
5339                 return bond_tlb_xmit(skb, dev);
5340         default:
5341                 /* Should never happen, mode already checked */
5342                 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
5343                 WARN_ON_ONCE(1);
5344                 return bond_tx_drop(dev, skb);
5345         }
5346 }
5347
5348 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
5349 {
5350         struct bonding *bond = netdev_priv(dev);
5351         netdev_tx_t ret = NETDEV_TX_OK;
5352
5353         /* If we risk deadlock from transmitting this in the
5354          * netpoll path, tell netpoll to queue the frame for later tx
5355          */
5356         if (unlikely(is_netpoll_tx_blocked(dev)))
5357                 return NETDEV_TX_BUSY;
5358
5359         rcu_read_lock();
5360         if (bond_has_slaves(bond))
5361                 ret = __bond_start_xmit(skb, dev);
5362         else
5363                 ret = bond_tx_drop(dev, skb);
5364         rcu_read_unlock();
5365
5366         return ret;
5367 }
5368
5369 static struct net_device *
5370 bond_xdp_get_xmit_slave(struct net_device *bond_dev, struct xdp_buff *xdp)
5371 {
5372         struct bonding *bond = netdev_priv(bond_dev);
5373         struct slave *slave;
5374
5375         /* Caller needs to hold rcu_read_lock() */
5376
5377         switch (BOND_MODE(bond)) {
5378         case BOND_MODE_ROUNDROBIN:
5379                 slave = bond_xdp_xmit_roundrobin_slave_get(bond, xdp);
5380                 break;
5381
5382         case BOND_MODE_ACTIVEBACKUP:
5383                 slave = bond_xmit_activebackup_slave_get(bond);
5384                 break;
5385
5386         case BOND_MODE_8023AD:
5387         case BOND_MODE_XOR:
5388                 slave = bond_xdp_xmit_3ad_xor_slave_get(bond, xdp);
5389                 break;
5390
5391         default:
5392                 /* Should never happen. Mode guarded by bond_xdp_check() */
5393                 netdev_err(bond_dev, "Unknown bonding mode %d for xdp xmit\n", BOND_MODE(bond));
5394                 WARN_ON_ONCE(1);
5395                 return NULL;
5396         }
5397
5398         if (slave)
5399                 return slave->dev;
5400
5401         return NULL;
5402 }
5403
5404 static int bond_xdp_xmit(struct net_device *bond_dev,
5405                          int n, struct xdp_frame **frames, u32 flags)
5406 {
5407         int nxmit, err = -ENXIO;
5408
5409         rcu_read_lock();
5410
5411         for (nxmit = 0; nxmit < n; nxmit++) {
5412                 struct xdp_frame *frame = frames[nxmit];
5413                 struct xdp_frame *frames1[] = {frame};
5414                 struct net_device *slave_dev;
5415                 struct xdp_buff xdp;
5416
5417                 xdp_convert_frame_to_buff(frame, &xdp);
5418
5419                 slave_dev = bond_xdp_get_xmit_slave(bond_dev, &xdp);
5420                 if (!slave_dev) {
5421                         err = -ENXIO;
5422                         break;
5423                 }
5424
5425                 err = slave_dev->netdev_ops->ndo_xdp_xmit(slave_dev, 1, frames1, flags);
5426                 if (err < 1)
5427                         break;
5428         }
5429
5430         rcu_read_unlock();
5431
5432         /* If error happened on the first frame then we can pass the error up, otherwise
5433          * report the number of frames that were xmitted.
5434          */
5435         if (err < 0)
5436                 return (nxmit == 0 ? err : nxmit);
5437
5438         return nxmit;
5439 }
5440
5441 static int bond_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5442                         struct netlink_ext_ack *extack)
5443 {
5444         struct bonding *bond = netdev_priv(dev);
5445         struct list_head *iter;
5446         struct slave *slave, *rollback_slave;
5447         struct bpf_prog *old_prog;
5448         struct netdev_bpf xdp = {
5449                 .command = XDP_SETUP_PROG,
5450                 .flags   = 0,
5451                 .prog    = prog,
5452                 .extack  = extack,
5453         };
5454         int err;
5455
5456         ASSERT_RTNL();
5457
5458         if (!bond_xdp_check(bond))
5459                 return -EOPNOTSUPP;
5460
5461         old_prog = bond->xdp_prog;
5462         bond->xdp_prog = prog;
5463
5464         bond_for_each_slave(bond, slave, iter) {
5465                 struct net_device *slave_dev = slave->dev;
5466
5467                 if (!slave_dev->netdev_ops->ndo_bpf ||
5468                     !slave_dev->netdev_ops->ndo_xdp_xmit) {
5469                         SLAVE_NL_ERR(dev, slave_dev, extack,
5470                                      "Slave device does not support XDP");
5471                         err = -EOPNOTSUPP;
5472                         goto err;
5473                 }
5474
5475                 if (dev_xdp_prog_count(slave_dev) > 0) {
5476                         SLAVE_NL_ERR(dev, slave_dev, extack,
5477                                      "Slave has XDP program loaded, please unload before enslaving");
5478                         err = -EOPNOTSUPP;
5479                         goto err;
5480                 }
5481
5482                 err = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5483                 if (err < 0) {
5484                         /* ndo_bpf() sets extack error message */
5485                         slave_err(dev, slave_dev, "Error %d calling ndo_bpf\n", err);
5486                         goto err;
5487                 }
5488                 if (prog)
5489                         bpf_prog_inc(prog);
5490         }
5491
5492         if (prog) {
5493                 static_branch_inc(&bpf_master_redirect_enabled_key);
5494         } else if (old_prog) {
5495                 bpf_prog_put(old_prog);
5496                 static_branch_dec(&bpf_master_redirect_enabled_key);
5497         }
5498
5499         return 0;
5500
5501 err:
5502         /* unwind the program changes */
5503         bond->xdp_prog = old_prog;
5504         xdp.prog = old_prog;
5505         xdp.extack = NULL; /* do not overwrite original error */
5506
5507         bond_for_each_slave(bond, rollback_slave, iter) {
5508                 struct net_device *slave_dev = rollback_slave->dev;
5509                 int err_unwind;
5510
5511                 if (slave == rollback_slave)
5512                         break;
5513
5514                 err_unwind = slave_dev->netdev_ops->ndo_bpf(slave_dev, &xdp);
5515                 if (err_unwind < 0)
5516                         slave_err(dev, slave_dev,
5517                                   "Error %d when unwinding XDP program change\n", err_unwind);
5518                 else if (xdp.prog)
5519                         bpf_prog_inc(xdp.prog);
5520         }
5521         return err;
5522 }
5523
5524 static int bond_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5525 {
5526         switch (xdp->command) {
5527         case XDP_SETUP_PROG:
5528                 return bond_xdp_set(dev, xdp->prog, xdp->extack);
5529         default:
5530                 return -EINVAL;
5531         }
5532 }
5533
5534 static u32 bond_mode_bcast_speed(struct slave *slave, u32 speed)
5535 {
5536         if (speed == 0 || speed == SPEED_UNKNOWN)
5537                 speed = slave->speed;
5538         else
5539                 speed = min(speed, slave->speed);
5540
5541         return speed;
5542 }
5543
5544 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
5545                                            struct ethtool_link_ksettings *cmd)
5546 {
5547         struct bonding *bond = netdev_priv(bond_dev);
5548         struct list_head *iter;
5549         struct slave *slave;
5550         u32 speed = 0;
5551
5552         cmd->base.duplex = DUPLEX_UNKNOWN;
5553         cmd->base.port = PORT_OTHER;
5554
5555         /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
5556          * do not need to check mode.  Though link speed might not represent
5557          * the true receive or transmit bandwidth (not all modes are symmetric)
5558          * this is an accurate maximum.
5559          */
5560         bond_for_each_slave(bond, slave, iter) {
5561                 if (bond_slave_can_tx(slave)) {
5562                         if (slave->speed != SPEED_UNKNOWN) {
5563                                 if (BOND_MODE(bond) == BOND_MODE_BROADCAST)
5564                                         speed = bond_mode_bcast_speed(slave,
5565                                                                       speed);
5566                                 else
5567                                         speed += slave->speed;
5568                         }
5569                         if (cmd->base.duplex == DUPLEX_UNKNOWN &&
5570                             slave->duplex != DUPLEX_UNKNOWN)
5571                                 cmd->base.duplex = slave->duplex;
5572                 }
5573         }
5574         cmd->base.speed = speed ? : SPEED_UNKNOWN;
5575
5576         return 0;
5577 }
5578
5579 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
5580                                      struct ethtool_drvinfo *drvinfo)
5581 {
5582         strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
5583         snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
5584                  BOND_ABI_VERSION);
5585 }
5586
5587 static int bond_ethtool_get_ts_info(struct net_device *bond_dev,
5588                                     struct ethtool_ts_info *info)
5589 {
5590         struct bonding *bond = netdev_priv(bond_dev);
5591         const struct ethtool_ops *ops;
5592         struct net_device *real_dev;
5593         struct phy_device *phydev;
5594         int ret = 0;
5595
5596         rcu_read_lock();
5597         real_dev = bond_option_active_slave_get_rcu(bond);
5598         dev_hold(real_dev);
5599         rcu_read_unlock();
5600
5601         if (real_dev) {
5602                 ops = real_dev->ethtool_ops;
5603                 phydev = real_dev->phydev;
5604
5605                 if (phy_has_tsinfo(phydev)) {
5606                         ret = phy_ts_info(phydev, info);
5607                         goto out;
5608                 } else if (ops->get_ts_info) {
5609                         ret = ops->get_ts_info(real_dev, info);
5610                         goto out;
5611                 }
5612         }
5613
5614         info->so_timestamping = SOF_TIMESTAMPING_RX_SOFTWARE |
5615                                 SOF_TIMESTAMPING_SOFTWARE;
5616         info->phc_index = -1;
5617
5618 out:
5619         dev_put(real_dev);
5620         return ret;
5621 }
5622
5623 static const struct ethtool_ops bond_ethtool_ops = {
5624         .get_drvinfo            = bond_ethtool_get_drvinfo,
5625         .get_link               = ethtool_op_get_link,
5626         .get_link_ksettings     = bond_ethtool_get_link_ksettings,
5627         .get_ts_info            = bond_ethtool_get_ts_info,
5628 };
5629
5630 static const struct net_device_ops bond_netdev_ops = {
5631         .ndo_init               = bond_init,
5632         .ndo_uninit             = bond_uninit,
5633         .ndo_open               = bond_open,
5634         .ndo_stop               = bond_close,
5635         .ndo_start_xmit         = bond_start_xmit,
5636         .ndo_select_queue       = bond_select_queue,
5637         .ndo_get_stats64        = bond_get_stats,
5638         .ndo_eth_ioctl          = bond_eth_ioctl,
5639         .ndo_siocbond           = bond_do_ioctl,
5640         .ndo_siocdevprivate     = bond_siocdevprivate,
5641         .ndo_change_rx_flags    = bond_change_rx_flags,
5642         .ndo_set_rx_mode        = bond_set_rx_mode,
5643         .ndo_change_mtu         = bond_change_mtu,
5644         .ndo_set_mac_address    = bond_set_mac_address,
5645         .ndo_neigh_setup        = bond_neigh_setup,
5646         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
5647         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
5648 #ifdef CONFIG_NET_POLL_CONTROLLER
5649         .ndo_netpoll_setup      = bond_netpoll_setup,
5650         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
5651         .ndo_poll_controller    = bond_poll_controller,
5652 #endif
5653         .ndo_add_slave          = bond_enslave,
5654         .ndo_del_slave          = bond_release,
5655         .ndo_fix_features       = bond_fix_features,
5656         .ndo_features_check     = passthru_features_check,
5657         .ndo_get_xmit_slave     = bond_xmit_get_slave,
5658         .ndo_sk_get_lower_dev   = bond_sk_get_lower_dev,
5659         .ndo_bpf                = bond_xdp,
5660         .ndo_xdp_xmit           = bond_xdp_xmit,
5661         .ndo_xdp_get_xmit_slave = bond_xdp_get_xmit_slave,
5662 };
5663
5664 static const struct device_type bond_type = {
5665         .name = "bond",
5666 };
5667
5668 static void bond_destructor(struct net_device *bond_dev)
5669 {
5670         struct bonding *bond = netdev_priv(bond_dev);
5671
5672         if (bond->wq)
5673                 destroy_workqueue(bond->wq);
5674
5675         if (bond->rr_tx_counter)
5676                 free_percpu(bond->rr_tx_counter);
5677 }
5678
5679 void bond_setup(struct net_device *bond_dev)
5680 {
5681         struct bonding *bond = netdev_priv(bond_dev);
5682
5683         spin_lock_init(&bond->mode_lock);
5684         bond->params = bonding_defaults;
5685
5686         /* Initialize pointers */
5687         bond->dev = bond_dev;
5688
5689         /* Initialize the device entry points */
5690         ether_setup(bond_dev);
5691         bond_dev->max_mtu = ETH_MAX_MTU;
5692         bond_dev->netdev_ops = &bond_netdev_ops;
5693         bond_dev->ethtool_ops = &bond_ethtool_ops;
5694
5695         bond_dev->needs_free_netdev = true;
5696         bond_dev->priv_destructor = bond_destructor;
5697
5698         SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
5699
5700         /* Initialize the device options */
5701         bond_dev->flags |= IFF_MASTER;
5702         bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
5703         bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
5704
5705 #ifdef CONFIG_XFRM_OFFLOAD
5706         /* set up xfrm device ops (only supported in active-backup right now) */
5707         bond_dev->xfrmdev_ops = &bond_xfrmdev_ops;
5708         INIT_LIST_HEAD(&bond->ipsec_list);
5709         spin_lock_init(&bond->ipsec_lock);
5710 #endif /* CONFIG_XFRM_OFFLOAD */
5711
5712         /* don't acquire bond device's netif_tx_lock when transmitting */
5713         bond_dev->features |= NETIF_F_LLTX;
5714
5715         /* By default, we declare the bond to be fully
5716          * VLAN hardware accelerated capable. Special
5717          * care is taken in the various xmit functions
5718          * when there are slaves that are not hw accel
5719          * capable
5720          */
5721
5722         /* Don't allow bond devices to change network namespaces. */
5723         bond_dev->features |= NETIF_F_NETNS_LOCAL;
5724
5725         bond_dev->hw_features = BOND_VLAN_FEATURES |
5726                                 NETIF_F_HW_VLAN_CTAG_RX |
5727                                 NETIF_F_HW_VLAN_CTAG_FILTER;
5728
5729         bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
5730         bond_dev->features |= bond_dev->hw_features;
5731         bond_dev->features |= NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX;
5732 #ifdef CONFIG_XFRM_OFFLOAD
5733         bond_dev->hw_features |= BOND_XFRM_FEATURES;
5734         /* Only enable XFRM features if this is an active-backup config */
5735         if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
5736                 bond_dev->features |= BOND_XFRM_FEATURES;
5737 #endif /* CONFIG_XFRM_OFFLOAD */
5738 #if IS_ENABLED(CONFIG_TLS_DEVICE)
5739         if (bond_sk_check(bond))
5740                 bond_dev->features |= BOND_TLS_FEATURES;
5741 #endif
5742 }
5743
5744 /* Destroy a bonding device.
5745  * Must be under rtnl_lock when this function is called.
5746  */
5747 static void bond_uninit(struct net_device *bond_dev)
5748 {
5749         struct bonding *bond = netdev_priv(bond_dev);
5750         struct bond_up_slave *usable, *all;
5751         struct list_head *iter;
5752         struct slave *slave;
5753
5754         bond_netpoll_cleanup(bond_dev);
5755
5756         /* Release the bonded slaves */
5757         bond_for_each_slave(bond, slave, iter)
5758                 __bond_release_one(bond_dev, slave->dev, true, true);
5759         netdev_info(bond_dev, "Released all slaves\n");
5760
5761         usable = rtnl_dereference(bond->usable_slaves);
5762         if (usable) {
5763                 RCU_INIT_POINTER(bond->usable_slaves, NULL);
5764                 kfree_rcu(usable, rcu);
5765         }
5766
5767         all = rtnl_dereference(bond->all_slaves);
5768         if (all) {
5769                 RCU_INIT_POINTER(bond->all_slaves, NULL);
5770                 kfree_rcu(all, rcu);
5771         }
5772
5773         list_del(&bond->bond_list);
5774
5775         bond_debug_unregister(bond);
5776 }
5777
5778 /*------------------------- Module initialization ---------------------------*/
5779
5780 static int bond_check_params(struct bond_params *params)
5781 {
5782         int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
5783         struct bond_opt_value newval;
5784         const struct bond_opt_value *valptr;
5785         int arp_all_targets_value = 0;
5786         u16 ad_actor_sys_prio = 0;
5787         u16 ad_user_port_key = 0;
5788         __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
5789         int arp_ip_count;
5790         int bond_mode   = BOND_MODE_ROUNDROBIN;
5791         int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
5792         int lacp_fast = 0;
5793         int tlb_dynamic_lb;
5794
5795         /* Convert string parameters. */
5796         if (mode) {
5797                 bond_opt_initstr(&newval, mode);
5798                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
5799                 if (!valptr) {
5800                         pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
5801                         return -EINVAL;
5802                 }
5803                 bond_mode = valptr->value;
5804         }
5805
5806         if (xmit_hash_policy) {
5807                 if (bond_mode == BOND_MODE_ROUNDROBIN ||
5808                     bond_mode == BOND_MODE_ACTIVEBACKUP ||
5809                     bond_mode == BOND_MODE_BROADCAST) {
5810                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
5811                                 bond_mode_name(bond_mode));
5812                 } else {
5813                         bond_opt_initstr(&newval, xmit_hash_policy);
5814                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
5815                                                 &newval);
5816                         if (!valptr) {
5817                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
5818                                        xmit_hash_policy);
5819                                 return -EINVAL;
5820                         }
5821                         xmit_hashtype = valptr->value;
5822                 }
5823         }
5824
5825         if (lacp_rate) {
5826                 if (bond_mode != BOND_MODE_8023AD) {
5827                         pr_info("lacp_rate param is irrelevant in mode %s\n",
5828                                 bond_mode_name(bond_mode));
5829                 } else {
5830                         bond_opt_initstr(&newval, lacp_rate);
5831                         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
5832                                                 &newval);
5833                         if (!valptr) {
5834                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
5835                                        lacp_rate);
5836                                 return -EINVAL;
5837                         }
5838                         lacp_fast = valptr->value;
5839                 }
5840         }
5841
5842         if (ad_select) {
5843                 bond_opt_initstr(&newval, ad_select);
5844                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
5845                                         &newval);
5846                 if (!valptr) {
5847                         pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
5848                         return -EINVAL;
5849                 }
5850                 params->ad_select = valptr->value;
5851                 if (bond_mode != BOND_MODE_8023AD)
5852                         pr_warn("ad_select param only affects 802.3ad mode\n");
5853         } else {
5854                 params->ad_select = BOND_AD_STABLE;
5855         }
5856
5857         if (max_bonds < 0) {
5858                 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
5859                         max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
5860                 max_bonds = BOND_DEFAULT_MAX_BONDS;
5861         }
5862
5863         if (miimon < 0) {
5864                 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5865                         miimon, INT_MAX);
5866                 miimon = 0;
5867         }
5868
5869         if (updelay < 0) {
5870                 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5871                         updelay, INT_MAX);
5872                 updelay = 0;
5873         }
5874
5875         if (downdelay < 0) {
5876                 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5877                         downdelay, INT_MAX);
5878                 downdelay = 0;
5879         }
5880
5881         if ((use_carrier != 0) && (use_carrier != 1)) {
5882                 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
5883                         use_carrier);
5884                 use_carrier = 1;
5885         }
5886
5887         if (num_peer_notif < 0 || num_peer_notif > 255) {
5888                 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
5889                         num_peer_notif);
5890                 num_peer_notif = 1;
5891         }
5892
5893         /* reset values for 802.3ad/TLB/ALB */
5894         if (!bond_mode_uses_arp(bond_mode)) {
5895                 if (!miimon) {
5896                         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");
5897                         pr_warn("Forcing miimon to 100msec\n");
5898                         miimon = BOND_DEFAULT_MIIMON;
5899                 }
5900         }
5901
5902         if (tx_queues < 1 || tx_queues > 255) {
5903                 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
5904                         tx_queues, BOND_DEFAULT_TX_QUEUES);
5905                 tx_queues = BOND_DEFAULT_TX_QUEUES;
5906         }
5907
5908         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
5909                 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
5910                         all_slaves_active);
5911                 all_slaves_active = 0;
5912         }
5913
5914         if (resend_igmp < 0 || resend_igmp > 255) {
5915                 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
5916                         resend_igmp, BOND_DEFAULT_RESEND_IGMP);
5917                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
5918         }
5919
5920         bond_opt_initval(&newval, packets_per_slave);
5921         if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
5922                 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
5923                         packets_per_slave, USHRT_MAX);
5924                 packets_per_slave = 1;
5925         }
5926
5927         if (bond_mode == BOND_MODE_ALB) {
5928                 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",
5929                           updelay);
5930         }
5931
5932         if (!miimon) {
5933                 if (updelay || downdelay) {
5934                         /* just warn the user the up/down delay will have
5935                          * no effect since miimon is zero...
5936                          */
5937                         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",
5938                                 updelay, downdelay);
5939                 }
5940         } else {
5941                 /* don't allow arp monitoring */
5942                 if (arp_interval) {
5943                         pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
5944                                 miimon, arp_interval);
5945                         arp_interval = 0;
5946                 }
5947
5948                 if ((updelay % miimon) != 0) {
5949                         pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5950                                 updelay, miimon, (updelay / miimon) * miimon);
5951                 }
5952
5953                 updelay /= miimon;
5954
5955                 if ((downdelay % miimon) != 0) {
5956                         pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5957                                 downdelay, miimon,
5958                                 (downdelay / miimon) * miimon);
5959                 }
5960
5961                 downdelay /= miimon;
5962         }
5963
5964         if (arp_interval < 0) {
5965                 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
5966                         arp_interval, INT_MAX);
5967                 arp_interval = 0;
5968         }
5969
5970         for (arp_ip_count = 0, i = 0;
5971              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
5972                 __be32 ip;
5973
5974                 /* not a complete check, but good enough to catch mistakes */
5975                 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
5976                     !bond_is_ip_target_ok(ip)) {
5977                         pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5978                                 arp_ip_target[i]);
5979                         arp_interval = 0;
5980                 } else {
5981                         if (bond_get_targets_ip(arp_target, ip) == -1)
5982                                 arp_target[arp_ip_count++] = ip;
5983                         else
5984                                 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
5985                                         &ip);
5986                 }
5987         }
5988
5989         if (arp_interval && !arp_ip_count) {
5990                 /* don't allow arping if no arp_ip_target given... */
5991                 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5992                         arp_interval);
5993                 arp_interval = 0;
5994         }
5995
5996         if (arp_validate) {
5997                 if (!arp_interval) {
5998                         pr_err("arp_validate requires arp_interval\n");
5999                         return -EINVAL;
6000                 }
6001
6002                 bond_opt_initstr(&newval, arp_validate);
6003                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
6004                                         &newval);
6005                 if (!valptr) {
6006                         pr_err("Error: invalid arp_validate \"%s\"\n",
6007                                arp_validate);
6008                         return -EINVAL;
6009                 }
6010                 arp_validate_value = valptr->value;
6011         } else {
6012                 arp_validate_value = 0;
6013         }
6014
6015         if (arp_all_targets) {
6016                 bond_opt_initstr(&newval, arp_all_targets);
6017                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
6018                                         &newval);
6019                 if (!valptr) {
6020                         pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
6021                                arp_all_targets);
6022                         arp_all_targets_value = 0;
6023                 } else {
6024                         arp_all_targets_value = valptr->value;
6025                 }
6026         }
6027
6028         if (miimon) {
6029                 pr_info("MII link monitoring set to %d ms\n", miimon);
6030         } else if (arp_interval) {
6031                 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
6032                                           arp_validate_value);
6033                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
6034                         arp_interval, valptr->string, arp_ip_count);
6035
6036                 for (i = 0; i < arp_ip_count; i++)
6037                         pr_cont(" %s", arp_ip_target[i]);
6038
6039                 pr_cont("\n");
6040
6041         } else if (max_bonds) {
6042                 /* miimon and arp_interval not set, we need one so things
6043                  * work as expected, see bonding.txt for details
6044                  */
6045                 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");
6046         }
6047
6048         if (primary && !bond_mode_uses_primary(bond_mode)) {
6049                 /* currently, using a primary only makes sense
6050                  * in active backup, TLB or ALB modes
6051                  */
6052                 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
6053                         primary, bond_mode_name(bond_mode));
6054                 primary = NULL;
6055         }
6056
6057         if (primary && primary_reselect) {
6058                 bond_opt_initstr(&newval, primary_reselect);
6059                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
6060                                         &newval);
6061                 if (!valptr) {
6062                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
6063                                primary_reselect);
6064                         return -EINVAL;
6065                 }
6066                 primary_reselect_value = valptr->value;
6067         } else {
6068                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
6069         }
6070
6071         if (fail_over_mac) {
6072                 bond_opt_initstr(&newval, fail_over_mac);
6073                 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
6074                                         &newval);
6075                 if (!valptr) {
6076                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
6077                                fail_over_mac);
6078                         return -EINVAL;
6079                 }
6080                 fail_over_mac_value = valptr->value;
6081                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
6082                         pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
6083         } else {
6084                 fail_over_mac_value = BOND_FOM_NONE;
6085         }
6086
6087         bond_opt_initstr(&newval, "default");
6088         valptr = bond_opt_parse(
6089                         bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
6090                                      &newval);
6091         if (!valptr) {
6092                 pr_err("Error: No ad_actor_sys_prio default value");
6093                 return -EINVAL;
6094         }
6095         ad_actor_sys_prio = valptr->value;
6096
6097         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
6098                                 &newval);
6099         if (!valptr) {
6100                 pr_err("Error: No ad_user_port_key default value");
6101                 return -EINVAL;
6102         }
6103         ad_user_port_key = valptr->value;
6104
6105         bond_opt_initstr(&newval, "default");
6106         valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
6107         if (!valptr) {
6108                 pr_err("Error: No tlb_dynamic_lb default value");
6109                 return -EINVAL;
6110         }
6111         tlb_dynamic_lb = valptr->value;
6112
6113         if (lp_interval == 0) {
6114                 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
6115                         INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
6116                 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
6117         }
6118
6119         /* fill params struct with the proper values */
6120         params->mode = bond_mode;
6121         params->xmit_policy = xmit_hashtype;
6122         params->miimon = miimon;
6123         params->num_peer_notif = num_peer_notif;
6124         params->arp_interval = arp_interval;
6125         params->arp_validate = arp_validate_value;
6126         params->arp_all_targets = arp_all_targets_value;
6127         params->missed_max = 2;
6128         params->updelay = updelay;
6129         params->downdelay = downdelay;
6130         params->peer_notif_delay = 0;
6131         params->use_carrier = use_carrier;
6132         params->lacp_active = 1;
6133         params->lacp_fast = lacp_fast;
6134         params->primary[0] = 0;
6135         params->primary_reselect = primary_reselect_value;
6136         params->fail_over_mac = fail_over_mac_value;
6137         params->tx_queues = tx_queues;
6138         params->all_slaves_active = all_slaves_active;
6139         params->resend_igmp = resend_igmp;
6140         params->min_links = min_links;
6141         params->lp_interval = lp_interval;
6142         params->packets_per_slave = packets_per_slave;
6143         params->tlb_dynamic_lb = tlb_dynamic_lb;
6144         params->ad_actor_sys_prio = ad_actor_sys_prio;
6145         eth_zero_addr(params->ad_actor_system);
6146         params->ad_user_port_key = ad_user_port_key;
6147         if (packets_per_slave > 0) {
6148                 params->reciprocal_packets_per_slave =
6149                         reciprocal_value(packets_per_slave);
6150         } else {
6151                 /* reciprocal_packets_per_slave is unused if
6152                  * packets_per_slave is 0 or 1, just initialize it
6153                  */
6154                 params->reciprocal_packets_per_slave =
6155                         (struct reciprocal_value) { 0 };
6156         }
6157
6158         if (primary)
6159                 strscpy_pad(params->primary, primary, sizeof(params->primary));
6160
6161         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
6162 #if IS_ENABLED(CONFIG_IPV6)
6163         memset(params->ns_targets, 0, sizeof(struct in6_addr) * BOND_MAX_NS_TARGETS);
6164 #endif
6165
6166         return 0;
6167 }
6168
6169 /* Called from registration process */
6170 static int bond_init(struct net_device *bond_dev)
6171 {
6172         struct bonding *bond = netdev_priv(bond_dev);
6173         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
6174
6175         netdev_dbg(bond_dev, "Begin bond_init\n");
6176
6177         bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
6178         if (!bond->wq)
6179                 return -ENOMEM;
6180
6181         if (BOND_MODE(bond) == BOND_MODE_ROUNDROBIN) {
6182                 bond->rr_tx_counter = alloc_percpu(u32);
6183                 if (!bond->rr_tx_counter) {
6184                         destroy_workqueue(bond->wq);
6185                         bond->wq = NULL;
6186                         return -ENOMEM;
6187                 }
6188         }
6189
6190         spin_lock_init(&bond->stats_lock);
6191         netdev_lockdep_set_classes(bond_dev);
6192
6193         list_add_tail(&bond->bond_list, &bn->dev_list);
6194
6195         bond_prepare_sysfs_group(bond);
6196
6197         bond_debug_register(bond);
6198
6199         /* Ensure valid dev_addr */
6200         if (is_zero_ether_addr(bond_dev->dev_addr) &&
6201             bond_dev->addr_assign_type == NET_ADDR_PERM)
6202                 eth_hw_addr_random(bond_dev);
6203
6204         return 0;
6205 }
6206
6207 unsigned int bond_get_num_tx_queues(void)
6208 {
6209         return tx_queues;
6210 }
6211
6212 /* Create a new bond based on the specified name and bonding parameters.
6213  * If name is NULL, obtain a suitable "bond%d" name for us.
6214  * Caller must NOT hold rtnl_lock; we need to release it here before we
6215  * set up our sysfs entries.
6216  */
6217 int bond_create(struct net *net, const char *name)
6218 {
6219         struct net_device *bond_dev;
6220         struct bonding *bond;
6221         struct alb_bond_info *bond_info;
6222         int res;
6223
6224         rtnl_lock();
6225
6226         bond_dev = alloc_netdev_mq(sizeof(struct bonding),
6227                                    name ? name : "bond%d", NET_NAME_UNKNOWN,
6228                                    bond_setup, tx_queues);
6229         if (!bond_dev) {
6230                 pr_err("%s: eek! can't alloc netdev!\n", name);
6231                 rtnl_unlock();
6232                 return -ENOMEM;
6233         }
6234
6235         /*
6236          * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
6237          * It is set to 0 by default which is wrong.
6238          */
6239         bond = netdev_priv(bond_dev);
6240         bond_info = &(BOND_ALB_INFO(bond));
6241         bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
6242
6243         dev_net_set(bond_dev, net);
6244         bond_dev->rtnl_link_ops = &bond_link_ops;
6245
6246         res = register_netdevice(bond_dev);
6247         if (res < 0) {
6248                 free_netdev(bond_dev);
6249                 rtnl_unlock();
6250
6251                 return res;
6252         }
6253
6254         netif_carrier_off(bond_dev);
6255
6256         bond_work_init_all(bond);
6257
6258         rtnl_unlock();
6259         return 0;
6260 }
6261
6262 static int __net_init bond_net_init(struct net *net)
6263 {
6264         struct bond_net *bn = net_generic(net, bond_net_id);
6265
6266         bn->net = net;
6267         INIT_LIST_HEAD(&bn->dev_list);
6268
6269         bond_create_proc_dir(bn);
6270         bond_create_sysfs(bn);
6271
6272         return 0;
6273 }
6274
6275 static void __net_exit bond_net_exit_batch(struct list_head *net_list)
6276 {
6277         struct bond_net *bn;
6278         struct net *net;
6279         LIST_HEAD(list);
6280
6281         list_for_each_entry(net, net_list, exit_list) {
6282                 bn = net_generic(net, bond_net_id);
6283                 bond_destroy_sysfs(bn);
6284         }
6285
6286         /* Kill off any bonds created after unregistering bond rtnl ops */
6287         rtnl_lock();
6288         list_for_each_entry(net, net_list, exit_list) {
6289                 struct bonding *bond, *tmp_bond;
6290
6291                 bn = net_generic(net, bond_net_id);
6292                 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
6293                         unregister_netdevice_queue(bond->dev, &list);
6294         }
6295         unregister_netdevice_many(&list);
6296         rtnl_unlock();
6297
6298         list_for_each_entry(net, net_list, exit_list) {
6299                 bn = net_generic(net, bond_net_id);
6300                 bond_destroy_proc_dir(bn);
6301         }
6302 }
6303
6304 static struct pernet_operations bond_net_ops = {
6305         .init = bond_net_init,
6306         .exit_batch = bond_net_exit_batch,
6307         .id   = &bond_net_id,
6308         .size = sizeof(struct bond_net),
6309 };
6310
6311 static int __init bonding_init(void)
6312 {
6313         int i;
6314         int res;
6315
6316         res = bond_check_params(&bonding_defaults);
6317         if (res)
6318                 goto out;
6319
6320         res = register_pernet_subsys(&bond_net_ops);
6321         if (res)
6322                 goto out;
6323
6324         res = bond_netlink_init();
6325         if (res)
6326                 goto err_link;
6327
6328         bond_create_debugfs();
6329
6330         for (i = 0; i < max_bonds; i++) {
6331                 res = bond_create(&init_net, NULL);
6332                 if (res)
6333                         goto err;
6334         }
6335
6336         skb_flow_dissector_init(&flow_keys_bonding,
6337                                 flow_keys_bonding_keys,
6338                                 ARRAY_SIZE(flow_keys_bonding_keys));
6339
6340         register_netdevice_notifier(&bond_netdev_notifier);
6341 out:
6342         return res;
6343 err:
6344         bond_destroy_debugfs();
6345         bond_netlink_fini();
6346 err_link:
6347         unregister_pernet_subsys(&bond_net_ops);
6348         goto out;
6349
6350 }
6351
6352 static void __exit bonding_exit(void)
6353 {
6354         unregister_netdevice_notifier(&bond_netdev_notifier);
6355
6356         bond_destroy_debugfs();
6357
6358         bond_netlink_fini();
6359         unregister_pernet_subsys(&bond_net_ops);
6360
6361 #ifdef CONFIG_NET_POLL_CONTROLLER
6362         /* Make sure we don't have an imbalance on our netpoll blocking */
6363         WARN_ON(atomic_read(&netpoll_block_tx));
6364 #endif
6365 }
6366
6367 module_init(bonding_init);
6368 module_exit(bonding_exit);
6369 MODULE_LICENSE("GPL");
6370 MODULE_DESCRIPTION(DRV_DESCRIPTION);
6371 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");